Deep Learning-Based Ultrasonic Testing to Evaluate the Porosity of Additively Manufactured Parts with Rough Surfaces

Ultrasonic testing (UT) has been actively studied to evaluate the porosity of additively manufactured parts. Currently, ultrasonic measurements as-deposited parts with a rough surface remain problematic because lowers signal-to-noise ratio (SNR) signals, which degrades UT performance. In this study, various deep learning (DL) techniques that can effectively extract features defects, even from signals low SNR, were applied UT, and their performance in terms evaluation surfaces was investigated. Experimentally, effects processing conditions additive manufacturing on resulting first analyzed using both optical scanning acoustic microscopy. Second, convolutional neural network (CNN), network, multi-layer perceptron models trained time-domain obtained specimens levels roughness. The experimental results showed all could accurately, specimens. particular, CNN delivered best at 94.5%. However, conventional not be SNR. generalization when newly high 90%.