Non-contact characterization of material hardness by deep learning-assisted electromagnetic acoustic resonance method

Abstract

This study introduces a cutting-edge approach for achieving precise non-contact characterization of material hardness by integrating electromagnetic acoustic resonance (EMAR) with a one-dimensional convolutional neural network (1D-CNN). EMAR is strategically utilized to address the challenge of low energy conversion efficiency in electromagnetic ultrasonic transducers for non-contact measurements. A 1D-CNN-based neural network is proposed, designed to dynamically extract features from the original signals and employ classification and regression techniques to directly forecast variations in material hardness. Furthermore, EMAR signals are meticulously compared to pinpoint the optimal input featuring specific resonant frequencies to enhance model performance. The viability of the proposed method is rigorously validated through experimentation on metallic specimens subjected to diverse heat treatments. The results underscore the efficacy of this approach in discerning alterations in material hardness induced by heat treatments, all achieved in a noninvasive manner.