Surface Defect Inspection of a Cutting Tool by Image Processing with Neural Networks

In this research, an image processing method and a system for inspection support of a rod figured cutting tool are developed. As is well known, the visual inspection of a cutting tool by image processing is not easy, because cutting blade have a helical blade structure. To cope with the problem, an experimental facility with rotation and longitudinal tool shift functions to enable acquisition of blade surface pictures along a cutting rod is developed. The type of the defect treated in this paper is the spot of coating on blade surface. To judge the quality of the processed image of blade surface, neural network with autonomous learning is used. The processed image of cutting tool is divided into 64 × 64 blocks used for the input to the neural networks. Before input, each block data is preprocessed applying a edge detection filter and a transformation by the discrete fourier transform (DFT). Using these technologies, the experimental inspection system is built and tested to check the capabilities of the inspection algorithms. The diagnostic performance of the surface defect of a cutting tool was confirmed. There remained a problem to mis judge the normal tools as the defect.