Quantitative evaluation of spheroidisation in 15CrMo steel based on deep learning

15CrMo steel has good mechanical properties and is widely used in high-temperature pressure components. After long term service, it is easy to cause material spheroidisation. The metallographic and mechanical property tests are used to evaluate the spheroidisation of specimens. Ultrasonic backscattering signal is sensitive to microstructure and contains more microstructure information, which can be used for the evaluation of spheroidisation. However, the ultrasonic backscattering signal is nonlinear, and feature extraction is difficult. In this study, ultrasonic testing is used to scan different spheroidisation specimens, and the ultrasonic backscattering signal is extracted as the input to the deep learning model, which is used to extract features from the backscattering signal. The models are evaluated using classification and regression evaluation metrics. The results show that the proposed CNN-LSTM model has good identification performance for the classification of spheroidisation and the prediction of mechanical properties. The classification accuracy, recall, precision, and F1-score are all 1. Additionally, the maximum predicted RMSE and MAE values are only 2.33 MPa and 1.70 MPa, and the minimum R2 is only 0.97. The worst prediction is the tensile strength, with an average value of 442.2 MPa and a maximum value of 481.1 MPa.