This data set is released with the publication REAL-WORLD ON-BOARD UAV AUDIO DATA SET FOR PROPELLER ANOMALIES, which is accepted at ICASSP 2022.

Propeller are an important subsystem of a UAV (also known as a drone). Audio data can be used to detect propeller damage, which improves the resilience and safety of the UAV system.

The audio data set consists of audio recordings of in-door missions, which were captured from take-off to landing. The number of broken propellers is varied from 0 to 4, inclusive. The data set contains samples of each propeller configuration.

The audio data is divided into 4 classes - Normal, 1-Broken, 2-Broken, 3_4-Broken Propellers. Each class directory contains sub-directory for each configuration. Each configuration directory is divided into 4 different mission types: square or figure eight (f8), clockwise (cw) or counter-clockwise (ccw). Each audio recording that belongs to a specific mission type is stored within these directories.

The configuration directories denote the positions of broken (1) and intact (0) propellers. The propeller position were labelled 1,2,3,4, and for example a configuration 0011 would have intact propellers in positions 1,2 and broken propellers in 3 and 4, making this configuration a member of 2-broken class.