Research And Application Of Online Robust Parameter Design Based On GP Model

There is an essential concept in quality management that product quality is designed first and manufactured second.It is a significant part of quality management to optimize the product or process through experimental design,and then control the fluctuating problems affecting product quality.One of the most effective ways to address the quality matters caused by fluctuations is robust parameter design(RPD).RPD is a quality improvement method that combines statistical theory and engineering technology.It is mainly used in the design stage of product or process,by adjusting the input level of controllable factors to meet the quality requirements while minimizing the response fluctuation.The traditional robust parameter design is offline design,i.e.,the optimal setting of controllable factors is obtained by one-time modelling optimization and does not change during the production process.If engineers find that the product quality does not meet expectations or fluctuates greatly during the production process,they need to stop and re-design the parameters.This is obviously unreasonable.In this paper,an online RPD method is proposed.The method can use the newly observed data samples to update the modelling until the optimal setting of the controllable parameters is obtained.In this paper,taking the robust parameter design based on response surface methodology as the subject of the research,it systematically studies the online robust parameter design based on GP model by means of system modelling,simulation experiment,empirical application research and data analysis,which synthetically uses Gaussian process(GP),online modelling technology,matrix fast inversion theory,mean square error optimization method and response variance optimization technology.The main research contents are summarized as follows:(1)In order to solve the problem that the response surface method cannot process streaming data,this paper uses online modelling technology to improve the GP model.An online GP model(IGP)based on incremental strategy is proposed by iteratively updating the covariance matrix.The experimental results show that the IGP model has good performance in prediction accuracy and modelling efficiency.When the number of data samples is too large,the IGP model has problems such as slow modelling speed and inability to respond in time.The paper uses the sliding window strategy and introduce the forgetting mechanism to improve the GP model,so that the historical sample with the least impact on the model are deleted while increasing the new sample,and an online GP model(OGP)based on the sliding window strategy is proposed.The experimental results show that the OGP model improves the modelling efficiency as much as possible while ensuring the prediction accuracy.It is more suitable for real-time control and fast response application scenarios.(2)In order to solve the off-line dilemma of RPD,this paper proposes an online robust parameter design framework by introducing the mean square error optimization strategy.In this framework,the current setting of controllable factors can be used to update the response surface model,and then the RPD process is iterated until a more optimal setting of controllable factors is obtained.It is suitable for industrial scenarios of real-time online control in product or process.The IGP-RPD method and OGP-RPD method suitable for two scenarios are proposed respectively based on the proposed IGP model and OGP model.The simulation results show that both methods can effectively find the optimal setting level of the controllable factors.They take into account the optimality and robustness of the output response,and the parameter design is more efficient.(3)This paper studies the application of online robust parameter design based on GP model.This paper expounds the industrial background of semiconductor device production and processing in chip industry under the background of intelligent manufacturing.The proposed IGP-RPD and OGP-RPD methods are used to solve the quality design problem in the product of semiconductor devices.The results show that the proposed online robust parameter design method can effectively find the optimal setting level of controllable factors in time.In addition,this paper offers the application suggestions of the proposed method from the perspective of engineers and practitioners.In summary,this paper uses the online strategy to improve the GP model.On this basis,the online robust parameter design method based on GP model and its application are studied.This paper provides a new idea and method for the research of RPD.The research shows that the proposed method can be better applied to the online quality design of product or process,effectively improve the efficiency of continuous improvement of product quality and reduce the fluctuation of product quality,which has important theoretical and practical significance for improving product quality.