Research On Multidisciplinary Collaborative Optimization And Its Uncertainty And Multiobjectivity

Multidisciplinary design optimization(MDO)is an essential technology for the design of complex engineering system,which can fully consider the coupling effect between disciplines and provide a more reliable solution for the design problem of complex product.But as a new subject,many issues about MDO remain to be further improved and perfected to adapt to the design requirement of increasingly complex engineering system.In this paper,the remaining defects of CO,as well as the solving methods considering uncertainty and multiobjectivity are studied.The main contents and contributions of this dissertation are describled as follows.(1)To solve the problem of premature convergence for the variable relaxation CO method based on discipline inconsistency,a fixed tolerance is added to the relaxation factor of variable relaxation method and the MR method is developed.In addition to preserving the global searching capability of the dynamic relaxation method,the fixed toleranc can ensure a certain size of feasible region for the optimization problem to search the optimal solution when the discrepancy is very small,thus improving the convergence performance of CO.(2)Variable relaxation method based on discipline inconsistency is not applicable to some multidisciplinary optimization problems in which shared variables and coupling variables are not complete coincident,and does not consider the difference of the magnitude of design variables in actual engineering fully.So a new method of magnitude of variables considered dynamic relaxation is presented.This method is not limited by the number of shared variables and coupling variables in each discipline,and the magnitude of design variables is taken into account by establishing the corresponding relaxation factor formulas for each discipline,so that the optimum solution can be searched within a more reasonable feasible region.(3)In view of the problem that the inconsistency of shared variables in improved CO methods may cause the final solution to not satisfy the constraints of the original optimization problem,an alternative method of modified collaborative optimization considering satisfiability of original constraints is proposed.In the new method,the constraints of subsystem level are corrected by the first order Taylor expansion based on sensitivity analysis,and the contribution of the shared-variables inconsistency to the subsystem-level constraints are included,so the feasibility of the system-level optimal solution can be enhanced powerfully.(4)In the robust CO method based on implicit uncertainty propagation,the global sensitivity equation needs to be solved many times,which lowers the computational efficiency of robust CO.A shared variables based robust collaborative optimization is developed,in which the state variables of each discipline and their uncertainty variations are all considered as shared variables,and the propagation of data and the solution of uncertainty variations are avoided during the optimization process.As a result,the computational complexity of multidisciplinary robust optimization can be reduced effectively.(5)Considering that the inconsistency of shared variables in improved CO methods would affect the accuracy of robust evaluation for subsystem constraints,a model error considered robust collaborative optimization is proposed in this paper.Not only the precision errors of design variables and prediction error of system analysis for state variables are reflected,but also the inconsistent variations of shared variables are included.As a result,the robust evaluation of subsystem constraints becomes more accurate and comprehensive,and the reliability of multidisciplinary robust optimization design is improved.(6)Aiming at the MDO problem with multiple objectives in system level,a multiobjective collaborative optimization method is developed by introducing the hybrid multi-gradient explorer algorithm(HMGE).HMGE combines the global searching ability of multi-objective genetic algorithm and the fast convergence ability of gradient-based algorithm,and employs a dynamically dimensioned response surface method for calculating local gradient information,so that it can search for the global Pareto solutions within a limited number of iterations,and improve the computational efficiency and accuracy of large-scale engineering optimization problems.(7)A multiobjective collaborative optimization method based on dynamic weighting is presented for the MDO problem with multiple physical objectives in subsystems.The compatibility objective and the physical objectives are combined into a single object by introducing variable weights and the expected value of physical objectives.The optimal solution of system-level problem is finally obtained through the dynamic adjustment of weights under the priority of the compatibility objective.The proposed method avoids additional selection the optimal solution for each subsystem and improves the solving efficiency of multiobjective optimization problems.The main innovations of this thesis are listed as followings.(1)The MR method and the magnitude of variables considered dynamic relaxation are presented for the system-level problem of CO,so the convergence performance of the system-level problem is improved,and the engineering applicability of the relaxation methods is widened.Besides,the constraints of subsystem level are corrected and an alternative method of modified collaborative optimization is developed,so the feasibility of the final solution is guaranteed for many improved CO methods.(2)A shared variables based robust CO model is establisted for MDO problems under interval uncertainty,in which the uncertainty evaluation is integrated into the optimization process,and the computational complexity of multidisciplinary robust optimization design is reduced.Additionally,considering the impact of interdisciplinary inconsistency on the robustness evaluation of constraints,a model error considered robust collaborative optimization is proposed.(3)Two multiobjective collaborative optimization methods based on hybrid multi-gradient explorer algorithm and dynamic weighting are constructed for the MDO problems with multiple objectives in system level and subsystem level respectively,so some more efficient solutions for large-scale multidisciplinary and multi-objective optimization problems are provided.