Research On Uncertainty-based Robust Design Optimization Methods For Electronic System

Electronic system consists of components,and the structure and function of electronic system are becoming complicated and diversified.Uncertainties in the production and use cause system's performance to deviate from the target value,even to fail.Considering the influence of all kinds of uncertain factors in the design period,it is urgent to improve the product quality by using robust design optimization.For mechanical engineering,civil engineering etc.,uncertainty-based robust design optimization has developed rapidly and been used extensively.However,the research for electronic systems is comparatively lagging behind,and the robust design method has many shortcomings applying to electronic systems.Several problems remain to be addressed,such as the computational complexity of uncertainty models,the limitations of traditional robust design optimization method,the low accuracy and efficiency of approximate model,the improper handling of discrete design variables in electronic systems,and lack of robust design method of smart electric meters.This dissertation makes an intensive study of the uncertainty-based robust design optimization method for electronic systems,and validates the method in the robust design of single-phase smart electric meter.Therefore,it has practical significance to improve the products consistency,balance cost and performance,and shorten the development cycle.With regard to the calculation complexity in the optimization process due to the implicit system response,a uncertainty modeling method based on Kriging model is presented,combining with experimental design method.A numerical function and a typical voltage sampling circuit from electric meter are taken as examples to verify the validity of Kriging model.Global sensitivity analysis based on Sobol' decomposition is proposed to determine key design variables,reduce approximate model's dimension,and solve the problem due to collaborative interaction by massive electronic components.For the voltage sampling circuit application,the proposed method is superior to traditional sensitivity method in terms of model's uncertainty allocation.To handle the contradiction between efficiency and accuracy of approximate model,the points adding strategy combining the expected improvement(EI)and mean square error(MSE)criterion is used on the basis of the Kriging model.Coupled with the strategy based on optimal frontier,the "two-step" sequential approximate modeling method is proposed to improve the accuracy of both global model and loal front point.According to the requirements of reliability index of electronic system,a reliability-based robust design optimization method is proposed by combining the conjugate gradient method and the variance criterion method.Consequently,this method ensures the robustness and reliability of the system performance at the same time.The proposed method is well applied to the structural optimization of manganin shunt used in electric meter.Considering that the parameters of electronic components are discrete,a branch and bound method for discrete variables optimization inspired by the idea of integer programming is proposed.And meanwhile,the constraint strategy is adopted to reduce the constraint space and solve the multi-objective optimization problems.For this stage,the parametric design,tolerance design and cost control could not be considered simultaneously during the optimization,and a synchronous "quality-cost" robust design optimization method is proposed on the basis of the restrictive relations between product quality and cost.The voltage sampling circuit example is taken to verify the effectiveness of the proposed method.The optimal results with significant improments illustrate that the proposed method is more applicable comparing with the traditional multi-objective genetic algorithm.To solve the out-of-tolerance problem caused by components' parameters drift and temperature change in the electric meters,the basic components research is taken as a starting point.The actual working temperature of the components is obtained using finite element method.On the basis of the components modeling,the electric energy measuring system Matlab/Simulink model is established to rapidly calculate the measurement error under various uncertain factors,and the accuracy of the model is verified through experiments.The “two-step” sequential approximate model is applied to establish the approximate models for measurement error and etc.,whereby the “quality-cost” robust optimization equation is designed.By using the multi-objective optimization algorithm based on constraint strategy,a set of optimal designs is obtained to meet different requirements of robustness and economy in the engineering application.