Robust Parameter Design Based On Gaussian Process Regression

In the face of the manufacturing system with complex action relationship and fluctuation of quality characteristics caused by noise factors,it is difficult and costly to realize the robust parameter design research.Therefore,the key to solve the above problems is to select appropriate model and optimization method.Existing robust parameter design methods are mostly based on second-order polynomials,support vector machines,artificial neural network and other traditional models.However,the above models are limited by factors such as insufficient fitting ability and less modeling samples,which make the parameter optimization effect not obvious.Therefore,gaussian process regression was introduced into robust parameter design.In the parameter optimization stage,the traditional particle swarm optimization algorithm was improved to make it suitable for the optimization process with more local extreme values.The main research contents were as follows:(1)A dual response surface method based on gaussian process regression was proposed.Firstly,several commonly used kernel function were listed,and their respective applicable ranges were introduced.Secondly,aiming at the kernel function suitable for the complex interaction relationship,this paper drew its generalization image,specifically discussed its global and local generalization ability,and selected the optimal kernel function form adopted in this paper.Then,the response surface models of controllable factors mean of quality,controllable factors and variance of quality characteristics were constructed,and the quasi-Newton algorithm was used to determine the hyperparameters to improve the fitting performance of the models.Finally,the validity of the models were verified through variance test.(2)A parameter optimization method based on improved particle swarm optimization algorithm was proposed.There were many local extremums in the process of complex interaction relation,the traditional particle swarm optimization algorithm was more likely to fall into the local optimal problem.Due to the size of the inertia weight determined the global optimization ability and partial ability,therefore,in this paper,dynamic adjustment of inertia weight,make its diminishing as the nonlinear iterative process,the early stage of the search,given its large weight,ensured the good at a large step to explore the local area,late search,given its small inertia weight,ensured precise search near the extreme point,in order to obtain a better parameter combination.(3)For the robust parameter design method based on gaussian process regression and improved particle swarm optimization algorithm,function simulation was used to verify the fitting performance of gaussian process regression and optimization ability of improved particle swarm optimization algorithm.On the other hand,this paper introduces 3D printing for empirical research,and compares it with the method based on second order polynomial,support vector machine.The research results showed robust parameter design method based on gaussian process regression and improved particle swarm optimization had significant advantages in terms of fitting performance,optimization ability and optimization efficiency.The method proposed in this paper enriches the selection of robust parameter design modeling model types and parameter optimization methods in theory,and plays a certain role in promoting enterprises to improve product quality in practice.