Weld crack detection and quantification using laser thermography, mask R-CNN, and CycleGAN

Abstract

Steel members are susceptible to cracking, and a number of non-destructive testing (NDT) techniques are used for crack detection. However, these NDT techniques are not only labor intensive but also time consuming. In particular, inspection of welded areas requires surface treatment, and data must be interpreted by experienced engineers to differentiate cracks from weld patterns. In this study, an automated weld crack detection and quantification system was developed by integrating laser thermography, Mask R-CNN, and CycleGAN. The developed system comprises a laser heat source, an IR camera, and a control unit. The laser applied heating to the target surface, and the resulting thermal radiation emitted from the surface was measured using an IR camera. Subsequently, the thermal images were processed for crack detection using Mask R-CNN, and for crack quantification using medial axis transform. The detection and quantification performance of the developed system were validated through laboratory and field tests.