Study Of Sequential Design And Modeling For Complicated Relationship Process With Mixed Parameters

It is very difficult to carry out design of experiment and modeling for the complicated relationship process with functional parameters and scalar parameters.First,due to the presence of mixed parameters and inhomogeneity change of quality characteristics in the domain of parameter definition,there is currently no suitable experimental design method.Second,at present,there is less researched modeling for process with mixed parameters,and the former studies more often adapt the linear regression model to functional parameter modeling.However,this method does not apply to the complicated relationship process.Therefore,this thesis proposes a sequential design method for mixed parameters,then selects appropriate regression algorithm to model the complicated relationship process with mixed parameters.First,the mixed parameters and inhomogeneity change of quality characteristics are taken into consideration in the experimental design scheme.Using a set of basis functions representing the function parameters,let the basis function coefficients as the scalar parameters of the process,forming a whole with other scalar parameters,thus the mixing parameters fusion;Through the sequential design gradually explore the significant of quality characteristics in each sub-area,and apply it to the each stages of sequential design,so the sequential design is completed.Then,the model of the complicated relationship process with mixed parameters is established.The Bezier curve was selected as a functional parameter design method.For comparative study,two algorithms,Gradient Boosting Design Tree(GBDT)and Least Squares Support Vector Regressions(LSSVR),were used as regression modeling methods for complicated relationship process with mixed parameters.Finally,the superiority of the method of this thesis is verified by simulation function and empirical research.Through the simulation function experiments,compared with the traditional method,the sequential design method proposed in this thesis reduces the mse by 15.40%,and increases the optimization result by 12.53%.For the data involved in this thesis,the generalization ability of GBDT is also Better than LSSVR,so in the empirical research,choose GBDT as the process regression modeling method.The empirical research of computer monitor panel plastic parts can also be concluded that the proposed method of this thesis has better generalization ability and optimization ability than traditional methods.In this thesis,the experimental design,modeling methods and implementation steps for the complicated relationship process with mixed parameters are proposed.Through simulation and empirical research,it is shown that the proposed method is more suitable for such processes than traditional methods.So,this study has significant theoretical and practical significance for the optimization of parameters and quality improvement through experimental design.