Ultrasonic non-destructive test method and system using deep learning, and auto-encoder-based prediction model training method used therefor

Abstract

An ultrasonic NDT method and system, which can extract and analyze a defect signal even when a signal reflected from a defect interferes with a unique initial pulse of an ultrasonic transducer or a signal reflected from the surface of a test object, and an autoencoder-based prediction model training method used therefor. The method may include acquiring a measured signal by transmitting an ultrasonic wave to a test object and receiving an ultrasonic wave reflected from the test object; inputting the measured signal to an autoencoder-based prediction model and predicting a reference signal which is to be expected to be measured from a test object with no defect; calculating a residual signal as the absolute value of a difference between the measured signal and the reference signal; and analyzing information on a defect contained in the test object by analyzing the residual signal.