Reliability-based Design Optimization Through Single-loop Strategy And Adaptive Surrogate Approximation

Nowadays,the structures of modern products tend to be more complex,and the working environment is stricter.As a result,high performance of modern products is required.But in practice,the performance is highly sensitive to different kinds of uncertainties.Due to the presence of uncertainties,the failure probabililiy of products may increase.Moreover,the intended function may not be achieved.Thus,it is important to perform reliability-based design optimization(RBDO)with the consideration of uncertainties during the stage of product design.However,there are still some challenges in the field of RBDO.For example,the effectiveness of the reliability analysis needs to be improved,the nested structure of RBDO needs to be efficiently decoupled,and new method for solving the time-dependent RBDO problem needs to be proposed.The researches in this dissertation focus on these challenges as well as try to make contribution to the development and application of RBDO.On one hand,some efficient reliability analysis methods and single-loop-based optimization strategies are proposed from the view of analytical methods.On the other hand,some advanced adaptive sampling and surrogate modeling techniques are developed for both the RBDO and time-dependent RBDO problems.The studies carried out in this dissertation can be divided into the following four parts.(1)A general optimization formulation for both decoupled strategy and single-loop strategy is constructed.Subsequently,an adaptive hybrid single-loop RBDO method with multiple minimum performance target points(MPTP)is proposed.Since both sequential optimization and reliability assessment method and single-loop method can not achieve a good balance between the accuracy and efficiency,a judgement criterion is developed to adaptively select the approximate MPTP or true MPTP.Moreover,an efficient iterative control strategy is proposed for searching the true MPTP while performing inverse reliability analysis.Finally,the effectiveness of the proposed reliability analysis method and RBDO method is demonstrated by some numerical examples and one engineering application.(2)A single-loop RBDO method is proposed based on a new concept of iterative reliable design space.Firstly,the advantages and disadvantages of reliable design space(RDS)are discussed in detail.Secondly,the IRDS-based RBDO method is proposed to overcome the disadvantages of the RDS-based RBDO method.In IRDS,an improved shifting vector strategy is developed to guarantee the accuracy of the equivalent probability constraints as well as the iterative reliable design space.Finally,the computational efficiency and accuracy of the proposed reliability analysis method and RBDO method are demonstrated by some numerical examples and one engineering application.(3)An active failure-pursuing sampling framework(FPS)is proposed,then the corresponding reliability analysis method and RBDO method using adaptive surrogate modeling technique are developed.Firstly,the FPS framework that can reasonably determine the location of sample points is built for approximating the computationally expensive limit state function.Secondly,the FPS framework is extended to RBDO with multiple probabilistic constraints by combining with simultaneous learning strategy.Finally,the performance of the proposed reliability analysis and RBDO method is validated by some numerical examples and one engineering application.(4)A wrong classification probability of Kriging model is derived for the timedependent probabilistic constraints,and the time-dependent reliability analysis and RBDO methods are proposed based on the probability.Firstly,the estimation error of failure probability can be estimated based on the wrong classification probability for the time-depenent limit state functions.With the estimation error of failure probability,the direct Kriging modeling strategy is developed with the random variables,stochastic processes and time parameters as inputs.Secondly,two strategies,named as global and local Kriging modeling strategies,are proposed for approximating the constraint boundaries.Finally,the effectiveness of the proposed methods is demonstrated by some numerical examples and one engineering application.Last but not least,the contribution and novelty of this dissertation are summarized,and the future work is also discussed.