基于深度卷积自编码器的风电机组故障预警方法研究

通过风电机组状态监测进行故障预警,可防止故障进一步发展,降低风场运维成本。为充分挖掘风电机组监控与数据采集（SCADA）各状态参数时序信息,以及不同参数之间的非线性关系,该文将深度学习中自动编码器（AE）与卷积神经网络（CNN）相结合,提出基于深度卷积自编码（DCAE）的风电机组状态监测故障预警方法。首先基于历史SCADA数据离线建立基于DCAE的机组正常运行状态模型,然后分析重构误差确定告警阈值,使用EMWA控制图对实时对机组状态监测并进行故障预警。以北方某风电场2 MW双馈型风电机组叶片故障为实例进行实验分析,结果表明该文提出DCAE状态监测故障预警方法,可有效对机组故障提前预警,且优于现有基于深度学习的风电机组故障预警方法,可显著提升重构精度、减少模型参数和训练时间。     

基于一维密集连接卷积网络的风电齿轮箱智能故障分类

人工智能技术的飞速发展为现代能源装备的精益化故障诊断与健康管理提供了可能。风电齿轮箱由多个齿轮、轴承组成,且长期在变速、变载荷工况下运行,依靠传统的故障特征提取结合机器学习方法进行故障诊断存在精度低、缺乏智能性等缺点。文章提出了基于一维密集连接卷积网络的风电齿轮箱故障分类方法:将原始振动信号直接送入网络模型,经过密集连接、合成连接与卷积运算,匹配对应的故障类型,迭代训练故障分类模型;振动信号输入模型后的分类结果决定所属故障类别。文章所提出的风电齿轮箱故障分类方法具有诊断流程简单、故障识别率高等特点,多工况试验台故障数据验证了该方法的有效性。     

Wind turbine fault diagnosis method based on diagonal spectrum and clustering binary tree SVM

Renewable energy sources like wind energy are copiously available without any limitation. Reliability of wind turbine is critical to extract maximum amount of energy from the wind. The vibration signals in wind turbine's rotation parts are of universal non-Gasussian and nonstationarity and the fault samples are usually very limited. Aiming at these problems, this paper proposed a wind turbine fault diagnosis method based on diagonal spectrum and clustering binary tree Support Vector Machines (SVM). Firstly, the diagonal spectrum is calculated from vibration rotating machine as the input feature vector. Secondly, self-organizing feature map neural network is introduced to cluster the fault feature samples and construct a cluster binary tree. Then the multiple fault classifiers are designed to train and test samples. The wind turbine gear-box fault experiment results proved that this method can effectively extract features from nonstationary signals, and can obtain excellent results despite of less training samples.     

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.     

Intelligent wind turbine gearbox diagnosis using VMDEA and ELM

Wind turbine gearbox diagnosis is a vital tool for maintaining wind turbine operation and safety. The gearbox vibration signal is invariably complex and variable, and useful information and features are difficulty of extraction. Recently, a new and adaptive signal decomposition method, known as variational mode decomposition (VMD), has been proposed, which helps to improve the efficiency and effectiveness of extracting features from gearbox vibration signals. However, the performance of the VMD method mainly depends on its input parameters, especially the mode number and balancing parameter (also called the quadratic penalty term). Hence, this paper proposes a selection method for an optimized VMD parameter using differential evolution algorithm (DEA), also called VMDEA. Firstly, the VMDEA is used to select optimized VMD input parameters for each of the vibration signals. Following this, VMD decomposes each vibration signal into sets of subsignals using the selected optimized parameter. Multidomain features are extracted from VMD reconstructed signals and are passed on to the extreme learning machine (ELM) for fault classification. This study can thus provide a good solution for determining an optimized VMD parameter for decomposing vibration signals and can also provide a more efficient and effective diagnostic approach to wind turbine gearbox maintenance.     

Application of the ensemble empirical mode decomposition and Hilbert transform to pedestal looseness study of direct-drive wind turbine

The fault signal problems of wind turbine are non-linear and non-stationary, thus it is difficult to obtain the obvious fault features. In this study, a time-frequency method based on EEMD (ensemble empirical mode decomposition) and Hilbert transform is presented to investigate the bearing pedestal looseness fault of direct-drive wind turbine. The real vibration signals are analyzed using IMFs (intrinsic mode functions) extracted by ensemble empirical mode decomposition and Hilbert spectrum in the proposed method. The experimental results indicate that the proposed method is effective to extract the fault features of bearing pedestal looseness of wind turbine. And the results also demonstrate that fault features of front bearing pedestal looseness are different from rear bearing pedestal looseness with the same looseness gap. The fluctuation of rotational frequency increases with the occurrence of front bearing pedestal looseness fault, especially the half rotational frequency and high-frequency components, and the shaft orbit is complex. Besides, we found that when the rear bearing pedestal is loosen, the fluctuation of rotational frequency also increases, and the half rotational frequency component can be found. But for the high-frequency components, it is not obvious, and the shaft orbit is an approximate ellipse. Although the fault features of front and rear bearing pedestal looseness are obvious, the powers generated by wind turbine generator only change slightly.     

基于优化VMD复合多尺度散布熵及LSTM的风力发电机齿轮箱故障诊断方法研究

以风力发电机齿轮箱加速度信号为研究对象,提出一种数据驱动的风力发电机齿轮箱故障诊断方法,该方法以灰狼优化的变分模态分解方法(AGWO-VMD)、复合多尺度规范化散布熵(NCMDE)及长短期记忆网络(LSTM)为基础,实现齿轮箱故障的快速诊断.首先将时域信号转换至角域;然后通过AGWO-VMD方法对角域信号进行自适应分解...     

基于S-SLLE的风电机组齿轮箱故障诊断方法研究

针对风电机组齿轮箱结构复杂、受交变载荷和恶劣工作环境影响容易出现故障导致停机的问题,提出基于统计学K-均值聚类理论的统计型监督式局部线性嵌入流形学习(S-SLLE)特征维数约简方法,首先通过对齿轮箱振动信号时频域故障特征提取,剔除冗余特征向量,减少诊断模型的复杂度和计算量,再利用RBF核支持向量机分类器建立诊断模型,对...     

基于深度学习与支持向量机的滚动轴承故障诊断研究

针对滚动轴承运行环境复杂,传统故障诊断方法难以从强非线性信号中提取有效故障特征,且无法充分利用信号自身特征的问题,提出CNN-LSTM-SVM故障诊断方法。以滚动轴承加速度寿命实验数据为研究对象,基于卷积神经网络(Convolutional Neural Network, CNN)与长短期记忆网络(Long Short Term Memory, LSTM)技术提取信号特征并结合支持向量机(Support Vector Machine, SVM)完成故障分类。结果显示：该方法具有良好外推性能,在变演变阶段下的平均准确率达到95.92%,与现有方法相比,至少高出11.34%,且在噪声环境下的诊断准确率均高于现有方法,稳定性更佳,体现良好的鲁棒性与泛化性。     

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.     

一种基于主成分分析与支持向量机的风电齿轮箱故障诊断方法

为解决风电齿轮箱状态监测数据样本量较少,特征指标间存在相互干扰且具有非线性难以分类等问题,本文提出了一种基于主成分分析结合支持向量机的风电齿轮箱故障诊断方法。首先,采用主成分分析法(PCA)对原始数据进行降维,做出第1,2主成分二维图及前3个主成分三维图,表明PCA对监测状态数据具有一定的分类效果。其次,提取累计贡献率80%以上的前5个主成分作为数据集。最后,采用支持向量机(SVM)比较4种不同核函数的诊断准确度,并加入噪声验证。分析结果表明：径向基核函数构建的支持向量机总体分类精度达到97%,准确率最高;在含噪的情况下,线性核函数与径向基核函数分类精度达到94%;与MLP神经网络进行对比发现,支持向量机更适应小样本分析且测试精度较高。实例分析表明,主成分分析结合支持向量机有较好的分类效果,适用于风电齿轮箱故障诊断的工程应用。     

Wind turbine SCADA alarm analysis for improving reliability

Previous research for detecting incipient wind turbine failures, using condition monitoring algorithms, concentrated on wind turbine Supervisory Control and Data Acquisition (SCADA) signals, such as power output, wind speed and bearing temperatures, using power‐curve and temperature relationships. However, very little research effort has been made on wind turbine SCADA alarms. When wind turbines are operating in significantly sized wind farms, these alarm triggers are overwhelming for operators or maintainers alike because of large number occurring in a 10 min SCADA period. This paper considers these alarms originating in two large populations of modern onshore wind turbines over a period of 1–2 years. First, an analysis is made on where the alarms originate. Second, a methodology for prioritizing the alarms is adopted from an oil and gas industry standard to show the seriousness of the alarm data volume. Third, two methods of alarm analysis, time‐sequence and probability‐based, are proposed and demonstrated on the data from one of the wind turbine populations, considering pitch and converter systems with known faults. The results of this work show that alarm data require relatively little storage yet provide rich condition monitoring information. Both the time‐sequence and probability‐based analysis methods have the potential to rationalize and reduce alarm data, providing valuable fault detection, diagnosis and prognosis from the conditions under which the alarms are generated. These methods should be developed and integrated into an intelligent alarm handling system for wind farms, aimed at improving wind turbine reliability to reduce downtime, increase availability and leading to a well‐organized maintenance schedule. Copyright © 2011 John Wiley & Sons, Ltd.     

Wind turbine fault diagnosis based on Morlet wavelet transformation and Wigner-Ville distribution

Based on the Morlet wavelet transformation and Wigner-Ville distribution (WVD), we present a wind turbine fault diagnosis method in this paper. Wind turbine can be damaged by moisture absorption, fatigue, wind gusts or lightening strikes. Due to this reason, there is an increasing need to monitor the health of these structures. Vibration analysis is the best-known technology applied in wind turbine condition monitoring, in which the time-frequency analysis techniques such as Wigner-Ville distribution (WVD) are widely used. Theoretically WVD has an infinite resolution in time-frequency domain. For early wind turbine fault signals, however, there are two main difficulties in WVD analysis. One is strong noise signals in the background and the other is cross terms in WVD itself. In this paper, continuous wavelet transformation (CWT) is employed to filter useless noise in raw vibration signals, and auto terms window (ATW) function is used to suppress the cross terms in WVD. In the CWT de-noising process, the Morlet wavelet, whose shape is similar to mechanical shock signals, is chosen to perform CWT on the raw vibration signals. The appropriate scale parameter for CWT is optimized by the cross validation method (CVM). An ATW based on the Smoothed Pseudo Wigner-Ville distribution (SPWVD) spectrum is taken to be a window function to suppress the cross terms in WVD. The new method can not only remove cross terms faraway from the auto terms, but also keep high energy close to every instantaneous frequency, the virtues such as high time-frequency resolution, and good energy aggregation etc. The wind turbine gear fault diagnosis experiment results indicate that the proposed method has a good de-nosing performance and is effective in suppressing the cross terms and extracting fault feature.     

小波包能量熵和改进的LSSVM在风力机轴承故障诊断中的应用

针对极端复杂工况下风力机轴承运行状态监测中的故障诊断问题,提出一种基于小波包能量熵故障特征提取并结合鲸鱼算法(WOA)优化最小二乘支持向量机(LSSVM)进行故障分类识别的风力机轴承故障诊断方法。通过小波包分解提取各频带成分的能量熵值构建故障特征集,同时针对LSSVM参数的选取依赖人工选择的盲目性问题,采用鲸鱼优化算法寻找LSSVM中最优的2个关键参数正则化参数和核函数参数,以此提高故障诊断模型的分类精度。通过不同工况下的试验数据集测试,实现了对不同故障状态特征参数的准确分类。结果表明,所提方法诊断结果优于遗传算法(GA)和粒子群算法(PSO)分别优化的LSSVM.远优于传统的LSSVM算法。     

Enhancement of damage-detection of wind turbine blades via CWT-based approaches

In this paper, a continuous wavelet transform-based approach is applied to enhance the damage-detection capability of wind turbine blades. With the time--frequency localization features embedded in wavelets, the time and scale information of the acquired signals can be presented as a visualization scheme, where the condition monitoring of turbine blades can be better realized. Based on these sensor signals, this proposed approach was applied to discriminate the damaged structure from the healthy one under several scenarios. Test results have demonstrated the practicality and advantages of the proposed method for the application considered.     

An engineering condition indicator for condition monitoring of wind turbine bearings

Condition monitoring (CM) of wind turbine becomes significantly important part of wind farms in order to cut down operation and maintenance costs. The large amount of CM system vibration data collected from wind turbines are posing challenges to operators in signal processing. It is crucial to design sensitive and reliable condition indicator (CI) in wind turbine CM system. Bearing plays an important role in wind turbine because of its high impact on downtime and component replacement. CIs for wind turbine bearing monitoring are reviewed in the paper, and the advantages and disadvantages of these indicators are discussed in detail. A new engineering CI (ECI), which combined the energy and kurtosis representation of the vibration signal, is proposed to meet the requirement of easy applicability and early detection in wind turbine bearing monitoring. The quantitative threshold setting method of the ECI is provided for wind turbine CM practice. The bearing run‐to‐failure experiment data analysis demonstrates that ECI can evaluate the overall condition and is sensitive to incipient fault of bearing. The effectiveness in engineering of ECI is validated though a certain amount of real‐world wind turbine generator and gearbox bearing vibration data.     

基于长短时记忆网络融合SCADA数据的风电齿轮箱状态监测

针对不具有时间记忆能力的机器学习方法融合风电机组数据采集与监控系统(SCADA)的时序数据而导致风电齿轮箱状态预测精度不高的问题,提出基于长短时记忆(LSTM)网络融合SCADA数据的风电齿轮箱状态预测模型。选择能表征风电齿轮箱运行状态的某个监测量作为模型的输出量,基于灰色关联度选择与该监测量关联密切的SCADA参数作为预测模型的输入量;使用正常状态下的SCADA数据训练LSTM预测模型,得出预测值和残差,通过3σ准则计算出上下预警阈值,用于风电齿轮箱状态监测和故障预警。某风电场风电齿轮箱的SCADA数据验证表明所提出的方法能有效预警风电齿轮箱故障。     

风电机组传动链典型故障特征提取方法

为解决风电机组传动链易发生故障的问题,文章阐述了风电机组齿轮箱特征频率的计算方法和基于振动信号分析的故障特征提取方法。结合实际情况,以行星级齿轮磨损、中间轴小齿轮崩齿、高速轴齿轮崩齿和发电机轴承电腐蚀等典型故障为例,通过齿轮箱特征频率和传动链典型故障振动信号基本特征分析,可较好地完成故障识别。结果表明,采用经典信号处理方法能对上述典型故障进行特征提取,验证了经典方法对单一、明显故障特征提取的有效性,为深入开展传动链故障特征提取方法研究奠定了基础,为风电机组故障检修维护提供了技术支撑。     

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.     

Development of a 20 kW wind turbine simulator with similarities to a 3 MW wind turbine

As the use of wind power has steadily increased, the importance of a condition monitoring and fault diagnosis system is being emphasized to maximize the availability and reliability of wind turbines. To develop novel algorithms for fault detection and lifespan estimation, a wind turbine simulator is indispensible for verification of the proposed algorithms before introducing them into a health monitoring and integrity diagnosis system. In this paper, a new type of simulator is proposed to develop and verify advanced diagnosis algorithms. The simulator adopts a torque control method for a motor and inverter to realize variable speed-variable pitch control strategies. Unlike conventional motor–generator configurations, the simulator includes several kinds of components and a variety of sensors. Specifically, it has similarity to a 3 MW wind turbine, thereby being able to acquire a state of operation that closely resembles that of the actual 3 MW wind turbine operated at various wind conditions. This paper presents the design method for the simulator and its control logic. The experimental comparison between the behavior of the simulator and that of a wind turbine shows that the proposed control logic performs successfully and the dynamic behaviors of the simulator have similar trends as those of the wind turbine.