TY - GEN
T1 - Machinery noise source identification with deep learning
AU - Antila, Marko
AU - Rantala, Seppo
AU - Kataja, Jari
AU - Lamula, Lasse
AU - Isomoisio, Heikki
AU - Zimroz, Radoslaw
AU - Wodecki, Jacek
AU - Wylomanska, Agnieszka
N1 - Funding Information:
The work was carried out in RockVader - Smart Hard Rock Mining System project (project number 16136) which has received funding from European Union’s EIT RawMaterials initiative. Special acknowledgements to Sandvik Austria for arrangements of the measurement possibilities and coordination of the RockVader project.
PY - 2019
Y1 - 2019
N2 - Machinery noise is produced by several noise sources. Our study focuses on hard rock mining machinery in harsh conditions. In such machinery, there are several independent and inter-dependent noise sources. They include cutting, airflow, drilling, and auxiliary devices such as hydraulics. In this study, we isolate the most annoying or harmful noise sources. This is done by automatic ranking of the noise sources. For the automatic ranking of the noise sources, we have tried deep learning, independent component analysis, and principal component analysis. Deep learning has produced the best results. It is used widely in detection of noise sources, but it has not been commonly applied to machinery noise source detection. The results show the possibility to separate the noise sources, if adequately long datasets without excessive random and impulsive components are available. A priori information about the individual noise sources improves the capabilities of the deep learning method, as well. The challenge is to use real-world datasets with slightly corrupted contents successfully and with adequate accuracy to improve the machinery noise properties.
AB - Machinery noise is produced by several noise sources. Our study focuses on hard rock mining machinery in harsh conditions. In such machinery, there are several independent and inter-dependent noise sources. They include cutting, airflow, drilling, and auxiliary devices such as hydraulics. In this study, we isolate the most annoying or harmful noise sources. This is done by automatic ranking of the noise sources. For the automatic ranking of the noise sources, we have tried deep learning, independent component analysis, and principal component analysis. Deep learning has produced the best results. It is used widely in detection of noise sources, but it has not been commonly applied to machinery noise source detection. The results show the possibility to separate the noise sources, if adequately long datasets without excessive random and impulsive components are available. A priori information about the individual noise sources improves the capabilities of the deep learning method, as well. The challenge is to use real-world datasets with slightly corrupted contents successfully and with adequate accuracy to improve the machinery noise properties.
KW - Deep learning
KW - Noise ranking
KW - Noise sources
UR - http://www.scopus.com/inward/record.url?scp=85084160414&partnerID=8YFLogxK
M3 - Conference article in proceedings
AN - SCOPUS:85084160414
T3 - NOISE-CON Proceedings
BT - InterNoise 19, Madrid, Spain
A2 - Calvo-Manzano, Antonio
A2 - Delgado, Ana
A2 - Perez-Lopez, Antonio
A2 - Santiago, Jose Salvador
PB - Institute of Noise Control Engineering
T2 - 48th International Congress and Exhibition on Noise Control Engineering, INTER-NOISE 2019 MADRID
Y2 - 16 June 2019 through 19 June 2019
ER -