Mother wavelet selection in the discrete wavelet transform for condition monitoring of wind turbine drivetrain bearings

Although the discrete wavelet transform has been used for diagnosing bearing faults for two decades, most work in this field has been done with test rig data. Since field data starts to be made more available, there is a need to shift into application studies. The choice of mother wavelet, ie, the predefined shape used to analyse the signal, has previously been investigated with simulated and test rig data without consensus of optimal choice in literature. Common between these investigations is the use of the wavelet coefficients' Shannon entropy to find which mother wavelet can yield the most useful features for condition monitoring. This study attempts to find the optimal mother wavelet selection using the discrete wavelet transform. Datasets from wind turbine gearbox accelerometers, consisting of enveloped vibration measurements monitoring both healthy and faulty bearings, have been analysed. The bearing fault frequencies' excitation level has been analysed with 130 different mother wavelets, yielding a definitive measure on their performance. Also, the applicability of Shannon entropy as a ranking method of mother wavelets has been investigated. The results show the discrete wavelet transforms ability to identify faults regardless of mother wavelet used, with the excitation level varying no more than 4%. By analysing the Shannon entropy, broad predictions to the excitation level could be drawn within the mother wavelet families but no direct correlation to the main results. Also, the high computational effort of high order Symlet wavelets, without increased performance, makes them unsuitable.     

Data‐driven multiscale sparse representation for bearing fault diagnosis in wind turbine

With the increase of the wind turbine capacity, failures occur on the drivetrain of wind turbines frequently. Since faults of bearings in the wind turbine can lead to long downtime and even casualties, fault diagnosis of the drivetrain is very important to reduce the maintenance cost of the wind turbine and improve economic efficiency. However, the traditional diagnosis methods have difficulty in extracting the impulsive components from the vibration signal of the wind turbine because of heavy background noise and harmonic interference. In this paper, we propose a novel method based on data‐driven multiscale dictionary construction. Firstly, we achieve the useful atom through training the K‐means singular value decomposition (K‐SVD) model with a standard signal. Secondly, we deform the chosen atom into different shapes and construct the final dictionary. Thirdly, the constructed dictionary is used to sparsely represent the vibration signal, and orthogonal matching pursuit (OMP) is performed to extract the impulsive component. The proposed method is robust to harmonic interference and heavy background noise. Moreover, the effectiveness of the proposed method is validated by numerical simulation and two experimental cases including the bearing fault of the wind turbine generator in the field test. The overall results indicate that compared with traditional methods, the proposed method is able to extract the fault characteristics from the measured signals more efficiently.     

Condition monitoring of spar‐type floating wind turbine drivetrain using statistical fault diagnosis

Operation and maintenance costs are significant for large‐scale wind turbines and particularly so for offshore. A well‐organized operation and maintenance strategy is vital to ensure the reliability, availability, and cost‐effectiveness of a system. The ability to detect, isolate, estimate, and perform prognoses on component degradation could become essential to reduce unplanned maintenance and downtime. Failures in gearbox components are in focus since they account for a large share of wind turbine downtime. This study considers detection and estimation of wear in the downwind main‐shaft bearing of a 5‐MW spar‐type floating turbine. Using a high‐fidelity gearbox model, we show how the downwind main bearing and nacelle axial accelerations can be used to evaluate the condition of the bearing. The paper shows how relative acceleration can be evaluated using statistical change‐detection methods to perform a reliable estimation of wear of the bearing. It is shown in the paper that the amplitude distribution of the residual accelerations follows a t‐distribution and a change‐detection test is designed for the specific changes we observe when the main bearing becomes worn. The generalized likelihood ratio test is extended to fit the particular distribution encountered in this problem, and closed‐form expressions are derived for shape and scale parameter estimation, which are indicators for wear and extent of wear in the bearing. The results in this paper show how the proposed approach can detect and estimate wear in the bearing according to desired probabilities of detection and false alarm.     

A new noise-controlled second-order enhanced stochastic resonance method with its application in wind turbine drivetrain fault diagnosis

Condition monitoring of a wind turbine is important to extend the wind turbine system's reliability and useful life. However, in many cases, to extract feature components becomes challenging and the applicability of information drops down due to the large amount of noise. Stochastic resonance (SR), used as a method of utilising noise to amplify weak signals in nonlinear systems, can detect weak signals overwhelmed in the noise. Therefore, a new noise-controlled second-order enhanced SR method based on the Morlet wavelet transform is proposed to extract fault feature for wind turbine vibration signals in the present study. The second-order SR method can obtain better denoising effect and higher signal-to-noise ratio (SNR) of resonance output by means of twice integral transform compared with the traditional SR method. Morlet wavelet transform can obtain finer frequency partitions and overcome the frequency aliasing compared with the classical wavelet transform. Therefore, through Morlet wavelet transform, the noise intensity of different scales can be adjusted to realize the resonance detection of weak periodic signal whatever it is a low-frequency signal or high-frequency signal. Thus the method is well-suited for enhancement of weak fault identification, whose effectiveness has been verified by the practical vibration signals carrying fault information. Finally, the proposed method has been applied to extract feature of the looseness fault of shaft coupling of wind turbine successfully.     

Fault diagnosis for a wind turbine transmission system based on manifold learning and Shannon wavelet support vector machine

Fault diagnosis for wind turbine transmission systems is an important task for reducing their maintenance cost. However, the non-stationary dynamic operating conditions of wind turbines pose a challenge to fault diagnosis for wind turbine transmission systems. In this paper, a novel fault diagnosis method based on manifold learning and Shannon wavelet support vector machine is proposed for wind turbine transmission systems. Firstly, mixed-domain features are extracted to construct a high-dimensional feature set characterizing the properties of non-stationary vibration signals from wind turbine transmission systems. Moreover, an effective manifold learning algorithm with non-linear dimensionality reduction capability, orthogonal neighborhood preserving embedding (ONPE), is applied to compress the high-dimensional feature set into low-dimensional eigenvectors. Finally, the low-dimensional eigenvectors are inputted into a Shannon wavelet support vector machine (SWSVM) to recognize faults. The performance of the proposed method was proved by successful fault diagnosis application in a wind turbine's gearbox. The application results indicated that the proposed method improved the accuracy of fault diagnosis.     

Sensor and actuator fault diagnosis for wind turbine systems by using robust observer and filter

In this paper, we consider sensor and actuator fault detection and estimation issues for large scale wind turbine systems where individual pitch control (IPC) is used for load reduction. The faults considered are the blade root bending moment sensor faults and blade pitch actuator faults. In the first part, with the aid of a dynamical model of the wind turbine system, a so‐called H_∞/H_? observer in the finite frequency range, is applied to generate the residual for fault detection. The observer is designed to be sensitive to faults but insensitive to disturbances, such as wind turbulence. When there is a detectable fault, the observer sends an alarm signal if the residual evaluation is larger than a predefined threshold. In addition to the fault detection, we also consider the fault estimation problem, where a dynamic filter is used to estimate the fault magnitude. The effectiveness of the proposed approach is demonstrated by simulation results for several fault scenarios. Copyright © 2010 John Wiley & Sons, Ltd.     

Structural health monitoring for delamination detection and location in wind turbine blades employing guided waves

Wind power is becoming one of the most important renewable energies in the world. The reduction in operating and maintenance costs of the wind turbines has been identified as one of the biggest challenges to establish this energy as an alternative to fossil fuels. Predictive maintenance can detect a potential failure at an early stage reducing operating costs. Structural health monitoring together with non‐destructive techniques are an effective method to detect incipient delamination in wind turbine blades. Ultrasonic guided waves offer possibilities to inspect delamination and disunion between layers in composite structures. Delamination results in a concentration of tensions in certain areas near the fault, which can propagate and create the total break of the blade. This paper presents a new approach for disunity detection between layers comparing two real blades, also new in the literature, one of them built with three disbonds introduced in its manufacturing process. The signals are denoised by Daubechies wavelet transform. The threshold for the denoising is obtained by a wavelet coefficients selection rule using the Birgé‐Massart penalization method. The signals were normalized and their envelopes were obtained by Hilbert transform. Finally, a pattern recognition based on correlations was applied.