Efficient Structural Reliability-based Design Optimization Method And Its Application In Aerospace Structure Design

The reliability of aerospace equipment is one of the most important parts of equipment quality evaluation,and it is also the basis for the optimal structural performance.As the main load-carrying component in aerospace equipment,the servicing performance of thin-walled stiffened structure is directly related to the success of the launch mission.Because uncertainties widely exist in the lifecycle of structure,it is of great practical significance to consider their influence and use the design optimization method to improve the structural performance.Reliability-based design optimization method is a significant way to quantify the structural failure probability and improve the structural reliability.How to guarantee the accurately.efficiently and robustly of the structural reliability-based design optimization process is the technical problem puzzling the engineering field for a long time.This dissertation mainly discusses the computational efficiency,robustness and accuracy of reliability-based design optimization method from aspects as epistemic uncertainty.stochastic uncertainty and multi-source uncertainty.Serval improved methods are proposed to enhance the performance of reliability-based design optimization and applied into the aerospace structural design.Firstly,enhance chaos control method and efficient adaptive-loop method are proposed in this dissertation,and the systematic research is carried out from two aspects of reliability analysis method and optimization strategy under epistemic uncertainty.The robustness of the design point search process and the efficiency of the optimization are improved by the proposed methods.On this basis,a hierarchical framework for curvilinearly stiffened panel with multi-cutouts is constructed,which significantly boosts the design efficiency.Secondly,for the stochastic uncertainty,this dissertation develops a global reliability-based design optimization method considering multiple design points.Combined with the active learning Kriging model,constraint boundary sampling technology and enhanced step length adjustment method,the proposed method can realize the efficient and robust solution of reliability-based design optimization under multiple design points.Finally,considering both the epistemic uncertainty and stochastic uncertainty introduced by the insufficient data,a confidence-based design optimization method using the single-loop strategy is established to simplify the optimization process and reduce the computational cost significantly.The proposed method provides the necessary theoretical basis for the application of reliability analysis and optimization method in the aerospace field.The main contents of this dissertation are as follows.1.The non-probability reliability-based design optimization framework using enhanced chaos control method is proposed to improve the robustness and computational efficiency.The oscillation detection is firstly involved into the reliability analysis based on the direction information of previous iterations.The control factor is adaptively calculated to control the convergence of the following iterations,which effectively improves the robustness of reliability analysis.According to Wolfe-Powell criterion,the control factor is rechecked and re-updated to ensure its rationality.The robustness and efficiency of the reliability analysis process is guaranteed by the updated control factor.Finally,the performance of the proposed method is verified by numerical examples and the reliability-based design optimization of stiffened cylindrical shell.2.The efficient adaptive-loop method is proposed based on the single-loop strategy and double-loop strategy sequence switching.It greatly reduces the computational cost of nonprobability reliability-based design optimization for practical engineering problems.Firstly,according to the target reliability index,the distribution of uncertain variables and the sensitivity information of the constraint function,the reliability constraint is transformed into deterministic constraint through the approximate transformation of design points.Thus,the deterministic solution of reliability-based design optimization problems is realized.In order to ensure the accuracy of reliability-based design optimization,this dissertation introduces accuracy detection and the enhanced chaos control method to verify the rationality of the current optimal solution.To further improve the computational efficiency,the mechanism of inactive constraint detection and deletion is introduced to realize the high-precision and efficient solution of reliability-based design optimization problems under multiple constraints.For the curvilinearly stiffened panel with multi-cutouts,this dissertation constructs a hierarchical framework based on efficient adaptive-loop method through the hierarchical solving of layout design and detail design and the separation of reliability-based design optimization and deterministic design optimization.The performance of this method is also verified by the reliability-based design optimization of curvilinearly stiffened panel with multi-cutouts.3.A global reliability-based design optimization method considering multiple design points is proposed in this dissertation.It significantly improves the utilization of sample information and the accuracy of reliability analysis.And also improves the applicability of reliability-based design optimization in engineering problems.In order to ensure the efficiency and robustness of the design point search process in reliability analysis,an enhanced step length adjustment method is proposed.It provides a powerful tool for reliability analysis considering multiple design points.Aiming at the poor accuracy of reliability analysis using the analytical method,this dissertation points out that multiple design points are one of the fundamental reasons for the error.For this reason,bulge method,FORM and SORM are also introduced to search design points on the limit state surface.Considering that multi design point search will bring extra computational cost,an improved constraint boundary sampling method is proposed based on the active learning Kriging model.It can improve the utilization of sample information by selectively fitting the limit state surface in the optimization.Finally,combined with the efficient global optimization strategy,a multiple design points oriented global reliability-based design optimization framework is established.The efficiency and applicability of the proposed method is verified by mathematical and numerical examples considering multiple design points.4.A confidence-based design optimization framework based on the idea of the single-loop strategy is proposed in this dissertation.It can comprehensively evaluate the impact of epistemic uncertainty and stochastic uncertainty on structural performance,and greatly enhance the efficiency of the confidence-based design optimization.For the insufficient data with known distribution types,this dissertation introduces the confidence of failure probability to quantify the propagation of the epistemic uncertainty.Considering that MCS method is needed to perform the reliability analysis and even confidence level evaluation in confidence-based design optimization,a complete performance measure approach is developed in this dissertation,which is inspired by the idea of performance measure approach.The original failure probability calculation of the inner loop is replaced by the design point search process based on secondorder reliability method.The computational efficiency of the whole optimization process is boosted by the proposed method under the premise of ensuring the accuracy.Finally,the singleloop strategy is used to integrate the reliability analysis into the confidence level evaluation process.Hence,the treble-loop nested confidence-based design optimization is transformed into a double-loop problem,which highly increases the computational efficiency.The performance of the proposed method is verified by several benchmarks.Compared with the traditional confidence-based design optimization method,the proposed method can greatly reduce the computational cost while ensuring the accuracy of reliability analysis and confidence level evaluation process.