Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive

Topias Turunen; Jesse Miettinen; Aleksanteri Hämäläinen; Aku Karhinen; Raine Viitala

doi:10.1007/978-3-031-45705-0_88

Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive

Topias Turunen^*, Jesse Miettinen, Aleksanteri Hämäläinen, Aku Karhinen, Raine Viitala

^*Tämän työn vastaava kirjoittaja

Sulzer Pumps Finland Oy

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference article in proceedings › Scientific › vertaisarvioitu

Abstrakti

Centrifugal pumps play a vital role in many industrial processes such as pulp and paper production and wastewater treatment. Condition monitoring has the potential to make pumps more reliable and energy efficient. However, implementing such a monitoring system is costly. This paper presents a novel deep learning application for fault diagnosis of centrifugal pumps using data from a variable frequency drive (VFD). This approach can potentially offer a sensorless pump condition monitoring method. For this study, a new dataset, consisting of two centrifugal pump fault conditions, namely cavitation and abrasive metal-metal contact, was recorded using torque estimate data from a VFD. The dataset was used to train a Convolutional Neural Network with Wide First-Layer Kernels (WDCNN). The trained model achieved a weighted classification accuracy of 78.4% with timeseries samples and 85.5% with corresponding frequency domain samples. The results demonstrate the potential for detecting multiple pump fault states from VFD data.

Alkuperäiskieli	Englanti
Otsikko	Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1
Toimittajat	Masafumi Okada
Kustantaja	Springer
Sivut	905-914
Sivumäärä	10
ISBN (elektroninen)	978-3-031-45705-0
ISBN (painettu)	978-3-031-45704-3
DOI - pysyväislinkit	https://doi.org/10.1007/978-3-031-45705-0_88
Tila	Julkaistu - 2023
OKM-julkaisutyyppi	A4 Artikkeli konferenssijulkaisussa
Tapahtuma	International Federation of Theory of Machines and Mechanisms World Congress - Tokyo, Japani Kesto: 5 marrask. 2023 → 9 marrask. 2023 Konferenssinumero: 16

Julkaisusarja

Nimi	Mechanisms and Machine Science
Vuosikerta	147
ISSN (painettu)	2211-0984
ISSN (elektroninen)	2211-0992

Conference

Conference	International Federation of Theory of Machines and Mechanisms World Congress
Lyhennettä	IFToMM WC
Maa/Alue	Japani
Kaupunki	Tokyo
Ajanjakso	05/11/2023 → 09/11/2023

Pääsy asiakirjaan

10.1007/978-3-031-45705-0_88

Muut tiedostot ja linkit

Linkki julkaisuun Scopuksessa

Siteeraa tätä

Turunen, T., Miettinen, J., Hämäläinen, A., Karhinen, A., & Viitala, R. (2023). Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive. teoksessa M. Okada (Toimittaja), Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1 (Sivut 905-914). (Mechanisms and Machine Science; Vuosikerta 147). Springer. https://doi.org/10.1007/978-3-031-45705-0_88

Turunen, Topias ; Miettinen, Jesse ; Hämäläinen, Aleksanteri et al. / Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive. Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1. Toimittaja / Masafumi Okada. Springer, 2023. Sivut 905-914 (Mechanisms and Machine Science).

@inproceedings{cd755101acf0408c9671906f7c81adda,

title = "Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive",

abstract = "Centrifugal pumps play a vital role in many industrial processes such as pulp and paper production and wastewater treatment. Condition monitoring has the potential to make pumps more reliable and energy efficient. However, implementing such a monitoring system is costly. This paper presents a novel deep learning application for fault diagnosis of centrifugal pumps using data from a variable frequency drive (VFD). This approach can potentially offer a sensorless pump condition monitoring method. For this study, a new dataset, consisting of two centrifugal pump fault conditions, namely cavitation and abrasive metal-metal contact, was recorded using torque estimate data from a VFD. The dataset was used to train a Convolutional Neural Network with Wide First-Layer Kernels (WDCNN). The trained model achieved a weighted classification accuracy of 78.4% with timeseries samples and 85.5% with corresponding frequency domain samples. The results demonstrate the potential for detecting multiple pump fault states from VFD data.",

keywords = "Cavitation, Centrifugal pump, Condition monitoring, Convolutional neural network, Deep learning, Fault diagnosis, Shaft deflection, Variable frequency drive",

author = "Topias Turunen and Jesse Miettinen and Aleksanteri H{\"a}m{\"a}l{\"a}inen and Aku Karhinen and Raine Viitala",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2023.; International Federation of Theory of Machines and Mechanisms World Congress, IFToMM WC ; Conference date: 05-11-2023 Through 09-11-2023",

year = "2023",

doi = "10.1007/978-3-031-45705-0_88",

language = "English",

isbn = "978-3-031-45704-3",

series = "Mechanisms and Machine Science",

publisher = "Springer",

pages = "905--914",

editor = "Masafumi Okada",

booktitle = "Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1",

address = "Germany",

}

Turunen, T, Miettinen, J, Hämäläinen, A, Karhinen, A & Viitala, R 2023, Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive. julkaisussa M Okada (Toimittaja), Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1. Mechanisms and Machine Science, Vuosikerta 147, Springer, Sivut 905-914, International Federation of Theory of Machines and Mechanisms World Congress, Tokyo, Japani, 05/11/2023. https://doi.org/10.1007/978-3-031-45705-0_88

Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive. / Turunen, Topias; Miettinen, Jesse; Hämäläinen, Aleksanteri et al.
Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1. toim. / Masafumi Okada. Springer, 2023. s. 905-914 (Mechanisms and Machine Science; Vuosikerta 147).

Tutkimustuotos: Artikkeli kirjassa/konferenssijulkaisussa › Conference article in proceedings › Scientific › vertaisarvioitu

TY - GEN

T1 - Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive

AU - Turunen, Topias

AU - Miettinen, Jesse

AU - Hämäläinen, Aleksanteri

AU - Karhinen, Aku

AU - Viitala, Raine

N1 - Conference code: 16

PY - 2023

Y1 - 2023

N2 - Centrifugal pumps play a vital role in many industrial processes such as pulp and paper production and wastewater treatment. Condition monitoring has the potential to make pumps more reliable and energy efficient. However, implementing such a monitoring system is costly. This paper presents a novel deep learning application for fault diagnosis of centrifugal pumps using data from a variable frequency drive (VFD). This approach can potentially offer a sensorless pump condition monitoring method. For this study, a new dataset, consisting of two centrifugal pump fault conditions, namely cavitation and abrasive metal-metal contact, was recorded using torque estimate data from a VFD. The dataset was used to train a Convolutional Neural Network with Wide First-Layer Kernels (WDCNN). The trained model achieved a weighted classification accuracy of 78.4% with timeseries samples and 85.5% with corresponding frequency domain samples. The results demonstrate the potential for detecting multiple pump fault states from VFD data.

AB - Centrifugal pumps play a vital role in many industrial processes such as pulp and paper production and wastewater treatment. Condition monitoring has the potential to make pumps more reliable and energy efficient. However, implementing such a monitoring system is costly. This paper presents a novel deep learning application for fault diagnosis of centrifugal pumps using data from a variable frequency drive (VFD). This approach can potentially offer a sensorless pump condition monitoring method. For this study, a new dataset, consisting of two centrifugal pump fault conditions, namely cavitation and abrasive metal-metal contact, was recorded using torque estimate data from a VFD. The dataset was used to train a Convolutional Neural Network with Wide First-Layer Kernels (WDCNN). The trained model achieved a weighted classification accuracy of 78.4% with timeseries samples and 85.5% with corresponding frequency domain samples. The results demonstrate the potential for detecting multiple pump fault states from VFD data.

KW - Cavitation

KW - Centrifugal pump

KW - Condition monitoring

KW - Convolutional neural network

KW - Deep learning

KW - Fault diagnosis

KW - Shaft deflection

KW - Variable frequency drive

UR - http://www.scopus.com/inward/record.url?scp=85177067462&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-45705-0_88

DO - 10.1007/978-3-031-45705-0_88

M3 - Conference article in proceedings

AN - SCOPUS:85177067462

SN - 978-3-031-45704-3

T3 - Mechanisms and Machine Science

SP - 905

EP - 914

BT - Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1

A2 - Okada, Masafumi

PB - Springer

T2 - International Federation of Theory of Machines and Mechanisms World Congress

Y2 - 5 November 2023 through 9 November 2023

ER -

Turunen T, Miettinen J, Hämäläinen A, Karhinen A, Viitala R. Deep Learning for Centrifugal Pump Condition Monitoring Using Data from Variable Frequency Drive. julkaisussa Okada M, toimittaja, Advances in Mechanism and Machine Science - Proceedings of the 16th IFToMM World Congress 2023—Volume 1. Springer. 2023. s. 905-914. (Mechanisms and Machine Science). doi: 10.1007/978-3-031-45705-0_88