Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

Sounding Industry: Challenges and Datasets for Industrial Sound Analysis

 
: Grollmisch, Sascha; Abeßer, Jakob; Liebetrau, Judith; Lukashevich, Hanna

:

Bugallo, Mónica F. (General Chair) ; Institute of Electrical and Electronics Engineers -IEEE-; European Association for Signal Processing -EURASIP-:
27th European Signal Processing Conference, EUSIPCO 2019 : A Coruña, Spain, September 2-6, 2019
Piscataway, NJ: IEEE, 2019
ISBN: 978-9-0827-9703-9
ISBN: 978-90-827970-2-2
ISBN: 978-1-5386-7300-3
5 S.
European Signal Processing Conference (EUSIPCO) <27, 2019, A Coruña/Spain>
Englisch
Konferenzbeitrag
Fraunhofer IDMT ()
acoustic detection systems; acoustic quality control; acoustic signal detection; acoustic signal processing; advanced quality control systems; airborne sounds struggle; audio; audio signal processing; audio-based analysis; datasets; deep learning; experienced machine operators; factory automation; freely available datasets; highly complex noise scenarios; industrial sound analysis; ISA; learning (artificial intelligence); machine conditions; machine learning; machine learning systems; Neural Nets; neural network based baseline systems; neural networks; production engineering computing; production lines; quality control; robust quality control; signal classification; signal processing; sounding industry

Abstract
The ongoing process of automation in production lines increases the requirements for robust and reliable quality control. Acoustic quality control can play a major part in advanced quality control systems since several types of faults such as changes in machine conditions can be heard by experienced machine operators but can hardly be detected otherwise. To this day, acoustic detection systems using airborne sounds struggle due to the highly complex noise scenarios inside factories. Machine learning systems are theoretically able to cope with these conditions. However, recent advancements in the field of Industrial Sound Analysis (ISA) are sparse compared to related research fields like Music Information Retrieval (MIR) or Acoustic Event Detection (AED). One main reason is the lack of freely available datasets since most of the data is very sensitive for companies. Therefore, three novel datasets for ISA with different application fields were recorded and published along with this paper: detection of the operational state of an electric engine, detection of the surface of rolling metal balls, and detection of different bulk materials. For each dataset, neural network based baseline systems were evaluated. The results show that such systems obtain high classification accuracies over all datasets in many of the subtasks which demonstrates the feasibility of audio-based analysis of industrial analysis scenarios. However, the baseline systems remain highly sensitive to changes in the recording setup, which leaves a lot of room for improvement. The main goal of this paper is to stimulate further research in the field of ISA.

: http://publica.fraunhofer.de/dokumente/N-575212.html