Reconfigurable Semi-Analytic Sensitivity Methods andMDO Architectures within the πMDO Framework

摘要

πMDO is a software framework which allows the automatic implementation of multidisci-plinary design optimization (MDO) architectures to find the optimum of any appropriately defined multidisciplinary problem. A significant improvement and extension of the irMDO framework is presented. Using the inherent advantages of πMDO's object oriented and highly flexible structure, several semi-analytic sensitivity methods are implemented for the MDO architectures within the framework. Their generalized nature allows the application of these powerful and efficient methods to any problem defined within πMDO, without modification to the existing structure of the problem. Further, exploiting information gathered by these methods, a new "meta" MDO architecture is proposed which dynamically reconfigures the problem to speed the optimization while maintaining the fidelity of the original analysis. This hybrid approach uses the existing architectures in πMDO en-capsulated within each other to reduce the dimensionality of coupling between disciplines. Again, due to the object oriented nature of πMDO, no modifications are required to the problem statement or the MDO architectures within the framework. Initial results suggest that both these additions produce valuable performance gains while maintaining the general flexibility and simplicity characteristic of πMDO.