PSI - Issue 17

Available online at www.sciencedirect.com Structural Int grity Procedia 00 (2019) 000 – 000 Available online at www.sciencedirect.com ScienceDirect Structural Integrity Procedia 00 (2019) 000 – 000 Available online at www.sciencedirect.com ScienceDirect

www.elsevier.com/locate/procedia www.elsevier.com/locate/procedia

ScienceDirect

Procedia Structural Integrity 17 (2019) 487–494

ICSI 2019 The 3rd International Conference on Structural Integrity Vibration-Based Fault Detection for Flywheel Condition Monitoring Takanori Hasegawa a, *, Mao Saeki a , Tetsuji Ogawa a , Teppei Nakano a,b a Department of Communications and Computer Engineering, Waseda University, 27 Waseda-machi, Shunjiku-ku, 1620042 Tokyo, Japan b Futaba Electric Co., Ltd., 2-22-19 Omori-nishi, Ota-ku, 1430015 Tokyo, Japan ICSI 2019 The 3rd International Conference on Structural Integrity Vibration-Based Fault Detection for Flywheel Condition Monitoring Takanori Hasegawa a, *, Mao Saeki a , Tetsuji Ogawa a , Teppei Nakano a,b a Department of Communications nd Computer Engineering, Waseda University, 27 Waseda-machi, Shunjiku-ku, 1620042 Tokyo, Japan b Futaba Electric Co., Ltd., 2-22-19 Omori-nishi, Ota-ku, 1430015 Tokyo, Japan Flywhe ls ar o e of t e promising energy storage devices for stabilizing the power quality of reusable energy, owi g to their fast response time and high cycle lifetime. However, it can be catastrop ic when th y fail, becaus they store kinetic energy that ca be released in a short amount of time. Data-drive mo itoring techniques h ve been propo ed to solve fault detection tasks in s veral types of rotation machinery with ball bearings. In contrast to traditional approaches using human-engineered features th t require a hig level of expertis , a dat -driven approach requires no such prior knowledge. However, flywheels differ from typical rotation machi ery because they use a magnetic or pivot bearing, and it is unclear whether a data-driven method can e used to detect a fault. In t e present study, the effectiveness of a data-driven fault detection system f r flywheels that use pivot bearings is evaluated. A flywheel fault progresses in several stages, and vibration data were coll cted for a flywhe l running at each of those stag s. A conv lutional eural network (CNN) was exploited t detect a f ult of th flywheel and identify a m d of the fault. Experimental comparisons conducted using vibration signals from an actual flywheel demonstrated that faulty operational state observed at an end of the flywheel’s life can be detected with high accuracy using a data-driven method. Abstract Abstract Flywheels are one of the promising energy storage devices for stabilizing the power quality of reusable energy, owing to their fast response time and high cycle lifetime. However, it can be catastrophic when they fail, because they store kinetic energy that can be released in a short amount of time. Data-driven monitoring techniques have been proposed to solve fault detection tasks in several types of rotation machinery with ball bearings. In contrast to traditional approaches using human-engineered features that require a high level of expertise, a data-driven approach requires no such prior knowledge. However, flywheels differ from typical rotation machinery because they use a magnetic or pivot bearing, and it is unclear whether a data-driven method can be used to detect a fault. In the present study, the effectiveness of a data-driven fault detection system for flywheels that use pivot bearings is evaluated. A flywheel fault progresses in several stages, and vibration data were collected for a flywheel running at each of those stages. A convolutional neural network (CNN) was exploited to detect a fault of the flywheel and identify a mode of the fault. Experimental comparisons conducted using vibration signals from an actual flywheel demonstrated that faulty operational state observed at an end of the flywheel’s life can be detected with high accuracy using a data-driven method.

© 2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers. © 2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers. © 2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers.

Keywords: Deep neural networks; Data-driven feature; Anomaly detection; Condition monitoring system; Flywheel Keywords: Deep neural networks; Data-driven feature; Anomaly detection; Condition monitoring system; Flywheel

1. Introduction 1. Introduction

A flywheel energy storage device (FESD) stores energy in kinetic form (Akinyele and Rayudu (2014)). Its advantages over other energy storage devices such as batteries and fuel cells are its high energy and power density, A flywheel energy storage device (FESD) stores energy in kinetic form (Akinyele and Rayudu (2014)). Its advantages over other energy storage devices such as batteries and fuel cells are its high energy and power density,

* Corresponding author. Tel.: +81-3-3203-4468. E-mail address: hasegawa@pcl.cs.waseda.ac.jp * Correspon ing author. Tel.: +81-3-3203-4468. E-mail address: hasegawa@pcl.cs.waseda.ac.jp

2452-3216 © 2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers. 2452-3216 © 2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers.

2452-3216  2019 The Authors. Published by Elsevier B.V. Peer-review under responsibility of the ICSI 2019 organizers. 10.1016/j.prostr.2019.08.064