An expert system applied to the controls tuning of digital force-reflecting servomanipulators

This thesis presents an expert system for the purpose of optimizing the controls tuning of a gear and shaft force-reflecting servomanipulator. Remote maintenance techniques have, for the first time, allowed the production of teleoperators that can function inside a hot cell environment without the need for hands-on repair. However, these manipulators are difficult to tune due to the conflicting priorities of maximizing operator sensation of force reflection and minimizing operator fatigue in combination with the complex nonlinear control algorithms and cross coupled motions. Due to the heuristic nature of this tuning problem and the emphasis on human perception of performance, an expert system is considered as an alternative to algorithmic optimization of gains.

The primary focus of this thesis is to produce a functional expert system that can be integrated with existing real-time control software. The development system is the Oak Ridge National Laboratory's Advanced Integrated Maintenance System. The control software for the dual arm master/slave manipulator and associated operator interface is written in FORTH. Therefore, the expert system "shell" utilized is the "FOrth-based Production System" (FORPS) developed by Chris Matheus at ORNL.

The emphasis in the content of this thesis is on the practical production of the expert system itself. The expert system development is carried from problem definition to knowledge representation to rule collection and integration into the expert system. Operator interfaces, real world testing, and suitability of approach in comparison with more traditional and orthodox optimization techniques are addressed.