Detection of Weld Root Crack in Metal Plates with Convolutional Neural Network

The detection of weld root cracks in metal plates using nonlinear ultrasonics and a convolutional neural network (CNN) was investigated. Lamb waves with a fundamental frequency of 0.5 MHz were transmitted into a carbon steel plate specimen, and the waveforms were recorded using an oscilloscope. Ratios of harmonic powers up to the third order, obtained through time–frequency analysis of the received waveforms, were used to train a one‐dimensional CNN. Evaluation experiments achieved a maximum recall of 89.5% when the crack was located near the center between transmitter and receiver, demonstrating the feasibility of nondestructive crack detection using the proposed method. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.