At Inter-Noise 2021, the most important conference for noise control and vibration technology, Fraunhofer IDMT will present two AI-based approaches for the assessment of acoustic measurement data - one from the field of acoustic traffic monitoring, the other for the acoustic monitoring of partial discharges. The 50th edition of the international congress will be offered as an E-Congress and will be extended by an additional event day.

 

Partial discharge monitoring using deep neural networks with acoustic emission
Gourishetti, Saichand; Johnson, David; Werner, Sara; Kátai, András; Holstein, Peter (SONOTEC GmbH)

The occurrence of partial discharge (PD) indicates failures in electrical equipment. Depending on the equipment and operating conditions, each type of PD has its own acoustic characteristics and a wide frequency spectrum.  To detect PD, electrical equipment is often monitored using various sensors, such as microphones, ultrasonic, and TEV, whose signals then are analyzed manually by experts using signal processing techniques. This process requires significant expertise and time, both of which are costly. Advancements in machine learning, aim to address this issue by automatically learning a representation of the signal, minimizing the need for expert analysis.  To this end, we propose a deep learning-based solution for the automatic detection of PD using airborne sound emission in the audible to the ultrasonic range. As input to our proposed model, we evaluate common time-frequency representations of the acoustic signal, such as STFT, CWT and Mel spectrograms. The extracted feature representations from the PD signal is used to train and evaluate the proposed neural networks for the detection of different types of PDs.  Our results show that classification of PDs using airborne sound in the audible and ultrasonic range is possible with deep learning techniques and is a promising non-invasive automated detection approach. Keywords: partial discharge, acoustic event detection, neural networks.

Session: 20.01 Artificial Intelligence for Noise and Vibration Control, Part 2
Tuesday, August 3, 2021, 6:40 AM - 7:00 AM

 

 

Investigating the influence of microphone mismatch for acoustic traffic monitoring
Gourishetti, Saichand; Abeßer, Jakob; Grollmisch, Sascha; Kátai, András; Liebetrau, Judith

The development of robust acoustic traffic monitoring (ATM) algorithms based on machine learning faces several challenges.  The biggest challenge is to collect and annotate a suitable data set for model training and evaluation, which must reflect a broad variety of vehicle sounds since their emitted acoustic noise patterns depend on a variety of factors such as engine noises at different speeds and road conditions. Additionally, the characteristics of the employed microphones have strong influence on the data.  If microphones with different directionality and frequency response are used during the model development and the final deployment phase, a data mismatch is caused which can have a deteriorating effect on the performance of machine learning algorithms. In this paper, the influence of mismatched recording conditions is investigated for deep learning-based ATM algorithms. To evaluate these effects, we implement state-of-the-art convolutional neural networks to detect passing vehicles, classify their type, and estimate their speed and direction of movement. The evaluated models show good performance on low, high, and mixed quality recordings at different vehicle locations when the same or mixed recordings are used for training and testing. Several normalization strategies are evaluated for the microphone mismatch scenarios including a transfer learning approach using OpenL3 embeddings.  Our results show that the vehicle detection task under mismatched conditions is best solved using Mel-spectrogram features combined with data normalization.  On the other hand, the vehicle speed and direction of movement detection tasks, which rely on cross-correlation features, are less sensitive to domain shift.


Session: 20.01 Artificial Intelligence for Noise and Vibration Control, Part 3
Tuesday, August 3, 2021, 11:20 AM - 11:40 AM