Comparative Study on Tree Classifiers for Application to Condition Monitoring ofWind Turbine Blade through Histogram Features Using Vibration Signals: A Data-Mining Approach

Wind energy is considered as a alternative renewable energy source due to its low operating cost when compared with other sources.The wind turbine is an essential system used to change kinetic energy into electrical energy.Wind turbine blades,in particular,require a competitive condition inspection approach as it is a significant component of the wind turbine system that costs around 20-25 percent of the total turbine cost.The main objective of this study is to differentiate between various blade faults which affect the wind turbine blade under operating conditions using a machine learning approach through histogram features.In this study,blade bend,hub-blade loose connection,blade erosion,pitch angle twist,and blade cracks were simulated on the blade.This problem is formulated as a machine learning problem which consists of three phases,namely feature extraction,feature selection and feature classification.Histogram features are extracted from vibration signals and feature selection was carried out using the J48 decision tree algorithm.Feature classification was performed using 15 tree classifiers.The results of the machine learning classifiers were compared with respect to their accuracy percentage and a better model is suggested for real-time monitoring of a wind turbine blade.

Wind-induced instabilities and monitoring of wind turbine

This paper presents real-time monitoring data and analysis results of the non-stationary vibrations of an operational wind turbine. The advanced time-frequency spectrum analysis reveals varied non-stationary vibrations with timevarying frequencies, which are correlated with certain system natural modes characterized by finite element analysis. Under the effects of strong wind load, the wind turbine system exhibits certain resonances due to blade passing excitations. The system also exhibits certain instabilities due to the coupling of the tower bending modes and blade flapwise mode with blade passing excitations under the variation of wind speed. An analytical model is used to elaborate the non-stationary and instability phenomena observed in experimental results. The properties of the nonlinear instabilities are evaluated by using Lyapunov exponent estimation.

Real-time Health Condition Evaluation on Wind Turbines Based on Operational Condition Recognition

针对大型风电机组运行工况和状态信息复杂，健康状态难以准确评价的问题，提出基于工况辨识的健康状态实时评价方法。该方法充分考虑机组运行工况的复杂性和多变性，采用工况辨识实现运行工况空间的划分。在各运行工况子空间，建立基于高斯混合模型（gaussianmixturemodel，GMM）多状态特征融合的健康状态评价模型。采用健康衰退指数（healthdegradationindex，HDI）作为机组健康状态评价指标，并给出健康衰退报警限的确定方法。该方法用于某1．5MW风电机组传动系统故障前的健康状态评价。结果表明，该方法提前监测到机组健康状态的衰退趋势，可实现故障的早期预报，避免严重故障发生，并为合理调整运行和安排维修提供依据。

Monitoring of Wind Turbine Blades Based on Dual-Tree Complex Wavelet Transform

Structural health monitoring(SHM)in-service is very important for wind turbine system.Because the central wavelength of a fiber Bragg grating(FBG)sensor changes linearly with strain or temperature,FBG-based sensors are easily applied to structural tests.Therefore,the monitoring of wind turbine blades by FBG sensors is proposed.The method is experimentally proved to be feasible.Five FBG sensors were set along the blade length in order to measure distributed strain.However,environmental or measurement noise may cover the structural signals.Dual-tree complex wavelet transform(DT-CWT)is suggested to wipe off the noise.The experimental studies indicate that the tested strain fluctuate distinctly as one of the blades is broken.The rotation period is about 1 s at the given working condition.However,the period is about 0.3 s if all the wind blades are in good conditions.Therefore,strain monitoring by FBG sensors could predict damage of a wind turbine blade system.Moreover,the studies indicate that monitoring of one blade is adequate to diagnose the status of a wind generator.

Vibration Signals and Condition Monitoring for Wind Turbines

Rolling element bearings are critical parts of modern wind turbines as they carry the loads of the turning structure and the wind force. The stochastic nature of the wind loads makes it difficult to estimate the useful operational life of the bearings. Condition monitoring of these bearings in a real time environment could be very helpful in estimating their performance and in scheduling maintenance actions when a condition-based maintenance strategy is followed. This procedure can be successfully implemented by using vibration analysis in the time domain or in the frequency domain, giving useful results about the current condition of bearings and the location of potential faults. Permanently located transducers on proper positions on the bearings’ housings can be used in order to collect, process and evaluate real time measurements and provide information about the bearing’s performance. In this work, a test rig is utilized in order to evaluate the performance of rolling bearings. The results of the experimentation are satisfactory and the progress of fatigue failures can be predicted through vibration analysis techniques showing that implementation in real scale may be useful.

Monitoring and Detection of Wind Turbine Vibration with KNN-Algorithm

Maintenance for wind turbines has been transformed using supervised machine learning techniques. This method of automatic and autonomous learning can identify, monitor, and detect electrical and mechanical components of wind turbines and predict, detect, and anticipate their degeneration. Using a machine learning classifier and frequency analysis, we simulate two failure states caused by bearing vibrations. Implementing KNN facilitates efficient monitoring, monitoring, and fault-finding for wind turbines. It is possible to reduce downtime, anticipate breakdowns, and import offshore aspects through these technologies.

A Comparative Study of Bayes Classifiers for Blade Fault Diagnosis in Wind Turbines through Vibration Signals

Renewable energy sources are considered much in energy fields because of thecontemporary energy calamities. Among the important alternatives being considered, windenergy is a durable competitor because of its dependability due to the development of theinnovations, comparative cost effectiveness and great framework. To yield wind energymore proficiently, the structure of wind turbines has turned out to be substantially bigger,creating conservation and renovation works troublesome. Due to various ecologicalconditions, wind turbine blades are subjected to vibration and it leads to failure. If thefailure is not diagnosed early, it will lead to catastrophic damage to the framework. In orderto increase safety observations, to reduce down time, to bring down the recurrence ofunexpected breakdowns and related enormous maintenance, logistic expenditures and tocontribute steady power generation, the wind turbine blade must be monitored now andthen to assure that they are in good condition. In this paper, a three bladed wind turbinewas preferred and using vibration source, the condition of a wind turbine blade is examined.The faults like blade crack, erosion, hub-blade loose connection, pitch angle twist and bladebend faults were considered and these faults are classified using Bayes Net (BN),Discriminative Multinomial Naïve Bayes (DMNB), Naïve Bayes (NB), Simple NaïveBayes (SNB), and Updateable Naïve Bayes (UNB) classifiers. These classifiers arecompared and better classifier is suggested for condition monitoring of wind turbine blades.

SVM-Algorithm for Supervision, Monitoring and Detection Vibration in Wind Turbines

With the implementation of supervised machine learning techniques, wind turbine maintenance has been transformed. A wind turbine’s electrical and mechanical components can be automatically identified, monitored, and detected to predict, detect, and anticipate their degeneration using this method of automatic and autonomous learning. Two different failure states are simulated due to bearing vibrations and compared with machine learning classifier and frequency analysis. A wind turbine can be monitored, monitored, and faulted efficiently by implementing SVM. With these technologies, downtime can be reduced, breakdowns can be anticipated, and aspects can be imported if they are offshore.

The Third-Order Viscoelastic Acoustic Model Enables an Ice-Detection System for a Smart Deicing of Wind-Turbine Blade Shells

The present work is based on the third-order partial differential equation (PDE) of acoustics of viscoelastic solids for the quasi-equilibrium (QE) component of the average normal stress. This PDE includes the stress-relaxation time (SRT) for the material and is applicable at any value of the SRT. The notion of a smart deicing system (SDS) for blade shells (BSs) of a wind turbine is specified. The work considers the stress in a BS as the one caused by the operational load on the BS. The work develops key design issues of a prospective ice-detection system (IDS) able to supply an array of the heating elements of an SDS with the element-individual spatiotemporal data and procedures for identification of the material parameters of atmospheric-ice (AI) layer accreted on the outer surfaces of the BSs. Both the SDS and IDS flexibly allow for complex, curvilinear and space-time-varying shapes of BSs. The proposed IDS presumes monitoring of the QE components of the normal stresses in BSs. The IDS is supposed to include an array of pressure-sensing resistors, also known as force-sensing resistors (FSRs), and communication hardware, as well as the parameter-identification software package (PISP), which provides the identification on the basis of the aforementioned PDE and the data measured by the FSRs. The IDS does not have hardware components located outside the outer surfaces of, or implanted in, BSs. The FSR array and communication hardware are reliable, and both cost- and energy-efficient. The present work extends methods of structural-health/operational-load monitoring (SH/OL-M) with measurements of the operational-load-caused stress in closed solid shells and, if the prospective PISP is used, endows the methods with identification of material parameters of the shells. The identification algorithms that can underlie the PISP are computationally efficient and suitable for implementation in the real-time mode. The identification model and algorithms can deal with not only the single-layer systems such as the BS layer without the AI layer or two-layer systems but also multi-layer systems. The outcomes can be applied to not only BSs of wind turbines but also non-QE closed single- or multi-layer deformable solid shells of various engineering systems (e.g., the shells of driver or passenger compartments of ships, cars, busses, airplanes, and other vehicles). The proposed monitoring of the normal-stress QE component in the mentioned shells extends the methods of SH/OL-M. The topic for the nearest research is a better adjustment of the settings for the FSR-based measurement of the mentioned components and a calibration of the parameter-identification model and algorithms, as well as the resulting improvement of the PISP.

基于残差网络的风电机组基础健康监测数据修复研究

为精准有效地修复连续性异常数据,提出一种基于残差块优化卷积神经网络的残差网络数据修复模型。以乳山风电场的风电机组基础健康监测数据为例对模型进行工程验证。同时选取具有修复功能的多种模型对实际异常数据进行修复验证,并对所有模型的性能以及自身的修复精度进行对比分析。结果表明:ResNet模型避免了FCN以及CNN模型存在的缺陷,提高了数据修复的精度;ResNet模型适用于缺失或异常比例低于30%的数据修复;ResNet模型修复实例的结果符合数据变化趋势,能较好吻合监测数据的原始曲线。

Model-based and Fuzzy Logic Approaches to Condition Monitoring of Operational Wind Turbines

It is common for wind turbines to be installed in remote locations on land or offshore, leading to difficulties in routine inspection and maintenance. Further, wind turbines in these locations are often subject to harsh operating conditions. These challenges mean there is a requirement for a high degree of maintenance. The data generated by monitoring systems can be used to obtain models of wind turbines operating under different conditions, and hence predict output signals based on known inputs. A model-based condition monitoring system can be implemented by comparing output data obtained from operational turbines with those predicted by the models, so as to detect changes that could be due to the presence of faults. This paper discusses several techniques for model-based condition monitoring systems: linear models, artificial neural networks, and state dependent parameter "pseudo" transfer functions.The models are identified using supervisory control and data acquisition(SCADA) data acquired from an operational wind firm. It is found that the multiple-input single-output state dependent parameter method outperforms both multivariate linear and artificial neural network-based approaches. Subsequently, state dependent parameter models are used to develop adaptive thresholds for critical output signals. In order to provide an early warning of a developing fault, it is necessary to interpret the amount by which the threshold is exceeded, together with the period of time over which this occurs. In this regard, a fuzzy logic-based inference system is proposed and demonstrated to be practically feasible.

Overview of condition monitoring and operation control of electric power conversion systems in direct-drive wind turbines under faults

Electric power conversion system （EPCS）, which consists of a generator and power converter, is one of the most important subsystems in a direct-drive wind turbine （DD-WT）. However, this component accounts for the most failures （approximately 60% of the total number） in the entire DD-WT system according to statistical data. To improve the reliability of EPCSs and reduce the operation and maintenance cost of DD-WTs, numerous researchers have studied condition monitoring （CM） and fault diagnostics （FD）. Numerous CM and FD techniques, which have respective advantages and disadvantages, have emerged. This paper provides an overview of the CM, FD, and operation control of EPCSs in DD-WTs under faults. After introducing the functional principle and structure of EPCS, this survey discusses the common failures in wind generators and power converters; briefly reviewed CM and FD methods and operation control of these generators and power converters under faults; and discussed the grid voltage faults related to EPCSs in DD-WTs. These theories and their related technical concepts are systematically discussed. Finally, predicted development trends are presented. The paper provides a valuable reference for developing service quality evaluation methods and fault operation control systems to achieve high-performance and high-intelligence DD-WTs.

基于改进PSO-LSTM算法的风电机组状态监测方法研究

通过改进粒子群算法(particle swarm optimization,PSO)优化长短期记忆神经网络算法(long short-term memory,LSTM)的参数,提出了一种基于改进PSO-LSTM算法的直驱式风电机组运行状态监测方法。首先将数据采集与监控系统(supervisory control and data acquisition,SCADA)采集到的数据利用随机森林的方法进行特征筛选,得到模型的输入参数;其次采用改进PSO-LSTM网络建立有功功率的预测模型,计算出预测值与实际值的残差,根据残差的分布来确实直驱式风电机组的状态;最后利用某风电机组SCADA数据对所提预测模型进行验证分析,结果表明,PSO-LSTM预测模型相比其他三种预测模型,具有较高的预测精度,并在状态异常后最短时间内发出故障警报,保证电场的健康稳定运行。

基于人工智能算法的风电机组状态监测和故障诊断技术研究综述

随着我国风电产业高速发展,风电机组服役时间延长,故障率和运维成本随之增加。利用人工智能算法对风电大数据进行数据挖掘,实现风电机组的状态监测与故障诊断,对风电产业提质增效具有重要的现实意义,近年来逐渐成为研究热点。文中介绍了风电机组数据采集与监控(Supervisory Control and Data Acquisition, SCADA)系统和振动信号数据的特性,阐述了风电机组状态监测和故障诊断智能算法的框架,归纳总结了相关研究成果,并对风电机组状态监测和故障诊断技术所面临的挑战和发展趋势进行了展望。

基于近邻元分析的风电机组状态监测特征选择方法

针对现有特征选择方法难以从大量的SCADA参量中挑选出重要变量的问题,基于近邻元分析算法提出一种专门适用于风电机组状态监测的特征变量选择方法。所提方法根据每个待选变量对回归精度的贡献率为各变量赋予相应的重要度权值,从而挑选出最重要的特征变量。通过分析SCADA数据中冗余变量的特点,针对性地提出了基于相关系数矩阵的去除冗余方法。采用Pearson相关系数、互信息和随机森林三种方法作为对比,以门控循环神经网络作为模型预测齿轮箱油池温度,用预测精度指标和残差控制图对各特征选择方法的选择结果进行评价和对比,结果表明所提方法的特征选择结果更加直观、冗余变量更少、预测精度更高。

Condition monitoring of a wind turbine generator using a standalone wind turbine emulator

The intend of this paper is to give a description of the realization of a low-cost wind turbine emulator （WTE） with open source technology from graze required for the condition monitoring to diagnose rotor and stator faults in a wind turbine generator （WTG）. The WTE comprises of a 2.5 kW DC motor coupled with a 1 kW squirrel-cage induction machine. This paper provides a detailed overview of the hardware and software used along with the WTE control strategies such as MPPT and pitch control. The emulator reproduces dynamic characteristics both under step variations and arbitrary variation in the wind speed of a typical wind turbine （WT） of a wind energy conversion system （WECS）. The usefulness of the setup has been benchmarked with previously verified WT test rigs made at the University of Manchester and Durham University in UK. Considering the fact that the rotor blades and electric subassemblies direct drive WTs are most susceptible to damage in practice, generator winding faults and rotor unbalance have been introduced and investigated using the terminal voltage and generated current. This wind turbine emulator （WTE） can be reconfigured or analyzed for condition monitoring without the need for real WTs.

An explainable AI framework for robust and transparent data-driven wind turbine power curve models

In recent years,increasingly complex machine learning methods have become state-of-the-art in modelling wind turbine power curves based on operational data.While these methods often exhibit superior performance on test sets,they face criticism due to a perceived lack of transparency and concerns about their robustness in dynamic,non-stationary environments encountered by wind turbines.In this work,we address these issues and present a framework that leverages explainable artificial intelligence methods to gain systematic insights into data-driven power curve models.At its core,we propose a metric to quantify how well a learned model strategy aligns with the underlying physical principles of the problem.This novel tool enables model validation beyond the conventional error metrics in an automated manner.We demonstrate,for instance,its capacity as an indicator for model generalization even when limited data is available.Moreover,it facilitates understanding how decisions made during the machine learning development process,such as data selection,pre-processing,or training parameters,affect learned strategies.As a result,we obtain physically more reasonable models,a prerequisite not only for robustness but also for meaningful insights into turbine operation by domain experts.The latter,we illustrate in the context of wind turbine performance monitoring.In summary,the framework aims to guide researchers and practitioners alike toward a more informed selection and utilization of data-driven wind turbine power curve models.

基于动态矩阵与特征相似度的AAKR风电机组状态监测

针对传统自组织核回归(AAKR)模型所选记忆矩阵冗余度较高、无法根据在线数据实时更新、计算相似度时未考虑特征参数权值不一的问题,提出一种基于动态矩阵与特征相似度的自组织核回归(DM-FS-AAKR)风电机组状态监测方法。首先基于样本间距离对原始数据集去冗余以降低运算复杂度,形成待选数据集;其次基于k-最近邻算法选取最符合当前运行条件的历史数据构建动态矩阵;为克服相似度计算时不良参数的偏差污染,提出一种特征相似度计算方法为不同参数分配相应权值进一步提高预测精度;最后以河北某风电场SCADA数据为例,对机组故障停机前工况进行验证实验。结果表明,相比于传统AAKR模型,所提算法平均绝对误差降低约15.6%,故障预警时能够提前35天实现预警,具有较高精度和实时性。

基于平衡分布自适应迁移学习的多风电机组运行状态监测方法

风电机组状态的准确监测对风电机组安全稳定运行和经济效益提升至关重要。但是,受不同风电机组运行数据分布差异的影响,现有状态监测方法在多风电机组应用场景下存在精度和效率难以兼顾的问题,而平衡分布自适应迁移学习(BDA)可以拉近数据距离,同化数据分布。因此,文章提出了一种基于BDA的多风电机组状态监测方法。首先,基于Copula熵的互信息法挖掘风电机组运行状态关键影响参量;然后,构建基于门控循环单元模型(GRU)和序贯概率比检验(SPRT)方法的单风电机组状态监测模型;最后,构建基于BDA的多风电机组运行数据分布同化模型,并用于多风电机组运行状态监测。算例结果表明,所提方法可以有效节省建模成本和计算成本,能够在保障多风电机组运行状态监测精度的前提下,显著提升监测效率。

基于改进实例学习算法的风电机组齿轮箱状态监测

风电机组齿轮箱的运行过程是复杂的非线性过程,采用实例学习(IBL)算法建立模型可有效对其进行状态监测。针对实例学习模型对训练数据质量敏感的特点,提出综合考虑多种性质的两步主动学习样本选择方法。首先提出一种基于拉丁超立方体抽样思想的网格划分初始样本选取方法,并基于z-score方法剔除其中的离群点。然后第一步基于信息性和代表性的综合得分选出候选样本来避免离群点影响,第二步基于多样性使第一步的候选样本稀疏化,从而避免冗余点影响。最后,基于指数加权移动平均控制图对实例学习回归模型输出的残差进行分析,并根据故障率对风电机组齿轮箱实现状态监测。利用某风电机组实际故障数据进行验证。结果表明:所提出的方法能选出优质样本,模型精度在验证集上较未改进前有所提升,且运算效率提升约50%,可实现齿轮箱异常的早期预警。