Simulation-based comparison and development of heuristic convergence strategies for multidisciplinary analysis.

摘要

The large nature of complex engineering systems, such as aircraft, makes coordinated system-level multidisciplinary design optimization (MDO) difficult to achieve in practice. Such methods will rely on computer-based tools to plan and execute the process, with the primary consideration being the time and cost involved. Programs such as DeMAID (Design Manager's Aide for Intelligent Decomposition) have arisen to begin to fulfill this need. The concepts embodied in DeMAID are extended and explored in this work.; One particular MDO problem formulation called multi-discipline feasible (MDF) requires repeated iterated solution of a coupled analysis. This work presents the development and comparison of heuristic convergence strategies for multidisciplinary analysis with the aim of finding strategies to reduce the cost and time in MDF. Consideration is given particularly to the effect of coupling strength (first-order sensitivities) in the development and operation of strategies.; A simulation technique related to discrete event simulation has been developed and implemented to operate on coupled system models built using the CASCADE (Complex Application Simulator for the Creation of Analytical Design Equations) methodology. Time and cost information is collected and statistical techniques are used to compare the results. Primary emphasis is given to parallel strategies as might be used in a cooperative design environment.; The results have demonstrated a profound impact of strategy choice on the cost and time of the convergence of a multidisciplinary analysis. They have also shown the importance of sequencing, and in particular sequencing for minimum feedback coupling strength. Results for the parallel strategies demonstrate a time/cost trade-off with important ramifications to the design process. Finally, strategies are considered in the context of optimization, and the relation of convergence properties to the relative step size through the design space is discussed.