Machine Learning Methods For Parameter Design Defect Identification Of Mechanical Product

Choices of design parameters for mechanical products are key factors thatdetermine quality of products. Extending product development cycle caused byparameter design defects is one of the main reasons that affecting the product marketcompetitiveness. In order to identify parameter design defects at early stages of design,models of parameter design defects identification based on support vector machine(SVM) and artificial neural network(ANN) are presented. Contents such as modelbuilding, parameter selection and result analysis are studied deeply and systemly in thisthesis.Aiming at design defects of high-speed motor car axle by irrational designparameters, design defect identification model is presented based on improved SVMclassification algorithm. Considering the influences of different design samples andcharacteristic parameters, a method combining sample weighting with feature weightingis presented to improve performance of model. Aiming at the difficult selection of SVMparameters, a algorithm based on support vector regression(SVR) is presented todetermine the parameters of model. The algorithm can find ideal model parametersquickly to improve classification accuracy and generalization ability of SVM.Considering design defects of axle wheel seat caused by irrational parameters. Designsamples are trained by SVM classification algorithm. Then SVM model is build whichcan achieve defects identification of new design parameters.Design defects of products are caused by irrational geometric design parametersand generalized design parameters such as product material selection, processparameters and structural parameters. In this thesis, mathod of defect identificationbased on BP neural network is presented. The design of car bumpers is studied. Andsome key parameters is selected from numbers of car bumpers design parameters. AndANN is trained in order to achieve identification of car bumper surface defects andoverall defects.Considering constraint relations such as assembly, processing of products occuredwith other parts, facts of design defects are caused by influences of external constraint parameters. In this thesis, model of the defect identification based on the constraintrelations of design parameters is build by using a-class neural network algorithm. Andinfluences of design results caused by changes of constraint parameters are studied.Concept of design parameters sensitivity is introduced for randomness of mechanicalproduct design parameters. To obtain the influence of design results caused by randomchanges of design parameters, sensitivities of random design parameters are calculatedon this basis of neural network model. And the sensitivities can guide. And purpose ofselections of design parameters, processing parameters and assembly precisionparameter is achieved.Method of design defect identification based on machine learning algorithms ispresented in this thesis will provide some effective theories, methodologys andtechniques for problems of mechanical products design parameters defect identification.