A performance-based investigation of parallel and serial approaches to multidisciplinary analysis convergence

For many real-life engineering systems, the costliest component of the MDO cycle is the coupled system analysis. The reason for this high cost is that the decomposition of a large system typically results in a coupled grouping of subsystems that are not hierarchical in nature. In the interest of identifying means for reducing the cost associated with multidisciplinary design synthesis, the present research effort provides a comparison of numerous approaches for the convergence of a multidisciplinary (coupled) system analysis. The comparison involves popular formal approaches for nonlinear analysis convergence, as well as a recently developed heuristic technique for analysis convergence, known as the Data Fusion Analysis Convergence (DFAC) procedure. This comparison focuses on two performance-based characteristics - the overall computational run time, and the total analysis cycles required for convergence. A majority of the nonlinear functions that are utilized for the comparisons are generated using the CASCADE simulator. To supplement these simulations, one Class II system from the MDO Test Suite is also examined. Using CASCADE, multidisciplinary analysis scenarios are generated to represent varying degrees of size, nonlinearity, and instability. The comparison takes place bom in a serial and in a parallel setting, using the Message Passing Interface (MPI) paradigm.