Many mechanical systems are sufficiently complex that it is impractical to describe their dynamics by exact mathematical models. In the presence of such modeling uncertainties, advanced controllers like adaptive controllers perform better than linear feedback controllers since they actively reduce the uncertainty by online parameter estimation. Unfortunately, the advanced control strategies, due to their lack of robustness, can become unstable in the presence of unpredictable external disturbances, and hence, there exists a need for a fault-tolerant approach to preserve the overall system integrity even at the cost of design performance. This motivated the research, presented in this paper, to investigate the suitability of the IDES (Influence Diagram Based Expert System) as an expert supervisory controller to predict incipient instability, a significant failure mode, and take corrective action in real-time when closed loop stability appears to be in danger. The expert supervisory control scheme is demonstrated on a model-referenced adaptive controller as applied to a robotic manipulator. The real-time expert system, with the information from sensors, dynamically optimizes the cost of control and as a result chooses between a robust auxiliary controller and the nonrobust adaptive controller depending on inferences made from the observable variables. IDES, as a real-time expert supervisory controller, preserves the stability of the system even under potentially destabilizing unexpected disturbances, exhibiting on demand a fault-tolerant behavior by trading design performance for overall system integrity. The results indicate the potential for influence diagram expert systems in monitoring and controlling mechanical systems where exact mathematical models are difficult or not practical to obtain.
An Approach to Self-adaptive Software Based on Supervisory Control
