Optimization of HBVD-EMAT Based on Orthogonal Experimental and Multilayer Perceptron Fusion Method

Abstract

HBVD-EMAT is an electromagnetic acoustic transducer (EMAT) composed of a Halbach magnet and a variable distance meander-line coil. By introducing a linear frequency-modulated (LFM) signal into the coil, wideband pulse compression surface waves can be generated. This article proposes an optimization method for HBVD-EMAT based on the fusion of orthogonal experiment and a multilayer perceptron (MLP) to enhance its performance in both the time and frequency domains. First, the finite-element simulation method is used to perform a four-factor, five-level orthogonal experiment on the size of the Halbach magnet. Then, the time- and frequency-domain response variables of the signal from the simulation results are extracted to analyze the orthogonal experimental results. The EMAT performance evaluation index is constructed based on this analysis. Finally, the MLP model is established with the performance evaluation index as the objective function. The orthogonal experimental results are used as training data to predict the optimal EMAT factor-level combination corresponding to the maximum objective function. The EMAT detection experimental results show that, compared with the nonoptimized HBVD-EMAT, the increase of incident surface wave generated by the optimized HBVD-EMAT in four response variables is 98%, 26%, 95%, and 10%, respectively. EMAT performance evaluation index is increased from 0.13 to 0.86. After optimization, the signal-to-noise ratio (SNR) of EMAT’s crack defect reflection signal and transmission signal is increased by 11.4 and 12.5 dB, respectively.