E-Congress  /  July 11, 2021  -  July 16, 2021

27th International Congress on Sound and Vibration

At the 27th Congress on Sound and Vibration Fraunhofer IDMT presents an AI-based approach for acoustic monitoring of electrical discharges. 


Acoustics of electrical discharges
Holstein, Peter (SONOTEC GmbH); Gourishetti, Saichand (Fraunhofer IDMT); Fuchs, Karsten (TU Ilmenau); Berger, Frank (TU Ilmenau);  Vyas, Daxeshkumar Harikeshbhai (TU Ilmenau/SONOTEC GmbH)

Partial discharges cause noise. Corona, arcing and tracking are typical discharge effects. Commonly, they are accompanied by characteristic acoustic emissions. This noise can be used in maintenance for monitoring the operation and/or safety of electrical equipment. The objective of acoustic condition monitoring is to detect the presence of faulty events and states. Discharges can be measured either by the acoustic transfer paths in a structure or by transmission via an air distance. In the paper, we concentrate on the air-transmitted detection of electrical caused faults. Commonly, there is a remarkable “background” noise originating by processes in the industrial environments. Consequently, the use of ultrasound can be helpful to “filter” the discharge acoustics from operational noise. The acoustic emission of the electrical source exhibits broad-band frequency behavior with a fast stochastic modulation of the amplitude in time. Traditionally, ultrasound is measured in a narrow frequency band (around 40 kHz). The signal is shifted down to the audible range by amplitude modulation (heterodyne technology by mixing with a carrier frequency) for listening and data processing. In the paper, a broad-band approach is proposed. The discharge noise is measured within a frequency range up to 100 kHz. The signal will be processed as a complete time-frequency pattern. Different signal processing techniques are combined preserving the complete frequency characteristics and information of the time fluctuation. FFT-/IFFT-, Wavelet-and correlation-procedures are used for the input in machine learning algorithms in order to provide an objective (and possibly automated) evaluation of electrically caused acoustic sources. A series of experiments have been carried out with some systematic variations of the electrical discharge situations.