A Fault Detection Approach for Robotic Systems Combining the Data Obtained from Sensor Measurements and Linear Observer-Based Estimations

摘要

The design of a fault detection device represents one of the major challenges that manufacturers of robotic systems face today. The detection process requires the use of a number of sensors to monitor the operation of these systems. However, the implementation costs and constraints of these sensors often lead designers to optimize the number used. This could accordingly induce a lack of necessary measures for the optimal detection of failures. One way to bridge this gap consists of realizing model-based estimations of non-measurable state variables describing the dynamics of the real system. This paper presents an approach based on mixed data (measured data and estimated data) for the detection of faults in robotic systems. The proposed fault detection approach is performed using a decision tree classifier. The data used to build this learning stage are obtained from the available measurements of the real system, according to its standard actions. Then, to improve the database classification with unmeasurable data, a linear observer is designed from an analytical model. From the estimations provided by the linear observer, new attributes are built, with the aim of enriching the knowledge used by the classifier and thus improving the rate of fault detection. Finally, an experiment on a robotized actuated seat is presented to illustrate the proposed combined linear observer and classifier approach.