Optimization Of Linear-Profile Response Experiment

The functional relationship of product quality characteristics is not uncommon during the complicated products process of vehicles,semiconductors.This quality characteristics with specific functional relationship is defined as profile.At the present stage,the study of profile stays at the stage of product quality control.Few researchers pay enough attention to the role of the Design Of Experiments(DOE)in optimizing the product quality.In addition,the traditional DOE mainly keeps its focus on single-response or multiple response.This dissertation focuses on optimizing design of profile with the optimization method of DOE.Taking practical problem as research orientation,profile model fitting and parameter optimization as starting point,this dissertation achieves the final optimization goal with the strategy of hierarchical optimization.The research work includes following:Firstly,a linear profile optimization model is established based on method of desirability function and TOPSIS aim at the unrelated parameters.Taking the optimal closeness degree of model parameter desirability function value as liner profile performance index,this method overcomes the conflicts caused by simultaneous optimization of multiple model parameters during the process of profile optimizing and the problem that traditional desirability function are easily affected by weight value and extreme value.Strategy of hierarchical optimization is proposed to solve complex linear profile optimization decision problem.Thus the complex strategy decision problem can be reduced to a multi-level sub-decision-making problem in a good order to refine the optimizing process.Secondly,a Seemingly Unrelated Regression(SUR)model is proposed for the strong correlation of model parameters.With this model,fitting error caused by traditional Ordinary Least Square(OLS)method due to its ignorance of the relevant information between model parameters can be reduced.For this reason,fitting method of control layer model in multi-level sub-decision-making strategy can be improved.Therefore,when it comes to the selection of optimization method on the optimal layer in the hierarchical optimization strategy,the approach combined the desirability function method with quality loss function method is proposed based on the SUR model to solve the problem of traditional method for its ignorance of the parameter correlation and uncertainty in parameter estimate and process economy.Finally,a virtual distance weighing method based on desirability function is proposed for prediction qualities of linear profile response model.The similarity between sample profile and target profile is selected as optimized subject.Response-based optimization model is established to optimize the linear profile.This model overcomes the vibration problem caused by fitting error and experiment error of tradition optimization method which is based on the experience model optimization method.In addition,taking some subjective factors into consideration such as personal preference,prior information and the objective information between subject and objective set,the method which is proposed in this thesis has the advantage of strong anti-interference.The study extends the application range of traditional method.It is helpful for engineering design personnel to learn more about the product and process.Furthermore,the results can be taken as a reference for the improvement of process and process.