Improvement of Inspection Training Tools and Validation of the Accuracy of Machine Learning Discriminant Models Using the Results

Recently, almost all the quality inspection work in the manufacturing industry has become automated. However, there are many products for which inspection cannot be automated. For example, because the tip of a dental treatment rotating tool (Diamond bar) is attached to diamond particles, all parts are slightly different. However, the judgments of the inspection are different by the operator. In previous studies, we have developed and improved operator training tools and created machine learning models to improve inspection accuracy by operator judgment and machine learning image judgment. As a result, it was found that there were images that operator judged differently after several training sessions. In addition, the previous tools did not provide a sufficient sample size of the data, so it was not possible to judge whether the students were proficient or not. This study investigated the improvement and proficiency of an inspection training tool with many sample images to reclassify samples that are judged differently by operators. We analyzed the results and extracted images that were prone to errors in judgment. We conducted experiments to see if reflecting the results in the machine learning model effectively improved the model's accuracy. The results showed that, in the case of the workers, no correlation was observed between the correct response rate and the time to answer in training, but the machine learning model using the training results showed an improvement in the correct response rate.