Title: Network Optimization Methods in Passivity-Based Cooperative Control with Application to Network Fault Detection and Isolation
10:00 - 11:00
NB 5113.0202 (Nijenborgh 4)
Network Optimization Methods in Passivity-Based Cooperative Control with Application to Network Fault Detection and Isolation
Cooperative control has recently become a trend in control research due to the popularity of multi-agent systems and the "system-of-systems" design philosophy. One important tool for analyzing and synthesizing cooperative control laws in the notion of passivity, as it allows one to guarantee stability of the closed-loop networked system by asserting that each individual agent is passive. Recently, maximal equilibrium-independent passivity was used to establish a connection between network optimization and cooperative control, allowing one to compute the steady-state limit of a network system by solving a convex network optimization problem. However, this network optimization framework was shown to hold only for passive SISO agents.
This talk will be divided into two parts. First, we present some recent extensions to the network optimization framework. We first explain how to extend it to MIMO agents. Later, we show that it does not hold for passive-short agents, due to the non-convexity of the achieved network optimization framework. Naturally, we consider a convexification of the network optimization problem, which results in a linear feedback term passivizing the networked systems. Both local and network-based convexification terms can be used, resulting in a local or network-based passivizing feedback term.
Second, we show how to use the network optimization framework to detect and isolate network faults in a multi-agent system. We do so by studying the effect of the faults on the associated network optimization problems, giving an asymptotic differentiation of the faultless system from its faulty versions, as long as the underlying graph is "connected enough". We then discuss how to assert that a networked system converges to a conjectured steady using measurements. We then combine these two methods to provide an algorithm for fault detection, which can isolate κ(G)-2 faults, where κ(G) is the vertex-connectivity of the graph. We will demonstrate the results by case study.