Classification and detection of cutting tools by deep learning method

In this thesis study, C, R. S, Q etc. used in turning operations and according to ISO standards. The so-called teams were classified and predicted using the deep learning method. For this purpose, a data set was created using images of cutting tools with different geometric shapes. The images in this dataset were augmented using image augmentation methods. Then, the images in these obtained data sets were trained, tested and verified using CNN, Xception, ResNet, LeNet, AlexNet and GoogleNet network architectures. Images of cutting tools were classified according to the codes in the ISO standard (C, R, S, Q, etc.) and the images were estimated accordingly. Validation values of cutting tool images trained with CNN, Xception, ResNet, LeNet, AlexNet and GoogleNet network architectures are respectively obtained as 91%, 99%, 13%, 60%, 97% and 13%. It has been seen that Xception, CNN and AlexNet give the best results in these network architectures. It was concluded that the success rate of images trained with ResNet, LeNet and GoogleNet was low. As a result, the images of the cutting tools were evaluated with the developed and trained deep learning method. Thus, a method was developed that predicts what type of cutting tool is used in turning and classifies it in accordance with the processing method. With this image processing method, the classification of cutting tools using their different features and their effective use in automation systems will contribute to the digitalized industrial field.