Abstract 

The Partial Discharge - Localisation Dataset, abbreviated: PD-Loc Dataset is an extensive collection of acoustic data specifically curated for the advancement of Partial Discharge (PD) localisation techniques within electrical machinery. Developed using a precision-engineered 32-sensor acoustic array, this dataset encompasses a wide array of signals, including chirps, white Gaussian noise, and PD signals. The data, meticulously recorded and annotated, provides a valuable resource for researchers and practitioners seeking to refine diagnostic algorithms, enhance acoustic localisation accuracy, and contribute to the reliability and maintenance protocols for electric machines and other critical electrical infrastructure. The PD-Loc Dataset invites the research community to explore innovative approaches to condition monitoring and fault diagnosis through acoustic analysis. 

Each audio file, sampled at 96 kHz, has a duration of 50 seconds and consists of a sequence where each sound type is played for 10 seconds and repeated five times. 

 

Instructions: 

A consistent naming convention was adopted across both sensor approaches, with the only variation being the sensor type (Condenser or MEMS) indicated in the header of the file name.

For ECM recordings, the format was:

MAX9814_Cx1_L_x2_x3 - Chx4.WAV

whereas for MEMS, it was:

ICS40180_Cx1_L_x2_x3 - Chx4.WAV

Here, 'x1' denotes the configuration number (1 to 9), 'x2' represents the speaker location number (1 to 90, accounting for all fault locations across 18 stator coils), 'x3' designates the audio file type (CH for cosine chirp, WGN for white Gaussian noise, and PD for partial discharge sounds), and 'x4' corresponds to the channel number (1 to 32).

Funding Agency: 
Irish Research Council
Grant Number: 
GOIPG/2021/1744