Anomaly Detection on Time Series with Wasserstein GAN applied to PHM

Modern vehicles are more and more connected. For instance, in the aerospace industry, newer aircraft are already equipped with data concentrators and enough wireless connectivity to transmit sensor data collected during the whole flight to the ground, usually when the airplane is at the gate. Moreover, platforms that were not designed with such capability can be retrofitted to install devices that enable wireless data collection, as is done on Airbus A320 family. For military and heavy helicopters, HUMS (Health and Usage Monitoring System) also allows the collection of sensor data. Finally, satellites send continuously to the ground sensor data, called telemetries. Most of the time, fortunately, the platforms behave normally, faults and failures are thus rare. In order to go beyond corrective or preventive maintenance, and anticipate future faults and failures, we have to look for any drift, any change, in systems’ behavior, in data that is normal almost all the time. Moreover, collected sensor data is time series data. The problem is then anomaly detection or novelty detection in time series data. Among machine learning techniques that can be used to analyze data, Deep Learning, especially Convolutional Neural Networks, is very popular since it has surpassed human capacities for image classification and object detection. In this field, Generative Adversarial Networks are a technique to generate data similar to a potentially high dimension original dataset. In our case, generate new data could be useful to enrich the learning dataset with generated abnormal data to make it less unbalanced. Yet we are more interested in the potential of such techniques to perform anomaly detection for high dimensional Mélanie Ducoffe et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 United States License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. data, comparing newly observed data with data that could have been generated from a distribution built from normal examples.