On Decentralized Optimization for a Class of Multisubsystem Codesign Problems

Abstract

Codesign refers to the process of integrating the optimization of the physical plant design and control of a system. In this paper, a new class of codesign problems with a multisubsystem architecture in both design and control is formulated and solved. Our work here extends earlier research on models and solution approaches from single system to multisubsystem codesign. In this class, the optimization model for the physical design part in each subsystem is assumed to have a convex objective function with convex inequality and linear equality constraints. The optimization model for the control part of each subsystem belongs to a class of finite time-horizon linear quadratic regulator (LQR) feedback control. A new multilevel decentralized method is proposed that can obtain optimal or near-optimal solutions for this class of codesign problems. Details of the model and approach are presented and demonstrated by a numerical as well as a more complex spring-mass-damper system example. The proposed decentralized approach has been compared with a centralized approach. Using a scalable test problem, it is shown that as the size of the problem is increased, the computation effort for the decentralized approach increases linearly while that of the centralized approach increases nonlinearly.