Sensors & Transducers · Sensors & Transducers Journal is a peer review international jo urnal ... Research of an Optoelectronic Current Transformer Based on a Designed Magneto-Optic

http://www.sensorsportal.com/HTML/DIGEST/Submission.htm

SSeennssoorrss && TTrraannssdduucceerrss

IInntteerrnnaattiioonnaall OOffffiicciiaall JJoouurrnnaall ooff tthhee IInntteerrnnaattiioonnaall FFrreeqquueennccyy SSeennssoorr AAssssoocciiaattiioonn ((IIFFSSAA)) DDeevvootteedd ttoo

RReesseeaarrcchh aanndd DDeevveellooppmmeenntt ooff SSeennssoorrss aanndd TTrraannssdduucceerrss

VVoolluummee 119966,, IIssssuuee 11,, JJaannuuaarryy 22001166

Editor-in-Chief Prof., Dr. Sergey Y. YURISH

IFSA Publishing: Barcelona Toronto

http://www.sensorsportal.com

Copyright 2015 IFSA Publishing, S. L. All rights reserved. This journal and the individual contributions in it are protected under copyright by IFSA Publishing, and the following terms and conditions apply to their use: Photocopying: Single photocopies of single articles may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copyright, copyright for advertising or promotional purposes, resale, and all forms of document delivery. Derivative Works: Subscribers may reproduce tables of contents or prepare list of articles including abstract for internal circulation within their institutions. Permission of the Publisher is required for resale or distribution outside the institution. Permission of the Publisher is required for all other derivative works, including compilations and translations. Authors' copies of Sensors & Transducers journal and articles published in it are for personal use only. Address permissions requests to: IFSA Publisher by e-mail: [email protected] Notice: No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Printed in the USA.

SSeennssoorrss && TTrraannssdduucceerrss

Volume 196, Issue 1, January 2016 www.sensorsportal.com

e-ISSN 1726-5479 ISSN 2306-8515

Editors-in-Chief: Professor, Dr. Sergey Y. Yurish, tel.: +34 93 4137941, e-mail: [email protected]

Editors for Western Europe Meijer, Gerard C.M., Delft Univ. of Technology, The Netherlands Ferrari, Vittorio, Universitá di Brescia, Italy Mescheder, Ulrich, Univ. of Applied Sciences, Furtwangen, Germany

Editor for Eastern Europe Sachenko, Anatoly, Ternopil National Economic University, Ukraine

Editors for North America Katz, Evgeny, Clarkson University, USA Datskos, Panos G., Oak Ridge National Laboratory, USA Fabien, J. Josse, Marquette University, USA

Editor for Africa Maki K., Habib, American University in Cairo, Egypt

Editors South America Costa-Felix, Rodrigo, Inmetro, Brazil Walsoe de Reca, Noemi Elisabeth, CINSO-CITEDEF UNIDEF (MINDEF-CONICET), Argentina Editors for Asia Ohyama, Shinji, Tokyo Institute of Technology, Japan Zhengbing, Hu, Huazhong Univ. of Science and Technol., China Li, Gongfa, Wuhan Univ. of Science and Technology, China Editor for Asia-Pacific Mukhopadhyay, Subhas, Massey University, New Zealand

Editorial Board

Abdul Rahim, Ruzairi, Universiti Teknologi, Malaysia Abramchuk, George, Measur. Tech. & Advanced Applications, Canada Aluri, Geetha S., Globalfoundries, USA Ascoli, Giorgio, George Mason University, USA Atalay, Selcuk, Inonu University, Turkey Atghiaee, Ahmad, University of Tehran, Iran Augutis, Vygantas, Kaunas University of Technology, Lithuania Ayesh, Aladdin, De Montfort University, UK Baliga, Shankar, B., General Monitors, USA Barlingay, Ravindra, Larsen & Toubro - Technology Services, India Basu, Sukumar, Jadavpur University, India Booranawong, Apidet, Prince of Songkla University, Thailand Bousbia-Salah, Mounir, University of Annaba, Algeria Bouvet, Marcel, University of Burgundy, France Campanella, Luigi, University La Sapienza, Italy Carvalho, Vitor, Minho University, Portugal Changhai, Ru, Harbin Engineering University, China Chen, Wei, Hefei University of Technology, China Cheng-Ta, Chiang, National Chia-Yi University, Taiwan Cherstvy, Andrey, University of Potsdam, Germany Chung, Wen-Yaw, Chung Yuan Christian University, Taiwan Cortes, Camilo A., Universidad Nacional de Colombia, Colombia D'Amico, Arnaldo, Università di Tor Vergata, Italy De Stefano, Luca, Institute for Microelectronics and Microsystem, Italy Ding, Jianning, Changzhou University, China Djordjevich, Alexandar, City University of Hong Kong, Hong Kong Donato, Nicola, University of Messina, Italy Dong, Feng, Tianjin University, China Erkmen, Aydan M., Middle East Technical University, Turkey Fezari, Mohamed, Badji Mokhtar Annaba University, Algeria Gaura, Elena, Coventry University, UK Gole, James, Georgia Institute of Technology, USA Gong, Hao, National University of Singapore, Singapore Gonzalez de la Rosa, Juan Jose, University of Cadiz, Spain Goswami, Amarjyoti, Kaziranga University, India Guillet, Bruno, University of Caen, France Hadjiloucas, Sillas, The University of Reading, UK Hao, Shiying, Michigan State University, USA Hui, David, University of New Orleans, USA Jaffrezic-Renault, Nicole, Claude Bernard University Lyon 1, France Jamil, Mohammad, Qatar University, Qatar Kaniusas, Eugenijus, Vienna University of Technology, Austria Kim, Min Young, Kyungpook National University, Korea Kumar, Arun, University of Delaware, USA Lay-Ekuakille, Aime, University of Lecce, Italy Li, Fengyuan, HARMAN International, USA Li, Jingsong, Anhui University, China Li, Si, GE Global Research Center, USA Lin, Paul, Cleveland State University, USA Liu, Aihua, Chinese Academy of Sciences, China Liu, Chenglian, Long Yan University, China Liu, Fei, City College of New York, USA Mahadi, Muhammad, University Tun Hussein Onn Malaysia, Malaysia

Mansor, Muhammad Naufal, University Malaysia Perlis, Malaysia Marquez, Alfredo, Centro de Investigacion en Materiales Avanzados, Mexico Mishra, Vivekanand, National Institute of Technology, India Moghavvemi, Mahmoud, University of Malaya, Malaysia Morello, Rosario, University "Mediterranea" of Reggio Calabria, Italy Mulla, Imtiaz Sirajuddin, National Chemical Laboratory, Pune, India Nabok, Aleksey, Sheffield Hallam University, UK Neshkova, Milka, Bulgarian Academy of Sciences, Bulgaria Passaro, Vittorio M. N., Politecnico di Bari, Italy Patil, Devidas Ramrao, R. L. College, Parola, India Penza, Michele, ENEA, Italy Pereira, Jose Miguel, Instituto Politecnico de Setebal, Portugal Pillarisetti, Anand, Sensata Technologies Inc, USA Pogacnik, Lea, University of Ljubljana, Slovenia Pullini, Daniele, Centro Ricerche FIAT, Italy Qiu, Liang, Avago Technologies, USA Reig, Candid, University of Valencia, Spain Restivo, Maria Teresa, University of Porto, Portugal Rodríguez Martínez, Angel, Universidad Politécnica de Cataluña, Spain Sadana, Ajit, University of Mississippi, USA Sadeghian Marnani, Hamed, TU Delft, The Netherlands Sapozhnikova, Ksenia, D. I. Mendeleyev Institute for Metrology, Russia Singhal, Subodh Kumar, National Physical Laboratory, India Shah, Kriyang, La Trobe University, Australia Shi, Wendian, California Institute of Technology, USA Shmaliy, Yuriy, Guanajuato University, Mexico Song, Xu, An Yang Normal University, China Srivastava, Arvind K., Systron Donner Inertial, USA Stefanescu, Dan Mihai, Romanian Measurement Society, Romania Sumriddetchkajorn, Sarun, Nat. Electr. & Comp. Tech. Center, Thailand Sun, Zhiqiang, Central South University, China Sysoev, Victor, Saratov State Technical University, Russia Thirunavukkarasu, I., Manipal University Karnataka, India Thomas, Sadiq, Heriot Watt University, Edinburgh, UK Tian, Lei, Xidian University, China Tianxing, Chu, Research Center for Surveying & Mapping, Beijing, China Vanga, Kumar L., ePack, Inc., USA Vazquez, Carmen, Universidad Carlos III Madrid, Spain Wang, Jiangping, Xian Shiyou University, China Wang, Peng, Qualcomm Technologies, USA Wang, Zongbo, University of Kansas, USA Xu, Han, Measurement Specialties, Inc., USA Xu, Weihe, Brookhaven National Lab, USA Xue, Ning, Agiltron, Inc., USA Yang, Dongfang, National Research Council, Canada Yang, Shuang-Hua, Loughborough University, UK Yaping Dan, Harvard University, USA Yue, Xiao-Guang, Shanxi University of Chinese Traditional Medicine, China Xiao-Guang, Yue, Wuhan University of Technology, China Zakaria, Zulkarnay, University Malaysia Perlis, Malaysia Zhang, Weiping, Shanghai Jiao Tong University, China Zhang, Wenming, Shanghai Jiao Tong University, China Zhang, Yudong, Nanjing Normal University China

Sensors & Transducers Journal is a peer review international journal published monthly by International Frequency Sensor Association (IFSA). Available in both: print and electronic (printable pdf) formats. Copyright © 2015 by IFSA Publishing, S. L. All rights reserved.


SSeennssoorrss && TTrraannssdduucceerrss JJoouurrnnaall

CCoonntteennttss

Volume 196 Issue 1 January 2016

www.sensorsportal.com ISSN 2306-8515e-ISSN 1726-5479

Research Articles

Gesture Based Control Using Windows Kinect and Robot Operating System Kunal Kaushik, K. Sriram and M. Manimozhi ....................................................................... 1 Metrological Array of Cyber-Physical Systems. Part 14: Basics of Metrology and Techniques for Accuracy Improvement Bohdan Stadnyk, Svyatoslav Yatsyshyn and Yaroslav Lutsyk ............................................ 7 Coherent Anti-Stokes and Coherent Stokes in Raman Scattering by Uperconducting Nanowire Single-Photon Detector for Temperature Measurement Annepu Venkata Naga Vamsi and Annepu Bhujanga Rao ................................................. 24 Dynamic Sensor Management Algorithm Based on Improved Efficacy Function Tang Shujuan, Xu Yunshan and Yang Tao ......................................................................... 30 Scattering Parameters of Broadband (18 – 40 GHz) RF MEMS Switch in -match Configuration Updesh Sharma, Shankar Dutta and E. K. Sharma ............................................................ 37 Packet Header Compression for the Internet of Things Pekka Koskela, Mikko Majanen and Mikko Valta ................................................................ 43 Collapse Mode Characteristics of Parallel Plate Ultrasonic Transducer Radiating in Air and Water Rashmi Sharma, Rekha Agarwal and Anil Aror .................................................................. 52 Improving Vibration Energy Harvesting Using Dynamic Magnifier Almuatasim Alomari, Ashok Batra and C. R. Bowen ........................................................... 57 Design and Implement of Pyroelectric Energy Harvester Experimental Measurement System Based on STM32F103VET6 Honghua Liao, Hao Fu, Weichuang Yu, Binbin Zhou, Ting Yu and Yongdan Zhu .............. 69 An Automatic Segmentation Method for Printed Circuit Board Welding Component under Stereo Optical Microscope Yi Liu, Mei Yu, Li Cui, Gang-Yi Jiang, Yi-Gang Wang and Sheng-Li Fan ........................... 75 Research of an Optoelectronic Current Transformer Based on a Designed Magneto-Optic Sensor R. El-Bashar, J. El-Azab, Y. Badr and R. Yousif ................................................................. 82

Authors are encouraged to submit article in MS Word (doc) and Acrobat (pdf) formats

by e-mail: [email protected]. Please visit journal’s webpage with preparation instructions: http://www.sensorsportal.com/HTML/DIGEST/Submition.htm

International Frequency Sensor Association (IFSA).


Sensors & Transducers, Vol. 196, Issue 1, January 2016, pp. 1-6

1

Sensors & Transducers

© 2015 by IFSA Publishing, S. L. http://www.sensorsportal.com

Gesture Based Control Using Windows Kinect and Robot Operating System

KUNAL KAUSHIK, K. SRIRAM and M. MANIMOZHI

School of Electrical Engineering, VIT University, Vellore, 632014, Tamil Nadu, India Tel.: +91-9159897511

E-mail: [email protected]

Received: 7 November 2015 /Accepted: 7 December 2015 /Published: 30 January 2016 Abstract: This paper deals with using a common gaming sensor Kinect in order to control a wheel chair using hand gestures to help speech disabled person. Lately there have been many attempts to make wheel chairs voice controlled or analog control, but gestures are natural way to communicate having a universal understandable meaning. Using gestures, we can control the speed and the direction of a wheelchair in a more intuitive way as the gestures significantly describe the intensity of the action desired. Various human body organs can be used to give input to the system. Copyright © 2015 IFSA Publishing, S. L. Keywords: Kinect, Skeleton tracking, Gesture recognisation, Robot Operating System, Wheelchair,depth tracking. 1. Introduction

The hand gestures are the most intuitive way of

human communication, which can covey message in a short and precise manner which is universal in nature, Now a days due to increase in robot human interaction and machine being so close in vicinity of humans, the need of a new human- machine interaction language which is more natural and easy to communicate is aroused, The earlier attempts were focused on the color detection techniques and voice recognition which had the limitations of fluctuations due to light conditions and need of frequent calibrations in case of color based approach and the difficulty of addressing the intensity of instructions in case of voice controlled models. The low accuracy and tough compositionality were also major drawbacks for these models.

Among the various techniques available for gesture recognition like image processing prove to be computationally very intensive and recognition using

an array of ultrasonic sensors might not be accurate for particular complex gestures.

Hence the Kinect provides a cost effective, less computationally intense and precise solution, The Kinect is a sensor developed by Microsoft which includes a RGB plus Depth camera, which prepares a density map of its surroundings in 3-D in form of colored point clouds, where each points have as many as 300,000 points. This data can obtain the real time position of a human body skeleton in 3-D space. Later the 3-D input data is fed to ROS for algorithmic computations and hence giving inputs for gesture recognition

2. Kinect & It’s Working

A Microsoft Kinect sensor includes a very high resolution RGB and depth sensing which is lately becoming a trend to recognize gestures and work with human computer interaction. It implements various

http://www.sensorsportal.com/HTML/DIGEST/P_2780.htm



2

tasks such as object detection and reorganization, tracking an object. It also helps in human activity analysis, hand gesture analysis and 3D mapping. It can be used to detect and tell apart different varieties of objects. Thus the Kinect was found to be an effective tool action recognition.

The Kinect camera includes an infrared projector, the color camera, and the Infrared detector. The depth sensor is made up of the IR projector combined with the IR camera, which is a monochrome CMOS sensor. The IR projector is a single IR laser that passes through a number of diffraction gratings to be turned into set of IR dots of a determined pattern.

The IR camera, emitter and different dots make up the vertices of a triangle and simple geometry is used to determine the depth. If a dot found out in an image matches with a dot in the predetermined pattern, rebuild it in 3D using triangulation. Because the pattern the dots make is relatively random, the match between the IR image obtained and the projector pattern can be done in a relatively easy way by comparing small neighborhood’s using, for example, normalized cross correlation.

In skeletal tracking, body of a human is represented using different joints representing body parts such as head, neck, shoulders, and arms. Each joint is in turn represented by using its 3D coordinates. The aim is to find out all the three parameters of these joints in real time to facilitate continuous interaction and with as less computation resources as possible. Xia et al. proposes a model based algorithm to detect humans by usage of the depth maps produced using the Kinect sensor.

Sung et al. extract the data obtained from the skeleton and uses a supervised learning to recognize activities from RGB and depth images obtained using a Kinect sensor. He used a supervised learning approach, where he extracts features from the skeleton joints, its positions, its orientation and arm movements, he developed a model from labeled instances. Xia et al. proposed an algorithm that recognizes the actions using a histogram of joints in 3D extracted from depth images. Theresia et al proposes an algorithm that recognizes activities with the Kinect using a logic-based approach, he developed a logic model with labeled instances. Wang et al. actionlets ensemble approach to recognize activities, according to Wang, the actionlet is defined as a conjunctive (or AND) structure on the base features, one base feature is defined as a Fourier Pyramid period of one joint of the skeleton. Ong et al presents an approach to extract features from Kinect based on human range of movement. Ong applied K-means clustering on the features extracted based on human range of movement, giving the results of improvement clustering performance. Maierdan et al. proposes a HMM approach for recognize human activities. Maierdan discusses two points in his approach, the first is the HMM application of HMM to recognize human activities, the second is the effect of K-means and fuzzy C-means. Rabiner et al. reviews the theoretical aspects of HMM and show how they

have been applied to problems in machine recognition speech.

All these are described works that uses the Kinect skeleton joints tracking to recognize the activity. Some of these papers use a supervised learning algorithm. In long term, the intelligent systems need to learn new activities that are not labeled, that is one of the many reasons there is an interest to decipher human activity discovery using unsupervised learning. Some of these works use K-means clustering to discover human activities and use HMM to recognize the activities.

The approach of this work is to discover human activities when they are seated, for this purpose, we are going to record data of a seated skeleton, the data recollected we are going to pass it to K-means algorithm because of it is an unsupervised learning algorithm, and we want to use it to discover human activities, also we are going to use HMM to recognize what activity the person is doing. 3. Skeleton Tracking

Before application of the action recognition approach, the depth maps obtained by using the Kinect sensor are fed to a skeleton-tracking algorithm. The depth maps mentioned were acquired using the OpenNI. The OpenNI high-level skeleton-tracking module is used for the tracking joints of a person’s body. More specifically, the OpenNI tracker detects the position of the following set of joints in the 3D space G = gi, i ∈ [1, I] ≡ T orso, Neck, Head, Lef t shoulder, Lef t elbow, Lef t wrist, Right shoulder, Right elbow, Right wrist, Left hip, Left knee, Left foot, Right hip, Right knee, Right foot. The position of joint gi is denoted by vector pi(t) = [x y z] T, where t denotes the frame for which the joint position is located and the origin of the orthogonal XY Z co-ordinate system is placed at the center of the Kinect sensor. The OpenNI skeleton-tracking module requires an initial user calibration in order for it to find out approximately several body characteristics of the person. This is done by performing a particular pose in front of the Kinect at the beginning of usage. In recent versions of OpenNI, this initial pose is eliminated by the ‘auto-calibration’ mode which enables user calibration without implying the person to stay in any particular calibration pose. Since a calibration pose was captured for the employed dataset, the OpenNI’s ‘auto-calibration’ mode is used not in this work. Fig. 1 shows he initial calibration pose required to identify the skeleton.

The experimental evaluation revealed that the skeleton-tracking algorithm being applied here is relatively robust. The position of the joints is usually detected accurately, although there were cases where the tracking was not completely correct. An example of the latter is the inaccurate detection of the joint positions when there is a very sudden and intense movement. (e.g. leg movement while a round house kick).


3

Fig. 1. The initial calibration procedure pose.

4. Depth Calculations

To calculate depth at a single point let us assume a point k. As the distance of speckle 2 with camera will be changed, its position in the focal plane will be changed in x direction. This change in distance d is called the disparity of that particular marker. This change in disparity is used to find the depth of the point.

Here, C is the camera that detects the markers, L is the laser source, D is the 3-D disparity, Z0 is distance of the camera from the Reference plane, Zk is the distance of the point k form the camera. Fig. 2 depicts the distance calculation for a single marker, with triangulation.

Fig. 2. Depth calculations.

By similarity of triangles,

By solving the above

1

The same process is repeated for all the markers.

Once, the distances of all the markers are triangulated, then a disparity map, shown below is produced. Fig.3 is an example of obtained disparity map.

Once the depth cloud is attained, an open-source software called NITE is used to interpret the point-clouds, that look similar to human figure, into human skeletons. Hence, the skeletal information is attained. The NITE toolbox publishes 3-D position information about various body parts, like hands, elbows, knees, etc. The tracked skeleton looks similar to the image below. In our case, we take the 3-D positional data of the right hand and train the algorithms.

Fig. 3. Disparity Map.

5. Action Recognition

Action recognition Action recognition can be further divided into three subtypes. 5.1. Pose Estimation

The aim of this step is to estimate an updated

orthogonal basis of vectors in real time for every frame t that represents the person’s pose. The calculation of the later is based on the assumption that the orientation of the person’s torso is the quantity that is most characteristic of the subject during the execution of any action and can be used as reference for the same reason. For pose estimation, the position of three joints is taken into consideration: Left shoulder, Right shoulder and Right hip. These make up the joints around the torso area, whose position relative to each other remains unchanged to the greater extent during the execution of any action. The reason behind the consideration of the three before mentioned joints, instead of directly approach to estimating the position of the torso joint and its normal vector, is to obtain a more accurate estimation of the person’s pose. It must be seen that the Right hip joint was preferred to the


4

obvious Torso joint selection. This was done so that the orthogonal basis of vectors to be estimated from joints with bigger distances in between that will be more likely to lead to obtain more accurate pose estimation. However, no significant deviation in action recognition performance was found when the Torso joint was used instead.

5.2. Action Representation

For getting efficient action recognition, a proper representation is required that will handle the differences in appearance, human body type and execution of actions among the individuals satisfactorily. For that the angles of the joints and relative position are used, which proved to be more discriminative than using normalized coordinates of the joints. Building on the fundamental idea of the previous section, all the angles are computed using the Torso joint as reference, i.e. the Torso joint position is used as the origin of the spherical coordinate system. For computing the proposed action representation, only a subset of the supported joints is used. This is because the trajectory of some joint contains redundant or noisy information mainly. To this end, only the joints0 corresponding to the upper and lower body limbs were taken into account after experimental evaluation, namely the joints Left shoulder, Left elbow, Left wrist, Right shoulder, Right elbow, Right wrist, Left knee, Left foot, Right knee and Right foot. The velocity vector is approximated by the displacement vector between two successive frames, i.e. vi(t) = i(t)−pi(t−1). The estimated spherical angles and angular velocities for frame t constitute the frame’s observation vector. Collecting the computed observation vectors for all frames of a given action segment forms the respective action observation sequence h that will be used for performing HMM-based recognition, as will be described in the next part.

5.3. HMM Based Recognition

Markov Models is stochastic model describing the sequence of possible events in which the probability of each event depends only on the state attends in the previous event. This model is too restrictive to be applicable to current problem and thus the concept of Markov model is extended to form Hidden Markov Model (HMM). HMM is doubly embedded stochastic process with the underlying stochastic process i.e. not observable (it is Hidden) but can only be observed through set of stochastic process that produce the sequence of observations. HMMs are employed in this work for performing action recognition, due to their suitability for modeling pattern recognition. In particular, a set of J HMMs is employed, where an individual HMM is introduced for every supported action aj. Each HMM receives as input the action observation sequence h (as described above) and at the

evaluation stage returns a posterior probability P (aj|h), which represents the observation sequence’s fitness to the particular model. The developed HMMs were implemented using the software libraries of Hidden Markov Model Toolkit (HTK). 6. Setup

The current project being discussed has a Kinect

sensor is placed in front of a wheelchair and is attached to a laptop running on LINUX and to the laptop attached is the Arduino board to control the motors of a wheelchair. The Arduino board is precoded to perform certain actions based on the inputs received from the laptop. The Kinect sensor captures the depth and RGB images and sends them to the laptop for processing. The laptop is upon receiving the images employs the above discussed algorithms to track the joints and determines the lengths which the right and left hands rise or fall. It then sweeps the values of the obtained lengths into limits both above and below which the Arduino can understand and move the motors accordingly. ROS system is used for the interface between laptop and the Arduino. 7. Process

The process is initialized by standing infront of the

Kinect in a particular pose in our experiment but auto calibration can be used when used for disables person in the chair. Fig. 4 here shows the process flow model of Skeletal Frame recognition by Kinect.

Initially Infrared Rays (IR) are emitted from the IR transmitter of Kinect sensor. Emitted rays are been received by Kinect receiver. Since the Kinect sensors monitoring for the human joints, it does not show any significant data until the human joints are recognized. If any object other than the skeleton joints are recognized it discards the frame and restarts the scanning of the next frame until joints are recognized. Black frame indicates that neither the object is been detected nor the skeletal joints are detected. This kind of image results into blackening of frame and the white spots on the black frame are due to noises present in the environment. Once the Joints are been recognized/detected Kinect uses HMM algorithm for joint estimation and predicts the future movements.

Fig.5 is the process flow model of Stage 2 which includes Calculation and implementation. This recognized joint information is read by the laptop and swept into limits and is sent to Arduino using Robot Operation System (ROS). There the signals are converted into PWM pulses by the programmed PWM pulse generator present on Arduino board. The generated PWM pulses which serve as input to the servo motors, are been made to perform the desired movement according to the action that has been captured. Since this is real time the entire process is been continuously repeated for each frame.


5

Fig. 4. Stage 1. Skeletal Frame recognition by Kinect.

Fig. 5. Stage 2. Calculation and implementation.

8. Wheelchair Operations

The motors run continuously in forward or

backward direction as per the input given by use of right hand. For the angular movements only one motor will be moving while other will be in stationary position.

For turning right the right motor will stop and left will rotate the wheelchair up to a certain extent and similarly for turning left , the left motor will stop allowing the right motor to rotate wheelchair in left direction. 9. Observations and Results

The observations are recorded considering the

initial calibration position as a zero reference. The

variations in the speed was recorded using tachometer and the angular movements the optical encoder in the feedback.

Table 1. Speed control using right hand gestures.

S.no

Right hand movement’s variation in

upward direction

(cm)

Speed in forward direction (km/h)

Right hand movement’s variation in downward direction

(cm)

Speed in reverse

direction (km/h)

1. 0 0 0 0 2 5 2.3 5 1.3 3 10 4.1 10 2.4 4 15 6.9 15 4.6 5 20 9.1 20 5.9 6 25 11.6 25 7.6 7 30 13.6 30 9.2


6

Table 2. Direction control using left hand gestures.

S.no

Left hand movement in upward direction

(cm)

Left rotation (degrees)

Left hand movement

in downward direction

(cm)

Right rotation (degree)

1. 0 6.7 0 6.9 2 5 13.6 5 14.5 3 10 20.8 10 19.4 4 15 32.6 15 33.8 5 20 40.7 20 39.8 6 25 51.6 25 50.7 7 30 58.3 30 59.4

10. Conclusions

The efficiency obtained by this system is good enough for a practical implementation. The hand gesture recognition on a repeated trials yielded as much as 86 % similar outcomes as its preceding trials which makes its behaviour in real life circumstances reliable and trusted.

The cost of this design is nearly 30-40 % cheaper than the other viable options as LIDAR, which makes it industrially feasible.

References [1]. Agarwal A., Triggs, B., 3D human pose from

silhouettes by relevance vector Regression, in Proceedings of the IEEE International Conference on Computer Vision and Pattern Recognition, Vol. 2, 2004, pp. 882-888.

[2]. Dnyaneshwar R. Uttaarwar, Motion Computing using Microsoft Kinect, in Proceedings of the National Conference on Advances on Computing, 2013.

[3]. Rodríguez, N. D., Wikström, R., Lilius, J., CuÃl’llar, M. P., Flores, M. D. C., Understanding Movement and

Interaction: An Ontology for Kinect-Based 3D Depth Sensors, in Ubiquitous Computing and Ambient Intelligence. Context-Awareness and Context-Driven Interaction, Springer, 2013, pp. 254-261.

[4]. J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman, and A. Blake, Real-Time Human Pose Recognition in Parts from a Single Depth Image, in CVPR, IEEE, 2011.

[5]. L. Xia, C.-C. Chen, and J. K. Aggarwal, Human Detection Using Depth Information by Kinect, in Proceedings of the IEEE Computer Society Conference on Computer Vision (HAU3D) and Pattern Recognition Workshops (CVPRW), Colorado Springs, CO, 2011, pp. 15-22.

[6]. Xia, L., Chen, C. C., Aggarwal, J. K., View invariant human action recognition using histograms of 3d joints, in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2012, pp. 20-27.

[7]. Sung, J., Ponce, C., Selman, B., Saxena, A., Human Activity Detection from RGBD Images, in Proceedings of the AAAI Plan, Activity, and Intent Recognition Workshop, 2011.

[8]. Maiike Johanna Theresia Veltmaat, Recognizing Acitivities with the Kinect A logic-based approach for the support room, MSc Thesis, Radboud University Nijmegen, The Netherlands, 2013.

[9]. J. Wang, Z. Liu, Y. Wu, J. Yuan, Mining Actionlet Ensemble for Action Recognition with Depth Cameras, in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012, pp. 1290 - 1297.

[10]. Ong, W. H., Palafox, L., Koseki, T., Investigation of Feature Extraction for Unsupervised Learning in Human Activity Detection, Bulletin of Networking, Computing, Systems and Software, 2, 1, 2013, pp. 30.

[11]. Maierdan, M., Watanabe, K., Maeyama, S., Human behavior recognition system based on 3-dimensional clustering methods, in Proceedings of the 13th IEEE International Conference on Control, Automation and Systems (ICCAS), 2013, pp. 1133-1137.

___________________

2015 Copyright ©, International Frequency Sensor Association (IFSA) Publishing, S. L. All rights reserved. (http://www.sensorsportal.com)

http://www.iaria.org/conferences2016/SIGNAL16.html


7

Sensors & Transducers© 2016 by IFSA Publishing, S. L.


Metrological Array of Cyber-Physical Systems. Part 14: Basics of Metrology and Techniques

for Accuracy Improvement

Bohdan STADNYK, Svyatoslav YATSYSHYN and Yaroslav LUTSYK National University ‘L’viv Polytechnic’, Institute of Computer Technologies, Automation

and Metrology, Bandera str.12, L’viv, 79013, Ukraine Tel.: +38-0322-37-50-89


Received: 27 November 2015 /Accepted: 30 December 2015 /Published: 30 January 2016 Abstract: It is examined the specific of metrology of Cyber-Physical Systems as well as the sequence of methods for accuracy enhancing. For precise operation of full-fledged CPS the metrological control of group of measurement parameters is needed. Different approaches of received results processing are studied aiming the improvement of accuracy of CPS operation. Copyright © 2016 IFSA Publishing, S. L. Keywords: Cyber-Physical System, Methods of Accuracy Improvement, Errors approach, Uncertainty approach, Hybrid approach.

1. Introduction Day after day the globalization of industry,

agriculture, transport, health-care and so on becomes more total. Certainly it contributes to continual development of Internet technologies, one manifestation of which is the occurrence of Cyber-Physical Systems (further CPSs). Realization of existing CPS development programs is impossible without taking into account the metrological aspects of designing and operating CPSs. Therefore current NIST program [1] mainly focuses on involving metrological science to resolve some CPS-problems at the design stage. Next row of problem seems to emerge the evident tasks in CPSs operating modes, firstly trying to provide traceable and quantitative data for validating the process models, calibrating in-process sensors, and determining the optimal process conditions, and furthermore endeavoring to obtain the objective quantitative information of technological

processes by measuring their parameters and at last to assess the quality of final CPSs products.

Just because measurement should be considered as a holistic process that starts from perception and transformation of object measurement data to its processing, storage, transmission and application for developing retroactivity in controlled technological objects. Therefore, one of the most important CPSs’ parameters is their general and metrological reliability due to continuously varying structure, modes, conditions, and environment of particular components and units. Additionally, the manufacturing CPSs would not cause environmental damage, greater from the acceptable standards. The problem of preventing the environmental and technogenic accidents and disasters should be noted also.

Current article tries to consider the classic metrology approach to CPSs operation, and to ensure their development in applied problems by studying:

- Verification and validation of the metrological units for parameters determining the controlled




8

equipment, process, materials through the develop-ment, implementation and realization of specific metrology and standardization methods and techniques, instruments and facilities, and etc. For instance, calibration could be performed remotely under condition of code access to CPS subsystems with implemented appropriate software;

- Aspects of metrological reliability, including its prediction, particularly of CPS integrated metrological subsystems remote nodes. This includes not only microwave research essential for the link at optical wavelengths, but low and ultralow frequency methods in which it can be detected successfully the hidden and latent defects of complicated CPSs.

2. Shortcomings

Unique and newly created CPSs often require checking and verification of metrology facilities to ensure their quality operation.

To reach the necessary metrological reliability of information and measuring subsystems of CPSs, in practice, try constantly to supervise the measurements. Reliable measurement information of required accuracy can be obtained only through technically informed choice of measuring instruments (further MIs) and includes the following data: availability of measured or monitored parameters of object; tolerance for deviations of these parameters and allowable measurement uncertainties; allowable probability of false and unidentified rejections for each of monitored parameters and the values of confidence for them; distribution laws of measuring parameters and their measurement errors that can arise while using the MIs; measuring conditions: mechanical loads (vibration, shock, acceleration, etc.), climatic impacts (temperature, humidity, pressure, etc.), and so on. The existing measures applicable for CPSs calibration lose their values (accuracy characteristics) by several orders while transferring them to the end user, and that is actually considered as a normal metrological practice. However, such practice cannot be deemed adequate for development of CPSs.

3. Aim of Paper

Goal of this paper is presentation and consideration of main trends in the branch of metrology of Cyber-Physical Systems which are becoming a key element of everyday life. Trying to highlight emergence and development of these systems, we present published in the current set of 15 works, as well as reviewed previously by means of 2 similar rows of articles, 2013, Sensors & Transducers, Issues 3-11 (common notion “Development of Noise Measurements”) and 2012 – “Research in Nanother-mometry” studies at an angle of CPS metrological specific. A new outcoming aim has emerged – metrological assurance that embraces the adjustment of necessary measurement precision by means of continuously self-checked, self-verificated and self-adjusted MIs, self-validated

metrological procedures, and finally obtaining the high-level metrological traceability and adequate metrological assurance of CPS’s final product.

4. Theoretical and Practical Consideration

There are considered traditionally that CPS is a system of collaborating computational elements controlling physical entities. Today, a precursor generation of CPSs can be found in areas as diverse as aerospace, automotive, chemical processes, civil infrastructure, energy, healthcare, manufacturing, transportation, entertainment, and consumer appliances [2]. The reliability of such systems is ensured by maintaining the operability of its components due to redundancy, regulatory replacement of components etc. Unlike more traditional embedded systems, a full-fledged CPS is typically designed as a network of interacting elements with physical input and output instead of as standalone devices. On the basis of available experience of metrology of the mentioned units operation, it seems appropriate to extend the understanding of aforesaid systems in the direction of natural affiliation, to the next concept: a full-fledged CPS with the metrological assurance of group of operating parameters as well as the basic characteristics of the intermediate product has to be designed as a network of interacting elements with physical input and output of every element that is controlled at each stage of operation providing a qualitative final product. Furthermore, the CPS can change over time, and a priori is known that the components and connections between CPS’s units are not 100 per cent reliable.

Firstly, such interacting elements may be sensors and actuators. Best modifications of them inherent in their own function-transformative computing proper-ties. The brief example seems to be a smart sensor. It is the analog or digital primary thermosensitive transducer combined with a processing unit and a communicating interface [3] and able to perform a row of smart metrological functions due to installed metrological software. This is intelligent sensor with a number of specialized algorithms provided in the design or installation stage, i.e. a sensor with such embedded algorithms that are necessary to provide implementation of the following specialized metro-logical functions. Namely, such functions include, f.i. the ability to realize automatic switching of sub-range of measurement, depending on input signal value; automatic self-validation, self-check, self-diagnostics and etc.; the introduction of adjustments when the action of impact factor takes place; linearization of metrological characteristic; compensation of cold-junction temperature for thermocouples and so on.

The major problem of CPSs operation is determined mostly by credibility of obtained infor-mation which depends as on sensors metrological reliability as well as on actuators precision and


9

accuracy. The latter has to be gauged and control by a set of different sensors whose participation in the management is determined in the design phase or changes automatically by adjusting. Unfortunately these units become obsolete, and more importantly metrological characteristics drift up to mechanical failure. Possible consequence of running processes affect in lowering the quality of service/product.

According to current practice of standardization the traceability of measurements is provided by periodic calibrations (graduations, verifications, etc.). Then duration of the intercalibration interval defines the period of operation of the mentioned unit with a certain, previously accepted probability of mainly metrological or total failure.

The state of MI is usually verified by comparing with measure or standard, or by supplying electrical signal of reference value to its input, or by verifying the installed metrological software versus the checked one. Since two from three failures of MIs are caused by metrological failures and they usually precede major failures, increases the need for cost calibration procedures of every sensor within calibration period (~2-3 years) or its substitution. The latter may be unrealized for instance for temperature and pressure sensors of nuclear power plants. The special issue seems to be a necessity to suspend the production cycle aiming to provide the calibration of sensor(s). Unreliable information received from the MIs with the considerable drift of characteristics degrades the quality of the final product.

CPS technologies companies have to utilize the sophisticated metrology equipment for production lines. This involves the estimation of the compa-rability of CPS component MI by verification. Development of portable, highly-precise devices is able to provide in-place precision measurements. Chip-scale devices could be directly integrated into equipment to provide continuous quality control and assurance, freeing manufacturers and customers from complex measurement traceability chains and lengthy calibration procedures.

For metrological calibration of MIs usually one applies the direct measurement by the verified MI of outgoing signal of multivalued measure with determi-nation of the error as a difference of its readout and mentioned signal. Correction methods of systematic error constituent are realized by operator impact or automatically in offline mode when, for example, self-calibration is carried out [4]. For CPSs operation is important not only equipping them with the MIs, but also providing CPSs by reliable information. For these information and measuring subsystems the periodic verifying the certain parameters is assumed.

4.1. Major Metrological Characteristics of CPSs and Their Units

Each of the following factors entails that the results

of measurements differ from the true values of the

measurands. The quality of measurements deteri-orates, and thus the quality of CFS gets worse. These factors are the next [5]: problem of object model and the measurand (it is due to simplification of measurement procedures as well as experimental and theoretical generalizations that results in idealization of object properties); mutual influence of object and MIs (for example caused by placing the sensor at the facility); imperfection of MIs (among all other possible factors deteriorating quality of measurement result, the instrumental factor is always available); calibration of MIs (is considered below); conditions of measurement (almost impossible to determine accurately the impact functions or their values as they may by unstable over time); dynamics of variables (significant influence on the dynamic characteristics of measurands is observed in nanotechnology); mathe-matical simplification of sensors transfer function; volume of measurement data and conjugated com-puting problems (too small array of experimental data can lead to misconceptions about the course of the considered process and, conversely, too big amount of data may result not only in low-quality changes weakening and in loss of reliability of controlled parameter, so it can be resolved involving cloud technologies).

4.2. Interpretation of Measurement Results within Different Approaches

Errors approach produces established way to the classification of errors based on their specific properties. This separation of errors defines methods of reducing their impacts and results assessment. Errors can depend or not depend on the value of measurand. In this regard the additive, multiplicative and nonlinear errors are distinguished. Additive one is independent on the value of the measurand, and the amendment is algebraically added to the measured value. Multiplicative error increases or decreases linearly with measurand increasing; it is proportional to the product of certain factor (positive or negative) and the measured value. Nonlinear errors nonlinearly depend on the measured value. The ultimate goal of the measurement errors analysis is just assessment of boundary errors in which they are located with a certain probability. Then measurement result with intervals determined by these error boundaries with given probability, covers the true value of measurand.

In uncertainty approach of measurement result [6] on the one hand does not use the concept of true quantity value because it is unknown, and, on the other hand, implements a unified approach to quantitative assessment of results quality regardless of origin and method of various factors impact on the measurement result. Another quantitative characterization of mea-surements quality, namely uncertainty of measure-ment result, is introduced. Although, most of the errors approach principles are successfully utilized in hidden form. Thus both methods rely on the use of source distribution density that causes the outcome. Standard


10

uncertainty is the uncertainty of result expressed by standard deviation. It may also be given in the form of dispersion as the square of the standard uncertainty. The standard uncertainty of type A is calculated by statistical processing of the results of series of successive observations. The standard uncertainty of type B is calculated other than in statistical way, for example basing on a priori specified source of uncertainty density distribution. The combined standard uncertainty is uncertainty that is determined in case if during measurement the effect of several uncertainty sources is simultaneously revealed or if obtained result is a certain function of other measurement results. The combined uncertainty is defined as the square root of sum of the squares of the particular standard uncertainties for the appropriate weight factors and eventual statistical relationship (correlation) between uncertainty components.

Hybrid approach of measurement result evalu-ation [7] that combines the error approach and uncertainty approach turned out to be the next step in the development of an integrated assessing the measurements accuracy. According to it, error is considered as the measurand with uncertainty determined by the assessment (evaluation or calculation). It reveals the possibility of simultaneous application of error and uncertainty approaches, which corresponds to hybrid approach of measurement result evaluation and assessment. To wit, an error is being calculated and evaluated as a physical value whose particular coefficients are defined with some uncertainty (Fig. 1).

Fig. 1. Threshold weight of summary error and its uncer-tainty of result (3): systematic component due to impacts of MI and fluctuation of its properties (2); similar factor due to influence of thermometry-processed object (1).

Hybrid-thermodynamic approach of measurement result evaluation [8] is an extension of the hybrid approach towards consideration of the origins of fluctuation deviations in metrological characteristics on statistical-thermodynamic basis. Hybrid-thermody-namic approach of measurement result evaluation implies researching the total error of a temperature transducer with involving Non-equilibrium thermodynamics. In this case the threshold value of cognizable component of instrumental error systema-tic constituent is determined as the additive totality of

multiplicative pairs of influence functions and their coefficients. Hereby, the pairs are formed so that one of the multipliers is determined by fluctuations of thermometric substance properties, and another – by that of applied outer field parameters. It corresponds to the content of Fluctuation dissipation theorem of Irreversible thermodynamics.

Hybrid-thermodynamic approach of measurement result evaluation has been developed consequently aiming to decrease the issues of MI intrusion and to improve the measurement accuracy of micro-, nanoobject temperature. It roots in the threshold value determination of an instrumental error systematic component as an additional totality of influence-functions’ multiplicative pairs (see below). However assessments of the origin of errors and uncertainties, based on thermodynamics, form the basis of hybrid-thermodynamic approach of measurement result evaluation. Its main reason roots in the next: the measurement result evaluation is quite good elaborated for macroobjects, having not been even established in the case of nanosamples. Nowadays the hybrid-thermodynamic approach of measurement result evaluation concerns with the study of origin sources of particular errors and influence functions and effectively applies in complicated cases of metrological reliability evaluation of measuring instruments. In particular, the research of energy-transmission processes, based on statistical thermodynamics, enables us to determine a methodical error component as well as cognizable part of systematic component of an instrumental error component, and thus to decrease substantially the guaranteed by the producer of thermometric means a total error of measuring the temperature in exploitation conditions.

Here the accepted IMC approach has been modified by the way of cognizing the certain compo-nents of an instrumental error through the extraction, study and evaluation of the factors influencing a MI, on the basis of statistical thermodynamic nature of their formation. The results of thermometric substance fluctuation concerning the summary influence function _ maxMetT K of thermoelectric transducers at

presence of external thermodynamic fields are determined as K=(KX+KM)KT, where KX; KM; KT are the chemical, mechanical and thermal influence functions respectively caused by specific transport processes created by the external effect in thermometric substance. At the availability of fluctuations, additional impact functions (temperature, density, strain and etc. gradients) multiplicity the influence actions related by the fluctuation effect of external environment up to: KΣ [F(T, p, V,…, t)] = (KXKP + KMK) KTKE, where KP; K; KE are the recrystallization, porous and entropy influence functions respectively.

Joining in the pairs, where one of the multipliers is defined by the fluctuations of thermodynamic substance properties, and another – by those of the parameters of the applied outer fields caused by the


11

thermometry-processed object, meets the content of the fluctuation-dissipation theorem of thermodyna-mics. This approach is quite precious and allows deter-mining the recognizable component of systematic error component of MI reducing significantly the guaranteed instrumental error.

The combined impact function of temperature measurement is defined within received by summation coverage interval:

from ( )P E

X M T

K K K

K K K

in the presence of two

independent systematic constituents;

by 2 2

( )P E

X M T

K K K

K K K

for the correlated

constituents;

to 2 2

( )P E

X M T

K K K

K K K

for uncorrelated values

(Ccor= –1). As result, thermotransducers with the foreseen and

managed value of an instrumental error are developed on this basis. Thus firstly, the decrement of unrecognizable error component of nanoobject tempe-rature measurement (absolute values, covering intervals and so on) has been reached, and secondly, the fluctuation restrictions of statistical physics for the improvement of metrological characteristics have been employed.

4.3. Reliability of Measurements

One of main metrological characteristics of MIs at periodic verification is reliability of measurement parameters that indicates the probability of unexceeding by measurement error value the permissible values with a certain probability

. .( )al confP Р , where P is an actual value of

probability; is the results total error obtained by

means of selected measurement units; .al is the

maximum allowable error of measurement result; Pconf. is the given value of confidence. Impossibility of establishing the measurand true value and accurate determination of measurement error as well as difficulty of taking into account all the possible destabilizing factors have contributed to the creation by IMECO the normative documents. To evaluate the measurement quality, the last applies the term "uncertainty" of received results, and also the recommendations to ensure the quality of both MI and of actual CPS performance or its final product.

5. Techniques for Accuracy Improvement

Universal techniques of errors identifying do not exist, because there is a wide variety of measurement methods, MIs, and conditions. Therefore, it should be carefully study the impact factors during the preparation of measuring experiment.

5.1. Methods to Improve the Accuracy, Errors and Examples of their Reduction

There are developed a lot of different methods for

improving accuracy that are divided into three groups: methods of prevention of errors arising; methods of reducing the current errors; techniques of methodic errors reduction.

The first group includes structural and technological, protective and preventive methods. They prevent the occurrence of the error or do not allow it exceeded the permissible value. These methods base on the use of elements and components of highest quality with the most stable parameters. F.i. to reduce the temperature error, apply temperature-independent resistors. Protective and preventive methods reduce the impact of external factors and consist in diminishing their impact on measuring instrument. Examples of such methods are: tempe-rature control; magnetic or electrostatic shielding; stabilizing the power supply.

5.1.1. Methodical Error Methodical error of electric noise research

caused by the improper technique or measurement means is one of determined components of methodical error that is due to the impossibility of increasing the integration time or bandwidth f for selective filter. It results in the dependence, f.i. in the close to cubic dependence of power spectral density (PSD) S(f) on frequency. The main reason is the Hrenander uncertainty principle: tΔf = Const. According to it, narrowing the filter bandwidth requires longer measuring duration, thus there remains the same referred component of an error. The shortening the bandwidth at fixed duration or reducing the duration at fixed bandwidth of filter results in the significant uncertainty of noise measured PSD.

- Methodical error of electric noise research caused by the performance linearization while pro-cessing is a component of measurement error due to imperfect method or object discrepancy of model adopted for the measurement. More precisely it is caused to insufficiently correct interpretation of expe-rimental results while further processing or to their imperfection.

Stochastic systems are characterized by PSD S(f), proportional to 1/f. This is the flicker-noise. Experimental data have revealed that PSD could be

defined as: ( )S ff

, where α is the constant;

γ=0…3. For instance, our research has concluded γ=2.8 at the frequency band 3-12 Hz and γ=0.5 at 12-17 Hz for Pt; and γ=2.28 γ=0.9 for oxide resistor respectively. Considering the problem of thermal and low-frequency noises, we discuss the peculiarities of electron-phonon interaction by applying different approximations, regarding the possible types of adequate descriptions.


12

The measuring and processing of experimental results suggest the invariance of PSD noise S'(f), cut by a filter within the certain bandwidth f. Thus, it does not take into account that PSD S(f) is

represented by the expression: 0

( , )( ) lim el

f

P f fS f

f

,

where ( , )elP f f is the PSD at the frequency band

from f-Δf/2 to f+Δf/2 reduced to approximation

equation: ( , )' ( ) elP f f

S ff

. As a result, the additio-

nal error appears caused by the linearization of previous expression by the last one. It strengthens significantly the character of PSD dependence at frequency approaching to 0.

Conducted analysis for PSD spectral distribution by Debye model approximation has shown that error S(f)=C/f2 is methodical one. That is, the measured dependence of PSD noise 1/f is quadratically related to the frequency. Einstein model approximation within which the temperature dependence of PSD is absent (the case of thermal electric noise) allows to get rid of the methodical error S(f)=0.

- Methodical error of temperature measuring in micro- and nano- world is an error caused by raising the significance of energy-transmission processes in the system “thermometer – controlled object” with decreasing sizes of object as well as thermometer.

Less the object we deal with, the more consi-derable methodical error of temperature measuring in micro- and nano- world is. Due to the intervention of sensor in energy exchange with controlled object it affects the gauge exactness, causing the emerging systematic component of methodical error. During prolong mutual contact of sensor and controlled object, while measuring, there was facilitated the determination of relative methodical error Tmet of temperature measurement, caused by heat transfer:

( )

( )sens

metob

abhT

ABH , where a, b, h are the linear dimen-

sions of sensor, and A, B, H are the same of object. Hence, the relatively smaller sensor of measuring instrument, the smaller relative methodical error of temperature measuring. As result of prolonged ther-mal contact of warm sensor and cold controlled object, the latter is heated and the sensor is cooled, fixing the situation of heat exchange:

0( ) ( )ob ob x sen sen sen xc m T T c m T T . Here T0 is the

temperature of controlled object before measurement; Tx is the temperature of controlled object, which has established thermal contact with the sensor; Tsens is the initial temperature of sensor; cob; mob; csen; msen are the specific heat and mass of the object and the sensor respectively. In this case, the sensor measures the averaged temperature of "controlled object – sensor" over the initial temperature of the first one.

Error depends on the ratio of volume or linear dimensions of sensor and controlled object. Let us consider that at comparable thermal characteristics of the object and the sensor ratio of the volumes will be 1:1 (Fig. 2(a)), 10:1 (Fig. 2(b)) and 1:10 (Fig. 2(c)).

Fig. 2. Temperature vs. time changes of sensor cooling and object heating during their prolonged thermal contact:

a) ob ob ob sen sen senc w V c w V ; b) 10 ob ob ob sen sen senc w V c w V ;

c) 10ob ob ob sen sen senc w V c w V .

Thus, the sensor changes smoothly over time its own temperature from Tsens to Tx measuring the temperature of the object with a certain error.

Expressing mass via specific density of matter w and its volume V and taking the object and the sensor uniform discoid shape (diameter D; d and height H; h, respectively), we obtain the equation of energy balance during prolonged contact of sensor and controlled object: 2 2 ( )ob ob met sen sen sen xc w D H T c w d h T T .

Dividing in 2ob obc w D H the left and right sides, we

receive a relative methodical error of measurement:

2

2( 1) ( 1)sen sen sen sen sen sen sen

metob ob ob x ob ob x

c w V T c w d h TT

c w V T c w D H T

(1)

Measurements of bulks assume by default that a

sensor linear size does not exceed 0.1 linear size of controlled object, and the ratio of their volumes – 0.001. That defines a relative methodical error of


13

measurement no higher than 0.1 %. This value loses in combined measurement error, including the instru-mental constituent. So therefore is possible not to consider methodical error of temperature measure-ment. For nanosized sensors and controlled objects with comparable thermophysical properties (Fig. 2(b)) relative methodical error is specified as

1senmet

x

TT

T F.i., while controlling microobject

temperature 270 K by commensurate-sized sensor of temperature 300 K, Tmet=0.11=11 % was received. This concerns the methodical errors in nanoobjects temperature gauging with help of nanosized sensors.

5.1.2. Random Error The notion “random” indicates that the measure-

ments are inherently unpredictable and their results vary nearby the true value and are inherent in average deviations equal to zero for repeated measurements, performed several times with the same MI.

- Random error of temperature measurement with help of gas sensor is random error which value is determined by volume of the gas thermosensitive substance. This error decreases to 0 if volume of thermosensitive substance of sensor increases, and vice versa it increases if volume decreases. As the latter is mainly known (at Avogadro number 6.02×1023 mol.-1) it enables to express the equation with indication of numbers of gas moles n in sensor

sensitive element:2

1[ ] wA

TD Q C m

nN , where m is the

mass of thermosensitive substance. In the case of conversion to standard units of volume the next

formula can be used 22.4

Vn , where V is the concrete

value of gas volume that is determined in m3. Then we change the form of last equation to:

21 22.4

[ ] wA

TD Q C m

N V .

Calculation of error that is specified by decreasing sensitive element chamber dimension can be done by the impact of temperature fluctuations or of heat quantity fluctuations. Root-mean-square deviation of heat quantity as function of chamber volume of

sensitive element is: 1

22.4[ ] w

A

C mQ T

VN . Relative

root-mean-square deviation is equal to:

1

2 1

[ ] 22.4[ ]

w A

TQQ

Q T T C mVN

. Having substituted

value of constants in it we simplify the equation to: 12

1

2 1

6.1 10 1[ ]

w

TQ

T T C mV

. Received results of de-

pendence of relative root-mean-square deviations of heat quantity on the volume of sensor element under different mass indexes of its copper walls are demon-strated in Fig. 3. Here is demonstrated that under significant decrease of fire sensor sensitive element

dimensions (to 4 ml), the relative root-mean-square deviation increases to ±0.007 %. Such value of random error is admissible for fire technology, where due to sensor sizes thermal inertia index is ≤ 1 s.

Fig. 3. Dependence of relative root-mean-square deviations of heat quantity δσ[Q] on volume V of sensitive element.

- Random error in noise measurement is an error that emerges in obtained results and its value varies while performing the repeated measurements of the same quantity.

The source of random error appearance could root in the influence of proper noise of measuring systems, interferences and etc. The estimation of random error could be made due to the variance D or standard

deviation D of the received results. One of the methods of reducing a random error is the averaging of measurement results, particularly with the N-fold increase in the quantity of gauges, the standard

deviation av at constx decreases: xav

N

. A

random error could be reduced by enlarging the number of gauges just to the some extent. The matter is that random error is also considerably influenced by the state of object of measurement, namely thermodynamic state when the values of object parameters are the functions of time.

All these processes are specially complicated at the reducing of object size to a nano-area. In the case of the single measurements of unique properties, especially in nanotechnology, the theory of uncer-tainty could become expedient. Here the evaluation of a result is supposed to be made with some uncertainty determined by the effect of the same impact factors. Within the framework of an uncertainty approach the expounded above results could be reduced to the extended standard uncertainty of the type A by introducing the factor 1 3 .

- Random error in temperature measurement. Let us consider the possible realization of concrete gauge of certain duration concerning the object that is characterized by the given relaxation time. The most trivial case seems to be the study of relaxing thermo-metric properties, e.g. the research on fluctuation-dissipation changes in thermoelectric thermometers


14

depending on annealing time at high temperature. Those changes are exponential, and we can reduce the coverage interval of transformation function drift by lowering the values of random error and increasing the reliability of measurement.

So, we may consider the response of substance as linear. Then the power spectral density (further PSD) S(f) of the fixed fluctuations is proportional to the spectral absorption coefficient (Debye model):

( ) ( )2

pp

kS f k f

f , where kp is the coefficient of

measuring system power transfer. To wit, in consequence of stipulated application of Debye model, the frequency dependence of PSD appropriate for 1/f noises is gained. Stationary random processes with 1/f–spectrum are characterized by the critical dynamics and scale-invariant fluctuation distribution. In those systems the energy of fluctuations could be accumulated at the low frequency bandwidth, increasing the probability of emergency emissions.

In the case of Einstein model, at the concentration of phonon energy on physically elementary volumes – tensile quasi-defects – an absorption coefficient is

found: ' '

2 B B

n n f

k T k T

, and spectral absorption

coefficient is proportional to PSD: '( )

B

nf

f k T

Here (f) is frequency-independent which corres-ponds to the case of thermal noise. Experimentally fixed square character of 1/f–noise PSD could be caused by an instrumental measurement error; then the higher level of degree dependence up to cubic one would probably be related to the restriction of frequ-ency-time analysis range and integration of gained signal at the measurement of substance remaining in a non-stationary disequilibrium thermodynamic state.

- Random error, dependent on quantity of noise measurements; it is a value that decreases with increa-sing the number of measurements in different ways depending on the type of noise. That is notified (Fig. 4) for “white” noise (WN), flicker-noise (FN) and “white” noise with a flicker-component (WN+FN). We can see there that the averaging of the results of 100 gauges produces the 10-fold reduction of the error in the case of “white” noise, ≈1.2-fold reduction in the case of flicker-noise and ≈1.4-fold reduction for “white” noise with flicker-component. Thus, the random error could be reduced to the negligible value only if the spectrum of the measured value is the same within the frequency bandwidth from 0 to super high frequencies.

The results of real measurements are represented below. Hereby, the interval of time between results of

measurement is chosen from condition: 1 1

2 h

tf f

,

here fh is the upper frequency in spectrum of measured value. Most gauges are made in the static mode of the measured value, hence fh→0 and the flicker-component becomes of importance in the spectrum. Hereby, the error of measurement could not be

reduced to infinitesimal value by the method of averaging the results of measurement.

Fig. 4. Random error dependence on the number of gauges at fh=10-6 Hz and Δt=1 s for different types of noise.

5.1.3. Duration of Noise Signal Gauging Duration of noise signal gauge is stipulated by

random nature of the measured voltage or current noise signal. Each one has a nature of homogeneous continuous random fluctuations concerning the ave-rage that is up to zero and constitutes a random ergodic stationary process. While studying it, any moment of time can serve as starting point. Measuring the parameters of stationary process within any period of time, we should receive the same values of characteristics.

Such integral characteristics of random process as the mean of a square (variance in statistical inves-tigations), root-mean-square (standard deviation) and spectral density of noise signal tend to be measured firstly. Since the noise signal is a random process, its true value could be gained during the infinite time of averaging. Any restriction on the averaging time leads to the appearance of a methodic error. In the ideal case, if there are no other noise signals excepting the measured one in the measuring circuit, the standard deviation of noise signal variance σ could be

calculated as: 1

t f

, here t is the time of measu-

rement, Δf is the bandwidth of noise signal. Results of modeling the dependence of standard deviation of noise signal variance on the time of averaging at the different values of bandwidth Δf are notified in Fig. 5.

To reach the relative root-mean-square value of the variance of noise signal 0.01 % for the bandwidth Δf = 100 kHz, we should conduct measurements for 1000 s., and for Δf = 1 MHz – 100 s.

Taking into account that other sources of noise signals (resistance of a connecting line, amplifiers, and feedback resistors) are present in the input circuit, the dependence of root-mean-square of noise signal vari-ance on time of averaging becomes more complicated. Time of measurement for reaching the equal error rises


15

as compared to an ideal case. Correspondently, measu-rement of integral characteristics of noise signals could take a lot of time for averaging – up to tenth – hundredth of seconds.

Fig. 5. Dependence of a standard deviation of a noise signal variance on the time of averaging.

If there is a necessity for measuring the integral characteristics within narrow bandwidth, the time of measurement rises considerably. So, to reach the relative root-mean-square of noise signal variance 0.1 % at the bandwidth f = 10 Hz, measurement should last approximately 30 hours.

5.2. Accuracy of Raman Thermometry Raman thermometry is elaborated insufficiently

due to the novelty and uniqueness of method. This problem is considered below basing on error, uncerta-inty and other approaches of metrology.

Within error approach, Raman thermometry consi-ders few components of the combined error of tempe-rature measurement. The first one is instrumental and could be determined by the accuracy of measuring device. The second one is methodical and is caused by heating the object during the process of measurement. There exists the third component that is also of instrumental type and is related to the changes in feeding parameters during the measurement. There is also the forth component caused by the instability of a surface and adjacent layers as result of light beam effect (close to drift error).

Instrumental error with systematic and random components is caused by the fluctuation of number and frequency of scattered, especially anti-Stocks quants. To reduce this error, is necessary to perform the signal time-averaging. At light exciting within Raman method with lasers of different wavelengths these effects are expressing themselves in various ways. Therefore the instrumental errors are distin-guished at different wavelengths. While using two lasers different wavelengths we gain two diverse instrumental errors with different results dispersion.

Errors of photocurrent measurement depend on metrological characteristics both of laser and spectro-meter. Moreover, their stability is quite important: there exist instabilities caused by the errors of setting

and determining the certain value of irradiation. Fortunately at serial measurements of anti-Stocks and Stocks signals the given error components compensate each other. Under the condition that photo detector sensitivity at Stocks and anti-Stocks frequencies slightly differs we could adopt that isias. By neglecting the slight deviation of SL 03/1 laser frequency with the wavelength 632.9910 ± 0.0002 nМ, we could advance that the instrumental error is defined as:

0

0

2 2

ln 3ln ias

s i

TiZ

i

, (2)

where ist and ias are the intensities of Stocks and anti-Stocks components of scattered radiation respectively, ν0 and νi are the wavenumbers of reflected phonons at the given number of dispersed phonons and used laser bunch respectively.

Methodic error related to heating the researched object by laser irradiation depends on its surface energetic luminosity. Even the effect of under-powered laser leads to surface heating and arising methodic error (~27 К), which could reach much larger values for small objects. We should indicate that methodic error as well as instrumental one encloses the cognizable and incognizable components (Fig. 6).

The drift is caused by the changes both in the chemical composition and surface shape of measured object as consequence of intensive irradiation. It is related to complicated transform processes in surface-adjacent object layers and is attributed to the systematic component of instrumental error. Apart from it, the latter is treated as partly cognizable due to its different influence factors. Their estimation is carried out involving the thermodynamics of irrever-sible processes: the more intensive irradiation and the larger methodical error are, the stronger entropy changes and the larger drift occur. To reduce this error, the known in metrology method of nearing to measuring point from both sides could be successfully applied.

Fig. 6. Optimization of methodic and instrumental components in decreasing the combined error of temperature measurement at laser power 2 mW.


16

Adopting that under the condition of linear alteration in time the calculated frequency of reflected anti-Stocks component becomes such a point. So in order to get rid of drift, becomes necessary to measure the Stocks frequencies before and after measuring the anti-Stocks frequency, and then to average the Stocks frequencies results.

Within uncertainty approach in Raman thermo-metry, the measurement of chips temperature in the process of their manufacturing has performed only once, since the next measurements are usually realized in other conditions. For one-time measuring an error approach is not quite adequate.

The estimating concept for measurement results could be based on the uncertainty approach. Here the methodic error is being calculated and evaluated as a physical value, which certain components and coeffi-cients could be estimated with a certain errors. To wit, this component of the combined error is considered with particular uncertainty. We take into consideration peculiarities of both MIs and standard patterns. For in-stance, with help of Raman method the measurement of CNTs temperature within the range 30…250 ºС is made. These tubes are treated as standard nanopatterns for testing and calibrating the nanotechnological means. Hereby, to study the action of seven and more possible influence factors (angle of light bunch incidence, distance to a photo-receiver, exposure time, duration of spectrum passing, power and mode of laser operation, drift characteristics and so on), 28000 gauges have been performed, enabling us to ascertain the following indices of the measurement accuracy.

Approach of errors is applied to results processing, consequently of which one of the gained results (with introduced correction to systematic error component) looks as Тreal=287.27 К±1.72 К (±0.6 %). At the same time, due to uncertainty approach, the gained result makes Тreal =287.27 К with expanded error 0.58 % and combined standard uncertainty 0.3 % at the credence level Р=0.95, expanded coefficient 1.96 and efficient value of freedom degrees 130.6.

5.3. Temperature Dependent Accuracy Threshold

Accuracy threshold is due to the fluctuation devi-ations in the processes which determine the metro-logical characteristics of the extra-sensitive MIs. Due to such great sensitivity, the accuracy threshold becomes temperature dependent since fluctuations intensity depends on temperature.

5.3.1. Accuracy Threshold of Sensitive Balance

Main constructive unit of torsion balance is a thin thread on which a light mirror hangs. It should be noticed the same part is the basis for a ballistic gal-vanometer construction. Molecules thermal motion of the environment leads to irregular in time molecules bombarding of the mirror that limits instrument sensi-tivity and not let to better the measurement accuracy.

Thread torsion module is 2 2

8

d Ga

l

, where G is the

shear modulus; d and l are the thread diameter and length. Then moment of force that effects on the thread is linked with rotation angle by the next ratio: M=a, and the potential energy of curled thread is

2

2aU . In accordance with Boltzmann formula,

the dispersion of value of angle close to which the mirror vibrates is equal to:

2

2

2 2

2

2

[ ]

aT

aT

e dT

D Da

e d

. Obviously the root-

mean-square deviation of this angle is equal to: 1

2

[ ]T

a

. At room temperature when а ~ 10-13 J,

the mirror rotation angle root-mean-square deviation is determined as ~10-4 radian. This is a real limit of single measurement sensitivity for practically all MIs in nanometrology.

By the same way the fluctuations impact on metrological parameters of a spring balance with co-efficient of elasticity k and equilibrium stretching Х0 is considered. Mass center oscillations occur in it as result of the temperature fluctuations presence. That’s why counting of equilibrium position of the pointer Х0 cannot be made more accurate than with the root-mean-square deviation of absolute value of the instrumental error random component:

20 0[ ] ( )

TX X

c , where с is the constant

that links mechanical qualities and sizes. On this basis let determine the root-mean-square deviations of absolute and relative value of instrumental error ran-dom component of mass determination:

0[ ]c k T

m Xg g c

,

[ ] 1[ ]

m Tm

m mg c

(3)

Hence the instrumental error random component is

smaller as the spring is weaker. However in this case the equilibrium stretching increases: Х0 = mg/k. It specifies practical inconvenience of balance construc-tion. Hereby temperature dependent fluctuations limit the metrological characteristics of balance.

5.3.2. Accuracy Threshold of Sensitive Ballistic Galvanometer

In electrical measurements, fluctuations specify

the absolute error independent of MI perfection state. So far as the ballistic galvanometer is used as supersensitive mean for small values of impulse current measurement, it is considered in details. The


17

galvanometer current I is measured by the mirror devi-ation angle φ. In equilibrium state when spring forces moment cφ is equal to electromagnetic forces effect

moment I, the mirror rotation angle is 0

I

c

( here

c, γ are the constants). Root-mean-square deviation estimation of the

mirror rotation angle is in line by its content with the similar estimation of the instrumental error random component of spring balance (see p.5.3.1). In such a way it was derived the root-mean-square deviations of absolute and relative values of the instrumental error random component of the current determination:

[ ]c T

Ic

, [ ][ ]

I cTI

I I

(4)

Hence for the galvanometer accuracy improve-

ment it needs to take smaller value of constant c and higher value of constant (otherwise to increase the number of winds in the galvanometer current coil). This leads to the equilibrium angles φ0 deviation that contradicts with springiness demands of the hanging thread deformation φ0<<π/2. Therefore temperature fluctuations limit the galvanometer accuracy: A=1/. Then the accuracy limit which can be gained in measu-rements is determined by assigned in advance sensitivity.

5.3.3. Dynamic Error It is considered (on example of thermotransducer)

as an error caused by heat inertia of transducer and inertia of measuring device, and is equal to the difference between transducer error in variable temperature mode and its static error.

Dynamic error emerges due to transducer not has enough time to follow the rapid temperature changes of controlled object. Exactly this delay characterizes thermal inertia index. If to consider that temperature in the cross area of transducer is uniform and heat removal and radiation exchange are absent, it becomes possible to submit the nature of change in temperature on the basis of elementary theory of thermal inertia for uniform transducer by expression:

0

1 Tdyn T

dTT T T

m d

, (5)

where TT() is the temperature of sensitive element of transducer, T0() is the object temperature, is the time, m is the parameter which characterizes the rate of heat exchange due to convection [9]. 5.3.4. Instrumental Error

- Instrumental error of noise thermometer is a

component of measurement error due to its intrinsic

properties. It may contain few components, including error of measurement and the error caused by the interaction of transducer with the object of measure-ment. For example, 100 (at 27.15 K) sensitive elements of noise thermometers were made from pure Ni, Pt, Cu; alloys (Ni-Cr and composites of various oxides). Research has been performed in reference points of ITS (4.2 K; 77 K…273.15 K) according to IMECO method, and at higher temperatures. Revealed deviations from linearity of calibration characteristics as the relative error δT increase from Cu (0.05 %) to Ni (0.26 %) sensitive elements. Mentioned deviations are not fixed for elements made from transition metal alloys and composites.

Analysis of measurement error of noise thermo-meter has shown the additional component existence that goes beyond a basic acceptable error. This error is caused by structural processes in thermometric ele-ment due to its manufacturing (bending, tension). The constancy of research temperature – 77 K – does not mean the thermodynamic equilibrium state of sensitive element and environment. The relaxation of nonequilibrium thermodynamic state depends on several factors (temperature, time, type and concen-tration of defects). Otherwise, condition of thermo-dynamic equilibrium, at which Nyquist formula is derived, has been broken in the case of a real noise thermometer wound at 300 K and used at low temperatures.

Temperature dependence of electrical noise power is derived directly from the basic equation of thermodynamics. In the stationary nonequilibrium state, the thermometer calibration characteristic non-linearity appears due to violation of energetic pro-cesses of environment – thermometric substance exchange. It is expressed by the instrumental (absolute ΔT and relative δT) error as:

( )c r c r elT T T b b P , ,c r

r r

b b bT

b b

(6)

where Tc and Tr are the estimated and real temperature, respectively; bc and br are the constants of estimated and actual calibration characteristics. The prolonged use of a thermometer leads to maximizing the constant b~(dS/dt)-1 at minimizing the entropy dissipative flow:

minp

dS S

dt

. Nonstationary nonequilibrium

thermodynamic state corresponds to the power change of nonequilibrium electrical noise in the elastic-plastic deformed thermometric substance.

Therefore its relaxation effects lead to error emer-gence. Substance of density ρ rapidly releases the previously accumulated elastic energy with appearing microcracks of length 2 l. Thus, relaxation constant 1

is estimated as 21

2~ El , where 2

2E is the

density of elastic energy. The latter can be transformed into surface microcracks energy with its relaxation


18

constant 2: 2 ~ ll , where is the specific energy

of surface tension; or removed from the relaxation place with constant 3, linked to thermal diffusivity a:

2

3 ~ la . At temperatures lower than 20-30 K, thermal

diffusivity in 100 and more times is higher than at 300 K. So 3 is much smaller in comparison with the relaxation constant of motion 4 ( 4 ~ ) or

reproduction 5 of dislocations (3

12

5 ~ ( )d d

l

nL E

). Here

is the effective length of dislocation run; ω is the dislocation velocity; Ld is the typical size; Ed is the dislocation energy, referred to one interatomic distance.

Effect of the mentioned above constants 1…5 is combined, and depends on the temperature and back-ground of substance, forming the total relaxation

constant .

1

11n st n

i i

. Consideration of competitive

effects of constant 2 due to microcrack formation, and constant 3 due to heat removal from the place of energy relaxation produces the modified constant

2 3

2 3st

. The joint effect of these two mechanisms

creates the reasons for changes in the electrical noise power and thus changes in the readings of noise thermometer. Hence, the error of thermometer, whose sensitive element is in stationary nonequilibrium state, is determined by the competitive action of two major in those conditions dissipation processes that form deviation from the calibration characteristics:

122

1( ) cT A ad , where C is the sound velocity;

a is the grain size; d is the atomic size, χ is the thermal diffusivity. Hence, the lower speed of sound and higher thermal diffusivity, the more efficient work mechanism for a heat removal and the less noticeable influence of dislocations on the electrical noise power and consequently on a noise thermometer error.

Described before concerns pure metals and is not related to alloys and composites due to significance of the process of dislocation multiplication (constant 5) that occurs in their blades and is accompanied with the microcrack formation. Finally, high temperatures up to a melting point are matched with diffusion removal at the relaxation constant 4. That is, in the high-temperature case, one should consider the competitive action of two relaxation mechanisms: diffusion mechanism and formation of microcracks in the deformed local substance microvolumes. Introduced before criterion is varied at the high temperatures to:

124

1~ ( ) cad D

. Here coefficient of diffusion D

which increases exponentially with temperature means deviations absence of calibration characteristics at high temperatures.

- Instrumental error of thermoelectric thermotransducer. Its study has been completed by

elaboration of algorithmic principles of thermotrans-ducer error minimization realized on basis of thermo-dynamic forces and fluxes consideration in sensitive substance (Fig. 7). Consequent evaluation of certain influence functions due to complicated transfer pro-cesses described by corresponding freedom degrees in basic equation of thermodynamics is realized. Preliminary algorithm settings comprise the values of:

1) Transformation function and its dispersion; 2) Temperature range, environment, exploitation

time and mode; 3) Peculiarities of sensor substances and

thermotransducers manufacturing. Moving along the chart, from magnetic freedom

degree, alternately estimate the effects of influence functions of all possible degrees on the transformation function.

Fig. 7. Transformation and influence functions in thermo-transducers: thermoelectric ΔU; resistive Δσ; noise ΔΡ.

6. Methods of Correction and Statistical Minimization of Errors

The mentioned methods are directed at reducing of already existing errors. Adjustment (or functional minimization) is considered to be the method that reduces errors, mainly systematic ones, by means of analytical or experimental study. Under statistical minimization we understand reducing the expected but not identified measurement errors; it is carried out both during and after the measurement (error reduction by spatial or temporal averaging). Examples include: reducing the random errors of the multiple measurement results by time or spatial averaging; reduction of quantization error [2].

For MIs calibration the direct measurement by verified MI of outgoing signal or by multivalued measure with determination of the error as a difference usually apply. Correction methods of systematic error


19

constituent are realized by operator or automatically in off-line mode when, f.i., self-calibration is carried out.

Errors adjustment with the operator participation can be fulfilled in 2 ways. The first one is the calib-ration of MI (Fig. 8).

Fig. 8. Measuring instrument error adjustment by calibration [5].

To correct the dominant error additive and multiplicative constituents the instrument calibration is usually performed in two points of scale: at shorted input and at supplying the measure’s output signal to input of verified MI. At shorted input, the operator sets the zero readouts of the mentioned device, then at connected measure with help of calibration unit sets the readout that corresponds to submitted value of measure.

Calibration without action on the measuring instrument is performed by introducing amendments, or during the measurement results processing. For example to correct error additive constituent, two measurements are performed; two indications are recorded: 1 0y kx and 2 0y , and the measu-

rement result is computed as 1 2y y y kx .

Automatic correction aims the amendments introduction into device structure or into measurement algorithm.

There exist some specific measurements which include the next methods of error adjustments.

Auxiliary measuring method is the version of invariance principle, according to which are needed as many additional channels measuring as the impact values exist. Iterative method consists in the multiple specifying of adjustment results performed by successive approximations. Therefore it requires the precise feedback transducer. Method of standards establishes the real conversion performance by con-necting a set of standards (or one multi-level standard) to input of MI.

When to turn off the physical quantity from device input or realize its set of measures is impossible, the test methods are applicable. The latest generate the test values involving both measured and model quantities.

6.1. Reducing the Methodical Errors Due to complexity of setting the correct expe-

riment, inevitably arise methodic errors caused by

inadequate of considered method to real conditions of measurement. They can include an incorrect transmis-sion function and mismatching the characteristics of different measuring instruments. To correct metho-dical errors, detailed study of conditions and nature of these instruments should be performed. To reduce some methodical errors, the special measurement methods have been developed: method of substitution, method of error compensation by sign, method of contradistinction, method of symmetrical observation, etc.

Method of substitution consists in submitting the initially measured value to input of measuring instru-ment. Subsequently this value is replaced by the appropriate measure of known value at which the readout of instrument remains unchangeable.

Method of error compensation by sign consists in double measuring of the same value at variable measuring conditions in a way that unchangeable systematic error would be included in the measure-ment result with the opposite sign.

Method of contradistinction consists in the double gauging the measured value; firstly it is compared with the value that is reproduced by measure; and before the second comparison these two values are mutually changed in measuring circuit. So, result of measurement becomes independent of the transfer factor of measuring circuit.

Method of symmetrical observation is as follows. First value X is measured, then after a time Δt full or partial substitution with measure of known value XM is performed; over the time interval Δt measurements is repeated again. This excludes the permanent and linearly-dependent systematic error constituents.

6.2. Eliminating the Systematic Errors

During analysis of adjusting methods, the absolute value of the combined error of MI is conveniently to divide in three components:

– Additive component a, independent of Х; it is also named “zero error” (it occurs if MI registers the certain readout when the latter should be zero) and causes concurrent shift of the MI characteristics; this kind of error can be easily detected at Х = 0;

– Multiplicative component м = s Х, proportional to Х. It is known as “sensitivity error” that causes the MI specifications rotation concerning the zero of coordinates; this kind of error be easily detected while applying the measure or scale transducer;

– Error non-lin of non-linearity of MI charac-teristics that non-lineary depends on Х and may be efficiently detected while applying the multi-level measure or scale transducer in measuring circuit.

6.2.1. Common Methods

Amendments. Action of systematic and other re-gular (e.g., linear-increasing over time error or drift) influences on the received result is reduced by using appropriate types of result adjustment or by


20

introducing the amendments. For this, a variety of methods is developed; most of them are based on per-formance of additional measurements with applying the so-called standard values, i.e. quantities with certi-fied magnitude. Adjusted result (xam) is obtained by adding to the measured value (x) amendment p which is equal to the corresponding systematic error component with opposite sign: xam = x + p.

Introduction of amendments is compulsory for each stage of results processing. However, it is virtu-ally impossible to fully explore the effects of syste-matic or regular impacts, at least due to the absence of ideal MIs, presence of random influences or time restrictions. Therefore complete correction of syste-matic effects is impossible. Error can be reduced by adjusting the results only if the relationship between the impact factor and output value is known.

Cold-junction compensator is a brief example of such device. It carries out the compensation of cold-junction temperature of thermocouple, or adjusts its shifted readouts. Electronic means can also compen-sate the similar errors for thermocouples of various types, and so reach the improvement of accuracy. Also, bridge scheme is designed so that, when chan-ging ambient temperature and therefore cold-junction temperature, it could provide adding the voltage pro-portional to mentioned temperature to thermo-EMF.

Processing the Measurement Results. When Y = kX + Yа, the additive error Yа can be excluded by performing one additional observation at X = 0, and the following subtraction. If the additive error exists at Х = Х1, the output value of device is equal to YD(X1) = kX1 + Yа. Then at X = 0: Y0=Yа. After subtraction we get adjusted value of measurand:

1 1

1 1

0

0

adj П

a

Y X Y Х Y

kX Y Y kX

(7)

Multiplicative error m is excluded via single-

channel fixed measure by calibrating the MI at the given value Х0 and subsequent dividing and

multiplying: 1 2 0;Y kX Y kX , whereof 10

2

YX X

Y .

If additive and multiplicative constituents of error in MI readings exist, they are also excluded via similar measure of fixed value by means of two additional measurements at X equal to 0; X0. So, correct result

corY is defined by subsequent computing:

0 2 0 1 0adjY X Y Y Y Y (8)

In the case of nonlinear transfer function (1 )n s aY k X X Y of MI, the problem arises

of selecting the optimal calibration value XCal. Firstly, you must reach the readout of MI at zero (Y = 0) for X = 0. During instrument calibration its transfer function is approximated by a linear dependence:

(1 ) (1 ) ,k n k k n s k kY k X k X X (9)

where a is the adjusted relative error at kX X . Let

us divide at n kk X both sides of the equality and bring

relative error of MI to input k s kX . Its

absolute error would be X (Fig. 9):

(1 ) (1 )

( ).

n s n k

n n

k

k X X k ХYХ

k k

Х Х Х

(10)

Fig. 9. Absolute error of MI reduced to input [5].

Minimum and maximum errors are respectively defined at the top of the parabola ( / 2kX X ) and at

the end of measuring range mX X :

min

max

2 2

( )

k kk

m m k

X XX X

X X X X

(11)

The condition of minimizing error over the entire

range of MI is min max ,X X and at k mX X we

obtain quadratic equation 2 + 4 - 4 = 0, the physical

meaning of which has just positive root: =2( 2 -1)=0,82. So, for a quadratic approximation of transfer function of MI its calibration should be performed at the point 0,82 .к mX X

Calibration of Measuring Instruments is performed by changing its sensitivity, or by altering the tilt of its characteristic. It is especially effective at predominance of error multiplicative component and can be implemented by operator for circuits with measures, with working standard MI, or with model reverse converter and calculating unit.

If there is multiplicative error component м, the equation for MI readout is presented as

1 0 0 0(1 )m mk X k X k X k X , here 0k is

nominal transfer factor. While calibrating (at applied to input of such device the measured value Х0), the operator changes factor k of device until at Х=Х0 readout becomes equal to 0. The last is usually marked on the scale with red tag, or corresponds to end value of scale. As a result of the calibration, coefficient becomes equal to 0 0 0/k X .

Calibrating the MI with help of the working standard is mostly performed at X-value close to Х0. During calibration, the coefficient is changed as long till readout of calibrated MI do not match the readout


21

of standard which (with model reverse converter and computing chip) is especially suitable for calibration of measuring transducers. While their calibrating, we change coefficient k till the error р in output of reader would be established at the zero. In this case,

0 0k

kXYX X k k .

6.2.2. Special Methods Special methods for improving the accuracy are:

by-sign-errors compensation; method of contradistin-ction; method of symmetrical observations; method of substitution [5].

- Method of Errors Compensation by Sign. Two measurements of the same value are fulfilled with such changeable measurement conditions, but in the second time, the unchangeable systematic error of measurement has to be included in result with opposite sign. So, in the measurement of voltage the result is received by two cycles (for identical and opposite polarities of additionally installed switches). Then if for measuring time the parasitic EMFs are immutable, the result becomes independent to their values.

- Method of Contradistinction. Measured value X is compared twice with adjustable measure Хк, and these two values (Х and Хк) are swapped before the second measurement. For instance, classic metrological task is the definition of mass in inaccu-rate balance. Here the result of the measurement

1 2x N NM M M is derived, considering the

system of two equations (Fig. 10):

1 1 2

2 2 1

x N

x N

M L M L

M L M L

(12)

L1 L2 L1 L2

Mx MN MxMN

Fig. 10. Improving the accuracy of masses definition with help of inaccurate balance.

- Recent example of method application in stepless Z-shift regulation of nanomashines. During the creation of hydraulic positioning unit along the axis Z of nanomachine with providing stepless shift and position control possibility along this axis, it is suggested to use hydraulic potential. To this purpose the U-shaped hydraulic construction is proposed with ends of large (D = 20 mm) and small (d = 0.3 mm)

diameter. Diameters ratio is equal to: 66.66D

d .

Spontaneous or enforced liquid level shift, for instance, in the small diameter tube, is detected by means of micrometer head (Fig. 11).

Fig. 11. The unit of nanosized objects hydraulic positioning: 1- shift micrometer head; 2 – cantilever;

3 – float with mounted research nanoobject.

In the motionless fixed tube its pivot sinks by the shift micrometer head pressing hydraulic liquid to the wide cylinder. Consequently the liquid level is increased in this end – on ∆h1. It results to weak but appreciable liquid level increase in the wide end. In accordance with joined vessels law the level increases on ∆H1 in the wide end. Floating plate mounted in this wide end lifts the same as the studied nanoobject mounted on it. In this case the condition of liquid quantity invariability under pouring from one end to

another one can be described by: 2 2

4 4D H d h

,

we use 2

24444.4

DA

d as constant for this unit

construction. So, we have got the formula specifying the level drops changes in wide (∆H1) and narrow (∆h1) device ends, as 1 1H h A . With respect to

the error approach [6] the relative errors of liquid level change in both device ends are linked between each other by: 1 1( ) ( )H h A , where 1( )h is the

relative error specified by the inaccuracy of the drop level measurement by the micrometer head; A is the relative error specified by the inaccuracy of value A.

The first error component is determined by the following way. So far as liquid level measurement drop in the narrow end is 5 μm, then the device measurement step is determined in the wide end is 5.0 μm /4444.4 = 1.1 nm. Micrometer head absolute error in accordance with passport is ± 2.5 μm. Its value included to the result of the level shift in wide end is ± 0.55 nm. The step measurement refinement result is equal to 1.10±0.55 nm. In the case of liquid level shift gauges in the narrow end with error ± 1.0 μm, it seems possible to reach the relative measurement error of hydraulic shift ± 20 %.

Hereinafter the unknown second component of the MI error – the relative error of constant A value determination – is considered below.

Metrological experience of error systematic component minimization. Method of contradistinction can be used for accuracy improving by multiplicative error component minimizing. Its peculiarity consists in that the measured quantity XH is compared twice


22

with regulated measure - 1 2;N NH H (before the 2nd

measurement it is rearranged with the measure). Consequently this quantity value with the eliminated measurement error multiplicative component can be

gained: 1 2X N NH H H .

To realize the multiplicative error component minimization we perform the hydraulic device calibration (Fig. 12).

Fig. 12. Calibration unit for hydraulic positioning: 1 – micrometer head; 2 – linear scale of the liquid

level shift.

This operation is fundamentally opposite to direct measuring operation. Thereby, device enables to set the liquid level, and nanoobject can be mounted on the floating platform with absolute error which is slightly more than atoms size.

- Method of Symmetrical Observations is applied for correcting the additive and progressive (linear-variable over time) components of errors. Three measurements are usually carried out at regular time intervals t:

1

2 1

3 1

,

,

2 .

a

k a

a

Y X Y

Y X Y t

Y X Y t

(13)

Errors of measurement results are determined by:

1 3 1

2 3 1

2 ;

2,a k

Y Y t

Y Y X Y Y

(14)

where aY is the additive error component of MI; 1is its rate of error change; Xk is the value of measure. The adjusted value of obtained result is determined from the 1st expression of mentioned system:

1 1 2 3 1 2a kX Y Y Y Y X Y Y (15)

- Method of Substitution is applied when the experimenter does not have a complete set of MIs to eliminate the errors that arise in them. Let the

transformation equation of MI is ( )Y f X and let’s consider two basic varieties: its replacement with variable measure and with adjustable scale converter.

The first one is used in the exact measurements. Method is implemented in two stages. At the first stage, signal X is fed to input, and output signal Y1 is fed to a memory element. At the second stage, from regulated measure’s output the signal of a variable

value NX is submitted; it changes as long as signal Y2 does not become equal to Y1. When using method of substitution, the additive and multiplicative errors of MI do not bring contribution in result [1, 6]. Request is imposed to factor k that consists only in temporary stability, since permanence of k must be provided within a small interval of time which is equal to expectancy of 2-stage measurement. Method with adjustable scale converter is realized on the basis of set of elementary means. It can be recommended if the unambiguous non-adjustable measure Х0, adjustable scale converter for value X are available, and comparison unit fits only for value kHX.

Method of substitution is widespread in the bridge circuit measurements, where firstly the resistance Rx is measured by bridge circuit; then it is substituted by multi-level measure RN. Under the theory of bridge circuits, the error of resistance measuring by method of substitution δx equals to δN (error of measure) at its full replacement by measure: RX = RN.

This method is considered quite valuable in metro-logy, especially during at the measures calibration. As example, the method of Ohm size transmission from State standard to 1 Ohm, 10 Ohm and 100 Ohm secondary measures is realized in 8 stages with help of ratios measure containing ten 10-Ohm resistors; the latter can be connected in parallel (1 Ohm), in series (100 Ohm) and in series-parallel (10 Ohms).

Moreover, importance of above method we can underline with next linked option, namely with implementation of exact measure of electrical resistance on the basis of conductance quantum in CPS self-checking operation. Such a standard is able to replace older one in the modern State standardization practice.

Consumers of metrological services of the State Institutes of Metrology and Standardization, who are in great interest in transfer of proposed Resistance unit to CPS working standard, aim the subsequent accuracy improvement of CPS’s products. We have considered earlier that the appropriate prototype of resistance measure (12906 Ohms) could be applied for calibrating the MIs of high accuracy.

Elsewise, we have obtained the reference point of Ohmmeter scale important for its calibrating in the high accuracy class. In this way, it can be realized the self-check, self-calibration of MIs and therefore self-validation of gauging data. Advantage of the similar methods of metrological self-check is evident; it was demonstrated in [10] on examples of checking the temperature, pressure and other kinds of smart sensors. By continuous controlling the reliability of metrological data and basing on the self-checking results for previous time duration, the forecasting of device’s metrological state is developed as well as CPS’s state.


23

7. Conclusions

1. Basing on the current metrological experience of complex technical objects we suggest the improve-ment of efficiency of Cyber-Physical Systems at equipping them with enhanced metrological subsystems. The latest have to provide exact measuring the performance including the control of actuators which actually ensure together with sensors the necessary mode of CPS operation.

2. Qualitative metrological instruments, their efficient metrological supervision and ensurance enable us to enhance slightly the accuracy of metrological subsystems. However to improve the CPS accuracy and to raise finally the quality of manufactured products by some orders, providing in-place the man-out-loop metrological procedures such as self-checking, self-validation, self-adjustment and so on becomes crucial.

3. Distinctive feature of such procedures could be introducing the set of special methods of minimizing the different kinds of random and systematic errors, for instance, through the introduction the methods of contradistinction, of substitution or others.

Acknowledgement

The scientific results, presented in this article, were obtained in the frame of research project number 0115U000446, 01.01.2015-31.12.2017, financially supported by the Ministry of Education and Science of Ukraine.

References [1]. NIST, National Technical Information Service,

FISCAL YEAR 2014, Budget Submission to Congress, 2014.

[2]. New Reports Define Strategic Vision, Propose R&D Priorities for Future Cyber-Physical Systems, From NIST Tech Beat, February 6, 2013.

[3]. S. Yurish, Sensors: Smart vs. Intelligent, Sensors and Transducers, Vol. 114, Issue 3, March 2010, pp. I-VI.

[4]. Yu. Yatsuk, M. Mykyychuk, V. Zdeb, R. Yanovych, Metrological array of Cyber-Physical Systems. Part 11. Remote Error Correction of Measuring Channel, Sensors and Transducers, Vol. 192, Issue 9, September 2015, pp. 22-29.

[5]. V. Yatsuk, P. Malachivski, Methods of improving the measurement accuracy, Beskyd-Bit, L’viv, 2008 (in Ukrainian).

[6]. ICGM 104: 2009, Evaluation of measurement data. An introduction to the “Guide to the expression of uncertainty in measurement” and related documents.

[7]. K. Ranev, Hybrid model of result processing, in Proceedings of the International Conference on Metrology, Minsk, Belarus, 2009, pp. 24-31.

[8]. S. Yatsyshyn, B. Stadnyk, Accuracy and metrological reliability enhancing of thermoelectric transducers, Sensors and Transducers, Vol. 123, Issue 12, December 2010, pp. 69-75.

[9]. S. Yatsyshyn, B. Stadnyk, Ya. Lutsyk, L. Buniak, Handbook of Thermometry and Nnothermometry, IFSA Publishing, Barcelona, Spain, 2015.

[10]. R. Taymanov, K. Sapozhnikova, I. Druzhinin, Sensor devices with metrological self-check, Sensors and Transducers, Vol. 10, Special Issue, February 2012, pp. 30-45.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Handbook_of_Measurements.htm





24



Coherent Anti-Stokes and Coherent Stokes in Raman Scattering by Superconducting Nanowire Single-Photon

Detector for Temperature Measurement

1 Annepu Venkata Naga Vamsi and 2 Annepu Bhujanga Rao 1 Dept. of Electronics and Instrumentation Engineering, Gitam University,

Visakhapatnam, India 2 Dept. of Instrument Technology, Andhra University, Visakhapatnam, India

1 Tel.: +91-9550417485 E-mail: [email protected]

Received: 20 November 2015 /Accepted: 30 December 2015 /Published: 31 January 2016 Abstract: We have reported the measurement of temperature by using coherent anti-Stroke and coherent Stroke Raman scattering using superconducting nano wire single-photon detector. The measured temperatures by both methods (Coherent Anti-Raman scattering & Coherent Stroke Raman scattering and TC 340) are in good accuracy of ± 5 K temperature range. The length of the pipe line under test can be increased by increasing the power of the pump laser. This methodology can be widely used to measure temperatures at instantaneous positions in test pipe line or the entire temperature of the pipe line under test. Copyright © 2016 IFSA Publishing, S. L. Keywords: Coherent anti-Stroke Raman scattering, Coherent Rayleigh-Brillouin scattering, Superconducting nanowire single photon detector, Quantum mechanics, Cryostat. 1. Introduction

In the process of light scattering, the most critical factor is the length scale of any or all of these structural features relative to the wavelength of the light being scattered. In all the light-scattering processes treated so far, Pan X., et al. has reported the utilization of the coherent Rayleigh scattering (CRS) to measure the temperature of the low density gases and weakly ionized plasma [1]. Later Graul J., et al. has reported the measurement of the temperature in the molecular gas by utilizing the coherent Rayleigh-Brillouin scattering method [2]. Gu Z. has reported the spectrometer for the measurement of spontaneous Rayleigh-Brillouin (RB) scattering line profiles at ultraviolet wavelengths from gas phase molecules [3]. In the recent work by Yu Zhang., et al. has reported

the coherent anti-Strokes Raman scattering with single molecule sensitivity using a plasmonic Fano resonance [4]. Zumbush A., et al. has reported three- dimensional vibrational imaging by coherent anti-Stroke Raman scattering [5]. Mathew P. Thariayn, et al. has reported the dual pump coherent anti-Strokes Raman scattering system for temperature and species measurement in an optically accessible high pressure gas turbine combustor facility [6]. Russel Lockett, et al. has come up with the similar ay of temperature measurement in an internal combustion engine by coherent anti-Stroke Raman spectroscopy [7].

This has motivated to work on the coherent stroke and coherent anti-Stroke Raman scattering for measurement of the temperature by utilizing the superconducting nanowire single photon detector in a single mode fiber.




25

In this paper we have worked on the single mode fibers by considering the incident radiation as one monochromatic wave of frequency ω1. We now consider the experimental situation illustrated in Fig. 1 where the incident radiation consists of two overlapping coherent monochromatic beams of frequencies ω1 and ω2, with ω1 > ω2. As the overlapping beams of radiation propagate through the material system, new radiation is produced with frequencies corresponding to various combinations of ω1 and ω2. From amongst the possible combinations we first consider the combination 2ω1 - ω2. If we vary ω2 while keeping ω1 constant we find that the intensity of the scattering increases dramatically when ω1-ω2=ωM, where ωM is a molecular frequency that can be observed in Raman scattering. When this frequency-matching condition is satisfied ωs=ω1+ωM, because ωs=2ω1-ω2=ω1+(ω1-ω2)=ω1+ωM.

Fig. 1. Diagrammatic representations of CARS (coherent anti-Stroke Raman scattering), CSRS (coherent stroke

Raman scattering).

The condition ω1 - ω2 =ωM can be regarded as a Raman resonance. This is quite different from the electronic resonances described earlier. The scattered frequency ω1 + ωM has the form of an anti-Stokes Raman frequency relative to ω1. As this scattered radiation is coherent, it is called Coherent anti-Stokes Raman Scattering, or CARS. By varying ω2 over a range of values that covers the desired values of ωM a CARS spectrum can be obtained. The CARS frequencies will be superimposed on a background of weak non-resonant scattering given by 2ω1 - ω2. If alternatively we consider the scattered frequency combination 2ω2 - ω1, then when ω1 - ω2 = ωM strong scattering now occurs at ω2 + (ω2 - ω1) = ω2 - ωM. This is Stokes radiation relative to ω2 and is called Coherent Stokes Raman Scattering or CSRS.

CARS and CSRS differ in many important respects from the Raman processes. They produce highly directional beams of scattered radiation with small divergences. The scattered intensity is proportional (a) to the square of the number of scattering molecules and (b) to the square of the irradiance of the incident radiation at ω1 and to the irradiance of the incident radiation at ω2. Consideration of the interaction of the waves of frequencies ω1 and ω2 involves the bulk or macroscopic properties of the material system which must then be related to the individual or microscopic properties of the molecules. From such considerations

emerge the special properties of CARS and CSRS radiation which we have just outlined energy change in the material system and the process is said to be passive or parametric. The material system acts as a facilitating agent as it were for the exchange of energy between radiation of different frequencies and this is very effective when ω1 - ω2 = ωM. Energy level diagrams for CARS and CSRS and given in Fig. 2 (a) and (b) respectively.

(a) CARS

(b) CSRS

Fig. 2. (a) Energy level diagrams for CARS,

b) Energy level diagrams for CSRS.

2. Methodology 2.1. Quantum Mechanics (QM) Domain

In the Quantum Mechanics (QM) domain, a harmonic oscillator oscillates at an angular frequency ωM with quantized energy levels:

= (n+1) ħ ω , n = 0, 1, 2, 3 … (1) According to the statistical mechanics if such a

quantum oscillator is in contact with a thermal reservoir of temperature T then this oscillator has probability Pn being in the energy level En given by:

= exp (-ħ

) / exp(− ń ħ )ń (2)


26

Here KB is the Boltzmann constant, and ħ= h/2 with h to be the Planck constant. Furthermore according to the QM the dipole transition strength from energy level En → En+1 is found proportional to quantum number n in the following: | | , ∝ √ + 1 (3)

Now we can evaluate the Stokes line strength from an ensemble of N identical oscillators connected to a thermal bath at temperature T

N ∑ ( + 1) = ( ħ ) (4)

Conversely, the dipole transition strength

from energy level En+1 → En is found in the following proportion | | , ∝ √ (5)

For the anti-Stokes line strength of an ensemble of N identical oscillators:

N∑ ( ) = ħ (6)

We could write down the strength of the Raman

Stokes line from an ensemble of identical QM oscillator ωM that is dominated by the induced electric dipole radiation:

IS = Io(ℓ ) ( ħ ) (7)

Here ℓ is a length scale and IO an intensity scale

proportional to the incident light strength. At the same time the associated anti-Stokes Raman line has the strength:

IAS = Io ( ℓ ) ħ (8)

We thus derived the expression that can be served

as the basis for the Raman temperature sensor: = ( )

exp − ħ (9)

The above formula is derived with the assumption

that each molecule is independent in the system and their mutual interactions are only represented by a statistical temperature parameter T. So by theoretical approach we have come to a conclusion that the coherent anti-Stokes and coherent stroke Raman scatterings, we can calculate the temperature inside a closed or open pipelines accurately. Hence by the above technique we report the measurement of the temperature experimentally by the Superconducting nanowire single – photon detector (SNSPD).

3. Experimental Setup

The light source is a fiber laser with a repetition frequency of 36 MHz and centered near 1550 nm. The sensing fiber is a standard low-loss single-mode fiber with a diameter of 10.1 μm and attenuation less than 0.2 dB/km at a wavelength of 1550 nm. We used filter for flittering the pump laser to a 1 nm line width at a wavelength of 1530.47 nm. We used a variable attenuator before amplifying the pump laser with an erbium doped amplifier (EDFA). The attenuator is used to minimize pulse distortion due to gain saturation in the amplifier. We used a pulse picking modulator applied to our pump laser for extending the test fiber in lengths on the order of kilometers.

The pump signal is again filtered post-amplification to remove the broad band from the pump, which could otherwise be backscattered in the fiber under test and create unwanted pulse counts at the SNSPD detectors, as the filters used before and after the amplifier can only provide 20-30 dB rejection of unwanted wavelengths. We thus injected the pump laser light into the fiber under test(FUT) via a fiber coupling (fiber-optic circulator).we have tested the fiber up to a length of 5 meters as it is limited by the repetition rate of our pump laser. This test fiber is passed through a pipeline of 5 meters in length for the experimentation work. This pipe is wrapped with electrical wire at 3 meter position for creating a localized hot spot externally by applying the current to the electrical wire for the temperature measurement inside the pipeline with respect to the wavelengths and the coherent stroke counts and coherent anti-Stroke counts. The backscattered light from the fiber under test is directed by the circulator to a series of filters that will reject the pump wavelength. Without these filters, we would have unwanted photons of the pump wavelength reaching the SNSPD detector. The 1350 nm filters are used to block any of the EDFA photons that might reach the SNSPD detectors.

Now the process of band splitters to separate and filter the Raman backscattered photons into an S-band channel (1450 nm-1490 nm) and an L-band channel (1550 nm – 1630 nm). The light from the S-band and L-band channels are coupled to SNSPDs, where accurate and precise temperature measurement and control is one of the key factors for this experiment.

We used Temperature Controller 340 (MAKE: Lakeshore) a device that measures and displays temperatures. We have installed the TC 340 at the position where the external electrical wire is wrapped around the pipe line, so as to compare the temperature at the 3 meter position by the proposed method and as well as to measure the temperature at the 3 meter position by the temperature controller (TC 340) to strengthen the results of our experimented method.

In the cryostat, the voltage passes through the low-pass filter that is part of the printed circuit board, and reaches the contact pads. The pulse signal is picked up from the sample holder with a coaxial cable, passes into a signal inverting cryogenic amplifier with a bandwidth of 30 MHz to 2.1 GHz and is brought


27

outside the cryostat. From there, a semi rigid coaxial cable passes the signal through a −3 dB attenuator to a second amplifier. The −3 dB attenuator damps reflections between the amplifiers and prevents the

formation of a standing wave. After the second amplifier the signal is damped by −6 dB before it reaches the oscilloscope. Before the signal cable enters the oscilloscope or the pulse counter.

Fig. 3. Measurement set up for wavelengths analysis’s of coherent anti-Stoke Raman Scattering and coherent stroke Raman Scattering in the fiber under test.

Fig. 4. Sketch of the electronic setup. The printed circuit board is mounted on the sample holder right next to the SNSPD chip with the amplification and bias.

4. Results and Discussion

We modeled the Raman backscattering by the strokes and anti-Strokes as follows:

Backscattered photons (Iu) ≈ ɳu∆ | | , (10)

where Iu is the backscattered photons per second, ɳu is the detection efficiency (DE) of the SNSPDs,∆ is the bandwidth of the stroke and anti-Stroke filters, is the peak pump power, L is the fiber length, is the Raman gain factor, is the duty cycle of the pump signal, N represents the photon population as given in the Equations (4) and (6) respectively.

By the theory of quantum Mechanics, the Raman gain generally increases, on the order of 0.8 . k . As our pump power is limited by gain saturation in our EDFA to around = 18 mW. This

configuration achieved count rates of approximately 3×105 Hz at room temperature. Thus with the ratio of strokes to anti-Strokes as given by Equation (9) and the instantaneous position in the fiber under test (FUT) is given by ( )( ) = ( ) exp − ħ ( ) , (11)

where is the strokes count, is the anti-Strokes

count, is the position in the FUT, and ( ) which

include the detection efficiency of the two SNSPDs detectors.

So by the Equation (11) we have calculated the temperature of the entire fiber under test as a function of desired position. We have calculated the input intensity and output intensity with respect to the power. The normalized intensity vs. the power (mW) is shown in the Fig. 5 where the input intensity is


28

represented by a blue line and the output intensity is represented by red color line.

Fig. 5. The normalized Intensity vs. Power.

Now by having the output intensity, the coherent anti-Stroke counts for the different wavelengths of the strokes and anti-Strokes. We have the coherent anti-Stroke counts vs. the wavelength as show in the Fig. 6.

Fig. 6. Coherent anti-Stroke counts vs. wavelength.

Thus as we already wrapped the electrical wire around the test pipe at 3 meters, which will create a localized hot spot in the test fiber and by this we were able to have the temporal profile of the single mode fiber under test at the desired position and the temperature spike at the 3 meters position is plotted and is as shown in the Fig. 7.

We have obtained the temperature at the 3 m position by the TC 340 and the measured temperature is 455 K and the measured temperature by proposed method is 450 K. The proposed methodology has a very good accuracy of ± 5 K. So by using the CARS and CSRS we were able to measure the temperature at a fixed position or we can also measure the entire temperature profile of a tunnel or a lengthily pipe line.

Fig. 7. The result of the temperature measurement from TC 340 (Red colour spike at 3 m position) and the temperature measurement by the proposed method (Blue

colour spike at 3 m position).

5. Conclusions

In this paper, we have experimented and demonstrated that the coherent stroke and coherent anti-Stroke Raman Scattering can be used for measurement of the temperature profile by using superconducting nano wire single – photon detector. We have tested the pipe line of 5 meters length by externally giving temperature at 3 meter distance and achieved the temperature profile at that position as shown in the Fig. 7 we have achieved the same temperature by TC 340 which strengthens the temperature measurement by CARS and CSRS using SNSPDs. We can increase the length of the test pipe by increasing the power of the pump laser. This methodology is therefore particularly suitable for applications to large or elongated structures; such as dams, large bridges and pipelines. References [1]. Pan X., Barker P. F., Meschanov A., Grinstead J. H.,

Shneider M. N., Miles R. B., Temperature measurement by coherent Rayleigh scattering, Optics Letters, Vol. 27, No. 3, 1 Feb. 2002, pp. 161-163.

[2]. Graul J., Lilly T., Coherent Rayleigh-brillouin scattering measurement of atmospheric and molecular gas temperature, Optics Express, Vol. 22, Issue 17, 2014, pp. 20117-20129.

[3]. Gu Z., Vieitez M. O., Van Duijn E. J., Ubachs W., A Rayleigh-brillouin scattering spectrometer for ultraviolet wavelengths, Rev. Sci. Instrum., Vol. 83, No. 5, May 2012, 053112.

[4]. Yu Zhang, Yu-Rong Zhen, Oara Neumann, Jared K. Day, Coherent anti-Stokes Raman scattering with single-molecule sensitivity using a plasmonic Fano resonance, Nature Communications, Vol. 5, June 2014, pp. 1-7.

[5]. Zumbusch A., Holtom G. R., Xie X. S., Three-dimensional vibrational imaging by coherent anti-Stokes Raman scattering, Phys. Rev. Lett., Vol. 82, No. 20, Aug. 1999, pp. 4142-4145.


29

[6]. M. P. Thariayn, Dual-pump coherent anti-Stokes Raman scattering system for temperature and species measurement in an optically accessible high pressure gas turbine combustor, Measurement Science and Technology, Vol. 22, No. 1, 2010.

[7]. Russel Lockett, Coherent anti-Stokes Raman spectroscopy temperature measurements in an internal combustion engine, Optical Engineering, Vol. 33, No. 9, 1994, pp. 2870-2874.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Nanosensors_IFSA.htm




http://www.sensorsportal.com/HTML/BOOKSTORE/Gas_lasers.htm





30



Dynamic Sensor Management Algorithm Based on Improved Efficacy Function

TANG Shujuan, XU Yunshan and YANG Tao

Institute of Aeronautics and Astronautics Engineering, Air Force Engineering University, Xi'an 710038, China

Tel.: 0081-02984787669, fax: 0081-02984787015 E-mail: [email protected]

Received: 9 July 2013 /Accepted: 9 September 2013 /Published: 30 January 2016 Abstract: A dynamic sensor management algorithm based on improved efficacy function is proposed to solve the multi-target and multi-sensory management problem. The tracking task precision requirements (TPR), target priority and sensor use cost were considered to establish the efficacy function by weighted sum the normalized value of the three factors. The dynamic sensor management algorithm was accomplished through control the diversities of the desired covariance matrix (DCM) and the filtering covariance matrix (FCM). The DCM was preassigned in terms of TPR and the FCM was obtained by the centralized sequential Kalman filtering algorithm. The simulation results prove that the proposed method could meet the requirements of desired tracking precision and adjust sensor selection according to target priority and cost of sensor source usage. This makes sensor management scheme more reasonable and effective. Copyright © 2016 IFSA Publishing, S. L. Keywords: Efficacy function, Sensor management, Kalman Filter, Error covariance.

1. Introduction

More requirements are requested about the environment sensory ability of the airborne information platform (AIP) under the diversified target, complex environments and multiple tasks situation. Many (kinds of) sensors are disposed to this kind of platform, so we can give full play on sensor coordinated work and sensor management technology. Faced with multiple targets and tasks, the limited sensor source has a core problem which is to optimize the distribution and allocation.

Document [1-2] set phased array radar as a study object for tracking task, the methodology for beam scheduling in multi-target tracking was proposed. Document [3-5] solved the target tracking task in centralized multisensory systems based on the method of covariance control. These methods above concentrated on high tracking precision, lacking of

effective control to task tracking. Document [6-8] was based on the needs of tasks requirement, established an allocation model according to the tasks requirement and brought out dynamic sensor source allocation algorithm. Document [9-10] established a sensor allocation plan which was independent from the tracking cycle system, and was an off-line predictive allocation substantially. Document [11-13] overcame the sensor management problem during tactical monitoring action, taking out studies mainly on task distribution, sensor-source pairing and some other aspects.

Concentrating on the AIP sensor and task feature, the tracking task precision requirements (TPR), target priority, sensor use cost and some other factors are considered comprehensively to establish an efficacy function. From the viewpoint of controlling covariance, normalized distance function and normalized filtering covariance are weighted to




31

assign the pairing coefficient and dynamically confirm the result of “target-sensor (combination)”. Under the multi-target tracking situation, simulation proves the efficiency of the sensor management method.

2. Sensor Management Model Under the practical multiple tracking situation,

the target tracking of sensor system only reflects on the tracking precision. Firstly, there is no need to maintain a high precision tracking on the low priority target. Secondly, due to the restriction from the sensory ability of sensor, it will cost differently while using different sensor. While improving sensory ability of one certain target, it surely has side effect on the other target tracking and task searching. Thirdly, using the active radar to maintain the targets measurement may bring vital threat to self-surviving.

Assuming that m sensors track n targets at the same time, there are m'=2m–1 kinds of sensor combinations which certain m'–m fake sensors. Hence, considering the precision need for target tracking task, target priority and the cost of sensor use, we describe the allocation as follows,

'

1 1

1

1

arg max ( ( ), , )

. , 1,2,......, '

1, 1,2,3,......,

m nopt

ij ij j i iji j

n

ij ij

m

iji

E ef pa pr c x

st x S i m

x j n

= =

=

=

= ⋅ ⋅

≤ =

= =

(1)

where E is the efficient value corresponding with the sensor optimal allocation strategy. x is the allocation case of sensors, xij means distribute sensor i to target j, conversely xij =0 means no distributing; in the equation, paij is the pairing function when sensor i and target j, which is related with the target task accomplishment condition. prj is the priority function for target j, ci is the cost of sensor use. efij is the efficacy of sensor i which allocation to target j, the better the accomplishment condition is, the higher the priority is, the smaller the cost of sensor use is, the bigger the value of efij will be.

Allocation restriction includes the maximum restriction of tracking ability (the tracking task allocation to each sensor can not be more than the limited capacity) and restriction of target cover (each one target should be allocation to one sensor at least).

3. The Establishment of Efficacy Function 3.1. Sensor-target Pairing Function

The pairing function paij when target j allocates

to sensor i is related to the TPR. Since the precision

of tracking period and collimation period are different, the expected TPR is different as well, even if considering one target. We can describe the tracking precision need through the differences between practical evaluating covariance matrix (PECM) and desired covariance matrix (DCM). We can use determinant of matrix, trace of matrix, or matrix metrics derived from different matrix norms. Different result coming from different method may bring different effect on TPR [8-9], we choose trace of matrix to measure the differences between expecting covariance Pd and practical covariance, that is to say,

[ ]( , )P d df f P P tr P P= = − − (2)

Since the covariance matrix is not negative, we

assume that A, B∈Rn×n, A=AT≥0 and B=BT≥0. Define the difference between two matrixes is M=A – B, so, M=MT, distance function f(A,B) can be expressed as,

1

( , ) ( )n

iii

f A B tr m m=

= = (3)

target-sensor (combination) pairing function

pa(x1, x2, ...; y1, y2, ...) should obey the equations Max pa( )=1, and Min pa( )=0, so TPR can not be used in this model directly. Evaluated paij( ) with the weighted sum of the normalized distance function 1–f(P,Pd) and normalized filtering error covariance norm ρ. The pairing function was defined as:

[1 ( , )]ij dpa f P Pα β ρ= × − Δ + × Δ (4)

where ρ=ǁPǁ, weight coefficient α, β reflect the effect on pairing function from filter covariance and different measure method between two matrices. Herein, α+β=1 (α＞0, β＞0) [8]. 3.2. Target Priority

Many factors affect target priority, but we usually

take the following factors into consideration: target identification (ID), information requirement (IM), target threat (FR), attack chance (CH), fire-control requirement external commands (EC), etc., so we can define a priority function pr(ID, IM, TR, CH, FR) whose value can be calculated with these five factors. According to the practical experience, we can figure out the concrete form of the function. No matter what the form of function is, we confirm the priority on the basis of the following requests:

1) Put the locked target on the very first position; 2) Then the target we can attack; 3) Next the enemy target and undefined state

follows; 4) Finally friend and ourselves; 5) The external commands owns the highest

priority, assuming pr( )=1.


32

The simplest and most practical expression is the linear sum of the five factors multiply its own coefficient

( , , , , )pr ID IM TR CH FR

ID IM TR CH FRχ δ γ ε η= × + × + × + × + × (5)

The higher the priority for the target is, the smaller the value of pr( ) is to meet the principle that the higher priority is, the bigger the value of efij, and the value is smaller than 1, we take reciprocal value for priority function, that is to say, as to the target j, the priority degree is 1j jpr pr= .

3.3. Sensor use Cost

Based on different practical situation, the definition of “sensor use cost” can be explained differently. As to different kinds of sensor allocated to AIP, when use active sensor to measure target constantly, it may bring threat to self-surviving. However, the passive sensors are in the passive state all the time, and will never be exposed. From this degree, the “use cost” of active radar is much bigger than passive radar and it is the same to active sensor (such as phased array radar), the use cost for different performance is not the same. Normally, the Radars improve their detecting precision by increasing the data updating rate, so improving the precision in certain domain of space will consume more time as “use cost”. The higher the detecting precision of sensor is, the larger the use cost is. We usually confirm the cost according to the sensor number, sensor type, sensor performance and some other factors in a comprehensive way.

The bigger the sensor use cost is, the bigger the value of ci is. To meet the principle that the value of efij increases as the use cost increases and the value of use cost is smaller than 1, we take negative normalized value for sensor use cost function, that is to say, as to the sensor i, the sensor use cost is

1

n

i i ii

c c c=

= − .

In conclusion, we establish efficacy function as

( , , )ij ij j i ij j ief pa pr c pa pr cω ξ λ= × + × + (6)

Herein, coefficient ω, ξ, λ reflect the effect on efficacy from pairing, priority and cost. The basic principle is that the priority has much greater effect on efficacy than what pairing coefficient does, and the pairing coefficient has much greater effect than what sensor use cost does.

4. “Sensor – target” Allocation Algorithms

When given DCM dP of the target, we can

iterate the Kalman filtering equation to figure out the

filtering error covariance, and then find the allocation efficacy. When couples of sensors are allocated to one target, covariance matrix for each sensor to measure targets are different, so we import a centralized sensor Kalman sequential algorithms to solve the target measuring problem with sensor combination.

4.1. Covariance Matrix Algorithms

Assuming target discretizing state equation expressed as

1( ) ( ) ( ) ( ) ( )k k k k kx t F T x t G T w t+ = + (7)

Herein, x(tk) is the state vector to time tk, w(tk) is the system noise vector, and its covariance matrix is Q(tk). F(Tk) is the transforming matrix to time tk, G(Tk) is the input matrix to time tk. Tk=tk+1 – tk is the sample interval to time tk.

The measuring equation for sensor is as follow:

( ) ( ) ( ), 1, 2,......, 'j k j k j kz t H x t v t j m= + = (8)

Herein, Zj(tk) is the measure vector of sensor j to timetk, vj(tk) is measure noise, and its covariance matrix is Rj(tk), Hj is observation matrix.

As to arbitrarily pseudo sensor Di from the pseudo sensor set D, the centralized sensor Kalman sequential algorithms is as follow,

1 1

1 1( ( )) ( ) ( ( )) ( ( ) ( ))k k k k kx t x t K t z t H x t− −= + − , (9)

1

1

( ( )) ( ( )) ( ( ))

( ( ) ( ))

K K Kk k k

KK k K k

x t x t K t

z t H x t

−

−

= + ⋅

− , (10)

( ) ( ( )) ,iNk k ix t x t K D= ∈ (11)

Herein the sequential gain can expressed as

1 1

1 1 1 1( ( )) ( ) ( ( ) ( ) )T Tk k k kK t P t H R t H P t H− − −= + , (12)

1

1 1

( ( )) ( ( )) ( ( ( )

( ( )) )

K K Tk k K K k

K TK k K

K t P t H R t

H P t H

−

− −

=

+ , (13)

filtering covariance matrixes (FCM)P(tk) are

1 11( ( )) ( ( ( )) ) ( )k n k kP t I K t H P t −= − , (14)

1( ( )) ( ( ( )) ) ( )K K K

k n k K kP t I K t H P t −= − , (15)

( ) ( ( )) ,iNk k iP t P t K D= ∈ (16)

Predictive value ( )kx t − and predictive error

covariance matrix ( )kP t − are respectively

1 1( ) ( ) ( )k k kx t F T x t−− −= , (17)


33

1 1 1

1 1 1

( ) ( ) ( ) ( )

( ) ( ) ( )

Tk k k k

Tk k k

P t F T P t F T

G T Q t G T

−− − −

− − −

= + (18)

Herein, (•)K means disposed by K sensors. Distance function of FCM and DCM is changed

by Dynamic state value of filtering time. During the Dynamic control procedure we can treat the FCM as PECM, and the TPR

Pf is

[ ]( ( ), ) ( )p k d k df t tr t= −P P P P (19)

4.2. Allocation Algorithms Steps

Based on the analysis above, we can bring the steps of “sensor – target” allocation algorithms into conclusion:

Step 1. Initialize the number of sensor i=1; Step 2. Calculate the predictive error covariance

( )kt−P matrix at kt

− according to (18);

Step 3. Calculate the FCM ( )ktP at kt after the

Di sensor measurement according to (13); Step 4. Measure the distance between FCM and

DCM at kt according to (19);

Step 5. Calculate the pairing coefficient paij( ) according to (4);

Step 6. Calculate allocation ijef according to (6);

Step 7. Loop the steps above, and then complete the dynamic allocation based on (1).

5. Simulation Analysis Considering 3 sensors track 4 targets at the same

time, and the coordinate for each target is (x,y). Assuming that the simulation cycle is 1 s, and the simulation time is 100 s. Herein, the tracking precision of sensor S1 at x axis direction is relatively higher. The tracking precision of sensor S2 at y axis direction is relatively higher, the tracking precision of sensor S3 at both x and y axis direction are relatively low. Tracking capacity is the total number a sensor can track at the same time, the correlation coefficient reflects the correlate degree of noise between two axes.

There were 23-1=7, seven sensor combinations should be considered coming from the randomly combination between sensors. In simulation of this paper, the coefficient in Formula (6) is ω=0.21, ξ=0.46, λ=0.33.

The particular set of measure noise parameter for the sensors and their tracking capacity was listed in Table 1.

The normalized sensors use cost was listed in Table 2.

We take filtering covariance (FC), ID, Target lock-on, Weapon state, type information (TI) and some other factors into consideration when measure

the target priority. We can check the current value of priority quantification in Table 3. The factors coefficient in the simulation is supposed to the same, the sequence of target priority can be calculated. The tiptop priority is target 2 and the lowest priority is target 4, the relationship is 2 1 3 4pr pr pr pr> > > .

Table 1. Measuring noise analysis of each sensor (standard deviation: meter).

Sensor

x axis direction standard deviation

y axis direction standard deviation

Correlation coefficient

Trackingcapacity

Sensor 1 9.23 20.9 -0.61 2 Sensor 2 22.1 8.2 0.82 3 Sensor 3 41.4 43.7 -0.94 4

Table 2. Sensor (combination) normalized use cost.

Sensor number Normalized use

cost ( ic )

S1 (Sensor1) -0.05 S2 (Sensor2) -0.075 S3 (Sensor3) -0.125 S4 (Sensor1+ Sensor2) -0.125 S5 (Sensor1+ Sensor3) -0.175 S6 (Sensor2+ Sensor3) -0.2 S7 (Sensor1+ Sensor2+ Sensor3) -0.25

Table 3. Current quantification of target priority.

FC ID Target lock-on

Weapon state

TI

Target 1 0.45 1 0 0 0.853 Target 2 0.50 0 0 0 0.106 Target 3 0.60 1 0 0 0.734 Target 4 0.80 1 0 0 0.808

Suppose the DCM in 1~50 s as Pd1=Pd2=Pd3=Pd4=diag(30, 1, 30, 1), set the DCM in 51~100 s as

Target 1: Pd1=diag(10, 0.3, 30, 1); Target 2: Pd2=diag(30, 1, 1 0, 0.3); Target 3: Pd3=diag(30, 1, 30, 1); Target 4: Pd4=diag(10, 0.3, 10, 0.3). The maneuvering noise covariance matrixes of

these four targets are listed as follows:

1

0.35 0

0 0.45Q

=

, 2

0.12 0

0 0.20Q

=

,

3

0.20 0

0 0.15Q

=

, 4

0.70 0

0 0.85Q

=

The filtering covariance for targets are given

in Fig. 1.


34

0 10 20 30 40 50 60 70 80 90 10015

20

25

30

35

40

45

simulation time/s

filte

ring

cov

aria

nce

x axis direction filtering covariance of target 1

0 10 20 30 40 50 60 70 80 90 1000

10

20

30

40

50

60

70

80

simulation time/s

filte

ring

cova

rianc

e

y axis direction filtering covariance of target 2

0 10 20 30 40 50 60 70 80 90 10010

20

30

40

50

60

70

80

simulation time/s

filte

ring

cova

rianc

e

x axis direction filtering covariance of target 4

0 10 20 30 40 50 60 70 80 90 10010

20

30

40

50

60

70

80

simulation time/s

filte

ring

cova

rianc

e

y axis direction filtering covariance of target 4

Fig. 1. Covariance control effect.

From simulation chart 1, we can see, due to improvement of the precision, the filtering covariance of each target can convergence to expectation degree, and can basically reach the requirement of adaptive covariance control. Since we import a target priority function in allocation efficacy, the source the highest target 2 get can perfectly reach the requirement of target tracking. From chart 3 we can see that, the tracking performance of target 2 is better than target 1.

The Pairing result of ‘Target-Sensor’ is shown in Fig. 2.

During 51~100 s, the tracking precision of x axis improves and the sensor S1 was allocated to target 1 increasingly. However the priority of target 1 is relatively low, and the tracking capacity of S1 is the weakest, so the target 1 can not sostenu to occupy the source of S1, and the result of source allocation is changeable all the time. When the tracking precision of target 2 on y axis improves, sensor combination S6 (includs S2) was allocated to the target increasingly. Due to the highest priority of target 2, it can occupy the source of S6 in the most time to meet the requirement of precision on y axis. For the tracking precision of target 3 is constant in the latter 50 s, the result of sensor source allocation acts steadily and the target gets the least sensor source. The expected covariance of target 4 on x axis and y axis were decreased, so the improvement of tracking precision leads sensor combination S4 (includes S1 and S2) was allocated to the target increasingly.

As for the influence of sensor use cost factor, the

pairing result of ‘Target-Sensor’ without cost is simulated, the result was shown in Fig. 3.

Comparing to the Fig. 2, it can be clearly seen in the Fig. 3 that pairing result of ‘Target-Sensor’ without cost of each target, the high cost sensor (sensor combination) were more allocated.

From the simulation above we can see that the algorithm we give in this paper could dynamically allocate the sensor source reasonably based on covariance matrix, target priority, and sensor use cost.

6. Conclusions

A dynamic sensor management algorithm based on improved efficacy function is proposed to overcome the sensor management difficulties during face to multi-tasks on AIP. This paper brings about the assignment of sensor to pairing coefficient and auto update. Meanwhile we introduce target priority function and sensor use cost to the efficacy function and give an adaptive allocation algorithm based on the linear programming. Furthermore we conduct the algorithm simulation. The result of simulation shows that this algorithm combines the advantages of covariance adaptive control and target weight. So this algorithm could adaptively allocate the sensor source based on expectation of tracking precision and dynamically adjust the sensor source according to the target priority degree and the sensor use cost. In summarize, this algorithm improves the rationality and effectiveness for sensor source allocation.


35

0 10 20 30 40 50 60 70 80 90 1000

S1

S2

S3

S4

S5

S6

S7

simulation time/s

sens

or(c

ombi

natio

n)

“target-sensor” pairing result of target 1

0 10 20 30 40 50 60 70 80 90 100

0

S1

S2

S3

S4

S5

S6

S7

simulation time/s

sens

or(c

ombi

natio

n)


0 10 20 30 40 50 60 70 80 90 1000

S1

S2

S3

S4

S5

S6

S7

simulation time/s

sens

or(c

ombi

natio

n)


0 10 20 30 40 50 60 70 80 90 1000

S1

S2

S3

S4

S5

S6

S7

simulation time/s

sens

or(c

ombi

natio

n)


Fig. 2. Pairing result of ‘Target-Sensor’.

0 10 20 30 40 50 60 70 80 90 1000

1

2

3

4

5

6

7

simulation time/s

sens

or(c

ombi

natio

n)

“target-sensor” pairing result without cost of target 1

0 10 20 30 40 50 60 70 80 90 1000

1

2

3

4

5

6

7

simulation time/s

sens

or(c

ombi

natio

n)


0 10 20 30 40 50 60 70 80 90 1000

1

2

3

4

5

6

7

simulation time/s

sens

or(c

ombi

natio

n)


0 10 20 30 40 50 60 70 80 90 1000

1

2

3

4

5

6

7

simulation time/s

sens

or(c

ombi

natio

n)


Fig. 3. Pairing result of ‘Target-Sensor’ without cost.


36

References [1]. Krishnamurthy V., Evans R. J., Hidden Markov

model multi-arm bandits: a method ology for beam scheduling in multi-target tracking, IEEE Transactions on Signal Processing, Vol. 49, No. 12, 2001, pp. 2893-2908.

[2]. V. Krishnamurthy, D. V. Djonin, Optimal threshold policies for multivariate POMDPs in radar resource management, IEEE Transactions on Signal Processing, Vol. 57, No. 10, 2009, pp. 3954-3969.

[3]. M. Kalandros, Pao L. Y., Covariance control for multisensory systems, IEEE Trans. on Aerospace and Electronic Systems, Vol. 38, No. 4, 2002, pp. 1138-1157.

[4]. M. Kalandros, Pao L. Y., Multisensor Covariance Control Strategies for Reducing Bias Effects in Interacting Target Scenarios, Trans. on Aerospace and Electronic Systems, Vol. 41, No. 1, 2005, pp. 153-173.

[5]. Lu J. B., Hu W. D., Xiao H., et al., Novel cued search strategy based on information gain for phased array radar, Journal of Systems Engineering and Electronics, Vol. 19, Issue 2, 2008, pp. 292-297.

[6]. Cui B. X., Xu Y. S., Xia H. B., et al., Dynamic multi-sensor management algorithm based on task control, Journal of Systems Engineering and Electronics, Vol. 34, No. 12, 2012, pp. 2473-2478.

[7]. Wang L., Yu L., Kou Y. X., et al., Multi-sensor resource management method drive by task

requirement, Journal of Systems Engineering and Electronics, Vol. 32, No. 9, 2010, pp. 1925-1930.

[8]. Kreucher C. M., Hero A. O., Kastella K. D., A comparison of task driven and information driven sensor management for target tracking, in Proceedings of the IEEE Conference on Decision and Control and the European Control Conference, 2005, pp. 4004-4009.

[9]. Boers Y., Driessen H., Nederland T., Results on the modified Riccati equation: Target tracking application, IEEE Trans on Aerospace and Electronic Systems, Vol. 42, No. 1, 2006, pp. 379-384.

[10]. Tong J., Shan G. L., Study of multi-sensor allocation based on modified Riccati equation and Kuhn-Munkres algorithm, Control and Decision, Vol. 27, No. 5, 2012, pp. 747-751.

[11]. C. J. Cohen, F. Morelli, K. A. Scott, A surveillance system for the recognition of intent within individuals and crowds, in Proceedings of the IEEE Conference on Technologies for Homeland Security, May 2008, pp. 559-565.

[12]. A. Benaskeur, H. Irandoust, Sensor management for tactical surveillance operations, Aerospace Science and Technology, Canada, Vol. 11, 2007, pp. 327-338.

[13]. Liu B., Zhao Y., Yi D. Y., Sensor Management based on Covariance Control Strategy for LEO Satellite Constellation, in Proceedings of the 4th International Symposium on Computational Intelligence and Design, Vol. 1, Oct. 2011, pp. 282-285.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Digital_Sensors.htm


37



Scattering Parameters of Broadband (18 – 40 GHz) RF MEMS Switch in π-match Configuration

1 Updesh SHARMA, 1 Shankar DUTTA and 2 E. K. SHARMA

1 Solid State Physics Laboratory, DRDO, Lucknow Road, Timarpur, Delhi, India – 110054 2 Department of Electronic Science, University of Delhi, South Campus, Delhi, India – 110021

1 Tel.: +91 1123903834, fax: +91 1123913609 E-mail: [email protected]

Received: 16 December 2015 /Accepted: 18 January 2016 /Published: 31 January 2016 Abstract: This paper discussed about radio frequency design of coplanar-waveguide (CPW) based π-matched shunt switch for broad-band (18-40 GHz) application. The effects of variation in membrane width (50 – 80 µm) of the switch and high-impedance transmission line length (200 – 600 µm) between the switch structures on scattering parameters are studied. The variation in beam width has very little effect on return loss and insertion loss of the switch in the up-state. The reduction in high-impedance transmission line length yields marginally improved return loss and insertion loss. In the down-state configuration, the return loss showed negligible change with the variation in beam width and high-impedance transmission line lengths. The isolation is found to be improved with the increase in beam width and high-impedance transmission line length in whole frequency range. The optimized high-impedance transmission line length (400 µm) and beam width (50 µm) yields insertion loss < – 1 dB and isolation better than – 40 dB in 18-40 GHz frequency range. Copyright © 2016 IFSA Publishing, S. L. Keywords: Radio frequency MEMS, Shunt switch, Scattering parameters.

1. Introduction

Radio frequency micro-electro-mechanical system (RF MEMS) components are receiving significant devotion due to their small size, lightweight, low insertion loss, and negligible power consumption [1-5]. The trade-off is somewhat high electrostatic actuation voltage, relatively lower switching speed, and the reliability issues that come along. Many RF-MEMS components approaches commonly used have also limited power-handling capabilities.

RFMEMS switches are fundamentally mechanical switches permitting either capacitive contact [6-8] or ohmic contact (dc) [7, 9] switching via electrostatic actuation. In past, a number of papers had been published on the development of capacitive RFMEMS switches of different materials [9-14] and

different configurations [6, 15-19]. The switch structure typically consists of a bottom electrode, a very thin dielectric layer, and a top moveable membrane/bridge. When voltage is applied, the membrane/bridge starts deflecting towards the bottom electrode and thus changes the capacitance between top and bottom electrode. The switch performance depends in terms of isolation (Cmax/ Cmin ratio).

Ka-band (18 – 40 GHz) has become the band of choice for many radio communication applications due to its increasing capacity availability and its applicability for broadband services. Ka-band is not just the next generation frequency band expansion to Ku-band, it encompasses a new type of architecture, new transmission and bandwidth management to provide higher quality, better performance and faster speed [1, 4-5, 7-8, 15].




38

In radio-frequency devices and components, impedance matching is absolutely essential to provide maximum power transfer between the source or RF energy and its load. This is especially important if one deals with low amplitude signals. Moreover, if RF circuit is not matched we get reflected power, which builds standing waves on the transmission line between the source and load. Depending on the phase between the forward and reflected waves, resultant can either subtract or add. Thus one can get places in the transmission line where either the voltage is the sum of both voltages or equals zero (maximum current).

The condition for impedance matching is that real part of the impedance should be equal to the real part of the load and reactance's should be equal and opposite in character. For example if our source impedance is R + jX to achieve matching our load should be R – jX. By using capacitors and inductances we can achieve impedance matching without power loss assuming the components are ideal. Real capacitors and inductors exhibit losses which need to be minimized during the match design. There are three primary impedance matching configurations – L match, T-match and π-match [4-6, 8, 13, 16]. Each of them has advantages and disadvantages. In RFMEMS switches, either T-match or π-match configuration is used.

In T-match configuration, the up-state capacitance is to use two short high impedance sections of t-lines before and after the switch. These sections behave as a series inductors and provide a good match at the design frequency. Whereas in π-match configuration, a short section of high impedance line is used between two shunt switches to result in an impedance match. The advantage of π-match configuration is twofold – it provides an excellent match in the up-state position over a wide bandwidth and wide isolation bandwidth. Improved isolation is achieved by using double shunt capacitance, which grounded the high frequency signal two times faster as compared to the T-match as there is only single shunt capacitance [4-5].

This paper focuses on the RF design of coplanar-waveguide (CPW) based MEMS shunt switch in π-match configuration for Ka band applications (18 – 40 GHz). The switch structure consists of two bridges separated by a high-impedance transmission line. Scattering parameters of the π-matched shunt switch is studied as a function of :

1) Width of the switch membranes; 2) The high impedance transmission line length.

2. RFMEMS Shunt Switch Structure in π-match Configuration

The model for the single MEMS membrane

switch can be used to design the high-isolation, low insertion-loss π-matched shunt switch. The π-matched switch consists of two single MEMS

shunt switches separated by a short length (d) of high-impedance transmission line as shown in Fig. 1.

Fig. 1. Schematic of RFMEMS shunt switch in π-match configuration.

The major characteristic parameters of RF MEMS switch can be broadly divided into two parts: RF and electro-mechanical. The RF characteristics are presented by return loss, insertion loss and isolation; whereas, modal patterns, resonant frequency, pull-down voltage, pull-down time and release time are clubbed in to electro-mechanical characteristics. In order to get broadband RFMEMS switch, we concentrate ourselves to analyze the frequency response of RF characteristics of the switch at different length of the midsection high-impedance transmission line.

RF performance of the switch structure is simulated using High frequency simulation software (HFSS) version 12 for full wave analysis. The CPW line with dimensions of G/W/G = 60 µm/ 100 µm/ 60 µm (50 Ω) was designed for Ka-band applications. In this study, width of the membrane bridge is varied from 50 µm to 80 µm; whereas, high-impedance transmission line length is varied from 200 µm to 600 µm. Each of the MEMS switch is designed with Si3N4 (k~ 7) as dielectric layer with isolation loss optimized (minimum) at 35 GHz. The frequency response of the structure is sweep 18 – 40 GHz range. 3. Results and Discussions

Simulated scattering parameters of the π-matched switch structure are presented in this section. Up state return loss (S11) behavior of the switch corresponding to different bridge width (50 - 80 µm) are presented in Fig. 2(a) – Fig. 2(d). Each figure corresponding to the S11 values for different high-impedance transmission line lengths (d) (200 µm - 600 µm). The up-state S11 parameters are found to vary between – 10 dB to – 45 dB in the 18 – 40 GHz frequency range. The return loss performance is found to be poorer with the increase in d values. In the present study, the switch structure (with w of 80 µm and d = 400 µm) showed the best possible return loss (> -15 dB) in full Ka-band. Up state insertion loss (S21) of the π-match switch are similarly presented in Fig. 3. Like return loss,


39

reduction of the distances between the two bridges (d) yield marginally improved insertion loss. At higher frequencies, the insertion loss is found to be increasing rapidly up-to – 2 dB with the increase in d value above 400 µm.

In the down-state configuration, return loss (S11) and isolation (S21) of the switch structure are shown in Fig. 4 and Fig. 5 respectively.

Return loss (S11) is found to vary -0.1 to 0.3 dB range in the Ka-band frequency range.

(a)

(a)

(b)

(b)

(c)

(c)

(d)

(d)

Fig. 2. Up-state return loss (S11) in (a) width – 50 µm (b)

width – 60 µm (c) width – 70 µm (d) width – 80 µm.Fig. 3. Up-state insertion loss (S21) in (a) width – 50 µm (b)

width – 60 µm (c) width – 70 µm (d) width – 80 µm.


40

(a)

(a)

(b)

(b)

(c)

(c)

(d)

(d)

Fig. 4. Down-state return loss (S11) in (a) width - 50µm

(b) width - 60µm (c) width - 70µm (d) width - 80µm. Fig. 5. Down-state insertion loss (S21) in (a) width - 50µm

(b) width - 60µm (c) width - 70µm (d) width - 80µm.

However, the S11 parameters do-not show any noticeable change with the width of the switch as well as with high-impedance transmission line lengths.

Effect of variation of beam widths and high-impedance transmission line lengths on the isolation can be seen in Fig. 5. The isolation values (– 30 dB to – 60 dB) seem to be decreasing with the increase


41

in the high-impedance transmission line lengths, while it increases with increase in the beam width. As the beam width increases, the downstate capacitance increases, which results in better isolation.

Thus, in totality for the π-matched switch configuration, the optimized distance between the two switches should be 400 µm with beam width of 50 µm. The optimized configurations showed improved broad-band scattering parameters (return loss < – 15 dB, insertion loss < – 1 dB and isolation better than – 40 dB) in the 18 – 40 GHz frequency range. Equivalent circuit model for each part of the

RFMEMS shunt switch in π-match configuration with d = 400 µm are modelled using Advanced Design Software (ADS). Fig. 6 shows the equivalent circuit of the switch. From the equivalent circuit modelling, the matching resistance, inductance and capacitance values for the each bridge of the optimized switch structure is found to be 1.5 Ω, 14 pH and 3.5 pF respectively. Fig. 7 shows the isolation (S21) of the switch structure in down state estimated by using ADS and HFSS simulation software.

CPW Sub1

Rough=0 milTanD=0T=0 milCond=1.0E+50Mur=1Er=11.9H=270 um

CPWSub

CPW 1

L=305 umG=60 umW =100 umSubst="CPW Sub1"

Term1

Z=50 OhmNum=1

Term2

Z=50 OhmNum=2

CPW 2


CPW 3


SP1

Step=1.0 GHzStop=40 GHzStart=18 GHz

S-PARAMETERS

L1

R=L=14 pH t

C1C=3.5 pF t

C2C=3.5 pF t

L2

R=L=14 pH t

R2R=1.5 Ohm t

R1R=1.5 Ohm t

Fig. 6. Equivalent circuit RFMEMS shunt switch in optimized π-match configuration (w = 50 µm; d = 400 µm).

Fig. 7. Comparison of HFSS simulated and ADS calculated isolation (S21 parameter) of the optimized RFMEMS shunt

switch in π-match configuration.

4. Conclusions

This paper presents parametric analyses of scattering parameters of the π-matched switch structure for broad band (Ka-band) application. In this study, width of the membrane bridge is varied from 50 µm to 80 µm; whereas, high-impedance transmission line length is varied from 200 µm to 600 µm. The variation in beam width has very little effect on up-state scattering parameters; whereas, the reduction in high-impedance transmission line length showed marginally improved return loss and

insertion loss. In the down-state configuration, the S11 parameters do-not showed any noticeable change with the beam width of the switch and high-impedance transmission line lengths (d). The isolation is found to be improved with the increase in beam width and high-impedance transmission line length in 18 – 40 GHz frequency range. The present configurations (w = 50 µm; d = 400 µm) showed improved broad-band characteristics (return loss < – 15 dB, insertion loss < – 1 dB and isolation better than – 40 dB) in the full Ka-band. For each bridge of the optimized switch structure, equivalent matching resistance, inductance and capacitance values are found to be 1.5 Ω, 14 pH and 3.5 pF respectively. Acknowledgements

Authors would like to thank Director SSPL for his kind permission to publish this work.

References [1]. M. Angira, K. Rangra, Design and investigation of a

low insertion loss, broadband, enhanced self and hold down power RF-MEMS switch, Microsystem Technologies, Vol. 21, 2015, pp. 1173-1178.


42

[2]. Z. Deng, H. Wei, S. Fan, J. Gan, Design and analysis a novel RF MEMS switched capacitor for low pull-in voltage application, Microsystem Technologies, Technical Paper, 2015, pp. 1-9.

[3]. C. L. Goldsmith, Z. J. Yao, S. Eshelman, D. Denniston, Performance of low-loss RF MEMS capacitive switches, IEEE Microw. Guided. Wave. Lett., Vol. 8, No. 8, 1998, pp. 269-271.

[4]. J. B. Muldavin, G. M. Rebeiz, High-Isolation CPW MEMS Shunt Switches – Part 1: Modeling, IEEE Trans. on Microwave Theory and Techniques, Vol. 48, 2000, pp. 1045-1052.

[5]. J. B. Muldavin, G. M. Rebeiz, High-Isolation CPW MEMS Shunt Switches — Part 2: Design, IEEE Trans. on Microwave Theory and Techniques, Vol. 48, 2000, pp. 1053-1056.

[6]. S. Dutta, Md. Imran, R. Pal, K. K. Jain, R. Chatterjee, Effect of residual stress on RF MEMS switch, Microsystem Technologies, Vol. 17, No. 12, 2011, pp. 1739-1745.

[7]. H. J. D. L. Santos, Introduction to Micro-electro-mechanical (MEMS) Microwave Systems, Artech House, Boston, London, 2004.

[8]. U. Sharma, S. Dutta, Study of scattering parameters of RFMEMS shunt switch with high-K dielectrics, Journal of Materials Science: Materials in Electronics, Vol. 25, No. 12, 2014, pp. 5546-5551.

[9]. S. Dutta, Md. Imran, A. Pandey, T. Saha, I. Yadav, R. Pal, K. K. Jain, R. Chatterjee, Estimation of bending of micromachined gold cantilever due to residual stress, Journal of Materials Science: Materials in Electronics, Vol. 25, No. 1, 2014, pp. 382-389.

[10]. X. J. He, Z. Q. Lv, B. Liu, Z. H. Li, High-isolation lateral RF MEMS capacitive switch based on HfO2 dielectric for high frequency applications, Sensors and Actuators: A. Phisical, Vol. 188, 2012, pp. 342-348.

[11]. C. F. Herrmann, F. W. DelRio, D. C. Miller, S. M. George, V. M. Bright, J. L. Ebel,

R. E. Strawser, R. Cortez, K. D. Leedy, Alternative dielectric films for rf MEMS capacitive switches deposited using atomic layer deposited Al2O3/ ZnO alloys, Sensors and Actuators: A. Phisical, Vol. 135, Issue 1, 2007, pp. 262-272.

[12]. Y. Liu, T. R. Taylor, J. S. Speck, R. A. York, High isolation BST MEMS switches, in Proceedings of the IEEE MTT-S International Microwave Symposium Digest, 2002, pp. 227-230.

[13]. J. Tsaur, K. Onodera, T. Kobayashi, Z. J. Wang, S. Heisig, R. Maeda, T. Suga, Broadband MEMS shunt switches using PZT/HfO2 multi-layered high k dielectrics for high switching isolation, Sensors and Actuators: A. Phisical, Vol. 121, Issue 1, 2005, pp. 275-281.

[14]. G. Wang, S. Barstow, A. Jeyakumar, J. Papapolymerou, C. Henderson, Low cost RF MEMS switches using photodefinable mixed oxide dielectrics, IEEE MTT-S International Microwave Symposium, Digest, 2003, pp. 1633-1636.

[15]. L. Y. Ma, A. N. Nordin, N. Soin, Design, optimization and simulation of a low-voltage shunt capacitive RF-MEMS switch, Microsystem Technologies, 2015 (in Print).

[16]. N. J. R. Muniraj, Design of wide range MEMS tunable capacitor for RF applications, Microsystem Technologies, Vol. 17, Issue 1, 2011, pp. 31-33.

[17]. N. J. R. Muniraj, K. Sathesh, Design of MEMS switch for RF applications, Microsystem Technologies, Vol. 17, Issue 1, 2011, pp. 161-163.

[18]. Y. Shim, Z. Wu, M. R. Zadeh, A multi-metal surface micromachining process for tunable RF MEMS passives, IEEE Journal of Microelectromechanical Systems, Vol. 21, No. 4, 2012, pp. 867-874.

[19]. A. B. Yu, A. Q. Liu, Q. X. Zhang, A. Alphones, L. Zhu, A. P. Shacklock, Improvement of isolation for MEMS capacitive switch via membrane planarization, Sensors and Actuators: A. Phisical, Vol. 119, Issue 1, 2005, pp. 206-213.

___________________


http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/UFDC_1.htm





43



Packet Header Compression for the Internet of Things

* Pekka KOSKELA, Mikko MAJANEN and Mikko VALTA VTT Technical Research Centre of Finland Ltd, P.O. Box 1100, FI-90571 Oulu, Finland

* Tel.: +35 8 40 751 390, fax: +35 8 20 722 2320 * E-mail: [email protected]

Received: 15 December 2015 /Accepted: 18 January 2016 /Published: 31 January 2016 Abstract: Due to the extensive growth of Internet of Things (IoT), the number of wireless devices connected to the Internet is forecasted to grow to 26 billion units installed in 2020. This will challenge both the energy efficiency of wireless battery powered devices and the bandwidth of wireless networks. One solution for both challenges could be to utilize packet header compression. This paper reviews different packet compression, and especially packet header compression, methods and studies the performance of Robust Header Compression (ROHC) in low speed radio networks such as XBEE, and in high speed radio networks such as LTE and WLAN. In all networks, the compressing and decompressing processing causes extra delay and power consumption, but in low speed networks, energy can still be saved due to the shorter transmission time. Copyright © 2016 IFSA Publishing, S. L. Keywords: Internet of Things (IoT), Compression, Energy efficiency, Wireless radios. 1. Introduction

Due to the extensive growth of Internet of Things

(IoT), the number of wireless devices connected to the Internet is increasing and will continue to increase remarkably in the near future. For example, Gartner estimates that the IoT, which excludes PCs, tablets and smartphones, will grow to 26 billion units installed in 2020, representing an almost 30-fold increase from 0.9 billion in 2009 [1].

In wireless IoT networks, the available bandwidth and energy is often restricted. Therefore, all means for saving those resources are welcome. The biggest resource consumption source is radio communication, including both transmission and reception [2]. One efficient way to control transmission and reception is the utilization of the duty-sleep cycle control of the radio with a MAC protocol [2]. Another effective way to reduce radio transmission, which supplements the previous approach, is the utilization of packet compression.

The packet compression can be targeted to cover only header or payload part, or both parts of the packet.

IoT devices, e.g., sensors, periodically report their current data values to the cloud services in the Internet. Thus, the transmitted data may be only couple of bytes, whereas the protocol headers of the packet (MAC, IP, TCP/UDP, etc.), are many tens of bytes. This big header overhead is the motivation behind the compression of packet headers and the design of lightweight protocols with small headers. For instance, the header of the traditional application layer protocol, Hypertext Transfer Protocol (HTTP), can take easily over 40 bytes, whereas replacing HTTP with Constrained Application Protocol (CoAP) [3] can drop the header size to less than 10 bytes.

In this paper, we first give a brief description of packet compression techniques in Section 2, which is followed by a more careful presentation of the evolution of header compression. In Section 4, we present the results of ROHC header compression for CoAP/UDP/IP protocols in the case of both fast




44

(LTE, WLAN) and low speed (XBEE) radio networks. Finally, Section 5 concludes the paper.

2. Packet Compression Techniques There are several techniques to compress packets,

see for example recent surveys in [4-7]. In this section, we shortly summarize the main compression techniques that are depicted in Fig. 1.

Fig. 1. Compression techniques.

2.1. Aggregation

Aggregation is an efficient method to compress data whenever it is suitable for the function in question. The problem of using aggregation is that it will always reduce information like by averaging, and the network structure and routing has to support aggregation. The other shortage is that if packet loss exists, then losing one aggregated packet means losing several original packets, especially when no packet caching is used. During decades, numerous aggregation schemes have been proposed, in which the most recent ones, e.g. [8] and [9], consider also security issues.

2.2. Network Coding Depending on the data information, there may be

a possibility to define sequences, which are repeating inside the message(s). By utilizing the repeated information, it is possible to describe the same information in a shorter way and to achieve compression in that way.

In traditional routing networks, packets are cached and forwarded separately downstream, even if they have the same destination. In network coding, the messages are merged and the code and the accumulated result are forwarded to the destination.

After receiving the accumulated message, it is decoded at the destination. Network coding techniques for wireless sensor networks (WSN) are discussed more carefully in [10].

There are several methods with varied complexity for describing the repeated information. Depending on the method, the decompression will return exactly the original data or it may have some losses or mistakes. In general, the method with more complexity will provide more lossless compression and better compression rate.

2.3. Distributed vs. Local and Symmetric vs. Asymmetric

Compression operation can be carried out locally

at the node or it can be distributed to several nodes in order to share the load between the nodes. Load sharing may help especially nodes that have scarce of resources. Usually, capabilities of network devices vary. Terminal nodes, like sensor nodes, have least resources whereas core network devices, like servers, can share their resources. A special case of distribution is an asymmetric system, where the most capable devices take care of the highest load and thus save the resources of the terminal nodes.

2.4. Lossy & Lossless Depending on the data reconstruction after the

decompression, compression methods can be divided into lossy and lossless techniques. Lossless techniques aim to return exactly the original data, whereas lossy techniques give only an approximation of the original data. Generally, lossy algorithms provide higher compression, but also higher information loss. Which approach will fit best depends on the requirements of the application. For instance, in video and voice applications, lossy compression may be accepted, but data loss or a wrong value may cause serious problems in the case of control measurements.

2.5. Stateless or Stateful Stateless compression does not require any per-

flow state, which could be corrupted during the change in wireless connection. The idea of the stateless compression is based on the assumption that some content between the sender and the receiver is well-known and thus can be assumed or extrapolated from the received information. That kind of information is for instance the static parts of the protocols’ header information, like the network prefix, which can be compressed into a single bit. If similar information is already available in several protocol headers, like checksum, then that information can be removed.


45

In the case of stateful compression (or shared-context), there will always be negotiation between the sender and the receiver. During the negotiation, the sender and the receiver agree on the semantics how the compression will be performed. When the compression is used, the sender and the receiver have to agree from time to time that the compression state is still valid.

The advantage of the stateful approach compared to the stateless is that it will allow much higher compression rate than the stateless. For instance, in a data flow, almost all header information can be compressed under one ID flag in the stateful approach, which is not always possible in the stateless case. Another advantage of the stateful approach is that it is more dynamic because there is no need to assume anything before the negotiation and so the packet formats can change freely.

Because of the nature of wireless communication, there will always be some packet loss and bit errors during messaging. If the losses and errors are significant, then stateless approach will outperform stateful, because the stateless approach does not need any pre-configuration before compression. In the case of stateful compression, if an error or a loss happens, as it often does in mobile ad hoc networks, then pre-configuration must be done again and again, which causes extra control traffic. So, maintaining the flow states will be difficult in a mobile ad hoc environment [11]. This makes pure stateful solution, despite it has a better compression rate, applicable only in good link conditions.

Resource-constrained devices must also consider memory usage and computational complexity. From this point of view, the stateless approach is more efficient, because it does not need any state establishment or management, and it has a simpler source code.

2.6. Non-adaptive and Adaptive In general, an adaptive compression can adjust

system to environmental changes, which can be for example changes in the data type or connection performance. The connection can be improved, for instance, by chancing the communication interface or operation of the link layer (L2), or utilizing multipath routing to get better connection. Adaptivity makes the system more complex, but at the same time, it provides better performance when the system is changing.

2.7. Header Compression The network packet can be divided into two parts:

the header information part coming from different protocols and the payload part containing the data. From the compression point of view, the header part is interesting because there is redundant information among different protocol headers and, especially,

between consecutive packets belonging to the same flow. This kind of redundant information can be elided [6, 12]. For example, many header fields remain constant between packets, or change according to a known pattern. Over a single link, not all that information is needed and part of it can be temporarily removed, i.e., the full IP packet will be re-created on the receiving side of the link. In the next section, we will take a closer look to different header compression solutions.

3. Development of Header Compression Header compression is not a new idea, but

compression standards are still evolving. The first header compression scheme, CTCP (i.e., VJ compression) [13], compresses TCP/IPv4 headers and it was presented in 1990. The evolution continued when IPHC [14] and CRTP [15] were presented in 1999 with wider protocol support (UDP, RTP and IPv6) and improvements in packet loss handling. The next evolution step was presenting ROHC [16] in 2001 and 6LoWPAN [17] in 2007. ROHC presents a robust compression scheme with modular protocol support, i.e., protocol profiles. 6LoWPAN presents a compact solution applicable only for wireless IEEE 802.15.4 technology. In the next chapters, we will discuss more carefully the above mentioned header compression solutions.

CTCP (Van Jacobson Header Compression): CTCP header compression defines compression for TCP/IP(v4) datagrams, which was specifically designed to improve TCP/IP performance over slow serial links, with speed around 300 bps. The primary idea behind the compression was to define per-packet information on which bytes are likely to stay constant over the lifetime of a connection, and which bytes are likely to change during the connection but do not all change at the same time. In the compression, the constant bytes can be omitted and the changed bytes can be indicated with a bitmask. The bitmask tells the difference between the previous and the current packet and that way only the differences in the changing fields are sent rather than the whole fields themselves. In practise, this is done by saving the states of the TCP connections at the both ends of the link, and sending only the differences in the header fields that changed. Van Jacobson compression reduces the normal 40 byte TCP/IPv4 packet headers down to 3-4 bytes in average, see Fig. 2. Because the scheme is designed only for low bandwidth connections, where bit errors and packet loss are not an issue, it works well there. When the bit error rate (BER) increases over 10-4, the scheme does not perform well [18]. This is mainly due to that the scheme does not have its own feedback and recovery mechanism concerning packet loss, instead it relies on TCP’s own recovery mechanisms. Nowadays, it is well known that TCP’s recovery mechanisms for packet loss do not perform well in wireless connections [19].


46

Fig. 2. CTCP header compression.

IPHC (IP Header Compression): In general, IPHC is based on a similar compression and decompression idea as CTCP, where known information is compressed and decompression is done based on the saved context information of the compression. The main development step that IPHC brought in was that it supports UDP, IPv6 and extension headers in addition to TCP. Moreover, IPHC improved error recovery mechanisms (important especially in lossy links) and supported multiple IP header compression (in case of tunnelling of IP packets) and allowed extensions for multi-access links and multicast. Two additional mechanisms that increase the efficiency of header compression over lossy links were also described. For non-TCP packets, compression slow-start and periodic header refreshments allow minimal periods of packet discarding after loss of a header that changes the context. When many packet streams (several hundreds) traverse the link, a phenomenon known as context ID (CID) thrashing can occur. In CID thrashing, headers cannot be matched with an existing context and have to be sent uncompressed or as full headers.

CRTP (Compressing RTP Headers): CRTP uses similar compression approach like CTCP and IPHC. As a new thing, it provides support for RTP protocol. As additional features, CRTP brings in a method for reporting packet loss information. The information contains a health report of packets and compression

level for sources capable adapting. CRTP replaces the IPv4, UDP, and RTP headers (40 bytes) with a 2-4-byte context ID (CID), as depicted in Fig. 3.

Fig. 3. CRTP header compression.

ROHC (Robust Header Compression) has similar idea like above, where compressed packets are decompressed based on the saved context information in the decompression side. One main difference for previous compression schemes is that the whole compression process is divided into different states depending on the link performance. The states are maintained within two finite state machines: one as a compressor and the other as a decompressor. The compressor states are Initialization & Refresh (IR), First Order (FO) (i.e., partial compression), and Second Order (SO) (i.e., full compression). The state machine aided compression makes compression process more robust and takes advantage of the link quality, but on the other hand, it increases complexity. The other remarkable difference is that new protocol headers can be presented as profiles, which makes a modular base for protocol implementation and development. Currently, ROHC supports, e.g., RTP, UDP, UDP-Lite, TCP, ESP, and IP protocols [20, 21, 22]. CoAP compression profile for ROHC was introduced in our earlier work [23]. Fig. 4 presents an example of CoAP/UDP/IPv4 packet header compression, where the headers are compressed from 37 bytes to 5 bytes.

Fig. 4. ROHC header compression with CoAP/UDP/ IPv4 profile.


47

6LoWPAN (IPv6 over low power wireless area networks) compression approach differs from previous ones as being a stateless compression. It uses only known or assumed headers in the compression process and thus there is no need to create decompression tables in the receiver side. This simplifies the compression process and allows stateless compression, which has a clear advantage in lossy communication like wireless communication with resource scare devices. 6LoWPAN is not only about header compression but also involves fragmentation and reassembly. Currently, 6LoWPAN supports only IEEE 802.15.4 devices. Based on the protocol stack, 6LoWPAN header compression involves the following header compression:

• Application layer header compression • Transport layer header compression (TCP,

UDP and ICMP) • Network layer header compression (IPv6

unicast and multicast, routing and other extension headers)

• Adaptation layer header compression (fragmentation, compression, mesh routing and IP routing headers).

6LoWPAN compresses IPv6/UDP headers (48 bytes) to 7 bytes compressed header (CH) and, respectively, IPv6/TCP headers (60 bytes) to 7-31 bytes compressed header, as depicted in Fig. 5.

Fig. 5. 6LoWPAN header compression.

4. IoT Compression Solution with ROHC Our IoT packet compression solution utilizes

ROHC and CoAP. In our earlier study [23], we extended ROHC by presenting CoAP compression profile for ROHC. The main result of our earlier study was that the CoAP packet size could be reduced by over 90 % at best. However, compression and decompression requires processing power, and that means increase in the delay. In our tests on Raspberry Pi (RPi) computer, the compression and decompression took about 1-3 ms depending on the compression state [23]. In the case of WLAN radio, the extra processing for compression and decompression outperformed the energy savings achieved during the shorter transmission time. However, smaller packet size and shorter transmission time can reduce packet loss in lossy links with high bit error rate. That reduces the need

for retransmissions, which improves the energy efficiency. In this paper, we extend our earlier work to find out the performance of ROHC compression with the CoAP profile in LTE and XBEE networks.

4.1. Test Bed and Testing Scenario

The test bed consists of a laptop and Model B+ Raspberry Pi (RPi) computers. A transmission link between RPi and laptop was made with XBEE, LTE and WLAN radios. VTT’s CNL laboratory’s Willab network was used to make it possible to connect RPi to the WLAN or LTE base station. In the XBEE test, the test link was made directly between RPi and laptop. The power consumption was monitored from RPi’s main power supply by using a current probe and an oscilloscope. During the test, the network traffic and power consumption data was logged.

The XBee test bed is shown in Fig. 6(a). The XBee radio shield was connected to RPi’s GPIO port. To make it possible to transfer IP packets through XBee network, the XBee-Tunnel-Daemon (libgbee, http://sourceforge.net/projects/libgbee/) was used. Libgbee is a daemon that creates a virtual network interface allowing transmission of UDP/IP packets over the XBee module.

CoAP client CoAP server

Virtual network interface


ROHC Compressor ROHC Decompressor

CoAP/UDP/IP packet

-profile and flow identification-compression-context update

-profile and flow identication-decompression-context update

CoAP/UDP/IP packet

ROHC packet encapsulated in UDP/IP

packet

RPi Laptop

XBee

libgbee libgbee

Fig. 6 (a). XBee test bed.

In the LTE radio scenario, the radio stick was connected to USB port (see Fig. 6(b)).

CoAP client CoAP server



ROHC Compressor ROHC Decompressor

CoAP/UDP/IP packet

-profile and flow identification-compression-context update

-profile and flow identication-decompression-context update

CoAP/UDP/IP packet

ROHC packet encapsulated in UDP/IP

packet

RPi Laptop

Willab

Fig. 6 (b). LTE test bed.


48

The LTE base station was connected to the Willab network. The WLAN test bed is described in more details in our earlier publication [23]. For the test traffic, we used the same CoAP messages as in our earlier study for WLAN in [23]. 4.2. Header Compression Results

Header compression reduces the size of the transmitted packets. During the transmission, these smaller packets have lower probability to have bit errors, and hence the packet loss will be smaller than for packets with the original size [23]. Thus, packet loss and response time decrease, which together provide better performance. This will realize especially in wireless and low bandwidth links, where bit errors and packet loss are common. The header compression is beneficial when the data amount is small compared to the header part, as illustrated in Fig. 7. The figure presents ROHC compression gain with UDP/IPv4, UDP/IPv6 and "theoretical" packet as the payload increases. In the theoretical packet, it is assumed that all headers (CoAP, UDP, IP, and also MAC header) are compressed to 4 bytes.

Fig. 7. Effect of payload size in CoAP packet for header compression gain with different profiles.

It can be seen that with small payloads the gain is at least 40 %. After 20 bytes of payload, the gain starts to decrease more rapidly. When the payload is over several hundreds of bytes, the advantage of the header compression gradually disappears. However, if the payload was also compressed, the header compression could become beneficial again.

Let's take a more careful look how compression affects to processing delay and radio transmission time in the case of a low bandwidth XBEE link. The whole process consists of packet creating and compression, transmission, and receiving and decompression. The compression and decompression processes create delay compared to uncompressed packet processing, as can be seen in Fig. 8.

Fig. 8. CoAP/UDP/IPv4 packet creating and receiving processes in RPi using XBEE radio.

The delays are 1.5 ms and 4.0 ms in compression and decompression phases, respectively. On the other hand, because of the smaller packets, compression will speed up radio transmission by about 1.2 ms, see Fig. 9. When the delays are calculated together, the total delay remains at 4.3 ms.

Fig. 9. XBEE transmission pulse when using ROHC with CoAP and Uncompressed profile.

Using Fig. 8 and Fig. 9, we can calculate the energy consumption by multiplying the time and the average power consumption during the packet creating & compression, transmission and receiving & decompression processes. Table 1 presents the total energy consumption in the case of CoAP/UDP/IPv4 packets utilising ROHC’s CoAP profile in SO, i.e., full compression state. Even if the time saved in the transmission was shorter than the time spent for compression/decompression processing, the total energy efficiency improved by 15.1 μJ/packet (0.184 %), when the packets were compressed. This is because during the transmission, the power consumption is on a much higher level


49

than during packet processing. The trade-off for saving bandwidth and for lower energy consumption was the extra delay of 4.3 ms.

Table 1. Energy consumption in XBEE using ROHC’s CoAP/UDP/IPv4 compression profile.

Packet processing

Uncomp. packet [ms]

Comp. packet [ms]

Time gap [ms]

Energy gap (µJ)

Creating 26.5 28 -1.5 -14.7 Receiving 22 26 -4 -40.0 Transmission 3 1.8 1.2 68.9 TOTAL -4.3 15.1

As depicted in Table 2, transmission in LTE and WLAN is more than 200 time faster than in XBEE

transmission. Thus, we could not detect the transmission peaks in our measurements, only delays of compression and decompression processing could be found out, see Fig. 10 and Fig. 11.

Table 2. Transmission times with different radios for compressed and uncompressed packet.

Radio Transmission

speed

(Mbit/s)

Transmission time (µs) Uncomp.

packet (736 bits)

Comp. packet

(440 bits) XBEE 0.25 2944 1184 WLAN 802.11b

54 14 8

LTE 100 7 4

Fig. 10. ROHC’s CoAP/UDP/IPv4 compression profile vs. uncompressed profile in WLAN [21].

Fig. 11. ROHC’s CoAP/UDP/IPv4 compression profile vs. uncompressed profile in LTE.

So, in the case of fast speed radios like LTE and WLAN, direct energy savings during transmission will be neglible and the cost of extra compression/decompression processing will rule and the total process will consume more energy than without compression. However, in lossy links, packet loss will decrease when the packet size decreases [24], i.e., when the packet is compressed. This will enhance transmission by avoiding retransmissions, and that way energy could be saved. However, the potential savings depend on how many packet losses actually can be avoided.

5. Discussion and Conclusions

Due to the extensive growth of Internet of Things (IoT), the number of wireless devices connected to the Internet is forecasted to grow to 26 billion units installed in 2020 representing an almost 30-fold increase from 0.9 billion in 2009 [1].

In wireless IoT networks, the available bandwidth and energy is often restricted. That resource can be effectively saved by reducing radio operation: transmission and reception. In the case of IoT devices like sensors, the transmitted data may be only couple of bytes, whereas the protocol headers of the packet (MAC, IP, TCP/UDP, CoAP, etc.) are many tens of bytes. This big header overhead serves as the motivation behind compressing the packet headers and that way decreasing the radio operation times.

The performance of ROHC header compression was studied with a low speed XBEE radio, and high speed LTE and WLAN radios in a real test bed environment. In our previous studies with WLAN radio, we found out that ROHC with CoAP compression profile can decrease the packet size by 90 % or more. The smaller packet size speeds up the radio operation during transmission and reception. This reduces the energy consumption during the radio


50

transmission. Smaller packet will also reduce packet loss in lossy links, which can further improve the energy efficiency. The trade-off for smaller packet size is the delay and extra processing power needed for the compression and decompression.

In the case of XBEE, the energy saving during radio transmission was bigger than energy consumption during compression/decompression processes. The total energy savings were 15.1 μJ/packet (0.184 %). The compression and decompression processes themselves consume roughly ten times less energy than the whole packet creating and receiving processes. However, the delay caused by the compression and decompression was bigger than the saved time during the radio transmission, so the trade-off for smaller packet and energy savings was the extra delay of 4.3 ms.

In the case of high speed radios (LTE and WLAN), the savings in transmission time and energy were neglible compared to the extra processing needed for compression and decompression. Thus, there were no energy savings during individual packet transmissions. However, the smaller packet size will reduce packet loss in lossy links and this way it could be possible to save energy, enhance throughput and decrease delay.

The compression and decompression processes cause quite long delays. They can possible be decreased by having a more optimal software implementation. Reducing the delay caused by the compression/decompression processing automatically improves also the energy efficiency.

ROHC has good compression gain and it is used in current mobile networks, which make it a promising candidate for IoT header compression solution. The disadvantage is the stateful operation and more complexity compared to 6LowPAN. One possible future work item could be to merge 6LowPAN functionalities to ROHC to have some stateless compression at the Initialization and Refresh (IR) state, and perhaps to extend the compression profiles to cover also MAC headers like 802.15.4 and Bluetooth Low Energy.

Acknowledgements

This work was supported by TEKES as part of the Internet of Things program of DIGILE. References [1]. Gartner Says the Internet of Things Installed Base

Will Grow to 26 Billion Units By 2020, (http://www.gartner.com/newsroom/id/2636073).

[2]. P. Koskela, M. Valta, T. Frantti, Energy Efficient MAC for Wireless Sensor Networks, Sensors & Transducers Journal, Vol. 121, Issue 10, October 2010, pp. 133-143.

[3]. Z. Shelby, K. Hartke, C. Bormann, The Constrained Application Protocol (CoAP), Internet Engineering

Task Force (IETF), Request for Comments: 7252, Category: Standards Track, 2014, pp. 1-112.

[4]. P. Koskela, M. Majanen, Robust Header Compression for Constrained Application Protocol, Internet of Things Magazine Finland, No. 1, 2014, pp. 36-39. http://www.internetofthings.fi/extras/IoTMagazine2014.pdf

[5]. Razzaque M. A., Bleakley C., Dobson S., Compression in wireless sensor networks: A survey and comparative evaluation, ACM Transaction on Sensor Networks, Vol. 10, No. 1, 2013, pp. 5-1–5-44.

[6] Shivare M. R., Maravi Y. P. S., Sharma S., Analysis of Header Compression Techniques for Networks: A Review, International Journal of Computer Applications, Vol. 80, No. 5, 2013, pp. 13-20.

[7]. T. Srisooksai, K. Keamarungsi, P. Lamsrichan, Araki K., Practical data compression in wireless sensor networks, Journal of Network and Computer Applications, Vol. 35, No. 1, 2012, pp. 37-59.

[8]. L. Yu, J. Li, S. Cheng, S. Xiong, H. Shen, Secure Continuous Aggregation in Wireless Sensor Networks, IEEE Transactions on Parallel and Distributed Systems, Vol. 25, No. 3, 2014, pp. 762-774.

[9]. S. Roy, M. Conti, S. Setia, S. Jajodia, Secure Data Aggregation in Wireless Sensor Networks: Filtering out the Attacker's Impact, IEEE Transactions on Information Forensics and Security, Vol. 9, No. 4, 2014, pp. 681-694.

[10]. P. Ostovari, J. Wu, A. Khreishah, Network coding techniques for wireless and sensor networks, in The Art of Wireless Sensor Networks, H. M. Ammari (Ed.), Springer, 2013, pp. 1-35.

[11]. B.-N. Cheng, J. Zuena, J. Wheeler, S. Moore, B. Hung, MANET IP Header Compression, in Proceedings of the IEEE Military Communications Conference (MILCOM’13), 2013, pp. 494-503.

[12]. Effnet AB WHITE PAPER Library, An introduction to IP header compression, 2004. (http://www.effnet.com/sites/effnet/pdf/uk/Whitepaper_Header_Compression.pdf).

[13]. V. Jacobson, Compressing TCP/IP Headers, Internet Engineering Task Force (IETF), Request for Comments: 1144, Category: Standards Track, 1990, pp. 1-46.

[14]. M. Degermark, B. Nordgren, S. Pink, IP Header Compression, Internet Engineering Task Force (IETF), Request for Comments: 2507, Category: Standards Track, 1999, pp. 1-47.

[15]. S. Casner, V. Jacobson, Compressing IP/UDP/RTP Headers for Low-Speed Serial Links, Internet Engineering Task Force (IETF), Request for Comments: 2508, Category: Standards Track, 1999, pp. 1-24.

[16]. C. Bormann, C. Burmeister, M. Degermark, H. Fukushima, H. Hannu, L.-E. Jonsson, R. Hakenberg, T. Koren, K. Le, Z. Liu, A. Martensson, A. Miyazaki, K. Svanbro, T. Wiebke, T. Yoshimura, H. Zheng, RObust Header Compression (ROHC): Framework and four profiles: RTP, UDP, ESP and uncompressed, Internet Engineering Task Force (IETF), Request for Comments: 3095, Category: Standards Track, 2001, pp. 1-168.

[17]. G. Montenegro, N. Kushalnagar, J. Hui, D. Culler, Transmission of IPv6 Packets over IEEE 802.15.4 Networks, Internet Engineering Task Force (IETF),


51

Request for Comments: 4944, Category: Standards Track, 2007, pp. 1-30.

[18]. M. Degermark, H. Hannu, L. Jonsson, K. Svanbro, Evaluation of CRTP performance over cellular radio links, in IEEE Personal Communications Magazine, Vol. 7, No. 4, Aug 2000, pp. 20-25.

[19]. Tian Ye, K. Xu, N. Ansari, TCP in wireless environments: problems and solutions, in IEEE Communications Magazine, Vol. 43, No. 3, March 2005, pp. 27-32.

[20]. Pelletier G., Sandlund K., RObust Header Compression Version 2 (ROHCv2): Profiles for RTP, UDP, IP, ESP and UDP-Lite, Internet Engineering Task Force (IETF), Network Working Group, Category: Standards Track, Request for Comments: 5225, 2008, pp. 1-124.

[21]. K. Sandlund, G. Pelletier, L.-E. Jonsson, The RObust Header Compression (ROHC) Framework, Internet

Engineering Task Force (IETF), Category: Standards Track, Request for Comments: 5795, 2010, pp. 1-41.

[22]. G. Pelletier, K. Sandlund, L.-E. Jonsson, M. West, RObust Header Compression (ROHC): A Profile for TCP/IP (ROHC-TCP), Internet Engineering Task Force (IETF), Category: Standards Track, Request for Comments: 6846, 2013, pp. 1-96.

[23]. M. Majanen, P. Koskela, M. Valta, Constrained Application Protocol Profile for Robust Header Compression Framework, in Proceedings of the Fifth International Conference on Smart Grids, Green Communications and IT Energy-aware Technologies (ENERGY’15), Rome, Italy, 24-29 May 2015, pp. 47-53.

[24]. N. Golmie, Coexistence in Wireless Networks: Challenges and System-Level Solutions in the unlicensed band, Cambridge University Press, Cambridge, 2006, pp. 40-41.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Human_Eye.htm


52



Collapse Mode Characteristics of Parallel Plate Ultrasonic Transducer Radiating in Air and Water

1 Rashmi Sharma, Rekha Agarwal and 2 Anil Anil Arora

1 Amity School of Engineering & Technology, New Delhi,110061, India 2 Thapar University, Patiala, 147004, India

1 Tel.: 9958810676 1 E-mail: [email protected]

Received: 6 December 2015 /Accepted: 8 January 2016 /Published: 31 January 2016 Abstract: A 2D finite element analysis of capacitive micromachined ultrasonic transducer (CMUT) is proposed taking into account radiation in air and water. Different CMUT element geometries circular, square and hexagonal have been considered for FEM simulations. FEM simulation software COMSOL is employed to determine the structural deflections caused by electrostatic forces. Since the structural deformation alters the electrostatic field, a coupled-field simulation is required wherein the electrostatic mesh is continuously updated to coincide with the deflection of the structure. In this paper the deflection profile, resonance frequency, material parameters and collapse mode characteristics are being compared with device in air and under water. The CMUT is an electromechanical system, therefore, the physics of electrical and structural mechanics is coupled to describe its dynamics. Maximum frequency of operation is obtained by deriving the time evolution of the device for several frequencies. Copyright © 2016 IFSA Publishing, S. L. Keywords: Ultrasonic transducer, Pull in voltage, Resonant frequency, Collapse mode.

1. Introduction

Capacitive micromachined ultrasonic transducers (CMUT) are a promising alternative to piezoelectric transducers and receive considerable attention due to their advantages such as wider bandwidth, higher sensitivity, ease of array fabrication and integration [1]. Capacitive micromachined ultrasonic transducers (CMUTs) were introduced as micromachined suspended plate structures with a moving top electrode and a rigid substrate electrode [2]. When immersed in a liquid medium, CMUTs are capable of generating wideband acoustical pulses with more than 100 % fractional bandwidth [3]. However, many applications require high transmitted pressures for increased penetration and signal quality. The power output capability of CMUTs can be increased by utilizing the collapsed state of the plates [4-6]. Accurate and fast

simulation methods are necessary for understanding CMUT dynamics and for designing high-performance CMUTs. CMUT models are based either on finite element method (FEM) models [7-8] or on equivalent circuits [9-12]. Precise modeling of capacitive micromachined ultra-sonic transducers (CMUT) is important for an efficient design process.

A CMUT structure comprises a capacitor which consists of two plates in which one of them is fixed and the other can deflect. Electrostatic forces act when a voltage is applied causing deflection of moving plate. The deflection of the movable plate is an important parameter that influences several basic CMUT parameters such as pull-in voltage and capacitance.

In this paper, we simulate the mechanical behaviour of a transmitting CMUT under electrical excitation. The model is dependent on plate




53

dimensions and mechanical properties and it can predict the plate movement in the collapsed state as well as in the uncollapsed state. The simulation results for deflections are compared with CMUT in air and under water for future design reference.

2. Principle of Operation 2.1. Physical Principle

A basic parallel plate capacitor cell made of a thin membrane is shown in Fig. 1.

Fig. 1. Cross sectional view of CMUT.

The electrostatic force generated on the membrane of the capacitor cell is proportional to the square of the applied voltage, the area of the capacitor and the permittivity of the material between the plates, and inversely proportional to the square of the separation between the plates [1]:

2

, (1)

where ε0 is the permittivity of free space, A is the area of the plates, V stands for the applied bias voltage between the plates, d0 is the initial gap height and x is the membrane displacement. Because the electrostatic force is proportional to the square of the bias voltage, linear cMUT operation requires DC bias voltage together with the AC excitation. Then the electrostatic force can be written as

2 2 (2)

The first term in the parenthesis represents the

static force, the second term represents the excitation force proportional to the applied AC voltage, and the last term represents the harmonic contribution of the AC voltage. When the DC bias voltage is much larger than the AC excitation, the harmonic contribution can be ignored. The membrane can be thought of as a mass clamped with a spring that opposes the electrostatic attraction force. The static force on the membrane is balanced by the mechanical restoring force.

, (3)

where k is the spring constant. The minus sign indicates different direction from the electrostatic force, the spring trying to pull upwards. When the sum

of the electrical force and the spring force equals zero, the following expression is obtained:

2 (4)

Expression (4) gives a relation between membrane

displacement x and applied bias voltage V. Pull-in or collapse occurs when dV/dx = 0. Making the calculation and substituting into (3) the pull-in voltage results as:

827

(5)

An important design parameter of a CMUT membrane is the collapse voltage, above which the attractive force can no longer be balanced by the restoring force of the membrane. This collapse voltage determines the operating point of the device. Therefore, it is crucial to calculate the collapse voltage accurately. The membrane of thickness tm is coated with a thin layer of conducting material on the top side, and the bottom electrode is separated from the membrane by a distance ta. The electrical capacitance can be written as:

, (6)

where ε is the dielectric constant of the membrane material, and is A the area of the membrane. The list of parameters used for simulation are listed in Table 1.

Table 1. Parameters used for Simulation.

Parameter Value

Plate Thickness 1 µm

Al Thickness 0.2 µm

Gap Height 1 µm

Insulation Layer 0.2 µm

DC Voltage 80 V

AC Voltage 10 V

2.2. Resonance Frequency

In a CMUT, a membrane is actuated by a time varying input voltage and the vibration of the membrane generates ultrasound waves. The resonance frequency increases with increasing intrinsic stress. At low stress levels the membrane behaves as a plate with the material parameters giving the plates own stiffness and frequency [13]. As intrinsic stress level increases, this stress will dominate over the flexural rigidity. The membrane will operate more as a membrane with no bending stiffness [14]. The wave equation is used for


54

solving the plate resonance frequency as a fourth order partial differential equation:

0, (7)

where x is the deflections, E is the Young’s modulus of the membrane, ν is the Poisson ratio, ρ is the density of the membrane and t is the thickness of the membrane. Solving with boundary conditions no deflection and rigid fastening at the border x(r) = 0 and dx/dr|r = R = 0 gives the resonance frequency of the first mode as:

0.471

(8)

In the plate model, intrinsic model is not

considered. It is only material dimensions and properties which determine the resonance frequency [10].

2.3. Results and Discussion

Resonant frequency calculated for different geometries are being shown in Table 2 with areas. Fig. 2 shows the simulation carried out in COMSOL.

Table 2. Resonant Frequencies for Different Geometries.

Geometry Area Resonant

Frequency Deflection

Circular 3.14 r2 1.31418e6 Hz 0.08 μm Square 2 r2 1.47245e6 Hz 0.083 μm Hexagonal 2.6 r2 1.363994e6 Hz 0.1 μm

Fig. 2 shows that if the area and thickness of the membranes are kept constant and silicon is selected as a membrane material then deflection is 0.1 μm for hexagonal membrane at Eigen frequency of 1.36 MHz, and deflection is 0.08 μm for circular membrane at Eigen frequency of 1.31 MHz.

Fig. 2. Eigen Frequency for Circular, Square and Hexagonal.

In case of square membrane, the deflection is 0.083 μm at the Eigen frequency of 1.41 MHz. In other words, for the same area and same material, deflection is lesser for square membrane and in order to obtain this deflection, it requires a higher frequency as compared to other membranes. Though circular

membrane produces nearly the same deflection at a lower frequency. Also, we can conclude that hexagonal membrane shows maximum deflection at a frequency of 1.36 MHz which is lower than square membrane but in close approximation with the circular membrane having frequency 1.31 MHz.


55

Fig. 3 illustrates the results of displacement versus DC voltage when transducer is placed in air. In this case, pressure of 1 atm acts on the membrane. At 0 V, the deflection of membrane is observed as 0.34 μm in case of circular geometry, 0.31 μm for hexagonal geometry and 0.25 μm for square geometry. Here, CMUT operates in collapse mode as expected and is able to generate and detect ultrasound more effectively than a CMUT operating in conventional mode.

Fig. 4 illustrates the results of displacement versus DC voltage when the transducer is placed in water at a depth of 5 m.

Fig. 3. Displacement with applied Voltage in Air.

Fig. 4. Displacement with Voltage under Water.

At zero volt, the observed deflection is 0.51 μm in case of circular geometry, 0.46 μm in hexagonal geometry and 0.38 μm in square geometry. Thus, we observed that the deflection for the discussed cases is highest for circular geometry. Although, hexagonal geometry is also in close approximation to this.

3. Conclusions

CMUT have been compared with geometries namely circular, square and hexagonal. Resonance

frequency is minimum for Circular and maximum for Square membranes. Maximum displacement is shown by Circular and minimum by Square when device is subjected in air as well as underwater with DC bias. CMUT offers best performance with circular geometry in terms of Eigen frequency, pull in voltage and in maximum deflection as a function of DC bias. But for array formation the results are not satisfactory because of voids, so area is not effectively utilized. So we can use hexagonal geometry for effective utilization of area. The deflection of membrane is more when immersed in water as compared to deflection in air. These results can be considered for further designing of CMUT for various applications. References [1]. I. Ladabaum, X. Jin, H. T. Soh, A. Atalar,

B. T. Khuri Yakub, Surface micromachined capacitive ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 45, Issue 3, 1998, pp. 678-690.

[2]. O. Oralkan, A. Ergun, J. Johnson, M. Karaman, U. Demirci, K. Kaviani, T. Lee, B. Khuri-Yakub, Capacitive micromachined ultrasonic transducers: next-generation arrays for acoustic imaging?, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 49, Issue 11, Nov. 2002, pp. 1596-1610.

[3]. R. O. Guldiken, J. Zahorian, F. Y. Yamaner, F. L. Degertekin, Dual-electrode CMUT with non-uniform membranes for high electromechanical coupling coefficient and high bandwidth operation, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 56, Issue 6, 2009, pp. 1270-1276.

[4]. Ö. Oralkan, B. Bayram, G. G. Yaralioglu, A. S. Ergun, M. Kupnik, D. T. Yeh, I. O. Wygant, B. T. Khuri-Yakub, Experimental characterization of collapse-mode CMUT operation, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 53, Issue 8, 2006, pp. 1513-1523.

[5]. Y. Huang, E. Hæggstrom, B. Bayram, X. Zhuang, A. S. Ergun, C.-H. Cheng, B. T. Khuri-Yakub, Comparison of conventional and collapsed region operation of capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 53, Issue 10, 2006, pp. 1918-1933.

[6]. S. Olcum, F. Y. Yamaner, A. Bozkurt, H. Köymen, A. Atalar, Deep collapse operation of capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 58, Issue 11, Nov 2011, pp. 2475-2483.

[7]. A. Bozkurt, F. L. Degertekin, A. Atalar, B. T. Khuri-Yakub, Analytic modelling of loss and cross-coupling in capacitive micromachined ultrasonic transducers, in Proceedings of the IEEE Ultrasonics Symposium, Vol. 2, 1998, pp. 1025-1028.

[8]. G. G. Yaralioglu, A. S. Ergun, B. T. Khuri-Yakub, Finite-element analysis of capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 52, Issue 12, 2005, pp. 2185-2198.


56

[9]. A. Lohfink, P. C. Eccardt, Linear and nonlinear equivalent circuit modeling of CMUTs, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 52, Issue 12, 2005, pp. 2163-2172.

[10]. A. Caronti, G. Caliano, A. Iula, M. Pappalardo, An accurate model for capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 49, Issue 2, 2002, pp.159-168.

[11]. S. Olcum, M. N. Senlik, A. Atalar, Optimization of the gainbandwidth product of capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 52, Issue 12, 2005, pp. 2211-2219.

[12]. H. Köymen, M. N. Senlik, A. Atalar, S. Olcum, Parametric linear modeling of circular CMUT membranes in vacuum, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 54, Issue 6, 2007, pp. 1229-1239.

[13]. Bayram B., Yaralioglu G. G., Kupnik M., et al., Dynamic analysis of capacitive micromachined ultrasonic transducers, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 52, Issue 12, Dec. 2005, pp. 2270-2275.

[14]. Rashmi Sharma, Rekha Agarwal, Anil Arora, Performance Analysis of MEMS-based

UltrasonicTransducer with Different Membrane Materials, Recent Trends in Sensor Research & Technology, Vol. 1, No. 3, 2014.

[15]. A. Caronti, R. Carotenuto, M. Pappalar, Electromechical coupling factor of capacitive micromachined ultrasonic transducers, J. Acoust. Soc. Am., Vol. 113, No. 1, 2003, pp. 279-288.

[16]. I. O. Wygant, M. Kupnik, B. T. Khuri-Yakub, Analytically calculating membrane displacement and the equivalent circuit model of a circular CMUT cell, in Proceedings of the IEEE Ultrasonics Symposium, 2008, pp. 2111-2114.

[17]. H. K. Oguz, S. Olcum, M. N. Senlik, V. Tas, A. Atalar, H. Köymen, Nonlinear modeling of an immersed transmitting capacitive micromachined ultrasonic transducer for harmonic balance analysis, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 57, Issue 2, 2010, pp. 438-447.

[18]. B. Bayram, Ö. Oralkan, A. S. Ergun, E. Hæggström, G. G. Yaralioglu, B. T. Khuri-Yakub, Capacitive micromachined ultrasonic transducer design for high power transmission, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 52, Issue 2, 2005, pp. 326-339.

___________________








57



Improving Vibration Energy Harvesting Using Dynamic Magnifier

1 Almuatasim Alomari, 1 Ashok Batra and 2 C. R. Bowen

1 Department of Physics, Chemistry and Mathematics (Materials Science Group), Alabama A&M University, Normal (Huntsville), AL 35762 USA

2 Department of Mechanical Engineering, University of Bath, Bath, BA2 7AY, UK

Received: 17 December 2015 /Accepted: 18 January 2016 /Published: 31 January 2016 Abstract: This paper reports on the design and evaluation of vibration-based piezoelectric energy-harvesting devices based on a polyvinylidene fluoride unimorph cantilever beam attached to the front of a dynamic magnifier. Experimental studies of the electromechanical frequency response functions are studied for the first three resonance frequencies. An analytical analysis is undertaken by applying the chain matrix in order to predict output voltage and output power with respect to the vibration frequency. The proposed harvester was modeled using MATLAB software and COMSOL multi-physics to study the mode shapes and electrical output parameters. The voltage and power output of the energy harvester with a dynamic magnifier was 2.62 V and 13.68 μW, respectively at the resonance frequency of the second mode. The modeling approach provides a basis to design energy harvesters exploiting dynamic magnification for improved performance and bandwidth. The potential application of such energy harvesting devices in the transport sector include autonomous structural health monitoring systems that often include embedded sensors, data acquisition, wireless communication, and energy harvesting systems. Copyright © 2016 IFSA Publishing, S. L. Keywords: Harvesting, Piezoelectric, Modeling, Resonance, MATLAB, COMSOL.

1. Introduction

In recent years there has been an increasing interest in employing piezoelectric energy harvesters (PEHs) in applications such as autonomous low-power electronics and wireless sensors [1-3]. Traditionally, the capture of ambient vibrational energy from the surrounding environment is an effective and popular technique to provide small amount of power (μW to mW) for low-energy electronics [4-5]. Piezoelectric harvesters based on cantilever configurations play a prominent role in this area, which convert mechanical energy into electrical energy. The devices can be considered as complex multi-physics systems requiring advanced methodologies to maximize their performance and convert ambient vibrations into useable electrical

power [6]. Piezoelectric materials are attracting interest for applications such as harvesting since they have strong electro-mechanical coupling and a high frequency response [7]. Conventional piezoelectric energy harvester devices operate as linear vibration resonators which are often limited to the capture of low vibration amplitudes that are close to the resonance frequency of the harvester [8]. However, energy harvesters with dynamic magnifier configurations are of interest due to their ability to work at different resonance frequencies as well as magnify the bandwidth of the harvester. One of the simplest configurations of a conventional piezoelectric energy harvester (CPEH) devices is a tip mass attached at the front of a piezoelectric cantilever energy harvester, as shown in Fig. 1 (a).




58

Energy harvester

Harvester massBase excitation

(a)

Magnifier Magnifier beam

Energy harvester

Harvester massBase excitation

(b)

Fig. 1. (a) Conventional Piezoelectric Energy harvester with mass (CPEH); (b) Energy harvester with dynamic

magnifier and mass (EHDM).

In order to enhance the maximum output power and operational bandwidth of piezoelectric energy harvesters, researchers have developed a variety of techniques based on, varying shape of structure beam using an L-shaped flexible structure [9-11], adding an additional impendence between the piezoelectric harvester and load resistance [12-15], using dual-mass systems [16], changing the cross-section of a dynamic magnifier [17] and using an energy harvester with a dynamic magnifier (EHDM) as shown in Fig. 1(b). The use of EHDM has been successful in amplifying the energy harvesting efficiency and widening the bandwidth of the device. A variety of EHDM models have been developed with the purpose of understanding the system and enhancing the device efficiency. Zhou, et al. developed a novel piezoelectric energy harvester with a multi-mode dynamic magnifier, which is capable of significantly increasing the bandwidth and the energy harvested from ambient vibrations [18]. Aladwani, et al. [19] and Aldraihem, et al. [20] have studied the piezoelectric harvester with a dynamic magnifier consisting of a spring-mass both experimentally and analytically where the system was able to amplify the electrical power output and enhanced the bandwidth of the harvester. Lee, et al. have investigated an innovative design platform of a piezoelectric energy harvester termed a segment-type energy harvester experimentally and analytically using finite element method (FEM) where the system showed excellent performance and generated sufficient power to operate a temperature wireless sensor [21]. A bimorph beam with a tip mass attached at the front of a dynamic magnifier was modeled and studied by Vasic, et al. where the numerical results showed that electric power produced by the harvesting beam is amplified for efficient energy harvesting over a

broader frequency range [22]. The potential application of such energy harvesting devices in the transport sector include autonomous structural health monitoring systems in trains or aerospace structures that typically include embedded sensors, data acquisition, wireless communication, and energy harvesting systems.

According to literature, limited research has examined the modeling of a unimorph beam attached at the front end of a magnifier with mass using the chain matrix technique for low electronic devices applications. The reason of choosing the chain matrix technique is that the model enables examination of the various electrical and mechanical boundary conditions by breaking the structure into elementary elements that can be readily adapted to represent different configurations of a dynamic magnifier energy harvester [22]. In this study, a mass was attached at the front of unimorph piezoelectric energy harvester to act as a conventional piezoelectric energy harvester, Fig. 1(a). A mass was then positioned between the copper beam and conventional piezoelectric energy harvester to form a dynamic magnifier energy harvester with mass (EHDM), as in Fig. 1(b). This research examined the influence of attaching a mass at the front of unimorph beam with dynamic magnifier on harvested power and bandwidth of the device. Analytical simulation of the harvesting system was carried out using MATLAB and finite element analysis was utilized for determination of mode shapes using COMSOL.

2. Mathematical Analysis of Dynamic Magnifier Model

2.1. Unimorph Beam

The piezoelectric unimorph beam in this study consists a piezoelectric layer which is sandwiched between two conducting electrodes and positioned on the top of shim layer, as shown in Fig. 2.

Lp x

zhp hs

w

Fig. 2. Piezoelectric unimorph beam.

The length and the width of the unimorph are denoted by Lp and w, respectively. The thickness of the piezoelectric layer and shim layer are hp and hs, respectively. The expressions of the bending moment M and the transverse shear force F relative to the


59

position x and according to the piezoelectric voltage V are written as [22]:

VNx

uKM uu +

∂∂−=

2

2

, (1)

3

3

x

uK

x

MF u ∂

∂−=∂

∂= , (2)

where Ku is the effective bending rigidity of the entire unimorph beam, Nu is the constant of the electro-mechanical conversion.

Both previous quantities are defined for the unimorph beam in Fig. 2 mathematically as [23-25]:

( ))2(3412

1 33pspssppssu hhhhEhEhEwK +++= , (3)

( )psp

spu hh

h

hEwdN 2

831 +−= , (4)

where Es, Ep are the Young's modulus of the shim layer and piezoelectric material, respectively, d31 is piezoelectric strain coefficient of the piezoelectric material.

The equation of the motion of the elastic unimorph beam is given as:

0)(2

2

4

4

=∂∂++

∂∂

t

uhhw

x

uK ppssu ρρ , (5)

where ρs, ρp are the densities of the shim and piezoelectric layer, respectively. By considering a harmonic mode, the general solution of displacement in Eq. (5) can be written as [26]:

xxxxu uuuu λαλαλαλα sinhcoshsincos 4321 +++= , (6) where 41 αα − are the coefficients determined from

the boundaries conditions and uλ is the wave

number:

4/12

= ωρλ

u

uu K

, (7)

where ω is the vibration frequency and uρ is the

linear mass:

)( ppssu hhw ρρρ += (8)

In order to model this element, we have

established a relationship between the bending moment denoted M1, the rotational velocity Φl, the transverse shear force F1, and the vertical velocity U1 at the position x=0 according to these same quantities (M2, Φ2, F2, U2) at the position x=Lp as shown in Fig. 3.

Fig. 3. Bending moments M1 M2, transverse shear forces F1 F2, vertical velocities U1 U2 and rotational velocities

Φ1Φ2 at the ends. Same here – narrow arrow ends otherwise it dominated the image.

By taking into account the following boundary conditions of the bending moments and the transverse shear efforts at the ends:

01 == xFF and 01 == xMM (9)

pLxFF =−=2 and pLxMM =−=2 (10)

Velocities at the ends of the beam corresponding

to linear displacements and rotation are defined by [27]:

01 =∂∂= xt

uU and 0

2

1 =∂∂∂−=Φ xxt

u (11)

pLxt

uU =∂

∂=2 and pLxxt

u=∂∂

∂−=Φ2

2 (12)

matrix relations can be establish as shown in Eq. (13) and Eq. (14).

−−−

−−=

Φ

Φ

4

3

2

1

2

2

1

1

00

0101

αααα

λλλλ

λλω

uuuuuuuu

ubub

uu

mncs

nmscj

U

U

(13)

and

−−−−

−−

=

4

3

2

1

2222

3333

22

33

2

2

1

1

00

00

αααα

λλλλλλλλ

λλλλ

ω

uuuuuuuu

uuuuuuuu

uu

uu

nmsc

mncsj

M

F

M

F

, (14)

where puu Lc λcos= , puu Ls λsin= ,

puu Lm λcosh= , puu Ln λsinh= .

By eliminating the coefficient α1 to α4, we establish a 4×4 matrix relationship (Eq. (15)) connecting the efforts and velocities of one end of the beam (x = 0) to the other (x = LP ) including the voltage V.


60

Φ1

1

1

1

U

M

F

= pLΓ

Φ

−−

2

2

2

2

U

M

F

+ VpLΠ , (15)

where pLΓ is called the 4×4 chain matrix, defined by:

+−+−−

++

−−−

−−−+

+−

−+−−+

=Γ

uuuuuuu

uu

uu

uu

u

uuuu

uu

uu

uu

uu

uuuuuuuuuu

u

uu

uuuuuuuuuuuuu

L

cmsnK

snj

K

cmj

sncm

K

cmj

K

snjj

snK

j

cmKcm

snj

cmK

j

snKsncm

p

)()()(

)()()(

)()()(

)()()(

2

1

2

23

2

23

λλ

ωλ

ωλλ

ωλ

ωω

λω

λλ

ωλ

ωλλ

(16)

and ΠLp is a matrix 1×4 defined by:

+

−

+−

−

=Π

uu

uuu

uu

uuu

uuuu

uuuu

L

K

snNjK

cmNj

cmNN

snN

p

λω

λω

λ

2

)( 2

)( 2

)(2

)(

2

(17)

2.2. Magnifier Beam

For a magnifier beam (shown in Fig. 1), the effective bending rigidity Km can be written as:

11

3

12 m

mm s

whK = (18)

The equation of the motion of the magnifier beam is given as:

02

2

41

4

=∂∂+

∂∂

t

uhw

x

uK mmm ρ (19)

By considering a harmonic mode, the relation

connecting the bending moments M1 M2, transverse shear forces F1 F2 and velocities of one end of the beam to the other end is obtained in a similar way to the previous case as in Eqs. (9-14) but the electro-mechanical constant is not used.

Φ1

1

1

1

U

M

F

= mLΓ

Φ

−−

2

2

2

2

U

M

F

, (20)

where mLΓ is the 4×4 chains matrix, defined by:

+−+−−

++−−−

−−−++−

−+−−+

=Γ

LLLLmmm

LL

mm

LL

m

LLLL

mm

LL

mm

LL

LLmmLLmmLL

L

LL

LLmmLLmmLLmLL

L

cmsnK

snj

K

cmj

sncm

K

cmj

K

snjj

snK

j

cmKcm

snj

cmK

j

snKsncm

m

)()( )(

)()( )(

)()()(

)()()(

2

1

2

23

2

23

λλ

ωλ

ωλλ

ωλ

ωω

λω

λλ

ωλ

ωλλ

(21)

where mmL Lc λcos= , mmL Ls λsin= ,

mmL Lm λcosh= , mmL Ln λsinh= .

4/1

2

= ωρλ

m

mm K

(22)

and the linear mass density:

mmm hwρρ = (23)

3. Composition of the Model

In the general case, the determination of the complete model of a particular device structure is undertaken by assembling simple elements that are


61

characterized by their chain matrix and their electro-mechanical conversion matrix thereby revealing the electrical variables (V and I). Thus, if we consider the particular case of the structure represented in Fig. 1, the separation of the system into simple elements is made in Fig. 4. The chain relation of each element of the harvester is:

Section 1:

Φ1

1

1

1

U

M

F

= LmΓ

Φ

−−

21

21

21

21

U

M

F

, (24)

Section 2:

Φ22

22

22

22

U

M

F

= LpΓ

Φ

−−

3

3

3

3

U

M

F

+ VpLΠ , (25)

where ΓLm is the chain matrix of the magnifier, ΓLp is the chain matrix of the energy harvesting beam with piezoelectric elements and the mass m1 at the end of the magnifier and mass m2 at the end of the harvester beam are taken into account with chain matrices Γm1 and Γm1, respectively.

Fig. 4. Simple cutting elements of the energy harvester with the dynamic magnifier.

Φ

Γ=

Φ

−−

22

22

22

22

1

21

21

21

21

U

M

F

U

M

F

m ,

Φ

Γ=

Φ

−−

3

3

2

3

3

3

3

0

0

UU

M

F

m (26)

The expression of the two mass matrices is:

=Γ

0000

0100

0010

0 01 1

1

mj

m

ω

,

=Γ

0000

0100

0010

0 01 2

2

mj

m

ω

(27)

Considering that the excitation at the fixed end of the harvester is a sinusoidal excitation, Aext, the

boundary conditions at the embedded end: ωj

AU ext=1

and Φ1 = 0, and at the free end: F3 = 0 and M3 = 0. The continuity of the mechanical quantities at the junction of the sections enables the matrix relation of the complete energy harvester with magnifier to be:

0

1

1

ωjA

M

F

ext = LmΓ 1 mΓ LpΓ 2mΓ

Φ3

3

0

0

U+ LmΓ 1 mΓ VpLΠ , (28)

=Γhar LmΓ 1mΓ LpΓ 2mΓ =

ΓΓΓΓ

harhar

harhar

43

21

``

``, (29)

harΠ = LmΓ 1mΓ pLΠ =

ΠΠ

har

har

2

1 (30)

We can extract two matrices relations Eq. (31)

and Eq. (32) from Eq. (28):

1

1

M

F= har

2Γ

Φ3

3U+ Vhar

1Π , (31)


62

0 ωjAext

= har4Γ

Φ3

3U+ Vhar

2Π (32)

This second relation Eq. (32) enables the

velocities at the end of the harvester beam to be calculated:

Φ3

3U= ( ) 1

4−

Γ har −

0 ωjAext

( ) 14

−Γ har Vhar

2Π (33)

The velocity Φ3 can be expressed as:

3Φ = ( ) 211

4 ]`[−

Γ har −ωj

Aext ( ) Vharhar22

14 ]`[ ΠΓ

−, (34)

3Φ = BVj

AA ext −

ω, (35)

where ( ) 21 1

4 ]`[−

Γ= harA and ( ) 2 21

4 ]`[ harharB ΠΓ=−

3.1. Electrical output Parameters

For calculation of the electrical output power of the energy harvester with a magnifier, firstly the electric current I is related to electric displacement D3:

∂∂=

A

edAt

DI 3 , (36)

where Ae is the area of the electrodes. The integration of Eq. (38) gives Eq. (39) [26]:

)( 322 Φ−Φ−= up NVCjI ω , (37)

where Cp is the internal electrode capacitance of the piezoelectric layers, and is given by [27]

p

pT

p h

wLC 33ε

= (38)

The boundary condition imposed leads to Φ22 = 0. If we consider a resistive load R connected at the electrodes of the piezoelectric layers, the voltage V can be expressed as:

31Φ

+−=

pRCj

RNV

ω (39)

Eq. (35) provides an additional relationship between voltage and velocity Φ3, thus relation Eq. (39) can be expressed as:

ωω j

A

BRNRCj

ARNV ext

bo

b

−+−=

1 (40)

Finally, the expression of the output power is:

R

VVP

*

= (41)

4. Results and Discussion

4.1. Experimental Design Results and Analysis

In this section, we present both theoretical and experimental results of:

1) A conventional piezoelectric energy harvester with mass;

2) An energy harvester with dynamic magnifier and mass (EHDM).

Energy harvesting measurements were carried out initially by attaching a mass at the front of a commercial polyvinylidene fluoride (PVDF) unimorph cantilever beam in the configuration shown in Fig. 1(a). The second configuration involved attaching the polyvinylidene fluoride (PVDF) unimorph cantilever with mass at the front of copper cantilever beam with magnifier end-mass, as shown in Fig. 1(b). Both devices were then connected to a shaker system. The dimensions, electrical, mechanical properties of cantilever beams and magnifier end mass are shown in Table 1. A parametric study was undertaken using the experimental setup shown in Fig. 5.

Fig. 5. Schematic diagram of the experimental set-up.

Fig. 6 shows the frequency response functions of output voltage and average output power for a CPEH and EHDM. It can be shown from Fig. 6 that there are three mode shapes of each proposed system, the blue circle line which represents the experimental results of CPEH has one peak at each mode shape, while the red circle line which represents the experimental results EHDM has two peaks at each


63

mode shape. The output voltage and the average output power of EHDM is larger than the CPEH and reached the maximum value at resonance frequencies of the first three modes. The energy harvesting bandwidth at every resonance frequency of the harvesting beam is widened in the EHDM configuration. It is convenient here to mention that the values of resonance frequency at 31 Hz and at 43 Hz will be of particular interests in this study due to high electrical output values. Fig. 7 and Fig. 8 show the experimental and theoretical data of the output voltage and average power for the conventional piezoelectric energy harvester and the EHDM, respectively; data is at a load resistance of 500 kΩ. It can be observed that the output voltage and average power in both cases have a good agreement between experiment and chain modeling.

The resonance frequencies, output voltage, and average power of the first three modes for the CPEH and the EHDM from the graphs in Fig. 7 and Fig. 8 are summarized in Table 2.

Fig. 9 shows the output voltage and average power for EHDM at a range of load resistances, RL, from 100 Ω to 10 MΩ. It can be seen from Fig. 9(a) that the output voltage increases with increasing RL and the resonance frequency of the EHDM harvester depends on the external load resistance RL. Moreover, the output power has a good agreement between experiment and modeling with an optimum power at the condition 2π⋅f⋅RL⋅Cp=1, where Cp is capacitance of the unimorph cantilever beam element (2.7 nF). The maximum peak of power can be observed between 300-600 kΩ; see Fig. 9(b).

Table 1. Properties of cantilever beams investigated. Magnifier end mass=5 g, unimorph end mass=1 g.

Type of beam Magnifier beam Piezoelectric unimorph cantilever beam

Copper PVDF layer Shim layer

Length (m) Lm=0.150 Lp=0.052 Ls=0.041

Width (m) wm=0.016 wp=0.016 we=0.016

Thickness (μm) hm=250 hp=48 hs=18

Young's modulus (GPa) Em=117 Ep=3 Es=5

Density (kg/m3) ρm=8960 ρp=1780 ρs=1820

Dielectric constant T33ε =12 -

Piezo strain coefficient (C/N) d31=-23×10-12 -

Capacitance (nF) Cp=2.7 -

Frequency (Hz)

0 20 40 60 80 100 120 140

Ou

tpu

t vo

ltag

e (

V)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

With magnifierWithout magnifier

Frequency (Hz)

0 20 40 60 80 100 120 140

Ou

tpu

t p

ow

er

( μW

)

0

2

4

6

8

10

12

14

16

With magnifierWithout magnifier

(a)

(b)

Fig. 6. Experimental data of CPEH and EHDM at load resistance 500 kΩ of (a) output voltage,

and (b) Output power versus frequency.


64

Frequency (Hz)

0 20 40 60 80 100 120 140

Ou

tpu

t vo

ltag

e (V

)

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

ExperimentModel

Frequency (Hz)

0 20 40 60 80 100 120

Ou

tpu

t p

ow

er (

μW)

0

1

2

3

4

5

ExperimentModel

(a)

(b)

Fig. 7. Experimental and theoretical data of (a) output voltage,

and (b) output power versus frequency of CPEH at load resistance 500 kΩ.

Frequency (Hz)

0 20 40 60 80 100 120 140

Outp

ut vo

ltage

(V)

0.0

0.5

1.0

1.5

2.0

2.5

3.0

ExperimentModel

Frequency (Hz)

0 20 40 60 80 100 120 140

Outp

ut pow

er

( μW

)

0

2

4

6

8

10

12

14

16

ExperimentModel

(a)

(b)


and (b) output power versus frequency of EHDM at load resistance 500 kΩ.

Load Resistance (×106Ω)

0 2 4 6 8 10 12

Outp

ut vo

ltag

e (V

)

0

1

2

3

4

5

1st peak EXP

2nd peak EXP

1st peak THE

2nd peak THE

Load Resistance (×106Ω)

0 2 4 6 8 10 12

Outp

ut pow

er (

μW)

0

1

2

3

4

5

6

1st peak EXP

2nd peak EXP

1st peak THE

2nd

peak THE

(a)

(b)


and (b) output power versus load resistance of EHDM at second mode shape.


65

Table 2. Resonance frequency, voltage and power output parameters of CPEHM and EHDM.

Cantilever beam type and mode shape

Resonance frequency (Hz) Output voltage (V) Output power (μW)

EXP. MAT. Error (%) EXP. MAT. Error (%) EXP. MAT. Error (%)C

PE

HM

First Mode 6.02 3.57 40.7 0.23 0.22 4.34 0.12 0.11 8.33

Second Mode

37.42 36.90 1.39 1.32 1.30 1.51 3.48 3.39 2.58

Third Mode

117.1 117.4 0.25 0.28 0.27 3.57 0.18 0.14 22.22

EH

DM

First Mode 5.02 5.24 4.38 0.45 0.29 35.55 0.52 0.18 65.38

8.69 8.96 3.11 0.82 0.53 35.36 1.35 0.43 68.15

Second Mode

31.04 30.64 1.28 2.62 2.45 6.48 13.68 11.88 13.16

42.72 43.76 2.43 1.92 1.81 5.72 7.47 6.54 12.45

Third Mode

95.04 95.64 0.63 0.35 0.25 28.57 0.32 0.11 65.62

107.3 109.2 1.77 0.33 0.25 24.24 0.27 0.18 33.33

4.2. Simulation and Analysis of EHDM Using MATLAb and COMSOL

By using Eq. (6) the displacement u(x, t) of the

uniphorm beam with magnifier can be obtained. The first three modes of CPEH system are shown in Fig. 10. Fig. 10(a) shows the theoretical results of model where the first three mode shapes happens approximately at 3.6 Hz, 37.5 Hz and 117.7 Hz, respectively. Fig. 10(b) shows the modeling results of CPEH using FEM where the first three mode shapes happens approximately at 9.3 Hz, 37.4 Hz and 108.4 Hz, respectively. Fig. 11 shows the theoretical results of EHDM using MATLAB R2015a, this figure indicates clearly that the dynamic magnifier energy harvester has an additional mode of vibration appearing at each resonance mode. For each mode, the first resonance (red line) is primarily due to the magnifier and the second resonance (blue line) is due to the unimorph beam [22].

For the more general case, FEM analysis is performed in COMSOL Multiphysics in two dimensions to investigate the resonance frequencies and modes of vibration of EHDM structure. As it is known the power density would be maximum when the vibration frequency matches the resonant frequency of piezoelectric harvester. A very fine mesh is used for generating accurate results of EHDM system. The natural frequencies and the corresponding mode shapes of EHDM of the first six mode shapes are shown in Fig. 12. The resonance frequency of each vibration mode in Fig. 10 – Fig. 12 are summarized in Table 3.

The columns in red line are represented an error between the measured, MATLAB and COMSOL data as can be seen in general, the percentage error in COMSOL or FEM is less than that calculated by MATLAB.

0.00 0.01 0.02 0.03 0.04 0.05 0.06-1.0

-0.5

0.0

0.5

1.0

1.5

2.0

1st mode frequency=3.57 Hz

2nd mode frequency=37.54 Hz

3rd mode frequency=117.66 Hz

(a)

(a)

(b.1)

(b.2)

(b.3)

Fig. 10. The first three mode shapes of CPEH obtained from the (a) theoretical analysis (MATLAB R2015a), where the first mode is 3.57 Hz, second mode is 37.54 Hz, and third mode is 117.66 (b) COMSOL software, where (b.1) the first mode is 9.29 Hz, (b.2) second mode is 37.44 Hz, (b.3) third mode is 108.36.


66

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

(a)

(b)

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

(c)

(d)

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

0.00 0.05 0.10 0.15 0.20 0.25-1.0

-0.5

0.0

0.5

1.0

(e) (f)

Fig. 11. The first six mode shapes of the EHDM structure obtained from the theoretical analysis (MATLAB R2015a), where the first resonance is mainly due to the magnifier and the second resonance is mainly due to the harvesting beam: (a) magnifier (fn = 3.58 Hz); (b) beam (fn = 6.81 Hz); (c) magnifier (fn = 30.95 Hz); (d) beam (fn = 42.69 Hz); (e) magnifier (fn = 95.08 Hz); (f) beam (fn = 119.54 Hz).

(a)

(b)

(c)

(d)

(e) (f)

Fig. 12. The first six mode shapes of the EHDM structure obtained from the FEM (COMSOL 5.0), where the first resonance is mainly due to the magnifier and the second resonance is mainly due to the harvesting beam: (a) magnifier (fn = 4.07 Hz); (b) beam (fn = 7.65 Hz); (c) magnifier (fn = 30.99 Hz); (d) beam (fn = 44.45 Hz); (e) magnifier (fn = 95.21 Hz); (f) beam (fn = 130.76 Hz).

5. Conclusions

This paper has examined the effect of attaching a piezo-unimorph beam with tip mass on the output electrical parameters and bandwidth of a CPEH and EHDM at both an experimental and theoretical level. The experimental results show maximum output voltage generated by a conventional energy harvester

at single frequency value of 37 Hz, where the maximum output voltage generated by EHDM are in frequency range of 31 Hz to 42 Hz. This result highlights the higher bandwidth of the EHDM compared to the CPEH and overlapping of modes of resonance of the system. The analytical model, which predicts the electrical parameters of this system is based on dividing the structure into elementary parts,


67

all these parts and proof mass have been written in the form of matrix representation with good agreement with the experimental data. The theoretical results of mode shapes of CPEH and EHDM have obtained and validated with

MATLAB software and the FEM (COMSOL). The modeling approach provides a basis to design energy harvesters exploiting dynamic magnification for improved performance.

Table 3. Experimental, MATLAB and COMSOL resonance frequency results of EHDM.

Cantilever beam type and mode

shape

Resonance frequency (Hz)

EXP. MATLAB Error (%) COMSOL Error (%)

CP

EH

M

First Mode

6.02 3.57 40.69 9.29 54.31

Second Mode

37.42 37.54 0.32 37.44 0.05

Third Mode

117.1 117.66 0.48 108.36 7.46

EH

DM

First Mode

5.02 3.58 28.68 4.07 18.92 8.69 6.81 21.63 7.65 11.96

Second Mode

31.04 30.95 0.29 30.99 0.16 42.72 42.69 0.07 44.45 4.05

Third Mode

95.04 95.08 0.04 95.21 0.18 107.3 119.54 11.40 130.76 21.86

Acknowledgements

The authors gratefully acknowledge support for this work through the National Science Foundation grant #EPSCoR R-II-3 (EPS-1158862). Authors thank Dr. Chance M. Glenn, Dean, College of Engineering, Technology and Physical Sciences and Dr. M. D. Aggarwal, Chairman, Department of Physics, Chemistry and Physics for their keen interest in this work. Authors thank Mr. Garland Sharp for fabrication of the sample holders. References [1]. J. Paradiso, T. Starner, Energy scavenging for mobile

and wireless electronics, IEEE Pervasive Computing, Vol. 4, Issue 1, 2005, pp. 18-27.

[2]. H. Yu, J. Zhou, L. Deng, Z. Wen, A Vibration-Based MEMS Piezoelectric Energy Harvester and Power Conditioning Circuit, Sensors, Vol. 14, Issue 2, 2014, pp. 3323-3341.

[3]. X. Zhang, Active Omni-directional Piezoelectric Energy Harvesting System for Wireless Monitoring on Electrical Traction Shearer, International Journal of Smart Home, Vol. 8, Issue 3, 2014, pp. 19-32.

[4]. D. Shen, J.-H. Park, J. H. Noh, S.-Y. Choe, S.-H. Kim, H. C. Wikle, D.-J. Kim, Micromachined PZT cantilever based on SOI structure for low frequency vibration energy harvesting, Sensors Actuators A: Physical, Vol. 154, Issue 1, 2009, pp. 103-108.

[5]. M. Ferrari, V. Ferrari, M. Guizzetti, D. Marioli, A. Taroni, Piezoelectric multifrequency energy converter for power harvesting in autonomous microsystems, Sensors Actuators A: Physical, Vol. 142, Issue 1, 2008, pp. 329-335.

[6]. C. R. Bowen, H. A. Kim, P. M. Weaver, S. Dunn, Piezoelectric and ferroelectric materials and

structures for energy harvesting applications, Energy & Environmental Science, Vol. 7, Issue 1, 2014, pp. 25-44.

[7]. J. Qiu, H. Ji, The Application of Piezoelectric Materials in Smart Structures in China, International Journal of Aeronautical and Space Sciences, Vol. 11, Issue 4, 2010, pp. 266-284.

[8]. H. Wu, L. Tang, Y. Yang, C. K. Soh, A novel two-degrees-of-freedom piezoelectric energy harvester, Journal of Intelligent Material Systems and Structures, Vol. 24, Issue 3, 2012, pp. 357-368.

[9]. Y. Ting, G. Hariyanto, B. K. Hou, S. Ricky, S. Amelia, C.-K. Wan, Investigation of Energy Harvest and Storage by Using Curve-shape Piezoelectric Unimorph, in Proceedings of the IEEE International Symposium on Industrial Electronics (ISlE), Seoul, South Korea, 2009, pp. 2047-2052.

[10]. J. W. Xu, W. W. Shao, F. R. Kong, Z. H. Feng, Right-angle piezoelectric cantilever with improved energy harvesting efficiency, Applied Physics Letters, Vol. 96, Issue 15, 2010, 152904.

[11]. A. Erturk, J. M. Renno, D. J. Inman, Modeling of Piezoelectric Energy Harvesting from an L-Shaped Beam-Mass Structure with an Application to UAVs, Journal of Intelligent Material Systems and Structures, Vol. 20, Issue 5, 2009, pp. 529-544.

[12]. N. A. Kong, D. S. Ha, A. Etrurk, D. J. Inman, Resistive impedance matching circuit for piezoelectric energy harvesting, Journal of Intelligent Material Systems and Structures, Vol. 21, Issue 13, 2010, pp. 293-302.

[13]. J. Liang, W.-H. Liao, Impedance matching for improving piezoelectric energy harvesting systems, in Proceedings of the SPIE Int. Conf. Active and Passive Smart Struct. Intell. Syst., San Diego, CA, USA, Vol. 7643, 2010.

[14]. W. J. Wu, Y. Y. Chen, B. S. Lee, J. J. He, Y. T. Pen, Tunable resonant frequency power harvesting devices, in Proceedings of the SPIE Int. Conf. Smart Struct. Mater. Damping and Isolation., San Diego, CA, USA, Vol. 6169, 2006.


68

[15]. A. Badel, D. Guyomar, E. Lefeuvre, C. Richard, Piezoelectric energy harvesting using a synchronized switch technique, Journal of Intelligent Material Systems and Structures, Vol. 17, Issue 8-9, 2006, pp. 831-839.

[16]. X. Tang, L. Zuo, Enhanced vibration energy harvesting using dual-mass systems, Journal of Sound and Vibration, Vol. 330, Issue 21, 2011, pp. 5199-5209.

[17]. V. K. Sharma, K. Srikanth, A. K. Viswanath, Influence of cross-sectional area of a dynamic magnifier for vibration energy harvesting, International Journal of Mechanical and Production Engineering, Vol. 2, Issue 5, 2014, pp. 82-85.

[18]. W. Zhou, G. R. Penamalli, L. Zuo, An efficient vibration energy harvester with a multi-mode dynamic magnifier, Smart Materials and Structures, Vol. 21, Issue 1, 2012, 015014.

[19]. A. Aladwani, M. Arafa, O. Aldraihem, A. Baz. Cantilevered piezoelectric energy harvester with a dynamic magnifier, Journal of Vibration and Acoustics, Vol. 134, Issue 3, 2012, pp. 031004-1–031004-10.

[20]. O. Aldraihem, A. Baz, Energy harvester with a dynamic magnifier, Journal of Intelligent Material Systems and Structures, Vol. 22, Issue 6, 2011, pp. 521-530.

[21]. S. B. Lee, B. D. Youn, B. C. Jung, Robust Segment-Type Energy Harvester and Its Application to a Wireless Sensor, Smart Materials and Structures, Vol. 18, Issue 9, 2009, pp. 095021-1–095021-12.

[22]. D. Vasic, F. Costa, Modeling of Piezoelectric Energy Harvester with Multi-mode Dynamic Magnifier with Matrix Representation, International Journal of Applied Electromagnetics and Mechanics, Vol. 43, Issue 3, 2013, pp. 237-255.

[23]. S.-L. Kok, Design, Fabrication and Characterization of Free Standing Thick-Film Piezoelectric Cantilevers for Energy Harvesting, University of Southampton, 2013.

[24]. H. Porwal, Piezoelectric Transduction Mechanism for Vibration Based Energy Harvesting, Master Thesis, National Institute of Technology, Rourkela, India, 2013.

[25]. R. Patel, Modelling, Analysis and Optimization of Cantilever Piezoelectric Energy Harvesters, Doctoral Dissertation, University of Nottingham, Nottingham, 2012.

[26]. J. R. Davidson, C. Mo, Piezoelectric Energy Harvesting with Frequency Tuning for Ventilation System Monitoring, International Journal of Engineering Science and Innovative Technology, Vol. 2, Issue 5, 2013, pp. 114-124.

[27]. Y. S. Cho, Y. E. Pak, C. S. Han, S. K. Ha, Five-port equivalent electric circuit of piezoelectric bimorph beam, Sensors Actuators A: Physical, Vol. 84, Issue 1-2, 2000, pp. 140-148.

[28]. A. Erturk, Electromechanical Modeling of Piezoelectric Energy Harvesters, PhD Thesis, Virginia Polytechnic Institute and State University, 2009.

___________________


http://www.sensorsportal.com/HTML/E-SHOP/PRODUCTS_4/USTI.htm






69



Design and Implement of Pyroelectric Energy Harvester Experimental Measurement System

Based on STM32F103VET6

1 Honghua Liao, 1 Hao Fu, 1 Weichuang Yu, 1 Binbin Zhou, 2 Ting Yu and 1 Yongdan Zhu

1 School of Information Engineering, Hubei University for Nationalities, Enshi Hubei 445000, China 2 School of Optical and Electronic Information, Huazhong University of Science and Technology,

Wuhan Hubei 430074, China E-mail: [email protected], [email protected], [email protected],

[email protected], [email protected], [email protected]

Received: 4 December 2015 /Accepted: 4 January 2016 /Published: 30 January 2016 Abstract: A design scheme of the pyroelectric energy harvesting and conversion experimental measurement system was proposed. It can be used to evaluate the harvesting and conversion properties of pyroelectric energy harvester. The STM32F103VET6 microcontroller is used as control core, and the variation temperature filed can be generated by using semiconductor chilling plate, which is controlled by Fuzzy-PID temperature module, to achieve the temperature rapid heating or cooling. In system, using DS18B20 to achieve the temperature real-time data acquisition, using TF TLCD 2.8 LCD to set the initial parameters and display the related real-time parameters. The measurement principle, system hardware and software architecture of experiment measurement system are specially described in article. Simultaneously, the energy conversion and harvesting properties of novel pyroelectric energy harvester, which is prepared by Pb[(MnxNb1-x)1/2(MnxSb1-x)1/2]y(ZrzTi1-z)1-yO3 (PMnN-PMS-PZT) ceramics, is discussed. The experiment results show that it can quickly realize the system temperature heating or cooling, and also can satisfy the experiment needs of the properties of pyroelectric energy harvesting and conversion characteristics measurement. Copyright © 2016 IFSA Publishing, S. L. Keywords: STM32F103VET6, Pyroelectric energy harvester, Semiconductor chilling plate, Fuzzy PID controlling, Pb[(MnxNb1-x)1/2(MnxSb1-x)1/2]y(ZrzTi1-z)1-yO3 (PMnN-PMS-PZT). 1. Introduction

In recent years, the quantitative and theoretical study of the pyroelectric effect aim to different materials is increasing day by day [1-5]. In particular, the micro energy harvesting, conversion, and storage research of the pyroelectric effect based on ceramic materials has become a research hotspot in the micro-energy development field of environmental [6-8]. In the research of the pyroelectric energy harvesting and

conversion, how to effectively study the relationship of the quantity of free electric charges by the pyroelectric device generated, the output voltage and pyroelectric coefficients of pyroelectric materials, the surface area of device, and the induced thermal frequencies, it has very important significance to the development and analysis of Pyroelectric energy harvester. Currently, there is little to establish a mature and effective method for the measurement of pyroelectric materials, and the special devices for




70

measuring the particular correlation parameters are less. In order to evaluate the energy harvesting and conversion properties of pyroelectric devices, therefore, a design scheme of the experimental measurement system for the pyroelectric device based on dynamic current method was proposed, and the energy conversion and harvesting properties of the novel pyroelectric devices, which is prepared by Pb[(MnxNb1-x)1/2(MnxSb1-x)1/2]y(ZrzTi1-z)1-yO3 (PMnN-PMS-PZT), is studied based on the experimental measurement system. 2. The Harvesting and Conversion

Principle of Pyroelectric Energy Harvester The charge can be generated on the pyroelectric

energy harvester surface with the temperature varying. The reason is that the intensity of spontaneous polarization will be changed with the temperature field variation [9-10]. That is to say, the harvesting and conversion of pyroelectric energy harvester is based on the principle of pyroelectric effect.

Pyroeledtric energy harvester, which can directly convert thermal energy to electrical energy, is a kind of pyroelectric device. Generally, the intensity of pyroelectric effect can be expressed by the pyroelectric coefficient, λ. And the pyroelectric coefficient, λ, can be defined as:

/sdP dTλ = ,

(1)

where Ps is the spontaneous polarization, T is the temperature.

By the polarization treatment, the spontaneous polarization of pyroelectric samples will be changed with the temperature field variation. When the output electrodes of Pyroeledtric energy harvester are connected into the measuring circuit, the pyroelectric

current, pi , can be is calculated as fellow:

/ / /

/

p s si S dP dt S dP dT dT dt

S dT dtλ= × = × ×

=, (2)

where S is the electrode area of pyroelectric samples, dT/dt is the temperature variation rate.

The principle of capacitance stored energy about Pyroelectric energy harvester is shown in Fig. 1.

The sample of Pyroelectric energy harvester can be equivalent to a resistance-capacitance parallel (Rx-Cx) circuit, and the charge on the electrode can be regarded as a current source, which can be expressed as a parallel connection of Cx and Rx. When the temperature increases (i.e. / 0dT dt > ), the load capacitance, CL, will be charged. When the temperature decreases (i.e. / 0dT dt > ), the load capacitance, CL, will be discharged. The stored

energy of capacitor can be calculated by the

formula 2 / 2E CU= .

Fig. 1. The capacitor energy storage measuring principle of Pyroelectric energy harvester.

3. The Hardware System of Pyroelectric Energy Harvester

3.1. The Module of Experimental

Measurement System and its Working Principle

The main functions of Pyroelectric energy

harvester experimental measurement system include realizing the temperature field variation by the specific law, quick real-time measurement of the temperature signal, the output-voltage signal sampling of pyroelectric energy harvester, etc. The module of experimental measurement system mainly includes pyroelectric energy harvester, semiconductor chilling plate, STM32F103VET6 core circuit board, the real-time temperature data measurement circuits, the data acquisition and processing module, PWM output module, LCD display module, and so on. The experimental measurement system structure diagram of pyroelectric energy harvester is shown in Fig. 2.

The main design ideas of experimental measurement system are that the STM32F103VET6 microcontroller is used as the control core, the related parameters of temperature variation are set by TF TLCD 2.8 LCD or infrared remote controller, the digital temperature sensor, DS18B20, is used to detect the real-time temperature data. Under the control of fuzzy-PID temperature module, the PWM controlling waveform can be output to control the solid state relay, which rapidly achieves the temperature changes by power supply switching. Thus the transient temperature field can be generated around the pyroelectric energy harvester by the semiconductor chilling plate heating or cooling control. And then, the output voltage of pyroelectric energy harvester can be acquired by the preamplifier circuits and ADC circuits. The related electric parameters of pyroelectric energy harvester can be calculated, such as the energy value, power densities, etc.


71

Fig. 2. The experimental measurement system structure diagram of pyroelectric energy harvester.

3.2. Fabrication of PMnN-PMS-PZT Ceramics Materials and Sample Preparation of Pyroelectric Energy Harvester

Pb[(MnxNb1-x)1/2(MnxSb1-x)1/2]y(ZrzTi1-z)1-yO3(0.3≤

x ≤0.7, 0≤ y ≤0.3, 0.85≤ z ≤0.96) (PMnN-PMS-PZT) pyroelectric ceramics with varied compositions were fabricated following the conventional solid-phase method. In brief, PbO, ZrO2, TiO2, Nb2O5, Sb2O3

powders with analytical purities were firstly mixed with Mn(NO3)2 solution according to the designed composition molar ratios, 94/6, 95/5 and 96/4. After ball-milling, drying and sieving procedures, the mixed PMnN-PMS-PZT powers were pre-sintered at 1123 K for 2 h, which was followed by the prilling and dry pressing processes to form circular disks of 17 mm in diameter and 1.5 mm in thickness. In the next step, the PMnN-PMS-PZT circular disks were burned out at 873 K for 2 h and sintered at 1503 K in sealed atmosphere for 2 h to form ceramics. After that, the ceramics were polished, rinsed and double-side coated with silver-paste electrodes. Finally, the fabricated ceramics with double-side electrodes were polarized in silicon oil of 373 K under an electric field of 3 kV/mm for 30 min.

To compare and analyze the dielectric and pyroelectric parameters of the PMnN-PMS-PZT ceramics, the Zr/Ti compositions is respectively 94/6, 95/5 and 96/4 after sample polarization. The experiment results show that the 95/5 composion of Zr/Ti posses best pyroelectric properties, such as smaller permittivity, ɛr, smaller dielectric loss and larger pyroelectric coefficient. It is suitable for the Pyroelectric energy harvester prepared with a large working temperature range. The samples prepared of

Pyroelectric energy harvester and its profiles are shown in Fig. 3.

Fig. 3. The samples prepared of Pyroelectric energy harvester and its profile:

(a) Single structure; (b), (c) Parallel structure; (d), (e) Overlap structure;

(f) The Pyroelectric energy harvester profile.

3.3. STM32F103VET6 Microcontroller

STM32F103VET6 has the high-performance ARM Cortex-M3 32-bit RISC core operating at a 72 MHz frequency, high-speed embedded memories (Flash memory up to 512 Kbytes and SRAM up to 64 Kbytes), and an extensive range of enhanced I/Os and peripherals connected to two APB buses. It can offer two 12-bit ADCs, three general-purpose 16-bit timers plus a PWM timers, as well as standard and advanced communication interfaces: up to two I2Cs, two SPIs, three USARTs, an USB and a CAN. The power supply can from a 2.0 to 3.6 V. A comprehensive set of power-saving mode allows the design of low-power applications. It can be suitable for a wide range of applications such as application control, handheld equipment, etc.


72

3.4. The Preamplifier Circuits of Pyroelectric Energy Harvester

The preamplifier circuits of Pyroelectric energy

harvester are shown in Fig. 4.

Fig. 4. The preamplifier circuits of Pyroelectric energy harvester. (a) Equivalent circuit of

Pyroelectric energy harvester, (b) The dual voltage followers, (c) The primary amplification circuits,

(d) The secondly amplification circuits.

The dual voltage followers are designed with the low noise and high input impedance JFET chip, TL072, and selecting rail-to-rail instrumentation

amplifier, AD8220, to design the primary amplification circuits. The secondly amplification circuits based on TL071 chip are designed to realize the gain intense adjustment. Using TL072, AD8220 and TL071 to design the preamplifier circuits of pyroelectric energy harvester, fault its reason, mainly lies in their advantages of much lower input leakage current, much higher input impedance and lower noise, etc. It can effectively solve the larger signal error in the precision measurement, and can improve the dynamic range and properties of system. 3.5. The Temperature Control Circuits of

Experimental Measurement System

In the system, using two chips, 10 A single-phase solid state relay MGR-1DD220D10, to achieve the temperature rapidly changes by power supply switching for semiconductor chilling plate, that is, the temperature will be rapidly heating or cooling. The structural diagram of temperature controlling circuits is shown in Fig. 5.

Fig. 5. The structural diagram of temperature controlling circuits.

The control signals of output power supply, which include the heating control signal and the cooling control signal, are used to control the analog electronic switch, CD4051, realizing power supply switching for Solid-state relay, MGR-1 DD220D10. Then the PWM controlling waveform by STM32F103VET6 controlling output, which is respectively PWM controlling waveform at the heating stage, PWM controlling waveform at the cooling stage, is used to achieve the temperature rapidly changes for the semiconductor chilling plate. 4. The Software Design of Experimental

Measurement System

The software design of experimental measurement system mainly includes the fuzzy-PID

temperature control algorithm program, the control system program, etc. The software flow diagrams of the fuzzy-PID control subroutine program and the control system program are shown in Fig. 6.

5. The Measurement Results Analysis

and Discussion

At the different temperature changing, the experimental measurement system is used to measure the changing curve of the pyroelectric energy harvester surface temperature, T, with the output voltage, U. The Zr/Ti compositions of PMnN-PMS-PZT are 95/5. While the temperature changing were respectively set at 45 mHz, 60 mHz, 35 mHz, 70 mHz, 85 mHz, the relationship curves about the temperature T(t) and the voltage U(t) of the pyroelectric energy harvester are shown in Fig. 7.


73

Fig. 6. The program software flow diagram of the fuzzy PID control subroutine and the control system.

Fig. 7. The relation curves about temperature T(t) and voltage U(t) at different

temperature changing frequency.

The results show that the output voltage of PMnN-PMS-a PZT pyroelectric energy harvester would be periodic change. When the pyroelectric energy harvester is heated by the semiconductor chilling plate, an induced voltage output can be observed.

In the temperature-increasing stage, the maximum output voltages reach 1.654 V, 1.897 V, 2.615 V, 2.632 V and 2.618 V for 35, 45, 60, 70 and 85 mHz, respectively. Further more, when the frequency is above 70 mHz, the output voltage does not significantly increase anymore, indicating that 70 mHz is the saturation frequency. In the temperature-decreasing stage, the output voltage does not change much. The voltage values for the five

frequencies are -0.529 V, -0.535 V, -0.538 V, -0.567 V and -0.576 V, respectively. 6. Conclusions

In this paper, a novel experiment measurement system of pyroelectric energy harvester was designed. It provides a novel scheme to study the performance and pyroelectric materials properties of Pyroelectric energy harvester, meanwhile, it also establish the analysis and test foundation for latter research.

Acknowledgements

This work is supported by National Natural Science Foundation of China (61463014, 61263030), Science Research Program of Eduation Bureau of Hubei Province (T201429), Innovation and Entrepreneurship Training Program for College Students in Hubei Province (201510517010).

References [1]. L. Kouchachvili, M. Ikura, Pyroelectric conversion –

Effects of P(VDF–TrFE) preconditioning on power conversion, Journal of Electrostatics, Vol. 65, No. 3, 2007, pp. 182-188.

[2]. S. Dalola, V. Ferrari, D. Marioli, Pyroelectric effect in PZT thick films for thermal energy harvesting in low-power sensors, Procedia Engineering, Vol. 5, 2010, pp. 685-688.

[3]. J. Xie, X. P. Mane, C. W. Green, K. M. Mossi, K. K. Leang, Performance of Thin Piezoelectric Materials for Pyroelectric Energy Harvesting, Journal of Intelligent Material Systems & Structures, Vol. 21, No. 3, 2010, pp. 243-249.

[4]. Z. W. Wu, W. Dong, L. Yao, et al., Study on pyroelectric characteristic of piezoelectric ceramic


74

excited by periodic thermal excitation, Journal of Thermal Science and Technology, Vol. 11, No. 3, 2012, pp. 277-281.

[5]. Z. W. Wu, W. Dong, Z. H. Qiao, et al., Characteristic analysis of small-scale pyroelectric energy conversion device, Journal of Thermal Science and Technology, Vol. 12, No. 2, 2013, pp. 162-167.

[6]. P. Mane, J. Xie, K. K. Leang, et al., Cyclic energy harvesting from pyroelectric materials, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 58, No. 1, 2011, pp. 10-17.

[7]. G. Sebald, D. Guyomar, A. Agbossou, On thermoelectric and pyroelectric energy harvesting,

Smart Materials and Structures, Vol. 18, No. 12, 2009, pp. 125006-125007.

[8]. A. Cuadras, M. Gasulla, V. Ferrari, Thermal energy harvesting through pyroelectricity, Sensors and Actuators A: Physical, Vol. 158, No. 1, 2010, pp. 132-139.

[9]. R. Kandilian, A. Navid, L. Pilon, The pyroelectric energy harvesting capabilities of PMN–PT near the morphotropic phase boundary, Smart Materials & Structures, Vol. 20, No. 5, 2011, pp. 130-138.

[10]. Q. Zhang, A. Agbossou, Z. Feng, et al., Solar micro-energy harvesting with pyroelectric effect and wind flow, Sensors and Actuators A: Physical, Vol. 168, No. 2, 2011, pp. 335-342.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/MEMS_Pressure_Sensors.htm


75



An Automatic Segmentation Method for Printed Circuit Board Welding Component

under Stereo Optical Microscope

1 Yi LIU, 1 Mei YU, 1 Li CUI, 1 Gang-Yi JIANG, 1, 2 Yi-Gang WANG and 1, 2 Sheng-Li FAN

1 Faculty of Information Science and Engineering, Ningbo University, Ningbo 315211, China 2 Department of Information, Ningbo Institute of Technology, Zhejiang University,

Ningbo 315010, China 1 Tel.: 13857476528, fax: 0574-87608138


Received: 28 June 2014 /Accepted: 18 August 2014 /Published: 31 January 2016 Abstract: Common image segmentation methods only consider the single image and cannot extract welded components automatically and exactly. GrabCut algorithm need draw a rectangle which labels possible foreground region and determines general location to segment desired object. In this paper, a novel automatic GrabCut algorithm is proposed for segmenting Printed Circuit Board (PCB) welding component under stereo optical microscope. The proposed method considers the characteristic of welded component on the foreground region of PCB and uses unparallel stereo microscopic image to obtain foreground mask of disparity map. To obtain parallel stereo microscopic image, Quasi-Euclidean epipolar rectification algorithm is utilized into original unparallel stereo microscopic image. Then, disparity map is obtained by using the non-local filter algorithm on parallel stereo microscopic image, and foreground mask of disparity map is extracted by applying the mean-threshold method. Finally, rectified microscopic image is segmented by initializing GrabCut’s trimap T with foreground mask of disparity map. The experimental results show that the proposed method can extract PCB welding component automatically and accurately without user intervention and is much better than adaptive binary method. Copyright © 2016 IFSA Publishing, S. L. Keywords: Image segmentation, Stereo microscopic image, Printed Circuit Board welding component, Foreground mask of disparity map, Automatic GrabCut.

1. Introduction

Intelligent detection and location analysis of Printed Circuit Board (PCB) welding components need to extract accurately PCB image’s external features [1-2], such as welding condition, wiring condition, and positioning hole. Therefore, the PCB image must be processed high efficiently. As we know, image segmentation is a process of dividing an

image into several characteristic regions, and plays a virtual role in detection and location of PCB welding components.

In recent years, many researches on image segmentation have been reported [3-4], such as threshold methods [5], region-based methods [6], edge detection methods [7], clustering methods [8], superpixel based methods [9], graph-based methods [10], and other hybrid methods [11]. Threshold




76

segmentation methods are widely used on account of their simplicity and efficiency. Hammouche K. proposed a multilevel thresholding method to determine the number of thresholds and the adequate threshold for a fast image segmentation [12]. However, conventional histogram-based threshold algorithms can only separate those areas which have obvious different gray levels. Moreover, they cannot perform well for images whose histograms are nearly unimodal. For edge detection methods, the most commonly applied edge detection operators include Canny, Sobel, Prewitt and Laplacian. They mainly focus on those pixels which locate on the boundaries of object [13]. Therefore, it is hard to yield closed contours and homogeneous regions. While region growing, splitting, merging and other region-based methods often deal with spatial repartition of image feature information to obtain closed and homogeneous regions. Geometric flows are inherently good at controlling geometric shape evolution. A geometric flow-based formulation and solution for image segmentation is proposed by Ye J. [13]. However, over-segmentation and under-segmentation are critical difficulties to be considered in those methods. Clustering methods can generate better segmentation effects with the consideration of viewing image as tremendous multidimensional data and classifying an image into different portions according to certain homogeneous criterion. Wang proposed an adaptive spatial information-theoretic fuzzy clustering algorithm to improve the robustness of the conventional fuzzy c-means clustering algorithms for image segmentation [15]. But over-segmentation is the problem that must be solved and feature extraction is also an important factor for clustering. Superpixel can provide meaningful grouping cues to guide segmentation and reduce the computational complexity. Yang introduces a novel 3-D geometry enhanced superpixels for RGB-D data [16]. But sometime the segmentation performance depends on the superpixel generation approach. Graph-based image segmentation methods are modeled to divide a graph into several sub-graphs such that each of them represents a meaningful object in the image, but they always need user intervention [17-18]. In conclusion, though much emphasis has been put on image segmentation and many methods have been proposed in recent decades, there is neither universal segmentation approach for all kinds of images nor an automatic and effective segmentation approach for PCB welding component under digital stereo optical microscope.

Common image segmentation methods only take consider of single image but they do not use the useful information offered by stereo image. Obviously, the location and appearance of the desired object is difficult to acquire. Consequently, welded components are hard to be segmented availably in the real complex PCB scenery. GrabCut algorithm [17] cannot extract desired object automatically. But it can segment the object effectually with manually drawing

a rectangle. Generally, PCB welding components locate at the foreground region of disparity map.

Considering the reasons of above, the proposed method uses unparallel stereo microscopic image to obtain foreground mask of disparity map for determining the general position of PCB welding component and initialing the trimap T of GrabCut to segment PCB welding component without any user intervention. 2. Proposed Automatic GrabCut Method

for PCB Welding Component Segmentation In this paper, an attempt to obtain a solution to

realize automatic PCB welding component segmentation is done. Firstly, the proposed automatic GrabCut (A-GrabCut) method rectifies original captured stereo microscopic image to parallel stereo microscopic image by employing the Quasi-Euclidean epipolar rectification algorithm. Secondly, to obtain disparity map of microscopic image, the non-local filter algorithm is utilized into rectified stereo microscopic image. Then, foreground mask is extracted from disparity map by using the mean-threshold method. Finally, extracted foreground mask initializes the trimap T of GrabCut algorithm, and automatic segmentation for PCB welding component starts on rectified stereo microscopic image. The whole procedure of proposed A-GrabCut method for PCB welding component segmentation of stereo microscopic image is illustrated as Fig. 1, which will be described in detail in the following subsections.

Fig. 1. Procedure of A-GrabCut method for PCB welding component segmentation.

2.1. Quasi-Euclidean Epipolar Rectification There are errors between inside equipment setup

of real microscope and ideal parallel stereo microscopic system. And small bias of parallel optical path leads to large offset of micro amplification imaging. Therefore, the same 3D point does not locate identical vertical coordinate in stereo microscopic image. It is very important to rectify


77

epipolar for requirement of parallel stereo microscopic image.

Because of the limitation of microscope’s depth of focus, the camera calibration of stereo optical microscopic system is tedious and difficult. The method that rectifies stereo microscopic image by utilizing camera’s internal and external parameters is not suitable for stereo optical microscopic system. However, the desired homograph matrixes of unrectified stereo microscopic image can be obtained by using Quasi-Euclidean epipolar rectification method [19], which find matching points of stereo microscopic image to make rectified error E approach 0. The formula can be expressed as follows:

1( , , , ) [ ] 0T Tl l r r l l r rE x y x y X H e H X×= = , (1)

where (xl,yl) and (xr,yr) are the matching points of the left and right views of the original stereo microscopic image, and Xl and Xr are homogeneous coordinates of (xl,yl) and (xr,yr). Hl and Hr are corresponding homograph matrix of the left and right views of the original stereo microscopic image for epipolar rectification. [e1]× is antisymmetric matrix of unit direction vector 1 (1,0,0)=e .

Finally, rectified left and right microscopic images rec

lI and recrI can be computed as follows:

=

=

rec orgl l lrec orgr r r

I H I

I H I

, (2)

where org

lI and orgrI are the left and right views of the

original stereo microscopic image respectively. Quasi-Euclidean epipolar rectification algorithm

can work well without calibrating camera’s internal and external parameters under stereo optical microscopic system. Therefore, it meets the requirement of stereo optical microscopic system’s epipolar rectification and reduces the difficulty of stereo microscopic image epipolar rectification.

2.2. Foreground Mask Generation This section mainly includes that disparity

map of microscopic image is solved by the non-local filter method and foreground mask is obtained from disparity map by using proposed mean-threshold method.

2.2.1. Non-local Filter Matching

The original stereo microscopic image has been rectified by section (2.1). To obtain disparity map of microscopic image, stereo matching is need to apply to parallel stereo microscopic image

reclI and rec

rI .Unlike common traditional approaches

which build similar function based on local window

or irregular shape, the non-local filter approach [20] connects all image pixels by using minimum spanning tree (MST) on which all weight sum is the most lowest.

( , )

( , ) ( , ) exp( )D p q

S p q S q pσ

= = − (3)

is the similarity between node p and q in the MST, where σ is a constant used to adjust the similarity between two nodes. ( , )= ( , )D p q D q p is the distance

between node p and q. The joint bilateral filter [21] can then be directly

extended to MST structure:

( ) ( , ) ( )

( , )exp( ) ( )

Ad d

q

dq

C p S p q C q

D p qC q

σ

=

= −

, (4)

where ( )dC q is the matching cost for pixel p at

disparity level d, and ( )AdC p is the aggregated cost.

Aggregation using MST relaxes two ambiguities to a single ambiguityσ .

Two nodes’ distance can be accumulated from leaf nodes tracking to root node in the MST. The formula can be expressed as follows:

2

( ) ( ( ), ) ( ( ))

[1 ( , ( ))] ( ( ))

A Ad d

Ad

C v S P v v C P v

S v P v C P v↑

= ⋅

+ − ⋅, (5)

where v is the node in the MST, and P(v) is the parent

node of v. ( ( ))↑AdC P v is the accumulated aggregation

cost from leaf nodes to node P(v). Consequently, its computation complex is very low and only requires a total of 2 addition/subtraction operations and 3 multiplication operations. This paper just considers the disparity map of the left microscopic image ( , )dm

lI i j .

2.2.2. Mean-threshold Foreground Mask

Extracting Supposed that PCB surface and camera’s image

plane are parallel, the disparity of same plane should be identical. Generally, it is a plane for PCB welding components, such as chips and resistances. The result is indeed after experimental verification. Obviously, the large disparity region that is foreground region is required to be extracted typically. And the proportion of those regions is large in microscopic image.

For this purpose, the histogram can be calculated from disparity map of microscopic image disp

lI . And

the probability of histogram can be formulated as follows:


78

( ) kP k n N= , (6) where N is the total number of an image pixels, and nk is the number of pixel at disparity level k. P(k) is the probability of occurrence at disparity level k.

Generally, mismatch and noise exist in disparity map of microscopic image. To reduce the influence of mismatch and noise of which the proportion is small in disparity map, the mean of disparities is solved by selecting fore M proportion of sorted nk. The formula can be expressed as follows:

1

1 M

ii

mean rM =

= (7)

Finally, foreground mask of disparity map is

obtained by selecting the disparity above mean disparity:

255, ( , )( , )

0,

dmfm l

l

if I i j meanI i j

else

>=

, (8)

where ( , )dmlI i j is the disparity of coordinate (i,j) in

disparity map of left microscopic image, and ( , )fm

lI i j is the disparity of coordinate (i,j) in

foreground mask image of left disparity map. 2.2.3. GrabCut Algorithm and PCB Welding

Component Segmentation GrabCut algorithm based on Gussian Mixture

Model (GMM) is an image segmentation method realized by combining with graph cut. The GrabCut technique considers the array z=(z1,…,zn,…,zN) of N pixels in an image where zi=(Ri, Gi, Bi), [1,..., N]∈i .

The segmentation is defined as an array a = (a1,…,ai,…,aN), 0,1∈ia and assigns a label to

each pixel of the image, indicating whether it belongs to the foreground or the background. A trimap T is given by the user, which consists of three regions: TF, TB and TU, each one including initial background, foreground, and uncertain pixels respectively. Pixels belonging to TF and TB are considered as foreground and background respectively, whereas those belonging to TU are labeled through by the algorithm.

A full covariance GMM of K components of foreground and background pixels is parameterized as follows:

( , ), ( , ), ( , ), 0,1, 1...a k a k a k a k Kθ π μ= Σ = = , (9)

where π , μ , Σ respectively represent the weight,

mean vector, covariance matrix of the GMM. The energy function for segmentation of GrabCut

technique is

( , , , ) ( , , , ) ( , )E k z U k z V zα θ α θ α= + , (10)

which consists of a region term U and edge term V. The region term U based on Gaussian probability distributions ( )p ⋅ and mixture weighting coefficients

( )π ⋅ computes the likelihood of a pixel to belong to

certain label. The formula is expressed as follows:

( , , , ) log ( , ) log ( , )i i i i iU k z p z a k a kα θ θ π= − − (11) Edge term V reflects the penalty of discontinuity

between neighborhood pixel m and n. The main problem of original GrabCut technique

is need to draw a rectangle by user to initialize the trimap T for computing the initial GMM. The proposed method locates welded component general position automatically to compute the initial GMM. The results of trimap T by using ( , )fm

lI i j are: = ∈iTU z mask , = ∉iTB z mask , where mask is

the region consisted of pixel value that is 255 of ( , )fm

lI i j . According to the lowest energy principle

and the initialized trimap T, the final segmentation is generated by graph cut, which is based on Min-Cut/Max-Flow algorithm [22].

3. Experiment and Discussion In order to verify the proposed method

sufficiently, a lot of experiments are conducted. Stereo optical microscope stage is used to capture stereo microscopic image. As shown in Fig. 2, the test stage includes a type of ZOOM460N stereo optical microscopic, which made in Nanjing Wavelength Opto-Electronic Technology Co., Ltd. The bottom is objective table on which a PCB is placed. Common objective transmits information of each to two top eyepieces via two optical paths. The system’s eyepieces are connected directly to two types of 902B camera produced by WATEC Company from Japan. And the size of 704×576 per frame is sampled by USB. The test environment: OPENCV2.4.8 and VS2012 in PC (Intel Core(TM) i3 CPU 3.19 GHz, 1.74G memory).

Fig. 2. Digital stereo optical microscope.


79

In this paper, original stereo microscopic image used in the experiment is shown in Fig. 3. And stereo microscopic image after epipolar rectification is shown in Fig. 4.

Fig. 3. Original stereo microscopic image.

Fig. 4. Rectified stereo microscopic image.

The left disparity map obtained by matching method is shown in Fig. 5 (a). The foreground mask image of left disparity map is shown in Fig. 5 (b). Proportion parameter M is 0.8 in this paper.

(a)

(b)

Fig. 5. (a) Left disparity map; (b) Foreground mask image

of left disparity map.

3.1. Comparison with User Intervention Segmentation

In the first contrast experiment, the intention is to

compare the segmentation effects between the proposed A-GrabCut method and common user intervention segmentation method. There have two realized user intervention segmentation methods in OPENCV’s samples. Therefore, our contrast experiment is between the proposed A-GrabCut method and watershed method [23] and original GrabCut method. The main difference between two user intervention methods is that watershed method selects two kinds of elements with user’s mouse at any position of image but original GrabCut method only need draw a rectangle.

Fig. 6 shows the manual marks and segmentation effects of user intervention segmentation methods and the visual effect of proposed A-GrabCut method. As shown in Fig. 6 (b), Fig. 6 (d) and Fig. 6 (f), the desired foreground object is segmented efficiently from complex background by three methods. However, proposed A-GrabCut method can generate satisfied result automatically without any manual marks on image in Fig. 6 (e) and Fig. 6 (f). Moreover, watershed method and original GrabCut method need mark image correctly in Fig. 6 (a) and Fig. 6 (c), and original GrabCut method requires many iterative computations to obtain final segmentation if the mark is non-ideal.

So in brief, the proposed A-GrabCut method can obtain the same good segmentation result as two common user intervention segmentation methods without any user intervention.

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 6. Segmentation results comparison of user intervention method and A-GrabCut method: (a) manual mark of watershed method; (b) segmentation result of watershed method; (c) manual mark of original GrabCut method; (d) segmentation result of original GrabCut method; (e) manual mark of A-GrabCut method: no mark; (f) segmentation result of A-GrabCut method. 3.2. Comparison with Binary Segmentation

As GrabCut method is popular foreground and

background method, it can also be viewed as a binary classification problem. So in the second experiment,


80

we evaluated the performance of A-GrabCut method and a typical adaptive threshold binary segmentation method [24].

Fig. 7. Segmented result of adaptive threshold binary method.

Fig. 7 illustrates the image segmentation result by using threshold method. It can be seen that the segmentation result of A-GrabCut is much better than that of the threshold method. The threshold method computes mean and variance of pixel value and can generate a balance gray to image segmentation. As shown in Fig. 7. Though the foreground welded component has been segmented, the similar grayscale pixels are considered as homogeneous region, such as holes and some region of PCB. And most objects segmented by threshold method are with noises or incomplete edges. It is hard to satisfy the segmentation requirement.

So in brief, the proposed A-GrabCut method is much better than the naive binary classification method for image segmentation. 4. Conclusions

Due to real PCB scenery is complex and common

image segmentation method cannot segmented automatically and efficiently welded components. In this paper, an automatic PCB welding component segmentation method based on GrabCut technique is proposed. The advantages of proposed A-GrabCut method include:

1) It exploits the character of welded components locating at disparity foreground region. The feature is not dependent on image intensive, shape, texture and other prior knowledge.

2) Welded component can be extracted correctly without any user intervention.

3) The proposed method is automatic, instead of processing certain region with user intervention.

In addition to these advantages, the proposed method has its limitation:

1) The trimap T of GrabCut method is dependent on the result of matching disparity map.

2) The color of mask region and non-mask region of original image cannot too similar.

References [1]. Benedek C., Krammer O., Janóczki M., et al., Solder

paste scooping detection by multilevel visual inspection of printed circuit boards, IEEE Transactions on Industrial Electronics, Vol. 60, No. 6, 2013, pp. 2318-2331.

[2]. Aghamohammadi A., Ang M. C., Prabuwono A. S., et al., Enhancing an automated inspection system on printed circuit boards using affine-SIFT and TRIZ techniques, Advances in Visual Informatics, 2013, pp. 128-137.

[3]. Cheng H. D., Jiang X. H., Sun Y., et al., Color image segmentation: advances and prospects, Pattern Recognition, Vol. 34, No. 12, 2001, pp. 2259-2281.

[4]. Estrada F. J., Jepson A. D., Benchmarking image segmentation algorithms, International Journal of Computer Vision, Vol. 85, No. 2, 2009, pp. 167-181.

[5]. Horng M. H., Multilevel thresholding selection based on the artificial bee colony algorithm for image segmentation, Expert Systems with Applications, Vol. 38, No. 11, 2011, pp. 13785-13791.

[6]. Garcia Ugarriza L, Saber E., Vantaram S. R., et al., Automatic image segmentation by dynamic region growth and multiresolution merging, IEEE Transactions on Image Processing, Vol. 18, No. 10, 2009, pp. 2275-2288.

[7]. Wang H., Oliensis J., Generalizing edge detection to contour detection for image segmentation, Computer Vision and Image Understanding, Vol. 114, No. 7, 2010, pp. 731-744.

[8]. Yu Z., Au O. C., Zou R., et al., An adaptive unsupervised approach toward pixel clustering and color image segmentation, Pattern Recognition, Vol. 43, No. 5, 2010, pp. 1889-1906.

[9]. Li Z., Wu X. M., Chang S. F., Segmentation using superpixels: A bipartite graph partitioning approach, Computer Vision and Pattern Recognition, 2012, pp. 789-796.

[10]. Boykov Y. Y., Jolly M. P., Interactive graph cuts for optimal boundary & region segmentation of objects in N-D images, in Proceedings of the International Conference on Computer Vision, Vol. 1, 2001, pp. 105-112.

[11]. Siang Tan K., Mat Isa N. A., Color image segmentation using histogram thresholding-Fuzzy C-means hybrid approach, Pattern Recognition, Vol. 44, No. 1, 2011, pp. 1-15.

[12]. Hammouche K., Diaf M., Siarry P., A multilevel automatic thresholding method based on a genetic algorithm for a fast image segmentation, Computer Vision and Image Understanding, Vol. 109, No. 2, 2008, pp. 163-175.

[13]. Ye J., Xu G., Geometric flow approach for region-based image segmentation, IEEE Transactions on Image Processing, Vol. 21, 12, 2012, pp. 4735-4745.

[14]. Arbelaez P., Maire M., Fowlkes C., et al., Contour detection and hierarchical image segmentation, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 33, No. 5, 2011, pp. 898-916.

[15]. Wang Z., Song Q., Soh Y. C., et al., An adaptive spatial information-theoretic fuzzy clustering algorithm for image segmentation, Computer Vision and Image Understanding, Vol. 117, No. 10, 2013, pp. 1412-1420.

[16]. Yang J., Gan Z., Gui X., et al., 3-D Geometry Enhanced Superpixels for RGB-D Data, Advances in


81

Multimedia Information Processing – PCM, 2013, pp. 35-46.

[17]. Rother C., Kolmogorov V., Blake A., GrabCut: Interactive foreground extraction using iterated graph cuts, ACM Transactions on Graphics, Vol. 23, No. 3, 2004, pp. 309-314.

[18]. Peng B., Zhang L., Zhang D., et al., Image segmentation by iterated region merging with localized graph cuts, Pattern Recognition, Vol. 44, No. 10, 2011, pp. 2527-2538.

[19]. Fusiello A., Irsara L., Quasi-Euclidean epipolar rectification of uncalibrated images, Machine Vision and Applications, Vol. 22, No. 4, 2011, pp. 663-670.

[20]. Yang Q., A non-local cost aggregation method for stereo matching, in Proceedings of the Conference on Computer Vision and Pattern Recognition, 2012, pp. 1402-1409.

[21]. Yoon K. J., Kweon I. S., Adaptive support-weight approach for correspondence search, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 28, No. 4, 2006, pp. 650-656.

[22]. Boykov Y., Kolmogorov V., An experimental comparison of mincut/max-flow algorithms for energy minimization in vision, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 26, No. 9, 2004, pp. 1124-1137.

[23]. Vachier C., Meyer F., The viscous watershed transform, Journal of Mathematical Imaging and Vision, Vol. 22, No. 2-3, 2005, pp. 251-267.

[24]. Otsu N., A threshold selection method from gray-level histograms, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, No. 1, 1979, pp. 62-66.

___________________


http://www.sensorsportal.com/HTML/BOOKSTORE/Advance_in_Sensors_Vol_2.htm


82



Research of an Optoelectronic Current Transformer Based on a Designed Magneto-Optic Sensor

1, * R. El-Bashar, 1 J. El-Azab, 1 Y. Badr and 2 R. Yousif 1 National Institute of Laser Enhanced Sciences, Cairo University, Egypt

2 Electrical Power and Machines Department, Faculty of Engineering, Cairo University, Egypt * Tel.: +201064088409

* E-mail: [email protected]

Received: 2 January 2016 /Accepted: 25 January 2016 /Published: 30 January 2016 Abstract: An Optical Current Transformer (OCT) with dual polarimetric configuration is evaluated using a magneto-sensitive glass SF-59 of 0.4 and 1.3 cm lengths located into an air gapped conventional current transformer (CCT) iron core. Using a 5 mW He-Ne laser light source at 632.8 nm, the maximum current measured is nearly 2800 A with maximum error of 0.6 % and 1 % for the two sensors lengths 0.4 and 1.3 cm, respectively. The results have been experimentally limited by the maximum output of Variac. The simulation of the electronic circuit of the current transducer is introduced and the results indicated the reliability and accuracy of using the OCT in measurement and protection. Copyright © 2016 IFSA Publishing, S. L. Keywords: Verdet constant, Optical current transformer, Optical current sensor, Bulk glass. 1. Introduction

Due to the recent demand of power consumption and expansions of electric power system network, a great number of protection zones become available and requires intelligent system. Current sensors are important devices as they are not only give the accurate power consumption but also used for the fast detection and identification of failures points in power systems. However, CCT have enormous limitations due to their saturation problems, huge size and weight, high insulation cost as well the sensitivity to electromagnetic interference (EMI) [1].

Recently, optoelectronic technology is one of the evolved solutions of CCT problems by converting the CCT output voltage into a corresponding frequency modulated optical signal. This signal is not influenced by EMI and is congruous with intelligent network of current measurement and protection.

However, this technology could not overcome the iron core (IC) saturation [2]. The development of the measurement circuit of Rogowski coil sensor which does not involve IC, for high current measurement is still in progress [3]. On the other hand, magneto-optic sensing method provided an alternate solution for solving the CCT problems [4]. It consumes less power as compared to the traditional Dc Hall Effect sensor used in plants such as aluminum factories [5]. OCT can be integrated with fiber-optic based Ethernet local area networks. Besides, replacing the traditional solution which digitizes the CCT output for relays, meters, and SCADA systems [6].

The effect of the magnetic field on the transmission of light through certain transparent optical material was first observed by Michael Faraday in 1845 [7]. He noticed that, when the transmitted linearly polarized light is subjected to a magnetic field in the direction of propagation, the

http://www.sensorsportal.com/HTML/DIGEST/P_RP_0208.htm



83

plane of polarization rotates. Based on this effect, optical current sensors were developed into mainly two types; bulk glass sensors and fiber optical sensors. In bulk sensors, an optical material sensitive to the magnetic field is positioned parallel to the lines of magnetic field induced by a conductor carrying the current to be sensed. This sensor has the advantages of small size, low cost, weak influence of the environmental conditions such as temperature, pressure and other noises and high sensitivity to magnetic field. The sensitivity of the sensor is directly proportional to the length of the material subjected to the magnetic field. To increase its sensitivity, the sensor was designed to surround the conductor. The geometrical structure promotes the propagation of light parallel to the magnetic field lines. It can be either rectangular or triangular and sometimes circular [7-9]. In this case, the sensitivity is based on the sensor structure and/or the multiple path of light within the optical material [10].

An alternative technique is to utilize a small bulk sensor which does not surround the conductor while compensating the reduced sensor length by using a material with a high Verdet constant [11-13] and a concentrator core to increase the magnetic field [14-16].

In this work, an evaluation of a dual quadrature polarimetric sensitive optical current sensor inserted into a gapped IC is carried out. In Section 2, the design of the optical current sensor (OCS) in a gapped IC will be introduced. The proposed electronic circuit of the OCT will be presented in section 3. The simulation of the electronic circuit and optical sensor experimental results are introduced in Section 4 and will be concluded in Section 5.

2. OCS System Design

2.1. Optical Current Sensor

The proposed OCS consists of a gapped toroid IC shown in Fig. 1(a). The air gap created causes a new magnetic drop in original magnetic circuit of the IC. Fig. 1(b) shows the magnetic representation circuit. In the equivalent circuit, the core and the gap are represented by and , respectively [17].

(a)

(b)

Fig. 1. (a) Toroid core with an air gap and (b) its magnetic

representation circuit.

According to Ampere's circuital law, the Ampere-turns Ni is given by:

, (1)

(2)

where ϕi is the flux, li is the mean length, Ai is the

cross section area where the flux pass through, and μi is the permeability. And the subscript i (= c and g) represents the core and the gap, respectively. The air gap permeability μg nearly equals the free space

permeability μ 4π10 T.m/A . For the core permeability μc μ0μcr , where μcr is the relative permeability of the core. By neglecting the flux lines that are diffused outside the common area of the core and the air gap, then Ac Ag A. Considering core

flux ϕcand the gap flux ϕg are equal, the Ampere-

turns becomes

(3)

The magnetic flux density becomes:

1

(4)

2.2. Polarimetric Detection OCS Setup [9-10]

A dual polarimetric schematic setup [17-18] is presented in Fig. 2. It consists from a He-Ne laser source (5 mW, 632.8 nm), a polarizer, polarizing beam splitter (PBS) and two photo detectors connected to an oscilloscope. The toroid IC with an appropriate number of turns is used to generate a magnetic flux corresponding to that the flux resulting from the power line conductor current.

The rotation in the polarization plane depends on the intensity of magnetic field B , the length of interacting material d , and the Verdet constant V, representing the medium sensitivity to magnetic field, through the following relation [7].

(5)

Fig. 2. Dual quadrature scheme.


84

Through Mauls’s law, the output from the PBS (Po is given by

P0=P1cos2(), (6)

where is the angle between transmission axes of the polarizer and the PBS, P1 is the power transmitted from the polarizer. The sensor sensitivity is optimized by adjusting the polarizer transmission angle to 45 degree with respect to the PBS. In the absence of the magnetic field (i.e. no electric current applied), the angle between polarizer and PBS remains 45o. The angle of rotation for the two outputs is altered in the presence of magnetic field to be:

45 ,

where is the angle of Faraday rotation. Thus, the light power outputs from the PBS falling on the detectors become:

12

12

2 (7)

12

12

2 (8)

The two optical detectors convert the optical

signal into electrical signal where the subtraction and addition are executed electronically. By dividing the two components (subtraction and addition) by an

electronic divider, the resulting signal is proportional to the applied current. The division of these components also removes the effect of the light power fluctuation [9]. The output signal of the electronic divider becomes:

And for small angles,

2 2 (9) In case of alternating currents, the output signal

can be expressed by:

8 101

(10)

3. The Proposed Electronic Circuit of the OCT

The schematic diagram of the electronic circuit of OCT is shown in Fig. 3. It represents the electronic processing of the detected signals as discussed in Section 2.B in order to obtain a signal proper for metering and relaying circuits.

Fig. 3. Schematic block diagram of the proposed optical current transformer.

Fig. 4 introduces the equivalent circuit of the amplifier namely transimpedance amplifier. The reversed bias is used to convert the photodetected current Ip into Vout controlled by the resistor Rf. The circuit output is generated and simulated on the simulator program as an input to the proposed electronic circuit [19].

The simulation of the electronic circuit is illustrated in details in Fig. 5, showing the amplifier,

subtractor [20], adder, and divider [21]. The subtractor removes of the dc component and increases the current signal sensitivity. The adder circuit is used to get the dc component of the detected signal. The divider plays the main job to extract the electric signal corresponding to the measured current. Finally, the resulting signal is amplified for current measurements systems.


85

Fig. 4. The Transimpedance amplifier circuit.

4. Simulation and Experimental Results

4.1. Measurement circuit Simulation

The simulation of the electronic circuit is carried out by Proteus 8. In Fig. 6, Ch. (A) is the optical

signal which equal 0.315 V, as adjusted experimentally, with noise fluctuations 4 % at 1.5 kHz and 2 % at 15 kHz. The simulation parameters of the OCS are as following; the Verdet constant V = 25.9 rad/T.m for SF-59 glass [22], glass length d=1.3 cm. For a current 1500 A, the magnetic flux density is measured experimentally to be 119 mT. At these values, the AC signal is simulated from equation (10) to be equal 0.08 V in amplitude (maximum value) at 50 Hz. The outputs of transimpedance amplifier circuits of the two detectors are Ch. (B) and Ch. (C) while the output of the adder, subtractor, and divider are shown in Fig. 7. The output of amplified current signal is shown at Ch. (D) for the metering system.

Fig. 5. The proposed electronic circuit.

4.2. Experimental Results of OCS

Experimentally, the conductor current is simulated by wounding the air gapped IC by an appropriate number of turns in order to obtain an equivalent magnetic field in the gap. As a result, the gap magnetic field becomes a function of the product of the winding current and the number of turns. The optical sensor behavior is studied experimentally by using 5 mW He-Ne Laser light at 632.8 nm wavelength with attenuator OD3. The toroid IC coil is supplied from the Variac to control the applied current. A Gauss / Tesla Meter model 4048 with T-4048-001 Probe from F.W. Bell model is used. Fig. 8 shows the relation between the magnetic field in the center of the gap and the applied current. The output signals from the two detectors are displayed on

HEWLETT PAKWARD HP 54503A – 500 MHz digitizing oscilloscope. The peak to peak voltage difference output signal of the two detectors is shown in Fig. 9 as a function with ampere-turns for the two sensor lengths 0.4 and 1.3 cm.

The experimental results showed a good agreement with the calculations. The long glass length (1.3 cm) provided better results as the light passing through the glass effectively interacts with the magnetic field in the gap and the sensitivity of the sensor is higher. However, the calculated R.M.S error of the long length is higher than the short length as it changes from 0.42 and 0.27. Thus, a trade-off between the strength of the obtained signal and the corresponding error is obtained and the selection will be user-defined.


86

Fig. 6. Oscilloscope output - Ch. (A): light signal, Ch. (B): detector 1 signal, Ch. (C): detector 2 signal

and Ch. (D): Subtractor O/P signal.

Fig. 7. Oscilloscope output - Ch. (A): Adder O/P signal,

Ch. (B): Subtractor O/P signal, Ch. (C): Divider O/P current signal and Ch. (D): Amplified current signal.

Fig. 8. The Ampere-turn versus

the magnetic field in the gap center of the toroid IC.

Fig. 9. The Peak to peak output voltage as a function

of the applied AC current (I=Ni) in the cases of d=0.4 and 1.3 cm.

5. Conclusions

This work presented an evaluation of an optical current transformer for high voltage network using a magneto-sensitive SCHOTT glass SF-59 inserted into a gapped core of a conventional current transformer. The OCS results were carried out experimentally while the validity of the sensor was verified numerically. The experiment was realized by using a 5 mW He-Ne laser at 632.8 nm for a two glass sensors of lengths 0.4 and 1.3 cm. Using a dual quadrature polarimetric configuration, the obtained

results showed that the sensitivity of the sensor is dependent on the sensor length with a trade-off with the R.M.S error. Therefore, it can be concluded that the optical current transformer can be utilized as a high sensitive sensor for high voltage networks.

Acknowledgements

The authors gratefully acknowledge Mr. Wahed Morsi for his collaboration in cutting and polishing the glass.


87

References [1]. Brigida A. C. S., Nascimento I. M., Mendonça S.,

Costa J. C. W. A., Martinez M. A. G., Baptista J. M., Jorge P. A. S., Experimental and theoretical analysis of an optical current sensor for high power systems, Photonic Sensors, Vol. 3, No. 1, 2013, pp. 26-34.

[2]. Popescu M., Manta A., Păduraru N., Marinescu A., Optoelectronic current transformer for medium voltage networks, Annals of the University of Craiova, Electrical Engineering Series, No. 34, 2010, pp. 103-108.

[3]. Bing Y., Yutian W., Hui L., Huixin W., Yiqiang C., Research of Measurement Circuits for High Voltage Current Transformer Based on Rogowski Coils, Sensors & Transducers, Vol. 165, Issue 2, February 2014, pp. 35-39.

[4]. Barczak K., Optical fibre current sensor for electrical power engineering, Bulletin of the Polish Academy of Sciences. Technical Sciences, Vol. 59, No. 4, 2011, pp. 409-414.

[5]. Bohnert K., Brändle H., Brunzel M. G., Gabus P., Guggenbach P., Highly accurate fiber-optic DC current sensor for the electrowinning industry, IEEE Transactions on Industry Applications, Vol. 43, No. 1, 2007, pp. 180-187.

[6]. Skendzic V., Ender I., Zweigle G., IEC 61850-9-2 process bus and its impact on power system protection and control reliability, in Proceedings of the 9th Annual Western Power Delivery Automation Conference, Spokane, WA, April 2007.

[7]. Silva R. M., Martins H., Nascimento I., Baptista J. M., Ribeiro A. L., Santos J. L., Frazão O., Optical Current Sensors for High Power Systems: A Review, Appl. Sci., No. 2, 2012, pp. 602-628.

[8]. Sato T., Takahashi G., Inui Y., Method and apparatus for optically measuring a current, Washington, DC: U.S. Patent and Trademark Office, U.S. Patent No. US 4564754 A, 1986.

[9]. Fisher N. E., Jackson D. A., Vibration immunity and Ampere's circuital law for a near perfect triangular Faraday current sensor, Measurement Science and Technology, Vol. 7, No. 8, 1996, pp. 1099-1102.

[10]. Ning Y. N., Chu B. C. B., Jackson D. A., Miniature Faraday current sensor based on multiple critical angle reflections in a bulk-optic ring, Optics Letters, Vol. 16, Issue 24, 1991, pp. 1996-1998.

[11]. Chen Q., Chen Q., Wang S., A new Faraday rotation measurement method for the study on magneto optical property of PbO-Bi2O3-B2O3 glasses for current sensor applications, Open Journal of

Inorganic Non-metallic Materials, Vol. 1 No. 1, 2011, pp. 1-7.

[12]. Westenberger G., Hoffmann H. J., Jochs W. W., Przybilla G., Verdet constant and its dispersion in optical glasses, Proceedings SPIE, Vol. 1535, November 1991, pp. 113-120.

[13]. Malak S., Suzuki I. S., Suzuki M. S., Faraday rotation measurement using a lock-in amplifier, 2011, https://www.researchgate.net/publication/269929805_Faraday_rotation_measurement_using_a_lock-in_amplifier

[14]. Mihailovic P., Petricevic S., Stojkovic Z., Radunovic J. B., Development of a portable fiber-optic current sensor for power systems monitoring, IEEE Transactions on Instrumentation and Measurement, Vol. 53, No.1, 2004, pp. 24-30.

[15]. Li G., Kong M. G., Jones G. R., Spencer J. W., Sensitivity improvement of an optical current sensor with enhanced Faraday rotation, Journal of Lightwave Technology, Vol. 15, No. 12, 1997, pp. 2246-2252.

[16]. Yi B., Chu B. C. B., Chiang K. S., Chung H. S., New design of optical electric-current sensor for sensitivity improvement, IEEE Transactions on Instrumentation and Measurement, Vol. 49, No. 2, 2000, pp. 418-423.

[17]. Wang S., Karri A., Harlev Y., Use of dual-frequency excitation method to improve the accuracy of an optical current sensor, Proceedings SPIE, Photonic Fiber and Crystal Devices: Advances in Materials and Innovations in Device Applications, Vol. 7420, August 2009.

[18]. Katsukawa H., Ishikawa H., Okajima H., Cease T. W., Development of an optical current transducer with a bulk type Faraday sensor for metering, IEEE Transactions on Power Delivery, Vol. 11, No. 2, 1996, pp. 702-707.

[19]. Photo diode Transimpedance Amplifier circuit, http://repository.asu.edu/attachments/56952/content/LaFevre_asu_0010N_10920.pdf

[20]. Operational amplifier - Available online: http://www.electronics-tutorials.ws/opamp/opamp_5.html

[21]. Lorsawatsiri A., Kiranon W., Silaruam V., Sangpisit W., Wardkein P., Simple and Accurate Frequency to Voltage Converter, in Proceedings of the International Conference on Electrical Engineering/Electronics Computer Telecommunica-tions and Information Technology (ECTI-CON), 2010, pp. 212-215.

[22]. Weber M. J., Handbook of optical materials, Laser and Optical Science and Technology Series, CRC Press. 2002.

___________________


Aims and Scope

Sensors & Transducers is established, international, peer-reviewed, open access journal (print and electronic). It provides the best platform for the researchers and scientist worldwide to exchange their latest findings and results in science and technology of physical, chemical sensors and biosensors. The journal publishes original results of scientific and research works related to strategic and applied studies in all aspects of sensors: reviews, regular research and application specific papers, and short notes. In comparison with other sensors related journals, which are mainly focused on technological aspects and sensing principles, the Sensors & Transducers significantly contributes in areas, which are not adequately addressed in other journals, namely: frequency (period), duty-cycle, time-interval, PWM, phase-shift, pulse number output sensors and transducers; sensor systems; digital, smart, intelligent sensors and systems designs; signal processing and ADC in sensor systems; advanced sensor fusion; sensor networks; applications, etc. By this way the journal significantly enriches the appropriate databases of knowledge.

Sensors & Transducers journal has a very high publicity. It is indexed and abstracted very quickly by Chemical Abstracts, EBSCO Publishing, IndexCopernicus Journals Master List (ICV=6.13 in 2011), Open J-Gate, ProQuest Science Journals, CAS Source Index (CASSI), Ulrich's Periodicals Directory, Scirus, Google Scholar, etc. Since 2011 to 2014 Sensors & Transducers journal was covered and indexed by EI Compendex (CPX) index (including a Scopus, Embase, Engineering Village and Reaxys) in Elsevier products. The journal is included in the IFSA List of Recommended Journals (up-dated 9.12.2014), which contains only the best, established sensors related journals. The global Impact Factor of the journal is 0.705 (2013).

Physical, chemical and biosensors Digital, frequency, period, duty-cycle, time interval, PWM, pulse number output sensors and transducers Theory, principles, effects, design, standardization and modelling Smart sensors and systems Intelligent sensors and systems Sensor instrumentation Virtual instruments Sensors interfaces, buses and networks Signal processing, sensor fusion Remote sensing Frequency (period, duty-cycle)-to-code converters, ADC Technologies and materials Nanosensors and NEMS Microsystems Applications

Further information on this journal is available from the Publisher's web site: http://www.sensorsportal.com/HTML/DIGEST/Submission.htm

Subscriptions

An annual subscription includes 12 regular issues and some special issues. Annual subscription rates for 2015 are the following: Electronic version (in printable pdf format): 590.00 EUR; Printed with b/w illustrations: 950.00 EUR; 25 % discount is available for IFSA Members.

Prices include shipping costs by mail. Further information about subscription is available through IFSA Publishing's web site: http://www.sensorsportal.com/HTML/DIGEST/Journal_Subscription.htm

Advertising Information

If you are interested in advertising or other commercial opportunities please e-mail [email protected] and your enquiry will be passed to the correct person who will respond to you within 24 hours. Please download also our Media Planner 2015: http://www.sensorsportal.com/DOWNLOADS/Media_Planner_2015.pdf

Instructions for Authors

Please visit the journal web page http://www.sensorsportal.com/HTML/DIGEST/Submission.htm Authors must follow the instructions very carefully when submitting their manuscripts. Manuscript must be send electronically in both: MS Word for Windows and Acrobat (pdf) formats by e-mail: [email protected]

http://www.sensorsportal.com/HTML/BOOKSTORE/Thermometry.htm

Documents

Sensors & Transducers · Sensors & Transducers Journal is a peer review international jo urnal ... Research of an Optoelectronic Current Transformer Based on a Designed Magneto-Optic