Development for Space Robotics with Emphasis on Fault-Tolerance, page 5

A sensor-based fault detection scheme that employs parity relations among temporal redundancies has been derived and implemented for the knuckle module. The failure of any one sensor (position or velocity) is masked, thus making FDI and sensor reconfiguration transparent to the control system . The scheme, as designed, can be configured in a triple or n-modular redundancy (NMR) setup so that faulty sensors can be isolated and dynamically configured out of the system. Actuator FDI for the knuckle is realized using parameter identification techniques based on the work of Isermann. In this technique, failure is detected and identified depending on the statistical variations of the on-line estimated values of the physical parameters of the actuator from their fault-free ones. The changes, in orientation and magnitude, of online estimated vectors in parameter space provide the necessary diagnostic information.

Reconfigurable Control - The problem of control of fault-tolerant systems, in particular those with redundant control inputs is of sufficient general interest across a variety of disciplines (robotics, flight control systems, etc.) that it warrants the investigation and development of a general theoretical framework. While preliminary efforts at deriving fault tolerance schemes that take into account the full dynamics of a manipulator have met with success, the absence of a formal mathematical framework restricts our ability to answer qualitative questions like existence and optimality of solutions to such problems. What is needed is and in-depth study of the theoretical issues underlying the utilization (under normal and failure modes) of redundant control elements in general nonlinear control systems. An operator-theoretic approach which treats the control problem in an input-output framework is also being pursued. The problem is then to solve appropriate operator equations that yield infinite solutions by optimizing a prescribed functional. The functional so prescribed will reflect useful criteria during normal operation and will reflect the failure status of control actuators when the FDI system signals a fault. The mathematical framework may be thought of as generalizing the familiar generalized inverse problem in robot kinematics to the more complex realm of operators that map, say, one Hilbert space into another. We expect to be able to fully and formally characterize the nature of redundancies in any control system, in a manner analogous to the concept and formulation of available redundancy in the kinematics of Level III manipulators.

Conclusions - A broad technical base for the development of component and system level technologies for space robotics. This paper gave an overview of our ongoing work in several key areas that are geared to address the needs of NASA's future missions. Our emphasis, driven by the needs of such missions, continues to be on enhanced accuracy and performance, condition monitoring and condition-based repair, modularity, criteria-based decision making and, above all, fault tolerance with reconfigurable control. To support the stated goals of our mission, and to fully integrate the component technologies described in this paper, we are also doing an in-depth study of micro-sensor technologies, sensor fusion and software architectures for failure-responsive control. This will allow us to realize the kind of machine intelligence that will animate the robots of the future.