Research Of Robust Parameter Design Based On The Bayesian Method

Robust parameter design applied in product and process design is the most effective method to reduce or control the variation in the product formation process.The uncertainty problem is currently the focus and one of difficulties of quality science research.In practical production or manufacturing processes,uncertainty factors such as model parameters and model structure significantly affect the modeling accuracy and analysis results of response surfaces due to the influence of internal and external noise factors.Bayesian methods can effectively quantify various uncertainties by considering basic principles of experimental design,experiential knowledge of the experimenter,and historical data information in the form of prior information,thus constructing more precise and realistic response surface models.Therefore,this thesis focuses on uncertainty robust parameter design for advanced manufacturing processes.Under the framework of Bayesian modeling and optimization,a comprehensive approach is used,including response surface methods,computer experimental design modeling,stochastic search techniques,and heuristic optimization algorithms,to systematically conduct research on robust parameter design based on Bayesian methods.The specific research contents include model construction,simulation analysis,experimental design,empirical research,and case studies.(1)Robust parameter design for non-normal response based on Bayesian generalized linear models.For the robust parameter design with non-normal response,generalized linear models are used to consider the basic principles(i.e.,effect sparsity,heredity and hierarchy)of experimental design through Bayesian prior information.Consequently,a response surface model is constructed between the output response and input factors.On this basis,Bayesian sampling techniques are used to obtain simulation sampling values of the output response.The mean square error function and conformity probability function are constructed using these sampling values,and a robust parameter design method for non-normal response considering noise factors is proposed.This method not only considers the uncertainty of model parameters and the impact of response fluctuations on optimization results but also effectively measures the impact of noise factors on optimization results.(2)Concurrent parameter and tolerance design based on a two-stage Bayesian sampling method.For the parallel optimization for robust parameters and tolerance design with uncertainty,a hierarchical Bayesian model has been constructed in the framework of the generalized linear model to establish the functional relationship between the output response and the design factors and tolerance variables.A twostage Bayesian sampling method was used to obtain the simulated sampling values of the output response,on which the rejection cost(i.e.,rework and scrap costs)function and quality loss function have been constructed.Finally,a genetic algorithm is used to optimize the total cost model(including tolerance cost,rejection cost,and quality loss).This method not only considers the parallel optimization for both robust parameters and tolerance design with uncertainty,but also considers the influence of model parameter uncertainty and output response fluctuation on optimization results.(3)Sequential robust parameter optimization design based on Bayesian information updating strategy.This method addresses the issue of limited sample information in small batch manufacturing processes,such as 3D printing,by using Latin hypercube sampling(LHS)designed through computer experiments to obtain initial experimental sample points.The physical experiments were then conducted using the relevant equipment in the additive manufacturing quality laboratory of the Jiangsu Provincial Engineering Research Center for High-end Equipment Quality Improvement to obtain output response experimental data.Based on these experiments,the Bayesian information updating strategy has been used to select or add appropriate experimental sample points,and then the optimal parameter design value has been obtained through sequential parameter optimization design method.This method considers how to improve the quality design level of small batch manufacturing processes through Bayesian information updating and sequential optimization design under the constraints of limited experimental resources and manufacturing costs.(4)Online robust parameter design based on Bayesian Gaussian process models.To address the high computational cost and low efficiency of online robust parameter design,this study combines Bayesian sampling techniques and cluster analysis to obtain the optimal design region of controllable factors.In this study,the Bayesian posterior probability method is used to obtain the optimal robust design region.Consequently,a new multi-stage online robust parameter design model is proposed by integrating global optimization algorithms,parameter updating strategies,and quality loss functions to obtain more robust parameter design solutions.This method gradually narrows down the design factor space required for online robust parameter design through an offline quality design approach,effectively reduces the computational cost and hence improve the efficiency of the online robust parameter design.Moreover,this method considers the robustness and reliability of the optimal parameter design through online updating strategies,to ultimately improve the quality design level of products and processes.Finally,based on a comprehensive summary of the above research results,further research directions are pointed out in the uncertain robust parameter design field that require in-depth investigation and exploration.