Anomaly Detection for Electro-Hydrostatic Actuators using LSTM Autoencoder

Researchers have developed two separate anomaly detection frameworks, one for Electro-Hydrostatic Actuators (EHAs) using a Long Short-Term Memory (LSTM) autoencoder and another for O-RAN networks using a semi-supervised deep contractive autoencoder called XAInomaly.

The EHA anomaly detection framework, reported in a study on arXiv^[1], achieved an average accuracy of 99.0% in detecting anomalies in temperature and pressure data from a controlled test bench. The LSTM autoencoder demonstrated high detection sensitivity and a low false-alarm rate across multiple fault-injection scenarios. EHAs are widely used in aerospace and industrial systems, where timely detection of sensor anomalies is crucial for safe operation. Conventional methods often fail to capture temporal dependencies in EHA signals, but the LSTM autoencoder effectively addressed this challenge. Future work will focus on adapting this framework for real-time environments. Meanwhile, a separate study on arXiv^[2] introduced XAInomaly, a framework for anomaly detection in O-RAN networks. XAInomaly is an explainable and interpretable deep contractive autoencoder that leverages generative modeling to learn normal network behavior and identify deviations indicative of anomalies. The XAInomaly framework also addresses the black-box nature of deep learning models using a reactive explainable AI technique called fastshap-C. The submission of the XAInomaly framework was recorded on 13 Feb 2025^[2].