Conference paper Open Access
Abstract - Automation in CFRP production poses multiple challenges. The material at hand is very un-isotropic and deformable, leading to various difficulties in handling. We believe that visual inspection and quality control are key technologies to improve automation in CFRP production. In this paper, we point out possible ways to exploit modern machine learning methods in the context of CFRP quality control. Taking the example of AFP, we show how to transform prior knowledge about the production process into a probabilistic model. By drawing samples from this model, we demonstrate how to infer hidden variables of the process efficiently. We show how to use the methodology to perform inline defect detection and to reconstruct global process parameters. We present results for artificial and selected real AFP monitoring data acquired during inline process monitoring.