An integrated monitoring scheme for wind turbine main bearing using acoustic emission

Highlights

• An integrated monitoring method based on acoustic emission is proposed for ultra-low speed bearing fault diagnosis.

• A rotating speed estimation approach is developed to recover the accurate operating speed of the main bearing.

• An adaptive spectral coherence is explored to identify faulty narrowband buried under multiple disturbances.

• An effective damage localization model is constructed to improve the health management efficiency of the main bearing in wind farms.

• The proposed method can not only effectively detect the incipient damage of the main bearing, but also accurately determine the damage location.

Abstract

The condition monitoring of the main bearing (MB) plays a crucial role in the maintenance of wind turbines (WT), especially for direct-drive wind turbines (DDWT). However, due to the harsh operating environment and ultra-low rotating speed, the condition monitoring of the MB is still a challenging issue. In this study, an integrated monitoring scheme using acoustic emission (AE) is proposed for incipient fault detection and localization of MB. First, an rotating speed estimation approach using high-frequency envelope autocorrelation (HFEA) is developed to recover the accurate operating speed of MB. On this basis, the adapted spectral coherence (ASC) is explored to identify faulty sources buried under multiple disturbances. Finally, an effective damage localization model is further constructed to improve maintenance efficiency in practical applications. The performance of the proposed methodology is evaluated through two engineering cases with natural damages. Compared with state-of-the-art approaches, the proposed method can not only effectively detect the incipient damage of the MB, but also accurately determine the damage location. With this scheme, the inspection efficiency can be improved, thus it may provide a promising tool for the health management of WT.

Introduction

With the rise of green energy, wind power has become the fastest developing renewable energy in the world. To achieve Net Zero Emissions by 2050, the annual installed capacity for WTs will be steeper, which is three times the power generation capacity in 2020 [1]. Generally, there are two types of WT in the market, namely the gearbox driven WT and DDWT. Compared with the former WT, DDWT eliminates gearbox failure and transmission losses. Therefore, the DDWT is more effective for higher power ratings, and it potentially becomes the dominant technology [2].

While the rapid development of wind power has brought huge opportunities for the wind power industry, it also presents higher requirements for the reliability of all components of WT [3]. In particular, the MB in DDWT is a critical component supporting heavy loads and handling large moments. As most wind farms are located in remote areas with harsh natural environments, the MB is prone to failure, which can result in unnecessary downtime, expensive maintenance, and even catastrophic accidents [4]. Thus, the condition monitoring of MB is vitally important to guarantee the safe and stable operation of DDWT.

Over the years, various approaches have been adopted to monitor the operating conditions of bearings and have achieved satisfactory results, especially with increasing usage and attention to vibration signals [5,6]. They ranged from the time domain to frequency domain to time-frequency domain, including variational mode decomposition [7], Kurtogram and its improved methods [8,9], wavelet and wavelet package transform [10,11], blind deconvolution [12,13], and cyclostationary approach [14,15], etc.

Nevertheless, different from traditional industrial bearings, the health monitoring of MB still faces many challenges. Firstly, the rotating speeds of MB are very slow (less than 20 rpm), which means that the fault features generated by collision are particularly slight because of low kinetic energy and are easily overwhelmed by noisy interferences [16]. Thus, it is difficult to detect weak faults by vibration signals. Secondly, accurate speed information is the core of periodic feature extraction and enhancement. Although the SCADA system collects the operating information of DDWT, the rotating speed is acquired in an intermittent manner (typically several seconds per sample), thus it is not synchronous with the high-frequency condition monitoring signal [17]. If the SCADA information is directly used for the health monitoring of MB, it will affect the accuracy and reliability of diagnosis results. Therefore, determining how to accurately obtain the rotational speed that synchronized with the monitoring signal is still a challenging problem. Finally, since there is no gearbox in DDWT, a generator with multiple magnetic poles is required to achieve a sufficiently high output frequency [18]. Although such an arrangement can simplify the mechanical structure, a side-effect is that weak damage signatures of MB are often overwhelmed by strong electromagnetic interference (EMI) from the alternating magnetic field. For this reason, it further increases the difficulty of monitoring and may result in rather poor monitoring performance. Therefore, it is urgent to explore an effective health monitoring scheme for the MB.

Recent studies have shown that AE technology is quite sensitive to low-speed bearing fault detection [19,20]. Compared with the vibration signal, the AE signals primarily reflect the rapid release of strain energy in solids material, which presents high sensitivity to identify early bearing defects and monitor the evolution of the bearing fatigue [21]. Furthermore, AE signals have the potential for detecting damage source localization in large-size bearings [22]. Accordingly, many researchers had adopted AE signals to monitor the operating conditions of the low-speed bearing. In [23], the AE energy was proven to be reliable and sensitive for detecting incipient cracks in low-speed bearing. Elforjani and Mba [24] utilized AE signals to diagnose low-speed bearings. Experimental cases indicated that AE signals can be utilized to monitor crack initiation and propagation. Additionally, Fuentes et al. [25] constructed a probabilistic model using AE signals to detect bearing sub-surface damage. But for MB operating under harsh working conditions, the AE signals have the features of low SNR and multi-modulation, which result in challenges to the incipient fault detection of MB.

The narrowband demodulation techniques have been considered as an efficient tool for bearing fault detection [26]. To determine the narrowband with rich failure information, Antoni [8] proposed the Kurtogram, which assumed the FB with maximum Kurtosis as the optimal one for demodulation. Whereafter, a series of Kurtogram-based methods had been constructed to improve the selection strategy of FB [27,28]. Barszcz and Jabłoński [29] pointed out that the kurtosis-guided strategy is susceptible to non-Gaussian noise and random impulses. To tackle this issue, they proposed the Protrugram for bearing diagnosis, where the optimal FB was selected according to the Kurtosis of envelope spectrum rather than the Kurtosis of the filtered signal. Furthermore, Infogram and Autogram were constructed to mitigate the influence of fault-irrelevant impulsive disturbances [30,31]. However, as the AE signals from EMI contamination contains similar and even stronger impulsive characteristics with bearing defects, the aforementioned demodulation methods tend to enhance EMI rather than fault symptoms.

Furthermore, the research on the cyclostationarity of AE signals had been widely employed in bearing fault diagnosis [32,33]. In [34], Spectral Coherence (SCoh) was proposed for bearing damage detection, where the signals can be described from two perspectives, including spectral frequency, which contains the natural frequency components, and cyclic frequency, which contains the modulation frequency components. When the SCoh is integrated along with the spectral frequency, the resulting spectrum of the demodulated signal can be used to detect bearing faults. To suppress noise interference, the improved envelope spectrum (IES) integrated along the spectral FB is proposed to enhance fault features [35]. In the following, the Integrated Spectral Coherence (ISC) was presented to monitor bearing failures under strong EMI and other impulsive disturbances [36]. Similar to Kurtogram, Mauricio et al. [37] constructed an adaptive FB selection mechanism for gearbox bearing diagnostics under time-varying operating conditions. Additionally, to improve the robustness to non-Gaussian impulses, the Generalized Spectral Coherence (GSC) was developed for the health monitoring of bearings operating in crushing machines [38]. Although these works had achieved excellent performance in laboratory environments, the AE signals in real applications are contaminated by strong noise and multi-interference, and their feature enhancement is still a challenging issue.

Additionally, it is necessary to mention that, in practical wind power plants, regular inspections are required for large and expensive equipment such as DDWT. The inspectors often use the endoscope to detect the location of bearing damage. However, the MBs in DDWT are relatively large (diameter greater than 2 m). Detecting the whole bearing is time-consuming and inconvenient in a small nacelle. Consequently, how to locate the damage of MBs based on AE signals is another issue to be addressed.

To address the aforementioned limitations, an integrated methodology based on AE signals is proposed for the health monitoring of MBs. In this paper, the HFEA is first constructed to accurately estimate the operating speed of the MB, which solves the problems caused by asynchronous collection. On this basis, the ASC is then established to adaptively distinguish the optimal FB containing rich failure symptoms. Lastly, the damage localization scheme is developed to improve the detection efficiency of large-size MBs. With the proposed method, the MB could be effectively monitored even under heavy noise and complex interference.

The innovations of this study are mainly summarized as follows:

Accurate rotational speed estimation: As previously mentioned, the rotating speed collected from SCADA and the monitoring information are often asynchronous. Adding an extra tachometer not only increases monitoring costs but also introduces instability. Accordingly, the proposed method directly estimates rotating speed from the AE signals itself, which not only realizes synchronous collection but also reduces the monitoring costs.

Adaptive narrowband selection scheme: On account of the weak fault response and strong interferences, traditional methods are difficult to accurately distinguish fault modulation sources. To address the limitation, the constructed ASC targets the fault source and determines the optimal FB through an adaptive bandwidth selection scheme, which provides a novel solution for narrowband demodulation.

Effective damage localization strategy: This work established an effective damage localization model to greatly improve the health management efficiency of the MBs in wind farms.

In the following, this work is arranged as follows. Section 2 reviews the typical SCoh-based bearing diagnosis methods. Subsequently, the proposed integrated monitoring scheme is elaborated in Section 3.

Next, two industrial cases in Section 4 are employed to validate the effectiveness of the proposed method. At last, Some conclusions are presented in Section 5.