Stuck Actuator Fault Detection and Isolation on the MMX Rover

Abstract

Since rovers for the exploration of celestial bodies are extremely expensive and long-term projects, it is crucial to avoid the collapse of these systems due to faults. Hence, the development of Fault Detection, Isolation, and Recovery (FDIR) systems for these rovers is indispensable. The MMX (Martian Moons Exploration) Rover is a small, wheeled rover created for the exploration of Mars moon Phobos as part of the MMX Mission by the Japanese Space Agency. In this thesis the problems of fault detection and fault isolation on one of the MMX Rover's locomotion units are discussed. The objective is to answer the question of how to detect an immobilization fault in the rover's locomotion unit and how to differentiate on which constructional element the fault occurs. To address this research topic, experiments are conducted to simulate the expected fault scenarios as well as normal operation. Specifically, the fault scenarios of a stuck wheel and a stuck leg are induced under various changing parameters, and torque and motor currents are measured. Additionally, baseline tests are performed to investigate the influence of Earth's gravity, motor drive velocity and direction of rotation. The test data is analyzed using Python scripts. The values of nominal and faulty operations are compared in order to establish thresholds for fault detection. Machine learning techniques based on logistic regression are employed to enable fault isolation. The results of the baseline tests show dependency of the torque and current measurements on all mentioned baseline test parameters. The evaluation using the methods described demonstrates reliable error detection through a torque threshold. Furthermore, the classification algorithm enables 95% accurate determination of the fault location.