Research On Uncertainty And Multiobjectivity Of Multidisciplinary Collaborative Optimization

Complex product design often spans over multiple disciplines. Multidisciplinary design optimization (MDO) is a concurrent engineering design tool, which divides the complex product design into several subsystems according to the existing engineering requirements. This decomposition form facilitates the discipline autonomy and the use of existing knowledge. MDO is a new subject and many issues remain to be further improved and perfected. The uncertainty and multiobjectivity of multidisciplinary collaborative optimization are studied. The main contents and contributions of this dissertation are summarized as follows.(1) The system-level optimization of collaborative optimization may be infeasible due to consistency equality constrains. It is difficulty to select the penalty factor when constrain handling method based on penalty function is adopted. In view of these two problems, a new collaborative optimization method based on genetic algorithm is presented. Genetic algorithm is employed in combination with strengthening constrain conditions step by step. The infeasibility degree of individual is calculated with the optimal results from subsystems. The feasibility of individual is determined by its infeasibility degree and the threshold. The threshold is adjusted via cyclic iteration steps to ensure that the system-level optimization will go towards decreasing the unsatisfiability of constrains on consistency equality, thus achieving the goal to enchance the interdisciplinary consistency.(2) To solve the problem that the collaborative optimization results are sensitive to the initial points and usually converge to the local extremum, a new collaborative optimization method based on two-phase optimization strategy is proposed. In the global optimization phase, a bigger slack factor is employed to reach the neighborhood of global extremum while ensuring the requirements of interdisciplinary consistency. According to the control mode of interdisciplinary compatibility, two methods called posteriori modification and prior constraint are presented. In the local optimization phase, the slack factor is reduced gradually and the interdisciplinary compatibility is strengthened so as to reach the global extremum.(3) In view of the uncertainties associated with design variable and design model, a simplified robust collaborative optimization model based on implicit uncertainty propagation (IUP) is presented. The auxiliary design variable is introduced to replace the variant value of coupling state variable, and the variant value is calculated in subsystems. The solution of global sensitivity equation is avoided. The proposed model accords with the structure characteristic of collaborative optimization, and its efficiency and application is improved. The rationality of this model is demonstrated from three angles of optimization structure, modeling uncertainty and interdisciplinary consistency.(4) Two solution methods based on ergodic combination and NSGA-II are proposed for robust collaborative optimization. In the solution method based on ergodic combination, the deterministic optimization result is used as the initial point to avoid the phenomenon that the optimization results are sensitive to the initial points. The weights are given by the ergodic combination mode. In the solution method based on NSGA-Ⅱ, the dynamic feasibility threshold of individual is set to avoid the phenomenon that objective function and its variant usually converge to the local extremum, which is caused by the two-level architecture of robust collaborative optimization. At the earlier stage of genetic evolution, the individuals with smaller value of objective function and its variant are more likely to be preserved so as to ensure the optimization process to reach the neighborhood of global extremum. At the later stage of genetic evolution, the individuals with smaller value of interdisciplinary inconsistency are more likely to be preserved so as to ensure the interdisciplinary compatibility.(5) The multiobjective collaborative optimization method based on linear physical programming is presented for the MDO problem with multiple physical objectives in subsystems, which is based on the precondition that the prior knowledge of all the objectives is known. In view of the fact that the interdisciplinary incompatibility function and physical objective functions have different priority levels, two transformation methods called closest distance and relaxed distance are proposed to transform the interdisciplinary incompatibility function into the disciplinary design constrain. The later method facilitates the physical objective functions to get better solutions.(6) Aiming at the MDO problem with multiple physical objectives in subsystems, the multiobjective collaborative optimization method based on NSGA-II is proposed and focuses on the multiobjective subsystem optimization. To improve the efficiency and accuracy of multiobjective subsystem optimization, a method for producing initial population with feasibility and diversity is presented. To obtain better solutions for the physical objectives, a method for setting feasibility threshold is presented for the interdisciplinary incompatibility function at the second priority level.