Intelligent machine vision for industrial inspection

Automating visual inspection in manufacturing is crucial to ensure product quality at high throughput. This paper presents an intelligent machine vision system for industrial defect detection using a YOLOv5-based deep learning model. The system is designed for real-time performance and high accuracy in detecting various defects on production parts. We describe the YOLOv5 architecture and its adaptation for defect inspection, including dataset preparation and training procedures. A case study on printed circuit board (PCB) defect detection is conducted, simulating an inline inspection scenario. The trained model achieves a mean average precision (mAP) of over 92% in identifying defect types such as missing holes, cuts, and surface blemishes, while maintaining real-time inference at 25+ frames per second (FPS). We also evaluate deployment latency on an edge GPU device and analyze the trade-off between model size and detection accuracy. Experimental results demonstrate that our YOLOv5-based system can reliably detect defects with high precision and low false alarm rates, meeting industrial requirements for both accuracy and speed.