Thermal inspection and quality assessment for AFP processes via automatic defect detection and segmentation

Automated fiber placement (AFP) technology, while highly beneficial, is susceptible to defects that compromise the final product’s mechanical integrity. Traditional manual inspection methods are labor-intensive, error-prone, and result in significant downtime. This study introduces an innovative framework for AFP process inspection and quality assessment using thermal imaging, machine learning algorithms, and computer vision techniques. The system comprises defect detection, defect segmentation, and quality assessment components. By providing real-time feedback, it offers qualitative defect attributes (shape, size, and location) and a novel quantitative defect impact metric. An active knowledge-driven decision support system (AFP-DSS) aids operator maintenance and repair decisions. Experimental validation shows the defect detection component achieving 96.4% accuracy with a 2.8% false negative rate, and the defect segmentation component attaining 93.2% mean pixel accuracy and a mean Intersection over Union (IoU) score of 0.72. Operating in real-time, the system effectively reduces machine downtime, enhances production quality, and improves the economic viability of AFP technology.