Intelligent monitoring of multi-axis robots for online diagnostics of unknown arm deviations

In the age of Industry 4.0, multi-axis robots are widely used in smart manufacturing thanks to their capacity of milling high complex forms and interacting with several systems in production lines. However, during manufacturing, the occurrence of small drifts in the robot arms may lead to critical failures and significant product quality damages and, therefore, high financial losses. Hence, this paper aims to develop an effective and practical methodology for online diagnostics of robot drifts based on information fusion of direct and indirect monitoring. The direct monitoring exploits the already installed encoders on each servomotor of the robot while the indirect monitoring uses heterogeneous sensors (current, vibration, force and torque) placed at the robot tool level. The sensor measurements of the robot tool are processed, in an offline phase, to build health indicators and fused to learn a classifier for drifts detection and diagnostics. Then, during the online phase and in the case of presence of new drift patterns, the encoder measurements are used to label these patterns and update the classifier learned previously to diagnose their origin. The efficiency and robustness of the proposed methodology are verified through a real industrial machining multi-axis robot that investigates different drift severities of its arms.