​Description of the use case

Use Case Brose (UC BRO) addresses a critical and recurring quality assurance challenge in high-volume automotive manufacturing, focusing on end-of-line optical testing (EOLT) of door modules. In current production environments, conventional rule-based vision systems are used to detect functional assembly defects such as cable crossings in cable-driven window regulator systems. However, these systems exhibit limited robustness under real production conditions, as variations in part positioning, contamination (e.g. grease), and fluctuating lighting conditions frequently require manual adjustments of parameters and inspection rules. As a result, defect detection reliability is not guaranteed, leading both to false non-OK classifications (pseudo-NOKs) and the risk of undetected defects.

UC BRO introduces an AI-enabled optical inspection system designed to overcome these limitations by leveraging data-driven visual AI approaches.

​

Description of the demonstrator

The AI-based optical inspection system operates in connection with the existing End-of-Line Tester (EOLT), enabling the continuous transfer of image data from the production environment to the AI system for evaluation. In addition, the AI-based evaluations show a stronger alignment with expert assessments, supporting more reliable defect detection in complex inspection scenarios.

​

A complete data pipeline has been established, allowing image acquisition, data transfer, and AI-based analysis to be performed in a stable and reproducible manner. The AI models demonstrate robust performance and improved detection stability compared to conventional rule-based vision systems, particularly under varying production conditions such as changes in part positioning, contamination, and lighting.

​

In parallel, the integration of the AI system into the production IT environment has been prepared through a container-based deployment approach, ensuring scalability and maintainability of the solution.

The demonstrator also provides explainability features, including heatmap-based visualizations, allowing users to understand which image regions contributed to the AI decision.