Hardware-in-the-loop simulation for quadrotor fault diagnosis enhancing airworthiness using OS-Fuzzy-ELM

With the rising adoption of multi-rotor UAVs, it has become ever more crucial that their airworthiness is ensured, especially for hobby-grade UAVs. For some UAVs, their onboard components might not be as reliable as required and as such, faults can occur during flight operations and may develop to c...

Full description

Saved in:
Bibliographic Details
Main Authors: Thanaraj, T., Sidharth, Sai, Ng, Bing Feng, Low, Kin Huat
Other Authors: School of Mechanical and Aerospace Engineering
Format: Conference or Workshop Item
Language:English
Published: 2022
Subjects:
Online Access:https://hdl.handle.net/10356/160590
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:With the rising adoption of multi-rotor UAVs, it has become ever more crucial that their airworthiness is ensured, especially for hobby-grade UAVs. For some UAVs, their onboard components might not be as reliable as required and as such, faults can occur during flight operations and may develop to cause catastrophe. Hence, these faults need to be accurately diagnosed and quickly mitigated. This paper presents a fault diagnosis model for a quadrotor subjected to partial actuator faults. Flight simulations, with actuator and GPS sensor fault injections, are performed on a hardware-in-the-loop experimental setup to gather flight data consisting of multiple sensors. Based on this data, a preliminary controllability threshold analysis is conducted for the quadrotor. After that, a fault diagnosis model using an online sequential fuzzy-extreme learning machine (OS-Fuzzy-ELM) is trained to locate the actuator faults on the quadrotor UAV. The trained model presents an average testing accuracy and macro-averaged F1 score of 80.2% and 78.0%. A subsequent study to isolate sensor and actuator faults presents the testing accuracy and macro-averaged F1 score to be 1.95% and 1.21%, marginally better than a fault diagnosis model based on a single-layer feedforward network