基于机器学习的4145H钢成分-淬透性预测及SHAP特征分析OA

4145H steel is widely used in high-load components such as drill collars due to its high hardenability and strength.Hardenability is a key indicator for the uniformity of cross-section properties during the quenching and tempering process of steel,which directly affects the strength and toughness.Traditionally,the Jominy end quenching test is the most commonly used method to evaluate hardenability.However,there are problems with the test process being complex,costly,and time-consuming.To this end,based on the steel production line data,the contents of 9 chemical elements are used as characteristic variables,and the hardness values at 1.5,25 and 50 mm at the quenching end of the standard Jominy end-quenched specimen are used as target variables(J1.5,J25,J50).Four machine learning algorithms are used to predict the hardenability,and combined with ten-fold cross-validation to evaluate the model performance.The results show that the CatBoost model has the best accuracy.The R2 of J1.5,J25 and J50 values are 0.979,0.961 and 0.970 respectively.The RMSE are 0.383,0.459 and 0.621 HRC.98.88%,97.09%and 89.26%of the samples fall within the confidence interval.In addition,Shapley additive explanation(SHAP)feature analysis is used to enhance the interpretability of the model,and it is determined that Mn,Cr,C and Mo elements are the key elements affecting hardenability.The research results achieve accurate prediction of the hardenability of 4145H steel and provide a reference for composition optimization.