Curvature and distance amplitude correction for ultrasonic testing of a component with curved surface using deep learning method

Surface curvature and distances from the transducer will have an effect on the amplitude of ultrasonic wave. When using the amplitude of ultrasonic signal to quantitatively size the flaws, a correction is necessary to assure all ultrasonic signal amplitude is based on the echo signal from flat surface. Hence, a deep learning method is proposed to adjust the signal amplitude which enters arbitrary curved surface. A deep learning model will adjust the signal amplitude as the radius of surface, distance from the transducer and attenuation coefficient of material. Then, the ultrasonic signal is imaged on the basis of corrected ultrasonic data. Compared to the ultrasonic imaging based on the raw data, the quantitative evaluation of flaw size is more precise by using the deep learning method. The deep learning model is trained by the CIVA simulation data. Ultimately, the model is verified by the comparisons between prediction and experimental data. The proposed method is adaptable to various curvature, distance, and material, which avoids manufacturing multiple standard specimens to satisfy lower costs and shorter cycle.