Data-driven ultrasonic signal analysis using empirical mode decomposition for nondestructive material evaluation

摘要

Phased array ultrasonic sensing is a well-known non-destructive evaluation approach and a lot of research efforts have been reported. In this paper, we study flaw identification and localization in coarse-grained steel components. To improve the detection effectiveness and performance, advanced ultrasonic signal processing plays a key role. We propose a non-parametric data-driven approach based on ensemble empirical mode decomposition (EEMD), an effective and powerful method to analyze the nonlinear and non-stationary characteristics of ultrasonic signals. In the EEMD approach, white noise is added and it will assist the sifting iterations to converge to the truly intrinsic mode functions (IMF) and cancel out the added noise as long as the iterations are sufficiently large. It is shown that the ultrasonic wavefront harmonics can be effectively represented by multi-mode IMFs, which have the well-defined local time scales and instantaneous frequencies. And the sifting iterations adapt to the varying physical process meaningfully. Numerical experiments are conducted and the presented results validate the effectiveness and advantages of our proposed approach over conventional methods.