AIAA 2004-4465 Improving the Performance of Design Decomposition Methods with POD

The use of decomposition methods for multidisciplinary design offer many advantages such as ease of implementation and scalability with increasing numbers of disciplines, but tend to be less computationally efficient than tightly coupled methods. However, by reducing the bandwidth of the interaction between the modules in a decomposition scheme significant gains in computational speed can be achieved. This has been recognized in methods such as Collaborative Optimization (CO), for example, where disciplines don't interact directly with the system level problem but through response surfaces or spline fits. In our past research we have found reduced order models based on Proper Orthogonal Decomposition (POD) to be of some practical use for modeling aerodynamics, but with useful mathematical properties such as incorporating the governing equations of the system into the approximation and its convergence in the limit of a large number of observations of the system. In this work we have implemented POD as a method of reducing the coupling bandwidth between disciplines in a decomposition method called Bi-Level Integrated System Synthesis (BLISS). By using POD the normally high band-width interaction between some disciplines, such as aerodynamics and structures where all of the surface pressures and structural displacements need to be exchanged, is replaced by a one time exchange of modes and a per iteration exchange of values that scale the modes. Results from applying this procedure to an aerodynamic panel code and beam structural model have shown that the results using the decomposition method with POD are identical to those computed using a Multidisciplinary Feasible (MDF) method that internally used the coupled adjoint method for gradients. Computational costs are competitive with the combination of BLISS and POD being less than twice as expensive as the MDF method.