Method And Its Application Of Parameter Correlation And Equivalent Simplification In Complex Product Design

With the complex structure, complex operation condition, numerous design variables and complex performance functions in complex product, the performance design is quite complex, and it’s hard to obtain the optimum design results. Model complexity and computational complexity are two difficulties in performance design of complex product. Basing on analysis and overview of the literature of complex product performance design, the technologies and methodologies including correlation strength calculating between design variables and performance functions, design model decoupling, design model equivalent simplification are deeply studied. According to the research achievements, the performance design software system for large scale air separation unit is developed and applied successfully to the relevant scientific projects.The main contents of the dissertation are organized as follows:In chapter 1, the research status of performance design of complex product is reviewed. The key technologies for correlation strength calculation between design variables and performance functions, design model decoupling, design model equivalent simplification, and performance calculation are discussed, and the shortages of present similar researches are proposed. Then the research content and organization structure of the dissertation are presented.In chapter 2, the correlation strength calculation technology between design variables and performance functions based on multiple data fusion is proposed. The multiple sources data are fused and performance design model is established based on response surface method and gray neural network method. Design variables are partitioned into strong correlative variables, weak correlative variables and changeable correlative variables according to sensitivity analysis method and information entropy theory. The changeable correlation matrix is established, then the correlation strengths between design variables and performance functions are calculated accurately, and the design variables are adaptive adjusted in the performance calculation.In chapter 3, the multiple couplings clustering analysis and design decomposition technology in the design model of complex product is proposed. The matrix of synthetic coupling degrees, which included couplings between design variables and performance functions, couplings among design variables, and couplings among performance functions, is created based on correlation strength analysis between design variables and performance functions. Clustering analysis of design variables and performance functions are accomplished through dimensional reduction of the synthetic coupling degrees matrix and comparison of coupling degrees among branches of the binary tree. The clustering degree is proposed to estimate the decoupling effect, and then the design model is decomposed to reach the greatest clustering degree.In chapter 4, the equivalent simplification technology for similar combined model is proposed based on parameter perturbation analysis. The characteristic of similar combined model is determined according to small disturbance analysis results of design variables, and the parameter perturbation analysis is put forward based on singular perturbation theory. Criterions for equivalent simplification of the similar combined model are proposed, which include partitioning criterion of the similar combined model, selecting criterion of sensitivity sub-model, and calculating criterion of correction factor. The equivalent simplification model is applied to product purity prediction of distillation column under open-loop and close-loop control.In chapter 5, the equivalent simplification technology for multiple data fusion model is proposed based on uncertainty analysis of multiple factors. The quantitative characteristics of multiple factor uncertainties are analyzed, and the product performance uncertainty function is developed considering all these uncertainties. The sample points of design variables are determined based on predictive interval of product performance, and the sample points of noise parameters are determined based on mean square error of the metamodel, then equivalent simplification model is constructed driven by performance uncertainty. The simplification model is applied to performance calculation of complex product, and structural robust design of a heat exchanger is used to demonstrate the effectiveness of the proposed method.In chapter 6, the software system for performance design of large-scale air separation unit is developed. The proposed technologies are used to performance design of air separation unit. The architecture and function modules of the software are described. The software are used to performance design of system level, critical equipment, and critical parts of large air separation unit in Hangyang Company.In chapter 7, the key research contents and conclusions are summarized along with recommendations for future research.