The Integrated Model And Method Research For Parameter Design And Tolerance Design

In the 1970s,Doctor Taguchi proposed three time design which contains system design,parameter design and tolerance design.Despite taguchi method has received much controversy in academic circles(such as signal-to-noise ratio,etc.),its quality philosophy(such as robust parameter design,quality loss function)has received wide recognition in both academia and industry.In the three time design,parameter design(i.e.,robust parameter design)constitutes the core content,and obtains the most extensive studies and applications.In previous researches or applications,researchers or quality engineers usually successively carry out the quality design work based on the taguchi's three time design.However,some researchers and engineers think that in the process of quality design for the product or process,the optimal parameter values may base on its tolerance value.If parameter design and tolerance design are considered separately,we can't obtain the optimal parameter design values from the economy angle.Secondly,the existing researches usually derive the function relation between parameter value and tolerance value according to the practical engineering experience,and construct the total cost model related to parameter value and tolerance values on the basis.However,in actual production process,the existence of factor tolerance values may cause factor parameter values to fluctuate,thus engineering experience formula can't accurately describe this kind of fluctuations;In addition,most of the existing researches focus on normal single response,in the actual production,however,we need to pay attention to more than one product or process characteristics,and in many cases production does not obey the normal distribution.To solve these problems,this paper puts forward the integrated modeling method for parameter design and tolerance design based on the response surface of bayesian modeling framework,and carries out the following research work:(1)For parameter design and tolerance design of parallel optimization in normal multi-response situation,in the framework of bayesian statistical analysis and modeling.This paper considers the existence of the product quality specification limit,builds the total cost model as optimization index which including tolerance cost,quality loss cost and rejection cost.When using bayesian method to study mulit-response problem,there are two situations for multiple responses:model structures are the same and model structures are different.Correspondingly,standard multiple regression method(SMR)and seemingly unrelated regression method(SUR)are used to get large sample data.Considering the influence of factor tolerance value on parameter values in the process of sampling,the sampling values are linked to factor parameter value and tolerance values at the same time.The useful information(i.e.,sample mean,variance)can be got from large sample data to get the optimization model.Minimize the total cost model by hybrid genetic algorithm to obtain the optimal parameter value and tolerance value.(2)For non-normal response situation,the overall research idea is similar to normal response situation.Optimization index is also the total cost function.For non-normal response problem,the generalized linear model is used to build the response model,and then function relation between the response variance and mean is built through the link function.Bayesian analysis of sampling method is also used to get the sample data considering the model uncertainty.Hybrid genetic algorithm is used to get the optimal value.Through the example analysis,we can find that the proposed method is adaptive for both normal response and non-normal response situations.Compared with the existing method,this new method solves the uncertainty problem effectively during the parallel optimization of parameter design and tolerance design.Besides,the product qualification ration has been improved by considering the existence of quality specification limits.At the same time,the total cost including quality loss cost,tolerance cost and rejection cost has also been reduced to a great degree.