Research On Quality Design Technology Considering Factorial Effect Principles

The main purpose of modern quality engineering research is to minimize and control fluctuations around the design target value as much as possible,which has become the core of continuous quality improvement activities.Generally speaking,continuous quality improvement activities based on quality design are mainly composed of three stages: variable screening,response modeling and parameter optimization.Many practical results show that quality design should start in the early stage of product production,and it is necessary to find the cause of defects in the product from the source(the variable screening stage).Considering the complexity of current product design and the diversity of customer needs,in the subsequent test phases,researchers should also pay attention to the various uncertainties and fluctuations in the product design process and how to consider multiple quality characteristics in the product quality design.In the previous research on multi-response quality design,only one of the above three stages was usually studied.For the problem of variable screening,most studies have not considered the factorial effect principles.For response modeling problems,most researchers ignore the uncertainty of model parameters and the fluctuation of predicted response values.In addition,the existing literature results mainly take the normal multiresponse as the research object,and take less attention on the optimization problem of non-normal multi-response.So this paper mainly studies the multi-response problem considering correlation,three-level fractional factorial experimental design and the quality design problem of non-normal multi-response.Combining Bayesian sampling technique,Bayesian posterior probability,genetic algorithm and parameter design,this paper based on the factorial effect principles studies the quality design problems of different production processes,by using system modeling(response surface modeling and generalized linear model)and example verification.The detailed research content is as follows.(1)Continuous quality improvement method combining SUR model and factorial effect principles.This paper uses SUR model to fit related multi-response problems.Firstly,a variable screening model is constructed combining the factorial effect principles and the Bayesian method.Then the uncertainty of model parameters and the fluctuation of predicted response value are displayed using Bayesian posterior trace plot and response posterior box plot.Finally,parameter optimization is implemented combining multivariate process capability index and genetic algorithm.(2)Three-level fractional factorial experiment and parameter optimization considering factorial effect principles.As for the experimental design with the number of variables larger than that of samples,this paper constructs a Bayesian Lasso model considering the factorial effect principles by using the mixed binary variable indicator,so as to select a small number of significant variables from many variables.A multivariate process capability index is constructed using the posterior sample values of the response model,to determine the optimal parameter level.(3)Non-normal multi-response quality design combining GLM and factorial effect principles.For the variable screening problem of non-normal response,this paper constructs a screening model of non-normal response based on empirical Bayesian hybrid prior and mixed binary variable indicators.Then,the repeatability of each optimization result is measured according to the Pareto optimal strategy.Finally,the grey correlation analysis is selected as the quality evaluation index to optimize the parameters.