Structure analysis in an octocopter using piezoelectric sensors and machine learning.

Area of Science:

• Engineering
• Robotics
• Structural Health Monitoring

Background:

• Existing unmanned aerial vehicle (UAV) diagnostics often focus on fault localization.
• There is a need to understand how localized motor faults propagate structurally within a UAV.

Purpose of the Study:

• To develop and validate a novel diagnostic methodology for assessing drive system damage and its spatial propagation in UAVs.
• To investigate the effectiveness of piezoelectric sensors and machine learning for UAV health monitoring.

Main Methods:

• Simulated motor damage (20%-80%) by varying PWM duty cycle on individual drone arms.
• Recorded voltage signals from piezoelectric sensors on each arm.
• Extracted time and frequency domain statistical features from sensor data.
• Utilized machine learning models (Random Forest, KNN) for damage classification.

Main Results:

• Achieved high classification accuracy (93%-94%) for the damaged motor arm.
• Identified lower accuracy (50%-57%) for the opposite arm, indicating damage propagation effects.
• Found that frequency-domain features improved diagnostic accuracy, especially for non-adjacent arms.

Conclusions:

• The proposed methodology effectively detects drive system damage and its propagation in UAVs.
• Frequency-domain analysis enhances diagnostic capabilities for distributed damage.
• This research contributes to robust health monitoring and structural reliability assessments for UAVs.