conférence industrie 4 - alliance-tt.ch · conférence industrie 4.0 data analysis to improve...
TRANSCRIPT
Conférence Industrie 4.0 Data analysis to improve production efficiencyProfesseur Pierre Bressy, 2 mai 2019
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
On the menu…
�2
Industry 4.0 and image analysis, a complex union:image analysis and AI,augmented reality,
Basics of machine vision:2D / 3D acquisition,image processing,evaluation and integration in the production tool,
Some projects:blister pack automated inspection,technical drawing digitalisation,3D automated inspection,Jewellery inspection,Traceability through blockchain.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
3 sites in Yverdon-les-Bains
HEIG-VD: University of Applied Sciences
�3
Route de Cheseaux Y-Parc
Saint Roch
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Academic side
University of Applied Sciences
�4
2000students
200professors diplomas/year
500
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Available technologies and hardware
University of Applied Sciences
�5
projects in 2017
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Research side
University of Applied Sciences
�6
13research institutes
300researchers projects in 2018
200+
for more information heig-vd.ch/rad
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Thematic groups
Institute for Industrial Automation
�7
HEIG-VD
Institute for Industrial Automation 13 research institutes…
Thematic groups
Electronics
14 professors25 engineers
projects
diploma work
Innovation
Automatisation 4.0
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Industry 4.0 - Image analysis: a complex union
�8
Industry 4.0 and image analysis, a complex union:image analysis and AI,augmented reality,
Basics of machine vision:2D / 3D acquisition,image processing,evaluation and integration in the production tool,
Some projects:blister pack automated inspection,technical drawing digitalisation,3D automated inspection,Jewellery inspection,Traceability through blockchain.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Industry 4.0 - Image analysis: a complex union
�9
Image Processing is resource consuming
- Image = 1D, 2D or 3D- Image flow = 2D, 3D or 4D
EmbeddedProcessing
Images
Relevantinformation
Pure Cloud computing
Production toolFog computing
Production tool
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Industry 4.0 - Image analysis: a complex union
�10
Image analysis and AI
DogCatCNN
Typical classification problem(Kaggle, 2013)
Classification (OK / Not Good) is related to:- position of the object in the image,- size,- orientation,- color…
Learning needs: - bank of good images (part without or acceptable default),- bank of bad images (part with default),- time.
OK NG NG NG
Training set: 25’000 imagesTesting set: 12’500 images
good classification >90%
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Industry 4.0 - Image analysis: a complex union
�11
Tools for Image analysis with AI
Prototyping:
- Matlab + machine learning toolbox,- Python + machine learning libraries
- pandas (data mangement),- scikit-learn,- tensor flow,- keras,- pytorch,- dearborn / matplotlib,- openCV,- Jupyter (SDK),
- Google Vision API,- Google AI platform.
Development:
- MLPACK (C++),- tensor flow (C++),- tinyCNN (C++),- Vision + CoreML (swift - Apple),- openCV (C++),- Microsoft Cognitive ToolKit (C++ / C#).
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
"Augmented reality" Bachelor diploma work
�12
Loading in progress
80 %
Status
Error - Tray unlocked
Microsoft "Hololens"
2D ‘tags’ reading
- space localisation, - image analysis, - auto-diagnostic.
Industry 4.0 and image analysis
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019 �13
Help for quality control and for machine
assembly
Guidance for machine maintenance
"Augmented reality" Research axes
Industry 4.0 and image analysis
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Basics of machine vision
�14
Industry 4.0 and image analysis, a complex union:image analysis and AI,augmented reality,
Basics of machine vision:2D / 3D acquisition,image processing,evaluation and integration in the production tool,
Some projects:blister pack automated inspection,technical drawing digitalisation,3D automated inspection,Jewellery inspection,Traceability through the blockchain.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Basics of machine vision
�15
2D / 3D image acquisition
Getting the image
- 1D: line camera (high speed),- 2D: standard camera (color, N&B, IR), interférométrie,- 3D:
- line / grid scanner ; construction of points cloud,- stereo vision (2x 2D camera) ; depth map,- time of flight sensor.
Lighting the scene
- flash or continuous,- flat panels, dome / tunnel light (LED),- chose appropriate colour,- frontlight or backlight,- collimated light.
Filtering the light
- reduce bad reflections,- keep specific colour.
Lens
- focus, focale length, aperture,- dimension of the scene,- purpose of analysis (telecentric lens for measures),- manage the image’s deformations.
Retrieve the image
- link between sensor and processing unit,- transfert duration,- acquisition sequence (multiple camera, multiple lightning, HDR).
Other points
- sensitivity to environment (external light pollution),- dust,- vibrations,- processing time,- calibration.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Basics of machine vision
�16
Image processing
Preprocess the image to enhance its content:
- denoising to remove artefacts,- contrast enhancement,- border detection,- adapt the image size (down/up scaling)- create regions of interest,- rotation, alignment. Content analysis:
- object characteristics:- height, width, area, boundaries,- geometric information,- form analysis,
- count objects,- comparison with models,- colorspace,- radiomics features (2D/3D):
- shape, - gray-gray-lever analysis, - neighbouring…
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Basics of machine vision
�17
Evaluation and integration in the production tool
Criteria for evaluation:
- specify a list of criteria to accept or reject the sample,- set limits for acceptance,- create a logic for criteria evaluation,- link with fabrication order for flexibility.
Output can be:
- logical: accepted / rejected / need manual check,- real information (size, precision of position, drift…),
Production tool integration:
- feedback for production analysis,- generation of statistics,- auto-adaptation according to production variants,- surveillance / alarms.
Phases:
- creation of the image database,- prototyping the application,- validation of the algorithms,- develop the industrial software,- test and integration,- qualification.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Some projects
�18
Industry 4.0 and image analysis, a complex union:image analysis and AI,augmented reality,
Basics of machine vision:2D / 3D acquisition,image processing,evaluation and integration in the production tool,
Some projects:blister pack automated inspection,technical drawing digitalisation,3D automated inspection,Jewellery inspection,Traceability through blockchain.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Blister pack automated inspection
Machine Vision Projects
�19
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019 �20
Blister pack automated inspection, 80 parameters, 30+ products variants
Machine Vision Projects
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Web-service, OpenCV and Convolutional Neural Network
"Technical drawing digitalisation" project
�21
PDF source file Cotation automatic identification Full annotated output file
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Web-service, OpenCV and Convolutional Neural Network
"Technical drawing digitalisation" project
�22
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
3D analysis framework
�23
Machine Vision Projects
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
3D analysis framework
�24
Machine Vision Projects
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Scratches Surface anomalies Color anomalies Material defects
Jewellery inspection
�25
Stain Cracks Edge default Edge default
Machine Vision Projects
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
Traceability through the blockchain
�26
Industry 4.0 Project
Hot project in brief
- record actions and documents in a blockchain,- help for the product conformity certificate delivery,- automatic reward of the certification entities,- decrease the travelling expenses,- speed up the whole process.
http://www.heig-vd.ch/tinIndustrial Technologies Department Prof. Pierre BRESSY - Institute for Industrial Automation – 2.V.2019
iai-vision.heig-vd.ch
+41 (0)24 557 64 10
HEIG-VDiAiRoute de Cheseaux 1CH-1401 Yverdon-les-bains
�27
Thank you for your attention.