A multidisciplinary design optimization algorithm with distributed autonomous subsystems

This work in progress presents a novel Multidisciplinary Design Optimization (MDO) algorithm that is tailored to provide maximum autonomy to the disciplines (subsystems). The link between system and subsystem levels is performed through shared variables (resources) whose nominal values are not imposed on the subsystems. As opposed to approaches such as Collaborative Optimization, the disciplines have the ability to not only find the optimal values of their own local design variables but also to inform the system level of optimal values of shared variables viewed from the subsystems. A distributed autonomous formulation of the fully integrated MDO problem is provided using a penalty decomposition method and a trust region approach. Disciplinary feasibility is achieved by the addition of artificial variables. Within each iteration loop, exactly one subsystem and the system are optimized. A first-order approximation of each discipline is used at the system level and the quality of this approximation is measured by an update scheme for the trust region. Computational results for this work in progress are provided for two test problems.