Physics-Informed Simulation for Aircraft Fuel Pump Fault Diagnosis

A team of researchers has created a high-fidelity co-simulation of a typical aircraft main fuel pump system utilizing MATLAB/Simulink Simscape Fluids. This simulation produces time-series data annotated with health and fault modes, addressing the shortage of training data necessary for anomaly detection and diagnostic algorithms in essential cyber-physical systems like airplanes, where data protection and limited observability hinder data access. To validate the benchmark's effectiveness, they employed an unsupervised Recurrent Variational Autoencoder (RNN-VAE) for detecting anomalies and a SOM-VAE for discretizing operating modes, trained to differentiate between normal and faulty states. This research is available on arXiv in the computer science and machine learning sections.