Surface quality control of face-milled parts using image processing

In this thesis study, a MATLAB-based image processing and deep learning system was developed to perform automatic quality control of metal surfaces processed on CNC machines. In the proposed method, surface images were analyzed using a convolutional neural network (CNN) based on the MobileNetV2 architecture to classify defective and non-defective regions. For images identified as defective, three separate masks were created to precisely detect the defects. These are the reference manual defect mask, the prediction mask automatically generated by the deep learning model, and the part edge mask that limits the analysis region. These masks were combined using appropriate image processing algorithms, and the system's classification performance and segmentation accuracy were evaluated. The results demonstrate that the proposed system achieved high performance with an accuracy rate of 94.98\%, significantly reducing the need for manual inspection by automating the surface quality control process. This study proves that MATLAB's integrated image processing and deep learning tools offer a reliable, flexible, and practical solution for industrial quality control applications.