Evaluation of Sensor Failure Detection, Identification and Accommodation (SFDIA) Performance following Common-Mode Failures of Pitot Tubes

摘要

Recent catastrophic air crashes have shown that physical redundancy is not a foolproof option for failures on Air Data Systems (ADS) on an aircraft providing airspeed measurements. Since all the redundant sensors are subjected to the same environmental conditions in flight, a failure on one sensor could occur on the other sensors under certain conditions such as extreme weather; this class of failure is known in the literature as "common mode" failure. In this paper, different approaches to the problem of detection, identification and accommodation of failures on the Air Data System (ADS) of an aircraft are evaluated. This task can be divided into component tasks of equal criticality as Sensor Failure Detection and Identification (SFDI) and Sensor Failure Accommodation (SFA). Data from flight test experiments conducted using the WVU YF-22 unmanned research aircraft are used. Analytical redundancy is provided through a least squares modeling based approach and an extended Kalman filter approach to handle the Sensor Failure Accommodation (SFA) task. From experiments, it is seen that both these approaches provide reasonable estimates of airspeed with an average estimation error of 0.533 m/s and standard deviation of 1.6446 m/s. Furthermore two approaches to the task of Sensor Failure Detection and Identification (SFDI) based on different fault detection filters were evaluated. A Cumulative Sum (CUSUM) detector and the Generalized Likelihood Ratio Test (GLRT) detector were evaluated for different failure conditions - a sudden step bias, a fast rising fault and a slow rising fault in the measured airspeed and compared in terms of sensitivity to the failure magnitude, detection delay, false alarms and undetected faults. It was determined that on an average, the CUSUM filter performed slightly better in terms of detecting failures than the GLRT based detection for the given set of data.