Travail de fin d'études et stage[BR]- Travail de fin d'études : Design and Development of a Computer-Aided Quality Inspection Station Using Hybrid AI and Rule-Based Image Processing Techniques[BR]- Stage d'insertion professionnelle : Eutomation & Scansys (Eupen, BE)

Abstract

In production settings where accuracy and consistency are key, visual quality inspection remains a core component of the production process. Traditional rule-based approaches, while being rapid and reliable, lack adaptability for coping with complex or unforeseen varieties of defects. Artificial intelligence techniques offer greater adaptability but are computationally and data-intensive, and at times, not interpretable. This thesis proposes the development of a hybrid quality inspection station which combines both approaches and takes advantage of their respective strengths.

The system is deployed using a modular Python platform so that both conventional image processing techniques and AI models can be integrated into an integrated, user-configurable framework. The approach is integrated with industrial Programmable Logic Controllers (PLCs) to provide for compatibility in real-world production environments and control systems. In particular, the system targets surface inspection of machined components, where the detection of over-machining defects or subtle cracks requires precision and context adaptation.

Development involved a detailed decomposition of hardware constraints, system latency, and model complexity versus inference time trade-offs. Different inspection pipelines were explored, ranging from rule-based threshold techniques to convolutional neural networks. The outcome shows that the hybrid architecture enables a balanced optimum between interpretability and strong detection, and modular upgrades or reconfiguration based on production needs are enabled.

In conclusion, this thesis provides a robust and scalable industrial real-time quality control platform that balances deterministic reasoning with intelligent inference methods. It provides a foundation for horizontally deployable and scalable expansion and integration of additional AI capability within visual inspection systems.