Automated Classification of Ultrasonic Signal via a Convolutional Neural Network

Abstract

Ultrasonic signal classification in nondestructive testing is of great significance for the detection of defects. The current methods have mainly utilized low-level handcrafted features based on traditional signal processing approaches, such as the Fourier transform, wavelet transform and the like, to interpret the information carried by signals for classification. This paper proposes an automatic classification method via a convolutional neural network (CNN) which can automatically extract features from raw data to classify ultrasonic signals collected of a circumferential weld composed of austenitic and martensitic stainless steel with internal slots. Experiments demonstrate that our method outperforms the traditional classifier with manually extracted features, achieving an accuracy rate of classification up to 0.982. Furthermore, we visualize the shape, location and orientation of defects with a C-scan imaging process based on classification results, validating the effectiveness of the results.