Insufficient Defect Data Deep Learning Techniques for Automotive Components Anomaly Detection

The rapid development of deep learning technology has provided new opportunities for the automotive manufacturing industry, particularly in the area of automotive component quality inspection. Traditionally, quality inspection of automotive components has relied on manual labor, which is time-consuming and labor-intensive, making it difficult to cope with the challenges of mass production and high-speed production lines. Additionally, due to the high yield rate of the production process, the initial implementation of defect detection technology may face difficulties in collecting sufficient defect data. The collected defect and normal data may be highly imbalanced, or only normal sample data may be available, making it challenging to provide enough defect samples for supervised learning methods to build defect detection models. This study aims to explore the feasibility and effectiveness of using the SimpleNet method for anomaly detection in automotive components. By training the SimpleNet with only normal sample data, it is possible to establish a defect detection model. Furthermore, we propose the multi-view enhanced normalization inference process to improve the model robustness performance for the small training data case, which will help improve production efficiency and product quality in the intelligent production lines of automotive manufacturing industry.