New Dynamic Model-Based Fault Detection Thresholds
                     for Robot Manipulators



       M. L. Visinsky, I. D. Walker, and J. R. Cavallaro

      Department of Electrical & Computer Engineering
            Rice University, Houston, TX  77251-1892


                            Abstract

    Autonomous robotic fault detection is becoming increasingly
    important as robots are used in more inaccessible and
    hazardous environments.  Detection algorithms, however, are
    adversely effected by the model simplification, parameter
    uncertainty, and computational inaccuracy inherent in robotic
    control, leading to an unacceptable number of false alarms and
    overzealous fault tolerance.  The algorithms must use
    thresholds to mask out these errors.  Typically, the thresholds
    are empirically determined from a specific robot trajectory.
    The effect of modeling inaccuracy, however, fluctuates
    dynamically as the robot moves and failures occur.  The
    thresholds need to be dynamic and respond to the changes in
    the robot system so as to differentiate between real failures and
    misalignment due to modeling errors.  This paper first
    summarizes the Reachable Measurement Intervals (RMI)
    method of computing dynamic thresholds and then, learning
    from the robot-oriented analysis of RMI, presents a more
    efficient threshold generation method using the manipulator
    dynamics property of linearity in parameters.