Agent-based real-time fault diagnosis

Theory and applications of model-based fault diagnosis have progressed significantly in the last four decades. In addition, there has been increasing use of model-based design and testing in automotive industry to reduce design errors, perform real-time simulations for rapid prototyping, and hardware-in-the-loop testing. For vehicle diagnosis, a global diagnosis method, which collects the diagnostic information from all the subsystem electronic control units (ECUs), is not practical because of high communication requirements and time delays induced by centralized diagnosis. Consequently, an agent-based distributed diagnosis architecture is needed. In this architecture, each subsystem resident agent (embedded in the ECU) performs its own fault inference and communicate the diagnostic results to a vehicle expert agent. A vehicle expert agent performs cross-subsystem diagnosis to resolve conflicts among resident agents, and to provide an accurate vehicle-level diagnostic inference. In this paper, we propose a systematic way to design an agent-based diagnosis architecture. A hybrid model-based technique that seamlessly employs a graph-based dependency model and quantitative models for intelligent diagnosis is applied to each individual ECU. Diagnostic tests for each individual ECU are designed via model-based diagnostic techniques based on a quantitative model. The fault simulation results, in the form of a diagnostic matrix, are extracted into a dependency model for fast fault inference by a resident agent. The global diagnostic inference is performed through a vehicle expert agent that trades off computational complexity and communication load. This architecture is demonstrated on the engine air induction subsystem. The solution is generic and can be applied to a variety of distributed control systems