Skip to Main content Skip to Navigation

Automation of quality control and reduction of non-compliance using machine learning techniques at Faurecia Clean Mobility

Abstract : Quality control applications in the automotive industry are numerous. Automotive companies are working on the automation of these applications and add a particular focus on securing their processes.In this thesis, the first contribution proposes an automated quality control for component presence. It allows to classify if the component is present, missing or has been replaced by another component. The algorithm can achieve an accuracy of 100% with live tests.The second contribution focuses on automating the quality control of welding seams that haven't been reached by leak tests, covering external aspects of welding defects. The images collected from the plants are not balanced, data augmentation techniques have been applied to reach more balanced dataset. In this contribution, a standard deep learning algorithm applied on raw data has been compared to data augmentation approaches. The target, defined by the plant, 97% of defected reference parts detection, has been reached on half of the welds. The challenge remains present on the other half.In the third contribution, deep learning model explainability and welding seams classification accuracy are combined. A hybrid approach of CNN-Machine learning classifier is proposed to improve the accuracy reached in the second contribution. This work presents a new model-driven optimization reaching an accuracy above 98% when applied on welding seams dataset.In the fourth contribution, ceramic monolith position and their relative breakages are studied. The quality control of these monoliths should be done during the production of exhaust pipes. A comparison between image processing filters for straight lines detection is presented. The tests were carried out by applying a rotation to the ceramic in a 5-degree step. The results show that canny filter applied with hough lines allows to achieve an accuracy of 99%.Finally, a Human Machine Interface (HMI) has been developed aiming to provide a Plug & Play system to the plants. The integration of this digital solution in the plant's cycle time will be discussed, as well as its architecture.
Document type :
Complete list of metadata
Contributor : ABES STAR :  Contact
Submitted on : Monday, September 12, 2022 - 2:52:12 PM
Last modification on : Tuesday, September 13, 2022 - 3:16:56 AM


Version validated by the jury (STAR)


  • HAL Id : tel-03775370, version 1


Charbel El Hachem. Automation of quality control and reduction of non-compliance using machine learning techniques at Faurecia Clean Mobility. Artificial Intelligence [cs.AI]. Université Bourgogne Franche-Comté, 2022. English. ⟨NNT : 2022UBFCD011⟩. ⟨tel-03775370⟩



Record views


Files downloads