CWRU

Birefringent Sensors for Motors

Dukki Chung, Francis L. MeratCase Western Reserve University

Fred M. Discenzo, James H. HarrisRockwell Automation

Presented at SPIE Photonics East �98: Metrology and Inspection � Three-Dimensional Imaging, Optical Metrology, and Inspection, Boston, November 1998.

CWRU

Introduction

� Strain Measurement

� strain gage� provides point-by-point data

� photoelasticity� provides point-by-point data or full-field data

� non-destructive measurement

� static and dynamic measurements

� higher bandwidth

CWRU

Photoelasticity

� Birefringent materials have the ability to resolve an impinginglight vector into two orthogonal circularly polarized componentswhich propagate with different velocities through the material.

� Transparent photoelastic materials such as some polymericplastics or glasses become birefringent when stress is applied.

� When linearly polarized light is transmitted through abirefringent plastic relative phase retardation will occur.

� The light exiting a birefringent plastic can be passed through alinear polarizer to convert the phase retardation into a two-dimensional intensity patterns.

CWRU

Optical Strain Analysis

CWRU

Photoelastic Strain Analysis

� primarily used to provide a qualitative analysis ofthe deformation or residual strain in a component

� typically needs human interpretation for properanalysis

� neural network image processing is proposed foranalyzing the fringe patterns of a shaft couplermade from birefringent plastic.

CWRU

Sensor Description

� polycarbonate plastic coupling with a high strain-opticalcoefficient

� coating of aluminum filled epoxy on the inner surface ofthe plastic coupling to reflect light back out

CWRU

Conceptual Optical Torque Sensor

PhotosensorArray

Neural NetworkTorque Estimator

EstimatedTorque

Polarized Light

PolarizationFilter

PhotoelasticPlastic

Coupling

LoadMotorShaft

DriveMotorShaft

CWRU

Neural Network

� nonlinear processing elements operating in parallel andarranged roughly similar to biological neural networks.

� processing elements (nodes) are connected via weights(synapses) that are typically adapted during training phase.

� incoming signals (stimulus) are multiplied by the weights,and summed at the processing elements, or nodes.

� can learn a mapping between the given inputs andcorresponding outputs through training samples.

CWRU

Neural Network

...

...

......

CWRU

Experimental Setup (Static Test)

CWRU

Typical Optical Torque Sensor Images

(a) 20 pound-inches (b) 40 pound-inches (c) 60 pound-inches

CWRU

Image Pre-Processing

(254,64)

15 32x32 virtualsensor cells

CWRU

Experimental Setup (Dynamic Test)

CWRU

Static Test Result

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70

Torque (lb-in)

To

rqu

e /

NN

Ou

tpu

t

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

14.0%

Err

or

(%)

Target Training Output Test 1 Output Test 2 OutputTraining Error (%) Test 1 Error (%) Test 2 Error (%)

CWRU

Dynamic Test Result (Low Speed)

0.2

0.3

0.4

0.5

0.6

0.7

0.8

30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70

Torque (lb-inch)

NN

Ou

tpu

t

0%

10%

20%

30%

40%

50%

60%

Err

or

(%)

Target Training Output @ 400 rpm Te st Output @ 400 rpm LB

UB Training Er ror @ 400 rpm Te st Error @ 400 rpm

CWRU

Dynamic Test Result @ 900 RPM

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70

Torque (lb-inch)

NN

Ou

tpu

t

0%

5%

10%

15%

20%

25%

30%

35%

40%

Err

or

(%)

Target Training Output Test Output LB UB Training Error Test Error

CWRU

Dynamic Test Result @ 1500 RPM

0.3

0.4

0.5

0.6

0.7

0.8

0.9

30 32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62 64 66 68 70

Torque (lb-inch)

NN

Ou

tpu

t

0%

5%

10%

15%

20%

25%

30%

35%

40%

Err

or

(%)

Target Training Output Test Output LB UB Training Error Test Output

CWRU

Conclusions

� accurately measure static and dynamic shaft torque values

� static test results showed less than 1% error.

� dynamic test results showed 3.3% error at 900 rpm and3.5% error at 1500 rpm

� for these dynamic tests, the torque fluctuation from thetest apparatus was reflected to the results.

� use non-contacting optical sensing to produce a motortorque sensor

� replace CCD camera by a linear array of photodetectorswith a considerable reduction in system complexity