Abstract

The main objective of this thesis is to develop a fault detection, isolation and identification (FDII) scheme based on Wavelet Entropy (WE) and Artificial Neural Network (ANN) for reaction wheels (RW) that are employed as actuators in the attitude control subsystem (ACS) of a satellites to perform the formation flying (FF) missions.
In this thesis two FDII approaches are proposed, i) Spacecraft-level fault diagnosis and ii) Formation-level fault diagnosis.

In the "spacecraft-level" fault diagnosis scheme in order to analysis faults, absolute attitude and angular measurements from a satellite are considered as diagnostic signals. In order to detect the fault, the wavelet-entropy technique is employed on diagnostic signals and the sum of the absolute wavelet entropies (SAWE) of the diagnostic signals are obtained and compared with an appropriately selected threshold. If the SAWE passes the threshold the faulty condition is established. In order to isolate the fault in a satellite the angular velocity measurements in a satellite are considered as diagnostic signals and the relative wavelet energy (RWE) of these signals are obtained and compared to a threshold. In our proposed fault identification scheme, the attitude measurements in a satellite are considered and the detail and approximation coefficients of the wavelet signals are obtained and these coefficients are used as inputs to an artificial neural network to identify the type of the fault in a satellite.
Using a confusion matrix evaluation system we demonstrate that our spacecraft-level FDII can detect, isolate and identify the high severity faults in a satellite however this scheme cannot detect low severity faults in a satellite.

Our proposed "formation-level" FDII scheme utilizes data collected from the relative attitudes and relative angular velocity measurements of the formation flying satellites. In this fault diagnosis scheme, the relative attitude and relative angular velocity measurements in a satellite with respect to each its neighbor's in a formation are considered as diagnostic signals. In order to detect the fault, the relative attitude measurements in a satellite are considered as diagnostic signals. The wavelet-entropy technique is utilized on diagnostic signals and the SAWEs with respect to each satellite's neighbor are obtained. These SAWEs are then compared with an appropriately selected threshold. The faulty satellite is determined if these SAWEs pass the thresholds. In order to isolate the fault in a faulty satellite, the relative angular velocity measurements are considered as diagnostic signals. The RWE of these signals are obtained and compared to a threshold. In our proposed fault identification scheme, the relative attitude measurements in a satellite are considered as diagnostic signals. In this scheme, the RWEs of the diagnostic signals are obtained and used as inputs to an artificial neural network to identify the type of the fault in a satellite.
According to the simulation results, our proposed FDII scheme can detect, isolate and identified both low severity and high severity faults in the reaction wheels of satellite.