C.W. de Silva and K. Wong


Fault detection, machine health monitoring, diagnosis, Kalman filter, robotic manipulators, redundant kinematics, multimodule deployable manipulator system


This paper presents a technique for online identification of faults in a multi-module deployable manipulator system (MDMS) and for task execution in the presence of faults. The MDMS consist of a chain of modules, having a prismatic joint and a revolute joint in each module. A nonlinear model of the MDMS is developed, which is cast in a discrete-time state-space form for use in the failure identification method. Bayes hypothesis testing is employed in the failure identification scheme. First, a possible set of failure modes is defined, and a hypothesis is associated with each failure mode. The most likely hypothesis is selected depending on the observations of the manipulator response and a suitable test. The test used here minimizes the maximum risk of accepting a false hypothesis, and accordingly the identification methodology is considered optimal. A bank of discrete Kalman filters is used for the computation of the hypothesis-conditioned information about the MDMS, which is required in the decision logic. Through this approach, the MDMS, using kinematic redundancy and control, is able to satisfactorily execute a task even in the presence of a failure. Computer simulations of one-module and two-module manipulators are used to demonstrate the effectiveness of the developed methodology for identification of three types of faults: sensor failure, locked joint, and free-wheeling joint; and for controlled manoeuvring in the presence of a locked joint. In particular, it is demonstrated that a two-module manipulator is able to successfully complete a targeting and pointing task in the presence of a joint failure.

Important Links:

Go Back