Automation of alignment of sensing & generation laser beams and defect classification for through-transmission ultrasonic imagers

Abstract

Nondestructive testing frequently requires laser generation and sensing. This is done in order to inspect the material properties. However, thick specimens pose a major obstacle for testing with ultrasound methods, like, pulse-echo ultrasonic propagation imaging systems that need the ultrasound wave to travel through twice the total thickness of the specimen. Hence, through-transmission ultrasonic testing is required. Alignment of laser ultrasonic beams in through-transmission mode, where both beams are hitting the surface of the specimen from the opposite sides, is not only crucial yet cumbersome for better visualization of the defects. In this thesis, we automatically align the generation and sensing laser beams for through-transmission ultrasonic propagation imaging systems, by studying the effects of laser pulse energy, time of arrival and amplitude of the signal on the laser-induced through-transmission ultrasound wave and their interrelationships with the material properties. Alignment of the beams would greatly improve the propagation depth and optimize the inspection in full-field through-transmission ultrasonic propagation imaging system for thick composites and pressure vessels. In this system, we have also developed an automatic defect visualization algorithm based on machine learning. After C-Scan inspection, the time series data was processed to extract features and converted into time-frequency images using STFT and power spectrogram. It was classified into artificial defect, natural defect and fiber by a machine-learning algorithm called LSTM network. The training set showed an accuracy of 93.2% and testing set reached the accuracy of 88.2%.