Study on Developing a Comprehensive Inspection System that Parallel Improves the Accuracy of Manual and Automatic Inspections

This study summarizes a three-year project targeting dental component manufacturing sites. The target inspection departments have always conducted manual inspections twice each. This department wanted to be performed automatically at least once by introducing an automatic inspection machine. We have two problems that need to be solved. The first is to equalize the judgment criteria that differ from one inspection operator to another, and the second is to develop an automatic inspection tool with the same accuracy level as the inspection operator’s judgment criteria. However, the target product, a rotary tool for dental treatment (diamond bar), has diamond particles attached to its tip; every part is slightly different. Therefore, creating an inspection tool with a simple threshold setting was impossible. In this study, we developed an automatic inspection tool using machine learning, and at the same time, we developed an inspection training tool to equalize operators’ skills. Each tool was repeatedly improved through verification experiments. In addition, we developed feedback rules for the results obtained from the training tools to the training data for the machine learning model to improve the accuracy of the discriminant model. Furthermore, we have proposed a labeling tool that establishes criteria for judging whether a product is quality or defective in consideration of the introduction of new products, thereby realizing the continuous introduction of products and the stabilization of inspection operations.