A Comparison of Formal and Heuristic Strategies for Iterative Convergence of a Coupled Multidisciplinary Analysis

data fusion, neural network A primary goal of Multidisciplinary Design Optimization (MDO) is decomposition- to break down a large and complex engineering system into a grouping of smaller, more tractable, yet still inter-related (coupled) subsystems. The problem with this procedure is that the resulting decomposition is typically not hierarchical in nature. Because of this lack of hierarchy, the subsystem outputs (behavior variables) must be initialized to some set of values, and iteratively converged thereafter, which can be a timely and computationally costly procedure. This research effort will compare various techniques for attaining a converged solution to a coupled multidisciplinary analysis. These techniques include formal strategies such as Fixed-point Iteration, which is an easily implemented but inefficient means for iterative convergence, and Newton’s Method, which is an efficient but costly and often times unstable means for iterative convergence. A new heuristic convergence strategy is under development which hopes to demonstrate characteristics that will serve as a “middle ground ” to those of the aforementioned formal strategies. Namely, the new convergence strategy hopes to demonstrate a higher level of efficiency than Fixed-point Iteration, and a higher level of robustness than Newton’s Method. The new heuristic strategy will exhibit neural network-like characteristics in its implementation. Further, the new strategy will utilize an approach that is analogous to a systems-based procedure known as data (sensor) fusion. Namely, information from numerous subsystems that require the same behavior variable as input will b