This study presents a novel diagnostic methodology for assessing drive system damage and its propagation in an unmanned aerial vehicle (UAV) using piezoelectric sensors mounted on each arm of the drone. In contrast to existing studies that focus solely on fault localization, this work investigates the spatial propagation of structural responses to localized motor faults under varying operating conditions. By varying the PWM control signal duty cycle on one motor, different degrees of damage (from 20% to 80%) were simulated. Voltage signals were recorded on each arm of the drone to identify damage and to optimize the number and placement of the sensors. Statistical features extracted in both the time and frequency domains were calculated within sliding time windows. These features (e.g., mean, variance, spectral skewness, spectral kurtosis) from voltage time-series were used as input data for machine learning models (e.g., Random Forest and K-Nearest Neighbors), which are widely applied in the diagnostics of rotary systems for binary classification problems (distinguishing between intact and damaged states of varying damage level). The highest classification accuracy was achieved for the arm where the electric motor failure was induced (from 93% to 94% depending on the degree of damage), while the lowest accuracy was obtained for the opposite arm (from 50% to 57% depending on the degree of damage). It was found that diagnostic accuracy increases when frequency-domain features of the signals are used, particularly for the opposite arms. The proposed methodology provides valuable insights into the structural behavior of the drone in both ground and flight conditions, illustrating the propagation of local damage to other components. The results contribute to the development of robust diagnostic techniques for health monitoring and structural reliability assessment of unmanned aerial vehicles (UAVs).