Research On Global Sensitivity Analysis Method For Structural Uncertainty Optimization Design

Due to factors such as manufacturing errors,inaccurate measurement data,and changes in operating conditions and environments,there are a large number of uncertain variables in actual complex equipment.Optimal design of structural uncertainty is to reduce the impact of these uncertainties on the deterioration of performance and improve equipment.Effective means of overall performance and reliability.Global sensitivity analysis is an effective way to measure the degree of influence of these uncertain variables on performance indicators and improve the confidence of predicted performance indicators.The thesis focuses on random design variables and their distribution parameters with uncertainties at the same time,and proposes a global sensitivity calculation method for uncertainty variables and their distribution parameters.While ensuring the accuracy of the global sensitivity index,it also effectively improves the calculation efficiency.And the application verification was carried out in the uncertainty optimization design of the head structure of the plate-fin heat exchanger of the air separation equipment.The research content of this article mainly includes the following four aspects:(1)A method for calculating the sensitivity of uncertain variables based on the unscented transform method is proposed.Aiming at the problem of only considering the uncertainty of the variables,in order to improve the computational complexity of the traditional Monte Carlo sampling method,a sensitivity analysis method based on the unscented transformation method is proposed.The integral interval segmentation is used to reduce the sub-interval performance function.The degree of non-linearity in the calculation of each sub-interval.The unscented transformation method is used to estimate the conditional expectation and variance of the output response.The singlelayer sampling is used to calculate the global sensitivity index of the uncertainty variable.Monte Carlo analysis method reduces the number of loop layers and the number of samples per layer,and reduces the calculation cost.(2)Considering the uncertainty of variables and their distribution parameters at the same time,a global sensitivity analysis method for the distribution parameters of random uncertainty variables based on variance sampling is proposed,and the Gaussian integral formula is improved by multiplying dimension reduction and unscented transformation methods.The technology simplifies the triple cycle of the Monte Carlo method into a double-layer cycle,reduces the number of samples,and satisfies the accuracy of the calculation result of the sensitivity index while effectively saving the calculation cost.(3)Aiming at the problem that the accurate design model of complex equipment is complex and takes a long time,which leads to the low efficiency of sensitivity analysis and uncertainty optimization,a global sensitivity analysis method based on adaptive Kriging model is proposed.Based on the prediction interval maximization criterion,new sample points of uncertain variables and their distribution parameters are selected,and the accuracy of the Kriging agent model is improved through adaptive sampling.The Kriging agent model is used for sensitivity analysis.In the optimization of complex equipment for engineering applications,the huge time cost is compared with the simulation model,which greatly improves the calculation efficiency of global sensitivity.(4)The application was verified in the uncertainty optimization design of the head structure of the plate-fin heat exchanger of the air separation equipment.Taking the uneven distribution of head fluid as the evaluation index,and taking the three structural variables of aperture d,baffle installation height h,and baffle length l as uncertain input variables,simulation calculations were performed to obtain the head performance under different head structure sizes.Assign unevenness,use Matlab to construct a highly accurate Kriging agent model;calculate the sensitivity of three uncertain input variables and their distribution parameters,and exclude variables that have a small impact on the output response;minimize the unevenness as the optimization goal The optimization was performed by using the NLPQL gradient optimization algorithm in Isight software.The accuracy of the proposed method was verified by comparing the prediction results with the simulation results,and the optimal size of the perforated baffle head of the plate-fin heat exchanger was obtained.