Ultrasonic signal classification and imaging system for composite materials via deep convolutional neural networks

Automated ultrasonic signal classification systems are finding increasing use in many applications for the recognition of large volumes of inspection signals. Wavelet transform is a well-known signal processing technique in fault signal diagnosis system. Most of the proposed approaches have mainly used low-level handcraft features based on wavelet transform to encode the information for different defect classes. In this paper, we proposed a deep learning based framework to classify ultrasonic signals from carbon fiber reinforced polymer (CFRP) specimens with void and delamination. In our proposed algorithm, deep Convolutional Neural Networks (CNNs) are used to learn a compact and effective representation for each signal from wavelet coefficients. To yield superior results, we proposed to use a linear SVM top layer in the training process of signal classification task. The experimental results demonstrated the excellent performance of our proposed algorithm against the classical classifier with manually generated attributes. In addition, a post processing scheme is developed to interpret the classifier outputs with a C-scan imaging process and visualize the locations of defects using a 3D model representation.