A novel defect detection system for complex freeform surface structures

In recent years, freeform surfaces have been widely used in various industries with the rapid development of manufacturing technology. However, the complex three-dimensional structures of the freeform surface products pose challenges to current visual inspection systems. To address these challenges, this paper proposes a novel and deep learning-based defect detection method for complex freeform surface structures in the injection industry. This method utilizes a system of Multi-Lights Multi-Cameras with Polarizer to eliminate glare caused by freefrom surfaces and capture high-resolution images. In addition, an improved YOLOv5 model is employed to detect small defects in images. Experimental results on our self-made dataset indicate that the improved model achieves a mean average precision (mAP) of 86.2% with 5.1% higher than the baseline, which demonstrates that the proposed method provides a promising approach to facilitate the development of visual inspection for products with complex freeform surfaces.