Analytical Redundancy and Fuzzy Inference in AUV Fault Detection and Compensation

Abstract : This paper will address the results of a recent study developing model based techniques using analytical redundancy in the production of observation residuals that are processed for the on-line, real time detection of dynamic faults during the operation of AUVs. AUVs are being developed for oceanographic survey, as well as military, missions. Now that underwater navigation to sufficient precision within cost limits is possible, the remaining technical issue is to improve the reliability of the mission completion. Increasing mission reliability implies many things-some having nothing to do with automatic fault detection. However, assuming that the vehicle is equipped with highly reliable sensor suites, connections, and computing components, it is of interest to see if unexpected faults could be detected reliably so that system reconfiguration could be accomplished to allow missions to be continued in some fashion. A survey of fault detection methods indicates that alarms can be easily monitored if signals are static. This is done using limits and trends' analysis with thresholds set for various levels of severity. With dynamic signals, such as those induced by an actuator fault in the form of a stuck fin, or a loose fin, or fouling of a propeller, the transient nature of the signal makes limits and trends analysis invalid. In these cases, signals are sought that include servo error, and the residual error in observation filters. Designing observation filters builds analytical redundancy into the decision making. Additionally, since there is always the difficulty of separating the fault response from disturbance response, operation near the surface under waves requires the development of a wave detector.