Konferensartikel

Developed by Elektra Solar, an official spin-off of the DLR Institute for Robotics and Mechatronics (DLR-RMC), the Elektra 2 is a solar-powered autonomous aircraft designed to endure long distances as well as high altitudes. The main motivation of this work is to develop a sensor fault detection and reconfiguration (FDR) approach to be applied to the Elektra 2 Aircraft. Currently, the Elektra 2 Solar aircraft provides a simple limit-checking of certain measurements such as aircraft angular velocities and pitch, roll and yaw angles. The most frequently method used to monitor a system is to evaluate the residuals against a predefined decision threshold. In this approach, the major drawback is due to the presence of noise, input variations and change of operating point which can cause an increasing number of false and missed alarms. Besides that, such methods also take into account additional expertise from the dynamic system and require massive experimentation in order to set suitable thresholds. In addition to that, residuals present a dynamic behavior that might result in a set of undesired false alarms during its transient response. In the proposed FDR approach, residual generation for both longitudinal and lateral dynamics of the aircraft is achieved based on the design of Kalman filters. The sequential probability ratio test (SPRT) is then used as a decision function to be evaluated with adaptive thresholds (ATLMS) for each of the aircraft measurements. Due to the critical aspects of IMU faults, the reconfiguration action is defined as switching from primary IMU to backup IMU in case of fault occurrence. Therefore, a hot standby reconfiguration scheme is used for sensor fault tolerant purposes. The ATLMS technique allows the threshold to be tuned by changing well known parameters independently of the application case and and the requirements to the dynamic behavior of such adaptive threshold are: (i) low sensitivity to control signal variation; (ii) low sensitivity to noise; (iii) high sensitivity to faulty residuals. For practical purposes, some fixed threshold approaches neglect the transient response of the residuals. This drawback could be also minimized with the use of the ATLMS. The simulation methodology consisted of eight flight scenarios with different additive faults (abrupt, incipient, extra noise) applied to roll and pitch angles and rates. Next steps include (i) integration of proposed fault detection and diagnosis approach to the Elektra 2 Solar aircraft and (ii) validation of proposed approach with real flight data.

sensor fault detection, fault diagnosis, autonomous aircraft systems