Master of Science in Electrical Engineering (MSEE)


Electrical and Computer Engineering

Document Type



This work studies a blind fault detection method, which only analyses a system's output signal for any change in the characteristics from pre-fault to post-fault to identify the occurrence of faults. In our case the fault considered to develop the procedure is change in time constant of an aircraft's aileron-actuator system and its simplified version - a position servo system. The method is studied as an alternative to conventional fault detection and identification methods. The output signal is passed through a filter bank to enhance the effect of a fault. The Short time Fourier transform is performed on the enhanced pre-fault and post-fault signals components to obtain indicators. Fault detection is approached as a clustering problem determining distances to fault signatures. This work presents two techniques to create signatures from the indicators. In the first method, the mean of the indicators is the signature. Tests on a position servo system show that the method effectively classifies the indicators by more than 85 % and can be used for online classification. A second method uses Principal Component Analysis and defines vector sub-space signatures. It is observed that for the position servo system, the pre-fault indicators had 14 % of false alarms and post-fault indicators the missed the faults by 17%. This second method was also applied to one axis model of an F-14 aircraft's aileron-actuator system. The results obtained showed around 80 % of correctly identified pre-fault indicators and post-fault indicators. The blind fault detection method studies has potential but needs to be understood further by applying it to more varied cases of faults and systems.



Document Availability at the Time of Submission

Release the entire work immediately for access worldwide.

Committee Chair

Jorge Aravena