An Overview of the Automated Optical Inspection Edge AI Inference System Solutions

The aim of the paper is to provide an overview of Automated Optical Inspection Edge AI Inference System solutions in the digital industry by discussing if and how they enable manufacturers to reach a satisfactory trade-off between customer needs and production costs. Numerous solutions can meet customer and factory needs, from inspection machines to testing boards equipped with cameras installed near the conveyor belt. In all the considered solutions we can implement effective defect detection algorithms, such as the latest YOLO versions based on Deep Learning, to obtain high KPIs, i.e. mean average precision, adequate process capability and high throughput yield. Parallel implementations of edge test systems allow us to further improve production yield, while repeated tests performed in sequence could allow us to approach the precision required by the constraint of near zero defects. The comparison of available solutions using KPI, FR and NFR points out that the advantage of using inspection machines is that they are equipped with user interface and data analysis which helps workers and managers to ensure respectively high quality of production process and effective order management. Their weakness, i.e., the cost of purchase and energy consumption, is the strength of solutions that use computing boards for defect testing at the edge. A demonstrator to evaluate the effectiveness of edge AI solutions based on the test boards available on the market and those developed by the EdgeAI project is outlined.