Vibration Anomaly Detection by Clustering in Unmanned Aerial Vehicles.

One of the critical factors affecting flight safety of unmanned aerial vehicles (UAVs) is the
amount of vibration they are exposed to during a flight. On one hand, external causes such as
wind gusts and turbulences or internal vehicle-centric faults such as incorrect sensor mounting
or propeller imbalances can cause high vibrations in UAVs. On the other hand, high vibration
itself may induce noise in the onboard miniature sensors of the UAV such as its accelerometers,
gyroscopes and GPS that can lead to uncertain state estimation causing the multi-rotor to drift
from its desired position or even result in loss-of-control. Hence, it is important to monitor the
vibration levels during a UAV flight. This paper specifically looks into vibrations recorded by
the autopilot system of a multi-rotor in presence of varying magnitudes of wind. Using data
from experimental flights conducted at two separate flight test regions under varying wind
conditions, we aim to classify between a nominal and anomalous vibration level for small UAV
systems. Further, we analyse other parameters of interest that affect vibrations in UAVs such as
UAV air speed and any propeller imbalance signatures. Analysis results from experimental
flights demonstrate the effect of wind on vibration magnitude in unmanned aircrafts.