Data-driven Rolling Mill Vibration Prediction Model Building And Interpretability Analysis

Hot strip mill is an important equipment in the steel industry,and the vibration generated in the operation process seriously affects the production efficiency and product quality.Modern mill equipment condition monitoring system is becoming more and more perfect,and a large amount of rolling process parameters and equipment condition data can be obtained,which brings opportunities for data mining.Therefore,taking a 1580 mm hot strip mill as the research object,machine learning algorithm combined with the rolling mechanism as a tool to establish the rolling process parameters-mill vibration relationship model,in-depth study of the internal correlation between rolling process parameters and mill vibration,reveal the role of process parameter fluctuations and mill operating conditions,to provide valuable basic theory and key technology to ensure the stable operation of the strip mill system.The main research contents are as follows:(1)A data-driven BO-XGBoost rolling mill vibration prediction model is established to achieve accurate prediction of rolling mill vibration through dynamic rolling process parameter fluctuations.The XGBoost algorithm can automatically learn the complex relationship between rolling process parameters and rolling mill vibration,and the Bayesian optimization algorithm(BO)is used to optimize the hyperparameters of the XGBoost algorithm to improve the model prediction accuracy.It is also compared with other benchmark models for validation.The results show that the prediction accuracy of BO-XGBoost prediction model is significantly better than other benchmark models.(2)The prediction accuracy and generalization ability of the BO-Stacking prediction model are verified using 10 different data sets.The results show that the prediction accuracy of the integrated BO-Stacking model is significantly better than that of the BO-XGBoost prediction model and has a strong generalization capability.(3)The SHapley Additive ex Planations(SHAP)algorithm is used to explain the calculation process and decision mechanism of the prediction model and reveal the dynamic response law of rolling process parameters-rolling mill vibration.The results show that: rolling force,rolling speed and entrance/outlet thickness have significant effects on rolling mill vibration,and back tension basically does not affect rolling mill vibration;the appropriate reduction of rolling force and depression rate can effectively reduce rolling mill vibration,and2.29m/s is the optimal rolling speed.