Prediction For Mechanical Properties Of Microalloyed Steel Driven By Data

As a structural material with better comprehensive performance,micro-alloyed steel is widely used in modern industrial fields.Mechanical property is one of the key indexes to measure the quality of steel,which affects directly the value of the product.However,there is a mutual coupling relationship between mechanical properties and process composition factors,and the mechanism of action is complex.This leads to high dimensional nonlinearity of the performance prediction model and increases the complexity of modeling the mechanism through metallurgical experiments.In this study,the prediction model of mechanical properties is established based on the data modeling.The prediction precision of the model should be high,and be suitable for multi-species micro-alloyed steel products.At the same time,the influence mechanism of mechanism parameters,composition parameters and carbonatite on mechanical properties need to be revealed.The main contents include the following three aspects:1)For the characteristics of large data volatility,low signal-to-noise ratio and uneven distribution,the sample data is preprocessed by data dispersion normalization and outlier detection to ensure the quality of data.Considering the related affecting factors such as carbonatite precipitate and composition parameters,the feature extraction method based on forward selection is proposed to reasonably select the performance influencing factors,so as to simplify the model structure and reduce the computational complexity.2)Based on the important parameters extracted and the performance analysis of the original Neural Network,intelligent algorithms such as GA,PSO are introduced to optimize the network structure parameters.And prediction model of GRNN Network with optimized kernel width is presented to increase the prediction accuracy and the range of applicable steel grades of the mechanical performance prediction model.On account of the constructed predictive model,the trend of output performance indicators is explored,and the reliability of the network predictive model is verified from the mechanism level.3)In order to construct a performance prediction model with display expression and easy-to-analyze characteristics,on the basis of the extracted important influence performance parameters,the operation process of gene expression programming(GEP algorithm)is designed based on the problem characteristics.By using the mutual transformation of genotypes and phenotypes,the complex functional relationship between the composition and process parameters and the mechanical properties is further explored.Finally,the prediction results of different modeling methods are compared to verify the effectiveness of the proposed model.