Nonlinear ultrasonic characterization of average grain size based on deep learning

The grain size of metal materials has a great impact on their mechanical properties, and ultrasonic testing has been widely used in characterizing the grain size of metal materials. However, linear ultrasonic testing represented by attenuation and sound velocity is difficult to distinguish small-sized grains, especially in the early stages of grain growth. Nonlinear ultrasonic testing technology has high sensitivity to microscale microstructure evolution, but the complex changes in acoustic nonlinear parameters inevitably lead to underdetermined problems when accurately characterizing the average grain size. This article proposes the use of depth studies to address the dilemma of characterizing the average grain size of metal materials using nonlinear ultrasound. Using characteristic parameters containing ultrasonic nonlinear effects as inputs for multi-layer perceptron and LSTM network, an implicit mapping of acoustic nonlinear effects and average grain size of polycrystalline pure copper is constructed to achieve high sensitivity characterization of small grains that cannot be measured by traditional ultrasonic testing.