Multi-source data modelling and graph neural networks for predictive quality

Le résumé fourni par la source

State-of-the-art predictive quality (PQ) applications use machine learning and deep learning methods to learn patterns and classify a product's quality. Models typically estimate quality control labels based on process data, such as machine control and sensor data, substituting time-consuming manual quality checks. Most current advancements in the field of PQ focus on machine-level quality prediction. The quality of the final product is, however, influenced by each manufacturing process of the production line. Performing PQ at the production line level involves challenges such as merging multiple data sources and considering process interdependencies. This paper thus explores a graph representation of manufacturing data and the use of graph neural networks to classify products' quality and develop a predictive quality solution. The graph representation of the data tackles the mentioned challenges and allows PQ at the production line level. Experiments have shown promising results for an assembly line use case.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.