Research On Multidisciplinary Robust Design Optimization Methods Based On Probabilistic Analytical Target Cascading

Along with the increasing importance of product quality and high complexity products, Multidisciplinary Robust Design Optimization(MRDO) which can effectively improve the product quality of complex system complied with the tidal current lives. MRDO integrates uncertainty analysis with Multidisciplinary Design Optimization(MDO). The MDO method can organize system uncertainties well, make full use of the role of the various disciplines, and achieve the consistency design by concurrent design of various disciplines and coordinate design of multi disciplines. Probabilistic Analytical Target Cascading(PATC) is a promising MRDO technology with good properties in decomposition-coordination, independent parallel optimization and convergence for complex hierarchical system. This dissertation studies the PATC method further to solve MRDO problems considering both objective robustness and feasibility robustness. The aim is to improve the efficiency and precision of the PATC method.Firstly, for the coordination strategy in PATC framework, various coordination strategies have been studied, especially the Exponential Penalty Function(EPF) formulation with better convergence property for target-response coordination in PATC. The PATC-EPF framework is established, and two examples are tested to verify the effectiveness of the PATC-EPF method.Secondly, in order to improve the accuracy of uncertainty analysis and efficiency of probabilistic optimization in PATC-EPF, a sequential PATC method based on Performance Moment Integration(PMI) and Performance Measure Approach(PMA) is proposed. The PMI method is applied to handle the objective robustness, the PMA method is to deal with the feasibility robustness. In both PMI and PMA, the Hybrid Mean Value(HMV) method is used to search the Inverse Most Probable Point(IMPP). Learning form Sequential Optimization and Reliability Analysis(SORA), the solution rapidly approaches its optimum by shifting the constraint. Two numerical problems are devoted to demonstrate the effectiveness of the proposed method.Finally, to reduce the computational cost of PATC when mathematical models are complex, especially when simulations are used, the PATC-EPF-KRG method is proposed. Based on the PATC-EPF framework, Kriging metamodels are used to replace the complicated models. A geometric programming problem is tested to verify the feasibility and high efficiency of the proposed method. The PATC-EPF-KRG method is also applied to the design optimization of the gear reducer, which highlights the great potential of the proposed method.