Defect detection for large-series automated fibre placement using a neural network-assisted machine vision approach

Tailored reinforcement structures from carbon fibre-reinforced thermoplastics show high potential for minimizing material volume, wall thickness and overall lightweight potential in plastic parts. Integrating small amounts of carbon fibre unidirectional tape material in injection-molded parts has the ability to maximize lightweight potential. Through a novel manufacturing approach developed by AZL, automated fibre placement becomes viable in large-series manufacturing by reducing cycle times down to 5 seconds. Therefore reliable, defect-free manufacturing becomes vital for economic competitivity. In this paper, a novel approach for defect detection using computer vision in combination with neural networks is developed and analysed for applicability in large-series manufacturing. Different occurring defects during processing are detected and classified. Subsequently, classification accuracy is analysed in regard to the predefined defect classes.