| In order to gain competitive advantage in global competition,manufacturing industry must pay attention to quality.Modern quality engineering theory believes that quality is designed and manufactured,and fluctuation is the root cause of quality problems.At present,quality design is widely used in product/process design stage,which is an effective way to reduce fluctuation.Robust parameter design(RPD)is a quality improvement method that can effectively reduce fluctuation,and response surface method(RSM)is one of the main methods of RPD.In view of the drawbacks of traditional response surface methods such as poor prediction effect and high modeling difficulty,a response surface modeling technique based on proxy model was developed.One popular approach is to use support vector regression(SVR)as response surface for an RPD.SVR model has the advantages of fast training speed,good generalization performance,high dimensional data processing,strong robustness and so on,and has been widely studied and applied.Existing RPD methods based on SVR model have been studied both offline and online,and can be used in actual industrial scenarios,but the parameter optimization of the model itself has received little attention.Bayesian optimization(BO)is a global optimization method,which can quickly obtain the approximate optimal solution of the complex objective function through relatively few evaluation times,and maintain high efficiency and accuracy in the parameter optimization of the model.In this paper,under the framework of SVR modeling technology,we take the RPD based on SVR as the main research object,use BO as the optimization method for parameter optimization,and construct more accurate response surface.We comprehensively apply the methods and technologies of computer experiment design,parameter optimization,system modeling,simulation experiment,etc.The robust parameter design based on Bayesian optimization support vector regression is studied systematically,which has important theoretical significance and application value for the robust parameter design research.The specific research content of this paper is summarized as follows:(1)An offline robust parameter design method based on BOSVR is proposed.Different from the traditional RPD method using SVR model as response surface,in the RPD strategy based on Bayesian optimization support vector regression(BOSVR)proposed in this paper,the parameters of SVR as response surface are optimized by using Bayesian optimization to ensure that a more accurate model can be used as response surface.The response surfaces constructed by the proposed method and other optimization methods are verified by simulation experiments.The experimental results show that,compared with other optimization methods,the response surface model obtained by the proposed method is superior to other parameter optimization methods in terms of accuracy,and can find the optimal setting for controllable factors more accurately.(2)An online robust parameter design method based on BOSSVR is proposed.Compared with the traditional offline RPD method,the difference in this paper lies in the online RPD strategy proposed based on the Bayesian optimization sequential support vector(BOSSVR).The response surface is constructed online by using the sequential support vector regression(SSVR),which can constantly update the optimal settings for controllable factors,to achieve online quality adjustment.At the same time,in the process of constantly adding new samples,BO is used to update the model parameters in real time to ensure that more accurate response surfaces can be obtained.The proposed method is verified by experiments.The experimental results show that,compared with the existing robust parameter design method without parameter optimization,the proposed method can find the optimal settings for controllable factors more accurately,and has rationality and effectiveness.(3)This paper applies the online robust parameter design method based on BOSSVR to a real case of TV signal transmission quality,to explore the practicability and reliability of the proposed method in real production.Experimental results show that the method proposed in this paper can effectively find the optimal settings for controllable factors,and because of the higher accuracy of the model,the speed of finding the optimal controllable factor is faster,which can effectively improve the efficiency of product design and improve the quality of product production.In manufacturing industry,efficient product design and optimization is the key to improve product quality and production efficiency.Therefore,the research proposed in this paper has important theoretical and practical significance for realizing this goal.To sum up,this paper focuses on the construction of response surface of support vector regression machine based on Bayesian optimization and sequential support vector regression based on Bayesian optimization,offline robust parameter design based on Bayesian optimization support vector regression and online robust parameter design based on Bayesian optimization sequential support vector regression and its application,which is a supplement to the existing robust parameter design theoretical system.It provides a new idea for the research of robust parameter design.At the end of the paper,the current research results are summarized,and some shortcomings of the paper are pointed out,and the direction of improvement and further research is put forward. |
