Research On The Prediction Of Shrinkage And Fluidity Of Aluminum Alloy Based On Support Verctor Regression

Shrinkage and fluidity,as two of the most important indexes of castability of aluminum alloys,plays a key role in casting quality.To quick get the shrinkage percentage and fluidity of an alloy is very important for casting production and development of a new alloy.In this thesis,Support Vector Regression(SVR)is adopted,combined with parameter optimization by Genetic Algorithm,to construct the prediction models for macro-shrinkage percentage and fluidity of aluminum alloys.Conclusions are obtained as following:(1)The datasets of shrinkage percentage and fluidity of a large number of aluminum alloys are obtained by our experiments.The Defect Rejection Model constructed by ourselves is used to optimize the raw dataset.By it,the availability of the dataset and accuracy of modeling is significantly improved with only small amount of data loss.(2)The prediction model for macro-shrinkage percentage of aluminum alloys is constructed by Support Vector Regression(SVR)combined with parameter optimization through Genetic Algorithm.When this model is trained to be optimal,the Model Hyperparameters are obtained as:C=3.07,?=0.12,?=0.03.Modeling results indicate the predicted data by this SVR model agrees well with the experimental data for the training dataset,and the Error Residual Distribution demonstrates the error randomness and reliability of this model.The prediction for testing dataset verifies that this prediction model for shrinkage percentage processes very excellent prediction ability,where,the maximum relative error is 2.7%,and the minimum relative error is 0.97%,and the mean absolute error is 2.29%.MIV significance analysis suggests that Si/Cu and Fe,in turn,are the most important elements to affect the shrinkage percentage of aluminum alloys.(3)The prediction model for fluidity of aluminum alloys is constructed by the same method.When this model is to be optimal,the Model Hyperparameters are gotten as:C=31.59,?=0.33,?=0.057.Whether for testing dataset,training dataset or total dataset,the predicted data by the developed SVR prediction model for fluidity are in good agreement with the experimental data.MIV significance analysis suggests that Si/Mg and Fe,in turn,are the most important elements to affect the fluidity of aluminum alloys.(4)Whether for binary alloys,ternary alloys or multicomponent alloys,the predicted data by the developed SVR prediction models for macro-shrinkage percentage and fluidity of aluminum alloys are not only in good agreement with the experimental data of ours,but also with the other data in literature.This indicates that the developed prediction models can accurately predict the macro-shrinkage percentage and fluidity of aluminum alloys,with high prediction accuracy and applicability.