Doctoral and Postdoctoral Research: Current Research | Doctoral Research | Past Research Decomposition-based design optimization methods, such as augmented Lagrangian decomposition (ALD), analytical target cascading (ATC), and collaborative optimization (CO), apply to design problems that can be partitioned into smaller subproblems, and require that a system partition and coordination strategy be defined a priori. While the partitioning and coordination strategy selection tasks have been studied at length independently, the interaction between them has not been addressed. I am investigating this interaction and developing methods for optimal partitioning and coordination decisions that account for this interaction. Studying tradeoffs that exist in partitioning and coordination decisions can also aid in quantitatively determining when it makes sense to use a decomposition-based approach. The following paragraphs describe the principal elements of my dissertation research. Partitioning and Coordination Decision Interaction and Tradeoffs Simple decision models were used along with exhaustive enumeration to confirm that partitioning and coordination decisions are coupled tasks. Investigation of the tradeoffs that exist when making these decisions was studied and used to quantitatively assess when systems are good candidates for decomposition-based optimization. Partitioning and Coordination Decisions for ALD All possible partitioning and coordination options for a class of ALD formulations were identified using graph-theory. This result was used to develop a coordination decision model specifically for ALD. ALD provides tremendous flexibility in coordination decisions over other formulation frameworks, and is therefore an excellent platform for studying partitioning and coordination decision interactions. Evolutionary Algorithm for Partitioning and Coordination Decisions The initial tradeoff and interaction studies utilized exhaustive enumeration for decision making, an approach that is only successful for small systems. A multi-objective evolutionary algorithm is under development for solving the optimal partitioning and coordination decision problem for large systems under the ALD framework. Design Examples The primary case study entails the integrated design optimization of an electric vehicle. Three vehicle systems are considered in the design: chassis, powertrain, and structure. Numerous interactions between these systems have been modeled, such as the effect of acceleration on vehicle pitch motion and the influence of battery size and position on vehicle dynamics. Several additional design examples have been developed for use in illustrating decomposition-based design optimization and partitioning and coordination decision-making:
|
|
|