OpenLMD, Multimodal Monitoring and Control of LMD processing
Jorge Rodríguez-AraújoAIMEN Technology Center, Porriño, Spain
Photonics WEST 2017, 1-2-2017
openlmd.github.io | [email protected] 2
Index
Index
1. Laser Metal Deposition
2. Motivation and innovative character
3. Open Laser Metal Deposition
4. ROS-based robot cell integration
5. Multimodal monitoring and virtualization
6. Image registration and data acquisition
7. Closed-loop laser power control
8. 3D geometrical monitoring
9. Adaptive LMD path planning
10.Conclusions and future work
openlmd.github.io | [email protected] 3
Laser Metal Deposition (LMD) Promising additive manufacturing technique
Parts are built up layer by layer directly from a 3D CAD model
The material is directly deposited on the previous surface
For repair and direct fabrication of pieces
Near-net-shape (close to the final shape)
Manufacturing of large metallic parts
LMD issues Complex setup and adjustment of parameters
Time consuming robot programming for repairing tasks
Thermal heating accumulation and dimensional distortion
Repeatability, rising defects, and metalurllical properties
Traditional off-line process (with constant parameters) becomes unsucessful for large metallic parts
LMD, Laser Metal Deposition
openlmd.github.io | [email protected] 4
Motivation and innovative character
MotionController
Off-line Path Programming
6-AxisRobot Laser Powder
Feeder
PowerController
FlowController
Motivation Lots of industrial robotized laser cells
Mostly manually operated
Lack of integration for monitoring and control
Innovation Retrofit current laser cladding facilities for LMD
Empower robotized laser cells for effective AM
Apply state of the art robotic software solutions
Goals Build a modular architecture for LMD full automation
Reduce heating accumulation for large parts
Adapt robot programming to the real part
Increase geometry accuracy and repeatability
Conventional Robotized Laser Cladding Cell
openlmd.github.io | [email protected] 5
Open Laser Metal Depositon
Concept and approach Open-source solution for on-line multimodal monitoring and control of LMD
Modular set of software components. Built on ROS (Robot Operating System)
Robotics, machine vision, embedded control, machine learning, big data
Full compatible with current robotized laser cladding cells
Focus on interoperability and standardization
ROS-based architecture Multiprocessing architecture based on message publishing
Multi-node and multi-machine
Modular (e.g. robot, laser, camera)
Synchronized data acquisition (common timestamp)
High bandwidth data management (i.e. images)
Visualization tools and components (e.g. rviz)
Advanced robotics environment
openlmd.github.io | [email protected] 6
ROS-based robot cell integration
PowderFeeder
Fiber Laser
6-Axis Industrial Robot
PC Controller
Cladding Head
ROS-Driver(ABB Rapid)
Geometrical Cell Description
(URDF)
STATEPUBLISHER
Laser Source(slave)
Powder Feeder(slave)
COMMANDSERVER
PowerSpeed
Powder flow
Motion path
States
Commands
ROBOT
Process parameters
AIMEN’s LMD robotized laser cell
ROS-based integration of modular laser cells The PC commands the robot integrating interfaces and modules with ROS
The robot controls all the cell elements
ROS components for robot integration Geometrical description (URDF)
ROS driver
openlmd.github.io | [email protected] 7
Multimodal monitoring and virtualization
Multimodal Cladding Head
3D System
TachyonMWIR
NIR
MWIR+NIRMultispectral
Imaging
LMD Cell Virtualization
Multimodal monitoring approach Coaxial SWIR/MWIR images (thermal monitoring): NIT microcore (1000fps) [1-3um]
Coaxial NIR images (surface monitoring): CMOS camera (100fps) [830-880nm]
Off-axis 3D system: on-line 3D point cloud scanning (50fps)
openlmd.github.io | [email protected] 8
Coaxial sensors registration Image registration: process of transforming different sets of data into one coordinate system
Data acquisition High throughput (28MB/s)
NIR + MWIR + 3D point cloud + robot
Data management and analysis
Bag files and Pandas DataFrames
NIT NIR(0, 0)
velx
y
Image registration and data acquisition
Calibration → Projection
openlmd.github.io | [email protected] 9
Closed-loop laser power control High speed SWIR/MWIR thermal meltpool monitoring (1000fps)
NIT Europe Tachyon 1024 microcore camera
Meltpool geometrical monitoring (elliptical approximation)
Increased geometry repeatability and reduced dilution and heat accumulation
Closed-loop laser power control
Meltpool, ellipse approach
Wid
th (m
m)
Time (s)
openlmd.github.io | [email protected] 10
3D triangulation
workingtable
nozzle
3D geometrical monitoring
Industrial Robotic Laser Cell
ROBOTROS-DRIVER
CAMERAIDS-DRIVER
State Publisher Peak Finder
Robot PoseTool-Camera
Laser TriangulationCalibration
3D ProfileCamera Pose
3D Point CloudWorking Cell Coordinate
On-line 3D geometrical reconstruction
On-line 3D point cloud registration Real-time point cloud registration
Actual metric measurement (mm)
Direct acquisition in robot coordinates
openlmd.github.io | [email protected] 11
Adaptive LMD path planning
Adapts the path to the real geometry 3D vision guided full automated laser cladding repair of complex metallic parts
Automatic generation of robot trajectories
1. Part scanning and filtering
2. Surface selection and path generation
3. Repair job generation and supervision
3D Filtering
Initialization(setup)
Scan layer
Depth mapTargetDepth map
Disparity
Data
Layer pathplanning
Layer Path Planning(geometrical control)
Laser Cellsupervisor
Robotized Cell
0 Finished
Repair Job
3D geometrical control
Coated surface
openlmd.github.io | [email protected] 12
Automatic coating of surfaces
3D scanning and robot path planningEnabled by the 3D point cloud directly provided by the
3D geometrical monitoring solution
1. Workarea scanning (direct part information in
1. cell coordinates) [mm]
2. 3D point cloud projection (2D Zmap image)
3. Surface selection directly in the 2D image
4. Segmentation of the Zmap image
5. Contours calculation from the segmented surface
6. Contours and Zmap feed the path planner
7. A new path is automatically calculated from that
1. information
A second scanning after the process enables an
adaptive path planning strategy in a full automatic way
Surface selection
3D scanned surfaces
openlmd.github.io | [email protected] 13
Conclusions Integration and data acquisition
Spatial reference system and temporally synchronized
Monitoring and control
Real-time closed-loop laser power control
Adaptive path planning
Work in-progress Embedded image monitoring and control
Big data and deep learning approaches
Data acquisition High throughput (28MB/s)
NIR + MWIR + 3D point cloud + robot
Data management and analysis
Bag files and Pandas DataFrames
Robot
Pose Process speed
3D geometry
Point cloud (<0.5mm)
SWIR/MWIR-NIR
2D melt pool geometry
Thermal distribution and texture
Conclusions and future work
Reconfigurable
Modular and reconfigurable
Interoperability
Large parts
Low-cost solution
Scalability
AIMEN – Central y Laboratoriosc/ Relva 27 A
36410 – O PORRIÑO (Pontevedra)Telf.+34 986 344 000 – Fax. +34 986 337 302
Thank you for your attention
Jorge Rodríguez-Araújo | Research EngineerPh +34 986 344 000 | [email protected]
www.aimen.es | [email protected]
This work has receive funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement Nº 637081. The dissemination of results herein reflects only the author’s view and the Commission is not responsible for any use that may be made of the information it contains.
http://openlmd.github.io