Fault Diagnosis of Aircraft Power Systems Based on Transients and Artificial Intelligence

Aerospace power systems are largely equipped with electrical circuits and electronic subsystems and components. Faults occurring within aircraft or spacecraft electrical systems may potentially put at risk the safety of the flight. This work presents a new approach based on transients caused by electrical failures and deep learning architectures to identify and diagnose faults of different components and subsystems within a conventional aircraft power system. The electrical components and subsystems to be diagnosed include: the power generation unit, primary and secondary electric distribution subsystems, transformer and rectifier units (TRU), the busbars that provide power to the bank of DC/AC loads that include motors which serve as actuators and resistive loads such as lamps or light bulbs. Two deep learning architectures were investigated: one based on LSTM neural networks and the other one based on one-dimensional convolutional neural networks (1D-CNN). Classification accuracies of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 6. 0 9 \%}$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 5. 5 4 \%}$</tex> were obtained with LSTM neural networks for the training and testing sets, respectively.