Sliding mode observers for vision-based fault detection, isolation and identification in robot manipulators

This paper proposes a possible scheme to detect, isolate and identify actuator and sensor faults which can occur in a robot manipulator. The scheme relies on the estimation of the robot configuration obtained by analyzing online the images captured from a camera during the robot motion. In this way, provided that no fault acts on the camera signal, it is possible to have a complete detection, isolation and identification of the faults acting on the inputs or on the outputs of the robotic system, provided that the faults are distinguishable from noise. To determine the residual signals, which indicate the presence of faults, three different observers having input laws designed relying on second order sliding modes, are proposed in the paper. The first observer enables the sensors Fault Diagnosis (FD), while the other two observers are specifically designed for the diagnosis of actuator faults. Experimental tests are made on a COMAU^® SMART3-S2 anthropomorphic rigid manipulator. To acquire the images from the workspace, an AXIS^® 207 camera is employed.