Unsupervised Fault Detection on Unmanned Aerial Vehicles: Encoding and Thresholding Approach

Abstract

Unmanned Aerial Vehicles are expected to create enormous benefits to society, but there are safety concerns in recognizing faults at the vehicle’s control component. Prior studies proposed various fault detection approaches leveraging heuristics-based rules and supervised learning-based models, but there were several drawbacks. The rule-based approaches required an engineer to update the rules on every type of fault, and the supervised learning-based approaches necessitated the acquisition of a finely-labeled training dataset. Moreover, both prior approaches commonly include a limit that the detection model can identify the trained type of faults only, but fail to recognize the unseen type of faults. In pursuit of resolving the aforementioned drawbacks, we proposed a fault detection model utilizing a stacked autoencoder that lies under unsupervised learning. The autoencoder was trained with data from safe UAV states, and its reconstruction loss was examined to distinguish the safe states and faulty states. The key contributions of our study are, as follows. First, we presented a series of analyses to extract essential features from raw UAV flight logs. Second, we designed a fault detection model consisting of the stacked autoencoder and the classifier. Lastly, we validated our approach’s fault detection performance with two datasets consisting of different types of UAV faults.

UAV Fault Detection Methods, State-of-the-Art

“…Park et al in [118] perform fault detection analysis using a stacked autoencoder. Using an unsupervised deep neural network, they perform a binary classification for different types of faults of UAV.…”

A Machine Learning Approach for Detecting GPS Location Spoofing Attacks in Autonomous Vehicles

“…To this end, some works have explored unsupervised ML methods and have approached the intended task through anomaly detection. In [17], an autoencoder, which is a special type of NN, has been proposed for fault detection on UAVs. Five categories of features are used, including internal measurements, location, position, orientation, system status, and control.…”

Enhancing the Security of Unmanned Aerial Systems using Digital-Twin Technology and Intrusion Detection

“…It generally requires deep knowledge and system understanding, and therefore the production of many specialist models for all sub-systems can be impractical [5]. Data-driven techniques like machine learning provide suitable alternatives, as shown by [6] - [12], who perform UAV-based anomaly detection. These techniques will be used within the proposed digital-twin architecture.…”

“…The open-source UAV attack dataset is used [31], which contains a mixture of normal flights and GPS-spoofed flights for six UAV platforms. A similar process is followed relative to [11] and [12], who also perform novelty detection on this dataset. However, improvements are made, including more generalisable models across all datasets, and enhanced interpretability of predictions.…”

“…This is known as novelty detection and is well-studied for UAVs, with classical-based approaches like One-Class Support Vector Machine (OC-SVM), Isolation Forest (IF) [10] and Local Outlier Factor (LOF) [11] being common. Recently, deep unsupervised learning architectures have emerged, including autoencoders for automatic UAV fault detection [12] and Long-Term Short-Term (LSTM) networks for detecting ADS-B manipulation [7]. These generally show significant performance gains compared to classical techniques, at the expense of increased complexity, computational overhead and data quantities.…”