A Weighted-Transfer Domain-Adaptation Network Applied to Unmanned Aerial Vehicle Fault Diagnosis

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Data Science

Background:

• Unmanned Aerial Vehicle (UAV) systems face complex fault diagnosis challenges due to intricate designs and diverse operational conditions.
• Traditional deep learning methods struggle with anomaly detection in UAVs because of data distribution variations and the difficulty of labeling fault data.
• Weak, nonlinear, and uncertain fault features in UAVs necessitate advanced diagnostic techniques for improved reliability.

Purpose of the Study:

• To introduce a novel weighted-transfer domain-adaptation network (WTDAN) for online anomaly detection and fault diagnosis of UAV electromagnetic-sensitive flight data.
• To address the challenges of data distribution imbalance and limited labeled fault data in UAV systems.
• To enhance the accuracy and transferability of fault diagnosis models across different UAV flight conditions.

Main Methods:

• Developed a weighted-transfer domain-adaptation network (WTDAN) employing unsupervised transfer learning.
• Integrated three novel multiscale modules: a feature extractor, a domain discriminator, and a label classifier.
• Utilized multilayer domain adaptation and weighted source domain samples to mitigate data distribution discrepancies.

Main Results:

• Achieved up to 90% classification accuracy in anomaly detection, even with scarce anomalous target data.
• Demonstrated superior transferability of the WTDAN model across cross-domain datasets compared to existing methods.
• The WTDAN method effectively extracts diagnostic information from weak, coupled, and noisy UAV flight data.

Conclusions:

• The WTDAN offers a robust solution for online anomaly detection and fault diagnosis in UAVs.
• This approach significantly improves the performance of fault diagnosis under data scarcity and distribution shifts.
• The proposed method enhances the prognostics and health management (PHM) of UAVs, contributing to increased system reliability, repairability, and safety.