Grain size prediction of low carbon steel based on Bayesian Optimized Gradient Boosting Regression Tree using multiple echo characteristics of laser ultrasonics

In this study, 20# carbon steel was used as the research object, and the microstructure information characterized by EBSD was used to calculate the average grain size of the sample. Based on the Rayleigh scattering, the grain size prediction model was established by fitting the attenuation coefficient in the frequency domain to the grain size. By extracting the signal amplitude in the frequency domain as the characteristic information of multiple echoes, taking it as inputs and the average grain size as output, the BO-GBRT prediction model was established. The average prediction error of the BO-GBRT model was 4.04%, which verified the feasibility and effectiveness of the BO-GBRT algorithm in grain size characterization and prediction.