Prediction And Optimization Of Steel Based On Machine Learning

With the economic globalization and the rapid development of national economy,the level of national iron and steel industry has been a huge leap.However,the quality problems such as large fluctuation of product performance and poor stability are still the major challenges.For example,the relationship between the chemical composition,process parameters,microstructure and properties of steel is very complex.And there is a strong coupling interaction between these variables.Therefore,it is difficult to establish an accurate mathematical model to restore the real production process and control its process parameters in real time.As a result,the quality of steel production is low and the cost is huge.Thus,it is of great significance to improve the quality of steel products by data mining and data-driven prediction.In view of the above problems,this thesis focuses on the mechanical properties(tensile strength,plasticity,fatigue strength and hardness)of steel and uses machine learning method to model,predict,optimize and analyze.This thesis firstly points out the deficiencies of the current iron and steel research.Then,we introduce the basic theory of machine learning and intelligent optimization,which lays a theoretical foundation for the establishment and optimization of the machine learning model.The detailed achievements are introduced as follows(1)Aiming at the problem of multivariable,nonlinear and strong coupling data in the process of steel production,an optimization modeling method based on machine learning is proposed,taking the tensile strength and plasticity of steel as the research object.Firstly,the mapping relationship between chemical composition,process parameters and mechanical properties is established by machine learning algorithm xgboost,which is used as fitness function.Then the data-driven multi-objective optimization model is established,and the improved particle swarm optimization algorithm is used to optimize the chemical composition and process parameters,so as to realize the construction of the integrated model of steel production process modeling and optimization.(2)Aiming at the problem that the fatigue failure mechanism is complex and difficult to explain,a fatigue strength mechanism analysis method based on snap theory is proposed.Firstly,in order to increase the reliability of the interpretation of the model results,a hybrid model of xgboost and lightgbm is established by using bagging idea.Grey wolf optimization algorithm is used to optimize the model parameters and predict the fatigue strength.Then,the black box model is interpreted by using a novel machine learning model interpretation method,the shap theory.Finally,the importance of the factors that affect the fatigue strength is sorted,the positive and negative correlation is analyzed,and the internal structure of the model is visualized,so that we can directly see how these variables affect the fatigue strength,so as to guide the production of the next generation of high fatigue strength steel.(3)Aiming at the problem that it is difficult to predict the mixed data set of multi materials,a new divide and conquer learning framework is proposed based on the hardness of ultra-high strength steel.First,the three sigma principle and cosine similarity principle are used to enhance the existing samples,and the lightweight density peak clustering algorithm is used to cluster the data,and the data samples of different materials are automatically divided.Then,a two-layer fusion model is established by using stacking idea to predict the clustered data samples respectively,so as to achieve the effect of "divide and rule".Finally,the model prediction results are compared and discussed on real data sets to verify the effectiveness of the proposed method.