This paper reports on work on a structural health monitoring system done for NASA at North Carolina A&T State University. The system combines a multiagent system as the "brains" and a workflow system as the "brawn." Agents here typically serve as proxies for techniques with intensive communication and computation requirements. Agents negotiate to determine a constellation of techniques for solving the task at hand and communicate it to a workflow engine, which actually carries out the tasks. A hierarchically structured collection of monitor agents determines efficient ways to use the computation resources of the workflow engine. A monitor agent at the lowest level in the hierarchy is associated with a wireless mote that transmits data from an acoustic sensor (or perhaps an optic fiber or strain gauges). It negotiates with specialized agents to find techniques for extracting feature vectors from signal samples and classifying the associated events. The classifiers (in the workflow engine) are trained by several machine-learning techniques. The regions of the structure for which the various monitor agents are responsible form an inclusion hierarchy, reflected in the hierarchy among the agents. This hierarchy is set up among the agents using the contract net protocol. As alternatives, we have considered the iterated contract net protocol and coalition game theory. The current prototype architecture is restricted to a single member, but the general principles generalize to structures as complex as aircraft (the ultimate target of this project).
展开▼
