Energy mapping deep transfer learning method for aluminum plate back damage detection using laser ultrasonic guided waves

Abstract

Laser ultrasonic guided wave detection has attracted extensive attention owing to its high sensitivity and non-contact advantages. Laser excitation can be divided into thermoelastic and ablation excitations according to energy intensity. The signal generated by the ablation excitation was stronger; however, the detection structure was damaged. The thermoelastic excitation did not damage the structure. However, the signal generated by thermoelastic excitation is unsuitable for detecting internal and back damage. In this study, a novel energy mapping transfer network (EMTN) detection method based on laser energy mapping deep transfer learning is proposed to align the thermoelastic signal feature space with the ablation signal feature space. The proposed method aims to improve the accuracy of laser nondestructive testing (NDT) by making the thermoelastic signal closer to the ablation signal. First, the overview waveforms of the source domain ablation and target domain thermoelastic signals were extracted by wavelet decomposition. The proposed mapping function was used to map the thermoelastic signal feature space to the ablation signal feature space to obtain the mapping thermoelastic signal. Thereafter, the feature space of the thermoelastic and ablation signals was aligned using EMTN. These signals shared the same feature extractor. The sum of the conversion and label errors was used as the feature space alignment error. The thermoelastic signal was detected after obtaining the detection model. Finally, the proposed method was verified by single point and scanning imaging detection experiments. Experiments demonstrated that the accuracy of aluminum plate internal damage detection under thermoelastic excitation is improved using the proposed method.