基于机器学习的薄板屈服强度与制耳率建模分析

针对1070铝合金薄板制耳率高、屈服强度不稳定等问题，基于生产数据，采用随机森林算法建立了“成分—工艺—性能”模型。选取冷轧率、热终轧温度、Fe含量、Fe/Si（铁硅质量比）等工艺和成分作为自变量，所建立的屈服强度模型精度（R2）为0.75，制耳率模型精度为0.87。利用模型定量分析各参数对屈服强度和制耳率的影响规律。通过对模型的解析，求解出各自变量的变量权重系数和Shap值。结果表明，对屈服强度影响最显著的因素为冷轧率，二者呈正相关关系，对制耳率影响最显著的因素为Fe含量，二者呈负相关关系。同时，根据模型进行了特定工艺下的性能预报并得出了较优工艺。 

Modeling analysis of yield strength and earing ratio of sheet metal based on machine learning algorithm

In view of the high earing ratio and unstable yield strength during 1070 sheet metal production, a "composition-process-performance" model was built with a random forest algorithm based on production data. The following factors were chosen as independent variables: cold rolling reduction, hot rolling finishing temperature, Fe content, Fe/Si, etc., and the accuracies of the yield strength model (R2) and earing model were 0.75 and 0.87, respectively. Based on those models, the influence of various parameters on yield strength and earing was quantitatively studied. The variable weight coefficients and SHAP values of their variables were calculated with the model. The most remarkable factor for the yield strength was the cold rolling reduction, which was positively correlated, and the most remarkable factor for earing ratio was the Fe content, which was negatively correlated. Models were used for property predictions with given process parameters, and optimized parameters were achieved.