Enhancing automated fiber placement process monitoring and quality inspection: A hybrid thermal vision based framework

Abstract

Automated Fiber Placement (AFP) has revolutionized composite manufacturing, yet quality assurance remains challenging due to the significant impact of emerging defects on part quality and the current reliance on time-consuming manual inspection protocols. This paper presents a comprehensive hybrid framework that enhances AFP process monitoring and quality inspection by integrating thermal vision with innovative methodologies. Our framework combines model-based and data-driven algorithms across three modules to address key AFP inspection tasks, including in-situ monitoring, dynamic tow identification, defect detection, segmentation, localization, and quantitative lay-up quality evaluation. The setup-independent spatial–temporal analysis algorithm estimates tow boundaries with sub-pixel accuracy. An optimized SVM classifier, trained on an extensive AFP defect database, achieves a defect detection accuracy of 96.4% and an F1-score of 96.43%, meeting industry standards. The active contours-based segmentation and localization module provides critical qualitative traits such as defect shape, size, and location. Moreover, the novel Defect Area Percentage (DAP) metric enables precise quantitative defect impact evaluation at both the course and tow levels. By consolidating qualitative and quantitative outcomes, the system offers real-time high-level feedback for informed decision-making, significantly improving process performance and reducing machine downtimes. This proactive approach advances AFP process monitoring and quality inspection and positions our framework as a promising solution for next-generation composite manufacturing.