PSI - Issue 17

Takanori Hasegawa et al. / Procedia Structural Integrity 17 (2019) 487–494 Takanori Hasegawa et al./ Structural Integrity Procedia 00 (2019) 000 – 000

488

2

long life as well as being environmentally friendly since they are made of non-hazardous materials (Wicki and Hansen (2017) and Nasiri and Hamidi (2018)). FESDs have been regarded as promising energy storage devices for stabilizing the power quality of reusable energy (Akinyele and Rayudu (2014) and Tanabe et al. (2009)) and also used in transport systems (Doucette and McCulloch (2011)). Although rarely reported, its failure can be quite catastrophic, because it stores kinetic energy that can be released in a short amount of time. In Awata and Mita (2006), diagnostic approach using amplitude modulation of the ultrasonic waves is proposed to detect faults in ball bearings. Finite element method is used in Liu et al. (2015) to simulate a crack, and Hilbert – Huang transform algorithm is applied for the FESD vibration analysis. Previous fault detection methods for FESDs, however, require high level of expertise and are not easily applicable to sites where expert knowledge is not sufficient or at sites that are not easily accessible. In contrast to such methods, data-driven approaches allow the feature extraction from a given dataset and require no domain knowledge. In Janssens et al. (2016) and Kateris et al. (2014), convolutional neural networks (CNN) (Hinton et al. (2012) and Krizhevsky et al. (2012)) are used to solve fault detection tasks in several types of rotation machinery with ball bearings. However, FESDs may 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. Hence, the aim of the present study is to evaluate the effectiveness of a data-driven, specifically a CNN-based fault detection system for FESDs that use pivot bearings. A fault in a FESD progresses in several stages. Initially, there is a loss of helium that is originally filled inside the FESD. Subsequently, degeneration of the oil around the pivot can be confirmed before a final breakdown. Attempts are made to detect these signs using vibration data that can be collected easily by attaching an accelerometer externally. Each of the fault stages can be identified by classifying unknown inputs into the fault stages. It serves for diagnosis of failure but needs preliminary recording of vibration data at all fault stages to be identified. If faulty data are not available for machine learning, novelty detection or outlier detection approach is applicable. In the present study, both approaches are examined on an actual FESD. The knowledge obtained from the present experiments could be useful to improve the condition monitoring of FESDs. The rest of the present paper is organized as follows. Section 2 describes the function and structure of FESD and data collection. Data-driven operational state classification and novelty detection are applied to condition monitoring of FESD in Sections 3 and 4. Finally, Section 5 provides a summary.

2. Flywheel energy storage device and data collection

This section describes functions and features of flywheel energy storage device (FESD) and vibration materials utilized in this study.

2.1. Flywheel

FESDs store energy in the form of rotational kinetic energy. FESDs typically consist of a spinning rotor, motor generator, bearings, a power electronics interface, and housing. FESDs consume electrical energy to increase and to keep the rotational speed of a rotor via motor-generator in the charging and the steady state. Contrary in the discharging state, electrical energy is generated by motor-generator using the rotational energy of the rotor, and electrical power is supplied to an electronics interface. A bearing system plays an important role in FESDs to reduce the friction against the rotation, and to achieve the main characteristics such as high cycle life, long lifetime and high round trip efficiency. Proper type of the bearing system is employed depending on the weight, lifecycle life, and lower losses (Amiryar and Pullen (2017)). The present study employed a FESD equipped with the pivot bearing system, which is one of the levitation bearing to avoid attrition (Kato et al. (2015)). Figure 1 shows the overview and specifications of the device, and Fig. 2 shows the internal structure and pivot bearing system. As illustrated in the right side of Fig. 2, the concave of the bearing is suffused with the oil. The pivot, mounted on the center of heavy rotor, levitates in the oil during the high-speed rotation to avoid attrition of the bearing. Besides, the housing case is fulfilled with helium gas to reduce the air resistance of the rotor. This helps to reduce the air temperature increase. With these features, high efficiency and long-lifetime energy storage system is achieved.