The authors already developed a user-friendly design and training application tool for deep convolutional neural networks (DCNNs) and support vector machines (SVMs). The application allows students and novice engineers to easily design and train DCNNs and SVMs even if they are not familiar with the software development using C++ or Python. For example, DCNNs generally have several blocks consisting of convolutional, ReLU and pooling layers to accept image files (or feature maps) and produce more characterized ones to the following latter hidden layers, which finally lead to fully-connected layers and a softmax function layer for output. Users can easily design such a DCNN by using the tool.
In this presentation, first of all, a binary classification method using DCNNs, SVMs and template matching techniques is introduced for the defect detection of resin molded articles with various small defects. Two types of SVM structures are firstly designed using the application, then they are trained using typical OK images without any defect in order to be able to distinguish images including defects from other normal images. It is assumed that the defects are crack, burr, protrusion, chipping, spot, fracture, etc. which appear in the manufacturing process of resin moulded articles. Two types of pretrained DCNNs, i.e., our proposed sssNet and well-known Alexnet, are severally incorporated into the fore parts of the two SVMs as feature extractors, in which convoluted feature vectors extracted by the DCNNs are used as input vectors to the SVMs. The performances of the SVMs incorporated with the two types of DCNNs are shown through training and classification experiments. In addition, a template matching technique is further applied to the SVM using AlexNet to narrow important target areas from original training and test images. This enhances the reliability and accuracy for binary classification using the SVM.
Besides the defect detection, an application using frequency analysis techniques is also introduced for anomaly detection of an NC machine tool for woodworking. The method of the frequency analysis is called the spectrogram, which provides the function of visual representation of spectrums with the frequency, time and strength. The spectrogram can be obtained based on the short-time Fourier transform (STFT), in which the short term sampling period can be changed, e.g., from one to ten seconds. It is assumed that the anomaly phenomena include unexpected sounds and vibrations due to, e.g., undesirable chipping of a router bit and/or variation of cutting force. The effectiveness and feasibility for anomaly detection of the NC machine tool are shown.