Abstract: In the process of ultrasonic non-destructive testing, the defects of different layers inside the aerospace components made of multilayer aluminum alloy plates of equal thickness by vacuum diffusion welding often overlap horizontally in the direction of sound wave propagation. This results in aliasing of echoes of the upper and lower defects, affecting identification of the defects of the lower layer. Therefore, separation of aliasing signals has great significance to detection of overlapping defects. In this paper, based on the traditional SL0 algorithm, the hyperbolic arctangent function and the modified Newton iteration method are used to approximate the minimum value of the L0 norm. Meanwhile, the optimal solution is obtained by convex quadratic programming with inequality constraints. Finally, the decomposition and reconstruction of noisy ultrasound overlapping signals are realized. The results reveal that compared with other sparse representation methods, the improved SL0 algorithm has better performance in both sparse ability and reconstruction effect for noisy signals. Besides, the parameters of the signal obtained by sparse decomposition are consistent with the real data obtained by actual measurement, which proves the accuracy of the algorithm.

Key words: vacuum diffusion welding, layered medium, ultrasonic testing, orthogonal matching pursuit, SL0 algorithm, convex optimizationhttp