COMPARATIVE STUDY OF IMAGE PROCESSING AND CNNS FOR SURFACE DEFECT DETECTION IN NUT PRODUCTION

In production, especially in industry, it is of great importance to increase the service life of products and maintain the safety of systems consisting of many components at the maximum level. Products must be analyzed using the correct methods and techniques. These analyses are performed using different techniques and procedures, particularly in metal and material production. Magnetic Particle Testing (MPT) is a test used for non-destructive inspection of ferromagnetic materials. However, defect detection here is performed through human visual inspection. This situation negatively affects the accuracy and completeness of the detection process. In this study, a system has been developed using a camera and a specially developed kit capable of running artificial intelligence models to detect defects that may occur on the surface or near the surface of nuts, one of the critical components widely used in the industrial field. Using this system, a balanced dataset consisting of two classes was created. This dataset is trained using deep learning-based models that are currently applied in different fields and have achieved successful results. The test results obtained are evaluated and compared using different metrics. In the evaluations, the VGG-16 approach achieved the best result with an accuracy rate of 89.33%.