An informational approach for sensor and actuator fault diagnosis for autonomous mobile robots

In this paper, a model-based fault detection and isolation (FDI) method is proposed, with the objective to ensure a fault-tolerant autonomous mobile robot navigation. The proposed solution uses an informational framework, which is able to detect and isolate both sensor and actuator faults, including the case of multiple faults occurrence. An information filter with a prediction model based on encoders data is adopted. For the diagnosis layer, a bank of filters are used. Residuals are generated by computing the Kullback-Leibler Divergence between the probability distribution of the predicted estimation with updated estimation obtained from sensors measurements. In order to isolate encoder and actuator faults, a secondary information filter with a prediction model based on a closed-loop controller is added. An additional bank of filters is developed, and extra residuals based on the Kullback-Leibler Divergence are generated. In the proposed method, the two designed filters supervise each other, which improves fault diagnosis, by taking into account all available information of the system, from control objective to multi-sensor data fusion. Actuator and sensor faults are treated within the same frame during the fusion process, and multiple faults occurrence is considered. A real-time experimentation on a real differential mobile robot is performed and demonstrates the efficiency of the proposed method.