Estimation strategies for fault detection in a 6-degree-of-freedom robotic manipulator

Condition monitoring and fault detection are used extensively in industry to predict the maintenance requirements of manufacturing robots and minimize unplanned downtime. These techniques generally fall into one of two categories. Data-driven methods use large quantities of historical operating data (such as joint position, frequency of vibration, operating temperature...) to detect abnormal behaviour, and frequently employ machine learning techniques. Model-based methods use physical or mathematical system models to establish a benchmark for expected behaviour, against which the system’s actual behaviour is evaluated. In the context of external space manipulators, condition monitoring is crucial because spare parts are not readily available, and otherwise simple repairs are made difficult by the harsh environment of space. Unexpected inoperability of an external space manipulator can be costly, resulting in the failure of a mission. Model-based condition monitoring relies on the use of estimation filters to predict physical states (position, velocity...) from measurements corrupted by sensor noise. In this paper, several estimation methods are compared in the context of a condition monitoring algorithm for a 6-degree-of-freedom robotic arm simulation. The studied estimation strategies include the unscented Kalman filter, the cubature Kalman filter, and the particle filter. The robotic manipulator is simulated using MATLAB Simulink. Simulated faults are introduced at the end effector, and the filters are evaluated on their ability to correctly detect faults based on the quality of the state estimates. The results of this study will inform the design of a robust and reliable condition monitoring strategy for use in on-orbit external manipulators, while also complying with the associated on-board computing limitations.