Multicopter Motor Damage Diagnosis Via Functionally Pooled Time Series Models: Experimental Assessment Via a Series of Flight Tests

This study presents a unified statistical framework for detecting, identifying, and quantifying damage on multicopter propellers through functionally pooled (FP) time series models. Unlike conventional autoregressive models, FP models feature parameters that are explicit functions of damage magnitude, enabling parsimonious representation of damage dynamics evolution while accounting for cross-correlation between different damage states. The methodology integrates three capabilities within a single framework: health detection, motor identification, and precise damage quantification with statistical confidence intervals. Experimental validation is conducted using a hexacopter executing figure-eight flight patterns under turbulent conditions, with blade damage ranging from 2 mm to 10 mm investigated across three different rotating motors. The proposed approach achieves effective damage detection and accurate motor identification using short data segments (4 seconds) from single-channel inertial measurement unit (IMU) measurements during flight. Damage quantification is demonstrated with corresponding confidence intervals, providing statistical reliability assessment. The framework’s effectiveness is evaluated through analysis of 160 s of independent test signals, demonstrating its potential for real-time fault diagnosis in autonomous aerial vehicles.