Research on ultrasonic defect imaging based on a neural network with Gaussian weight function fusion model

Abstract

Ultrasonic testing is widely used in the industrial field because of its high sensitivity, safety and low-cost, which plays an increasingly important role in detecting defects on industrial. However, at present, defect detection only stays in scanning surface defects and classifying the internal defects of the specimen. The accuracy of quantitative imaging of the morphology of the internal defects of the specimen is still an issue to be improved. In this paper, the simple defects in several specimens are detected by ultrasound based on finite element simulation software and the accuracy of the signal can be verified by ultrasonic experiment. Then, the features can be extracted by signal processing method. The neural network imagines the inversion results with the Gaussian weight function fusion model. Finally, this paper uses the signals of simple shape defects as the training set to detect the shapes of complex defects and obtain the corresponding imaging results.