A simulation based approach to detect wear in industrial robots

In this paper we simulate, detect, and locate wear in industrial robots operating over long time-periods on repetitive tasks. We simulate wear using the MATLAB robotics toolbox by increasing the Coulomb friction coefficient on select robot axes. Disturbance and measurement noise are included in each axis while computing the inverse and forward kinematics. To compare change across all configurations within a repetitive cycle, the output torque data is transformed into a normalized distribution sampled once every hour. We identify wear as significant change in torque values from normal operating conditions, and detect wear using principal component analysis; the source axis is located by cross-correlating the output data from all axes through time. Simulation results with a range of friction values tested for three different tasks across all axes show that wear detection on the basis of comparison with normal operating condition depends on the manipulator task, whereby onset of wear was detected in five out of six axes for one task and two out of six axes in the rest. Wear location was correctly identified in all successful detections except one.