UAV Propeller: Fault Detection, Characterization, and Calibration: A Comprehensive Study

The growing use of Unmanned Aerial Vehicles (UAVs) demands higher reliability and safety. Propeller faults are a major risk for small UAVs, yet many studies do not calibrate severity or test early faults that preserve nominal flight. We investigate three propeller fault groups (edge cuts, cracks, surface unbalance), each introduced at three calibrated severity levels as small perturbations during normal flight without controller retuning. Severity levels are referenced to ISO 21940 balance-quality limits and quantified by measured residual unbalance and effective damaged area. We propose a physics-based residual detector that identifies a simple inner-loop model directly from flight dynamics data. A single residual state is tracked, with decisions based on a percentile threshold and a latency gate tuned on healthy data. Validation over 12 indoor flights (9 faulty and 3 healthy) shows perfect case-level precision and recall (no false alarms), while interval-level precision is 1.0 and recall is 0.48. Using two residual features, maximum margin and signature duration, the three fault groups occupy distinct regions of the feature space, showing clear separation for fault classification. A one-way ANOVA across fault types shows significant mean differences, and post-hoc tests confirm pairwise separations, indicating that the residual features provide discriminative power for fault type classification. Severity analysis reveals surface unbalance scales linearly with measured unbalance, while edge cuts and cracks show minimal variation due to aerodynamic stabilizing torques and persistent stiffness asymmetry. The study provides severity-calibrated evidence, a physics-based lightweight detector, and a consistent protocol for early propeller-fault assessment in real flight.