Incremental Learning Approach for an Industrial Inspection System - - PowerPoint PPT Presentation
Incremental Learning Approach for an Industrial Inspection System MATLAB Expo 2019, Bern Dr. Jianyong Wen & Ralph Stephan, Stubli Sargans AG MATLAB Expo 2019, Bern Stubli at a Glance 2 2019-05-23 Stubli Sargans AG MATLAB Expo
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 2
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 3
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 4
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 5
MATLAB Expo 2019, Bern
Our Textile manufacturing products and services range from ▪ Shedding solutions for frame weaving and Jacquard weaving to ▪ Carpet weaving systems, ▪ Weaving systems for technical fabrics, ▪ Automation solutions for sock knitting machines, ▪ Automated weaving preparation systems.
2019-05-23 Stäubli Sargans AG 6
MATLAB Expo 2019, Bern
▪ Located in Sargans, SG, since 1994 ▪ ~120 employees (~30 in R&D) ▪ Product lines ▪ Drawing-in ▪ Tying ▪ Leasing ▪ Inspection
2019-05-23 Stäubli Sargans AG 7
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 8
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 9
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 10
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 11
MATLAB Expo 2019, Bern
Idea Model Design Model Verification SoC Integration Integration Test Field Application Data Collection Feedback
▪ Real-time industrial fabric inspection systems face challenges of a great number of pattern variations, fast and easy training process. ▪ Strongly imbalanced datasets ▪ Limitation of hardware resources in embedded systems
2019-05-23 Stäubli Sargans AG 12
MATLAB Expo 2019, Bern
▪ The incremental model developed at Stäubli Sargans AG consists of two process stages, combining machine learning and deep learning models.
2019-05-23 Stäubli Sargans AG 13
MATLAB Expo 2019, Bern
Incremental Learning Sub- model
Image Pre - processing Initial Learning Sub-model
2019-05-23 Stäubli Sargans AG 14
MATLAB Expo 2019, Bern
Developing models Optimization parameters Functionality test HDL Code generation Verification Efficiency and Speed Test
2019-05-23 Stäubli Sargans AG 15
MATLAB Expo 2019, Bern
Filtering Segmentation Feature extraction Debayer Resizing
▪ Classification based on extracted features from pre-processing
2019-05-23 Stäubli Sargans AG 16
MATLAB Expo 2019, Bern
Corners Statistical Features Edges Gray Level Co- Occurrence Matrix Lawmasks
Classifier
2019-05-23 Stäubli Sargans AG 17
MATLAB Expo 2019, Bern
Input Image Conv. Batch Norm Relu Class Output Maxpool FC Soft- max
2019-05-23 Stäubli Sargans AG 18
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 19
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 20
MATLAB Expo 2019, Bern
▪ Applicability
❖ The CNN sub-model improves classification accuracy during production process
(incremental learning). ▪ Efficiency
❖ Based on the initial sub-model, the complexity of the CNN sub-model can be
reduced (resource and speed efficiency).
2019-05-23 Stäubli Sargans AG 21
MATLAB Expo 2019, Bern
2019-05-23 Stäubli Sargans AG 22
MATLAB Expo 2019, Bern
Idea Model Design Model Verification SoC Integration Integration Test Field Application Data Collection Feedback
▪ Challenges ▪ matching of tools and data sets ▪ debugging with blackbox IP ▪ limited computing power ▪ large databases
2019-05-23 Stäubli Sargans AG 23
MATLAB Expo 2019, Bern
Idea Model Design Model Verification SoC Integration Integration Test Field Application Data Collection Feedback
▪ Benefits ▪ fast and simple code generation ▪ step-by-step implementation and verification (controlled progress) ▪ consistent and comparable intermediate results ▪ validation of results together with customers
2019-05-23 Stäubli Sargans AG 24
MATLAB Expo 2019, Bern
Idea Model Design Model Verification SoC Integration Integration Test Field Application Data Collection Feedback
▪ There are no limit to imagination ▪ Limits are given by ▪ Our knowledge and implementation capacity ▪ The capabilities and limitations of our tools
2019-05-23 Stäubli Sargans AG 25
MATLAB Expo 2019, Bern
Idea Model Design Model Verification SoC Integration Integration Test Field Application Data Collection Feedback
www.staubli.com