| At present,the characteristics of steel enterprises are large-scale,continuous and intensive.The characteristics of consumer demand are personalized,diversification and high quality.Production technology and organizational management of steel enterprises are in contradiction with consumer demands.In order to solve the contradiction,the technology of hot rolling process optimization design is developed.In consideration of the complex constraints in the actual production process and the consumer demands,combining the microstructure and mechanical property prediction technology with the multi-objective optimization algorithm,the optimal production process parameters are quickly obtained.In the technology of hot rolling process optimization design,a reasonable microstructure and mechanical property prediction model is the key factor that can determine the success during the optimization calculation.In industry,the data driven model is widely applied to predict microstructure and property because of its high accuracy and the capability of self-learning.However,researchers mostly pay attention to improving modeling techniques to elevate the accuracy of the model.They establish the model using the data with simply processed such as normalization or standardization and ignore the investigation of the regularity of the model.During the model application,it may produce a deviation from the fact when the regularity of the model is focused.Therefore,appropriate data processing is very important for industrial data modeling.In view of the practicability and rationality of the microstructure and property prediction model of hot rolled strip in big data circumstance,this research is carried out based on the industrial big data processing.This research aims at improving the quality of data for data mining,elevating the accuracy of the model for online prediction,developing efficient intelligent algorithm for reverse optimization calculation,enriching the microstructure prediction of the model and finally realizing the industrial application of the microstructure and property prediction and optimazition technology.The innovation and main work of this research are as follows.(1)Research on data processing method in iron and steel industry:data collected from hot rolling production line have been matched and stored.Combined the theory of mathematical statistics with the theory of rolling process,the data processing methods,such as filling vacancy value,merging steel coil,clustering of similar process and data balancing,were proposed to make the processed data present a reasonable regularity.The relationship model of chemical composition,process parameter and mechanical property of C-Mn steel was established by Bayesian regularization neural network,and the influence of chemical composition and process parameter on mechanical property was compared with the model established without data processing.The result shows that the model deviates from the measured value when the model is established on the data without processing.However,after data processing,the model reflects the reasonable principle of physical metallurgy under the premise of ensuring a certain prediction accuracy.(2)Research on mechanical property prediction model based on artificial intelligence theory:The random forest algorithm was introduced to the mechanical property prediction of hot rolled strip.The resampling technique was used to select the modeling data and a large number of classification regression trees were constructed to predict the mechanical property of Q345B steel.The comparison investigation of traditional stepwise regression model and random forest model was carried out.It is shown that the random forest model obtains strong robustness and high prediction accuracy.The mean error and standard deviation between the predicted yield strength and the measured yield strength are-0.61 MPa and 25.10 MPa,respectively.The mean error and standard deviation between the predicted tensile strength and the measured tensile strength are 0.548 MPa and 23.05 MPa,respectively.The mean error and standard deviation between the predicted elongation and the measured elongation are 0.0088%and 2.09%,respectively.(3)Research on intelligent optimal design of hot rolling process:an multi-objective optimization algorithm,called ε-ODICSA,was developed by introducing the orthogonal experiment design theory and ε dominant strategy into the immune clonal selection algorithm to accelerate convergence.The comparison between the ε-ODICSA algorithm and the traditional multi-objective optimization algorithm such as IBEA,NSGA2 and SPEA2 was carried out on the ZDT test function.The result shows that compared with the traditional IBEA,NSGA2 and SPEA2,the ε-ODICSA shows obvious superiority.Combined with the relationship model of chemical composition,process parameter and mechanical property of hot rolled strip,the ε-ODICSA was implemented to the optimal design of hot rolling process for 380CL steel.By decreasing the coiling temperature from 600 to 510℃,the content of Mn was reduced by half under the condition of meeting the requirement of mechanical property of 380CL steel,successfully saving the cost.(4)The development of intelligent optimal design system of hot rolling process based on big data:based on industrial data mining technology,high accuracy mechanical property on-line prediction technology and high efficient multi-objective optimization algorithm,the intelligent optimal design system of hot rolling process was developed by using C++and C#.The system included data query and filter module,data mining module,high accuracy mechanical property online prediction module and intelligent optimal design of hot rolling process module.Based on 2150ASP hot rolling production line,this system was used to predict the mechanical property of typical steels(500L-Z,L485M,SS400Cr,S275JR,SPHC,Q235B and Q345B).The prediction accuracy showed that the relative error between predicted yield strength and measured yield strength was within the range of ±8%,the relative error between predicted tensile strength and measured tensile strength was within the range of ±6%,the absolute error between predicted elongation and measured elongation was within the range of ±6%.Finally,the mechanical property stability control of HP295 steel was realized in industry by using the system,saving the production cost and improving product quality.(5)Development of big data driven model based on physical metallurgy principle:based on the 2150ASP hot rolling production line,the model describing the microstructure evolution of the workpiece in the process of hot rolling and continuous cooling was established.The whole model included temperature field model,austenite recrystallization model,austenite grain growth model,phase transformation model,precipitation model and final mechanical property prediction model.Based on the model framework,genetic algorithm was applied to optimize the key parameters of the model,realizing the development of the big data driven model based on physical metallurgy principle.Based on the temperature field models,the micro structure and property prediction models of Q235B and X70 were established respectively.Compared with the traditional model based on physical metallurgy principle,the big data driven model based on physical metallurgy principle has achieved a higher prediction accuracy.For Q235B,the relative error between predicted yield strength and measured yield strength was within the range of ±10%,the relative error between predicted tensile strength and measured tensile strength was within the range of ±6%,the absolute error between predicted elongation and measured elongation was within the range of ±6%.For X70,the relative error between predicted yield strength and measured yield strength was within the range of ± 10%,the relative error between predicted tensile strength and measured tensile strength was within the range of ±4%,the absolute error between predicted elongation and measured elongation was within the range of ±6%. |
PDF Full Text Request