Fault Detection and Identification Method for Quadcopter Based on Airframe Vibration Signals

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Quadcopter safety relies on real-time fault detection, but limited airframe space restricts the use of numerous sensors.
• Existing data-driven fault detection methods often require extensive sensor data, which is challenging for sensor-limited quadcopters.

Purpose of the Study:

• To propose a novel Fault Detection and Identification (FDI) method for quadcopter blades using only onboard acceleration sensors.
• To leverage airframe vibration signals and Long and Short-Term Memory (LSTM) networks for effective quadcopter blade fault diagnosis.

Main Methods:

• Collected triaxial accelerometer data from quadcopter flight experiments to capture airframe vibration signals.
• Employed wavelet packet decomposition to extract features from vibration data, forming feature vectors using standard deviations of coefficients.
• Developed and implemented a Long and Short-Term Memory (LSTM) network model for fault detection and identification (FDI).

Main Results:

• The proposed LSTM-based FDI method demonstrated effective detection and identification of quadcopter blade faults.
• Achieved superior FDI performance and higher model accuracy compared to traditional Back Propagation (BP) neural network models.
• Validated the method's efficacy through practical flight experiments.

Conclusions:

• The developed FDI method successfully identifies quadcopter blade faults using only airborne acceleration sensor data and LSTM networks.
• This sensor-efficient approach enhances quadcopter safety and reliability by enabling real-time fault diagnosis.
• The study highlights the potential of AI-driven vibration analysis for critical component monitoring in unmanned aerial vehicles.