Mechanical Properties Prediction Of Hot Rolled Carbon Steel Based On Intelligent Algorithms

Steel is the most widely used structural engineering material in construction,automobile,ship,railway and other fields.The mechanical property is the key index of steel quality,and the accurate prediction of the mechanical property is of great guiding significance to the control of product quality and the optimization of product process.Because the production of hot rolled steel plate is a physical metallurgical process with complex mechanism and highly coupling factors,the prediction of mechanical properties is a complex,multi-dimensional and nonlinear problem,and the traditional mechanism model and regression model are difficult to achieve accurate and reliable prediction,and can not meet the needs of actual industrial production.In view of the above problems,based on the actual production data of hot rolled carbon steel plate in a steel mill,this topic adopts intelligent algorithm to establish a prediction model of mechanical properties of hot rolled steel plate.The main research contents are as follows:(1)Taking production data of hot rolled carbon steel plates of multiple grades of carbon manganese steel produced by a steel mill as data samples,Pearson correlation coefficient method was used to evaluate the influence of chemical composition and rolling process parameters on mechanical properties of hot rolled carbon steel plates.From 12 in the parameters selection for a greater influence on the mechanical properties of seven parameters(the contents of carbon,silicon,manganese,outlet temperature of the outlet temperature of the roughing and finishing,coiling temperature,finishing thickness)as input parameters of the model,select the yield strength,tensile strength,elongation as the output of the model,three parameters,such as input and output structure of performance prediction model is established,It lays a foundation for the establishment of performance prediction model based on intelligent algorithm.(2)The production data of hot rolled carbon steel plate were removed by using Rayda criterion and Grubbs criterion.Taking Sigmoid function as activation function,coefficient of determination(R2)and mean relative error(MRE)as comprehensive evaluation indexes,a BP neural network model for predicting mechanical properties of hot rolled carbon steel plate with 7 neurons in input layer,8 neurons in hidden layer and 3 neurons in output layer was established.The results show that the prediction time of the neural network model is about one minute,the prediction accuracy of yield strength is 93.7%,and the relative error is 4.6%.The prediction accuracy of tensile strength is 93.9%and the relative error is 3.3%.The prediction accuracy of elongation is 89.5%and the relative error is 8.6%.In addition,k-fold method is used to test the reliability of BP model,and the results show that the random split data sets have a great influence on the accuracy of the model.(3)In order to improve the generalization ability of prediction model,support vector machine(SVM)algorithm is adopted to establish the hot rolled steel plate mechanical properties prediction model,first of all,choose different kernel function model,then analyzes the penalty parameter C epsilon,loss function parameters and kernel function parameter sigma precision of support vector machine(SVM)model,the influence of second contrast prediction accuracy of different kernel functions,Finally,the radial basis function was used as the kernel function to establish the prediction model of mechanical properties of hot rolled carbon steel.The prediction results show that compared with the neural network,the prediction accuracy of yield strength and tensile strength is improved by more than 2%,the prediction time is reduced,and the reliability is increased significantly.(4)In order to improve the prediction speed and reduce the calculation cost,XGBoost algorithm was used to establish the prediction model of mechanical properties of hot rolled carbon steel plate.The sensitivity analysis method was used to optimize the XGBoost model parameters and analyze the sensitivity of input to output of the model to simplify the model and determine the key variables affecting the mechanical properties of the steel plate.Compared with BP neural network and support vector machine,XGBoost algorithm has the characteristics of faster training speed and higher accuracy.The prediction time of XGBoost model is 2.5 seconds,the prediction accuracy of yield strength is 99.6%,and the relative error is 0.51%.The prediction accuracy of tensile strength is 99.8%and the relative error is 0.74%.The prediction accuracy of elongation is 97.9%and the relative error is 2.17%.By comprehensive comparison of the prediction accuracy,running time and reliability of the three models,XGBoost is more suitable for predicting the mechanical properties of hot rolled steel and can control the rolling process in real time.