Methods for Advanced Wind Turbine Condition Monitoring and Early Diagnosis: A Literature Review

Abstract

Condition monitoring and early fault diagnosis for wind turbines have become essential industry practice as they help improve wind farm reliability, overall performance and productivity. If not detected and rectified at early stages, some faults can be catastrophic with significant loss or revenue along with interruption to the business relying mainly on wind energy. The failure of Wind turbine results in system downtime and repairing or replacement expenses that significantly reduce the annual income. Such failures call for more systematized operation and maintenance schemes to ensure the reliability of wind energy conversion systems. Condition monitoring and fault diagnosis systems of wind turbine play an important role in reducing maintenance and operational costs and increase system reliability. This paper is aimed at providing the reader with the overall feature for wind turbine condition monitoring and fault diagnosis which includes various potential fault types and locations along with the signals to be analyzed with different signal processing methods.

1. Introduction

Wind energy has become a superior renewable energy resource that plays a vital role in the power sector. Above 54 GW of new wind turbine plants have been installed worldwide since the year 2016 according to the statistics of Global Wind Energy Council (GWEC) which includes more than 90 countries. These new installations represent a cumulative global market increase of more than 12.6% that reaches a total of 486.8 GW [1]. In Australia, the penetration of renewable energy level has also been increased during the past decade. Approximately 17.3% of Australia’s electricity was generated from renewable energy sources in the year 2016 [2]. Wind power is one of the dominating renewable energy resources worldwide. According to the clean energy council, the current proportion of wind power generation to the total renewable energy generation in Australia is 30.9% as shown in Figure 1. This represents 5.3% of Australia’s total electricity. Wind turbines are usually installed in remote areas or offshore in a very harsh environmental condition. This makes wind turbines more prone to failures. According to some statistical studies [3], the maximum wind turbines failure rate takes place within the gearbox followed by faults within the power electronic converters. Some of these faults can be catastrophic with significant loss or revenue and they can interrupt businesses that rely mainly on wind energy. To avoid such consequences, implementation of reliable condition monitoring and fault diagnosis techniques has become essential for all critical components in the electricity grid including the wind turbine [4,5,6,7,8].

This paper is organized as follows: Section 2 presents the basics of the wind turbine system and Section 3 presents the faults that occurred in the wind turbine system. Signals to be analyzed and signal processing methods are presented in Section 4 and Section 5 respectively. Conclusions are drawn in Section 6.

2. The Basics of Wind Turbine System

A wind turbine is a complex electromechanical system that consists of several components and subsystems. The major components include rotor, bearings, mechanical shaft, gearbox, generator, power electronic interface and sensors as shown in Figure 2 [9]. Various types of generators such as wound rotor induction generator, squirrel-cage induction generator, and synchronous generator can be employed with wind turbines [10]. The squirrel-cage induction generator runs at a fixed speed and the variable wind speed may cause faults for this type. Adaptive control scheme is presented in [3] to ensure the reliable operation of the permanent magnet synchronous generator under different operating conditions. This type of generator is interfaced with the grid through full-scale power converters, which increases the implementation cost [11]. The wound rotor induction generator (also called doubly fed induction generators (DFIG)) based wind turbine system has become popular due to the superior features of this type which include low cost, low converter rating, active and reactive power control, reduced losses and high efficiency [12]. Moreover, this type can operate within a wide range of variable speed that overcomes the drawbacks of fixed speed synchronous generator [13]. In the DFIG-based wind energy conversion system (WECS), 70% of the generated power is directly fed to the grid through the stator terminals while the remaining 30% is fed to the grid via back to back converters that comprise rotor side converter, DC-link capacitor and grid side converter as shown in Figure 3 [14,15,16,17].

3. Faults of Wind Turbine System

The wind turbine system is subject to several types of faults within various components as shown in Figure 4. These faults are discussed below:

3.1. Rotor

The rotor of the wind turbine consists of blades and hub. The wind turbine rotor is subject to various mechanical faults such as rotor asymmetry, fatigue, crack, increased surface roughness, reduced stiffness, and deformation of blades [19]. Incorrect design of blades pitch angle and blades mass imbalance are the main causes of rotor asymmetry [20]. Fatigue is caused by material aging and variable speed of wind on the blades. Long-term fatigue causes a reduction of the stiffness of the blades and leads to crack on the surface. Blade surface roughness is usually caused by icing, pollution, exfoliation and blowholes. As rotor faults are accompanied with a change in the blade material structure, these faults can be diagnosed using acoustic emission and vibration sensors. Acoustic emission is able to detect emerging structural changes by inserting sensors on the blades [21]. If these faults develop to a certain level that contributes abnormal vibrations of the blades, then signals information acquired from vibration sensors can be utilized for fault diagnosis.

3.2. Gearbox

Gearbox faults represent approximately 35% of the overall faults in wind turbines [22]. Failure in the gearbox and bearing depends on various factors such as material defects, design, manufacturing and installing errors, surface wear, torque overloads, misalignment and fatigue. The most common gear box failures include tooth abrasion, tooth crack, breakage, fracturing and surface fatigue initiated by the debris due to bearing failures [23]. These faults may cause an abnormal temperature increase of the bearing and the lubrication oil which can be used as an indication for such types of faults.

3.3. Main Shaft

The failures of the mechanical shaft include corrosion, misalignment, crack and coupling failure [24]. These faults affect the normal rotation of the shaft as well as other subsystems connected to the shaft. Hence the torque transmitted via the drivetrain will be affected and may lead to vibrations at certain characteristic frequencies in the gearbox, rotor, and generator [25]. Shaft misalignment fault affects the amplitude of the fundamental frequency of the vibration of the gearbox, rotor, and generator. Hence, shaft faults can be detected and analyzed by capturing vibration, torque, and electrical signals [24]. The analysis is done with the methods of frequency analysis such as fast Fourier transform (FFT).

3.4. Hydraulic System

A hydraulic system delivers hydraulic power to drive the motors that is used to adjust the blade pitch angle [26], maximizes wind power generation by adjusting yaw position and controls the mechanical brake to ensure the safety of wind turbine [27]. This system is subject to oil leak and sliding valve blocking faults. Pressure and level sensors’ signals are used to diagnose these faults.

3.5. Mechanical Brake

A mechanical brake typically has three main components such as disk and calipers, hydraulic mechanism and three-phase ac motor. Hydraulic mechanism is used to drive the calipers and motor is used to power the hydraulic mechanism [28]. The brake is usually mounted on the main shaft. It is used to prevent the over speed of the rotor and even force the shaft to stop in case of failures of critical components. The brake is also applied for yaw subsystem to stabilize the bearing. The disk may be cracked due to overshoot of mechanical stress on the brake and overheating. Faults of the mechanical brake can be diagnosed through temperature and vibration monitoring.

3.6. Tower

The wind turbine tower faults are mainly occurred due to structure damages such as cracks and corrosions. These faults may be caused due to several factors including improper installation, loading, poor quality control during the manufacturing process, lightning, fire and earthquakes. Time and frequency domain analysis techniques can reveal the health condition of the tower [29].

3.7. Electric Machine

There are two types of faults that may take place in the electrical machine: mechanical and electrical faults. The electrical faults comprise open circuit, stator/rotor insulation damage and electrical imbalance. On the other side, broken rotor bar, air gap eccentricity, bent shaft, bearing failure and rotor mass imbalance are the main mechanical faults. The most common fault reported in the literature is the short circuit turns of coils in the wind turbine generator [30]. These faults can be detected through shaft displacement detection, torque measurement, and vibration analysis. Temperature sensor can be used to detect winding faults [31]. Stator open-circuit faults alter the spectra of the stator line currents and instantaneous power [32]. Since rotor electrical imbalance causes shaft vibration, vibration signals can be used to monitor electrical imbalance [25]. Stator electrical imbalance can be detected from the change in the harmonic content of electrical signals [30].

3.8. Power Electronic Converter

The reliability of the power electronic converter becomes more complex with the increase of wind turbine power rating. According to literature, about 20% of WECS is due to power electronic converters failure as can be seen in Figure 4 [25]. Temperature, vibration and humidity are the three major factors that cause failures in power electronic converters [33]. There are three major converter components in which faults normally occur: capacitors, printed circuit board (PCB) and insulated gate bipolar transistor (IGBT) as shown in Figure 5. Capacitor faults include open/short circuit, electrode materials migrating across the dielectric and forms conductive paths, dielectric breakdown and increased dissipation factor. PCBs faults include broken buried metal lines, corrosion or crack of traces, component misalignment, board delamination and cold-solder joints. The failure modes of the IGBT modules include bond wire liftoff, short circuit, gate misfiring, solder fatigue and cracks [34]. Thermo-sensitive electrical parameters, such as the collector-emitter saturation voltage, gate-emitter threshold voltage, on-state resistance and internal thermal resistance are used to monitor the degradation of IGBT modules.

3.9. Sensors

Faults within the WECS may also take place in the sensors mounted at various locations to measure some condition monitoring parameters such as temperature, voltage, current and torque transducers. More than 14% of WECS faults occurs in these sensors [35]. Faults such as malfunction/physical failure, malfunction of the data processing/communication software [21] may cause performance degradation to the wind turbine, failure to the control system, mechanical and electrical subsystems and may lead to shut down the wind turbine. Encoder faults in an induction motor drive are detected by measuring the mean and standard deviation of the rotor speed signal [36]. The correlation between rotor position and stator current are used to detect encoder faults by using the wavelet transform [37].

3.10. Control System

The control system plays a vital role in regulating the operations of wind turbine. Faults in the control system can be categorized into hardware and software failures. Hardware failures include sensor faults, actuator faults, failure of control board and communication links. Model-based methods can be used to detect hardware failures. The software failures include buffer overflow, resource leaks and out of memory. These faults can be detected using diagnosing codes in the software.

4. Signals of Wind Turbine Condition Monitoring System

4.1. Vibration

Vibration analysis is one of the powerful tools that is currently used to monitor the mechanical integrity of wind turbines [38]. Vibration sensors installed on the casing of the wind turbine are used to detect faults within various wind turbine components such as gearbox, bearing, rotor and blade, tower, generator and main shaft [39]. There are three main types of vibration sensors: displacement sensors, velocity sensors and accelerometers. The signals acquired from accelerometer carry out the accelerated fault information and the amplitude of the accelerated signal demonstrates the fault severity level [40]. Installation of vibration sensors and required data acquisition devices increase the wiring complexity and capital cost of the technique. In addition, it is quite difficult to insert the sensors on the surface or into the body of the components. Moreover, if the sensors and data acquisition devices fail to provide signals, it may lead to the failure of wind turbine control, mechanical and electrical subsystems. Vibration signals are not capable of detecting incipient faults due to low signal to noise ratio (SNR).

4.2. Acoustic Emission

Acoustic emission sensors mounted on critical areas emit sound signals which are used to detect the structural defects of blades, gearbox, and bearings [41]. The signal acquired from small structural changes indicates an incipient structure damage or defect such as fatigue, crack, reduced stiffness and increased surface roughness [42]. Unlike vibration analysis, this technique is able to detect incipient faults at early stages. However, it requires a large number of sensors which increases the cost and complexity of the technique.

4.3. Strain

Fiber optic strain sensors are normally mounted on the surface of the wind turbine blade. The signal collected from the sensors provides information about the structural defects of the blades [43]. This technique has the ability to detect small structural changes, requires low sampling frequency, and has consistent performance over transmitting distance.

4.4. Torque and Bending Moment

There are two torque sensors, namely rotary torque and reaction torque sensors. Rotary torque sensors measure the torque signals whereas reaction torque sensors measure the bending moment signals. Electrical signals of a generator can also be used to calculate the torque that omits the requirement for sensors and reduces the capital costs [44]. Signature extraction of torque signal identifies mechanical failures. The frequency spectrum derived from the generator torque signal is also used to detect the generator mechanical faults [45].

4.5. Temperature

Temperature data are mainly used to detect faults within generators, bearings, gearbox and power converters [39]. A cost effective wind turbine thermal model is developed to diagnose the faults through temperature analysis based on the Supervisory control and data acquisition (SCADA) signal [46,47]. The SCADA signal provides reach information regarding faults of the wind turbine system with suitable signal processing methods. An electro-thermal model of DFIG is presented that reduce the cost of fault prevention and diagnosis for the wind turbine system [48]. SCADA data control sets are used to prevent and diagnose wind turbine faults by analyzing of the machine’s temperature [49]. A generalized model is presented based on SCADA data analysis of ambient temperature and wind speed to predict the faults for wind turbine [50]. The performances of the model are compared with a real 1.5 MW DFIG based wind turbine system that shows more effective results than the conventional one. Although the technique is considered reliable and cost effective, its implementation is a bit complex. This technique cannot identify incipient faults and it is hard to identify the root cause and source of the temperature variation. Temperature detected by thermal sensors may rise due to nearby faulty components. Hence, the technique is unable to detect accurate fault locations. In addition, the thermal sensors are quite fragile in harsh environments.

4.6. Lubrication Oil Parameters

Lubrication oil parameters such as viscosity level, water content, particle count and identification, pressure and temperature are measured to detect defects of gearbox and bearing at early stages [44]. The condition of lubrication oil can be monitored in two ways; offline or online [51]. In the offline monitoring, oil sample is taken for condition testing, typically every six months [48]. This may cause interruption to the overall system. In the online oil condition monitoring, oil sensors to detect level, dissolved particles, viscosity and temperature are utilized to reflect the oil condition in real time [48].

4.7. Non-Destructive Testing

Non-destructive testing (NDT) techniques such as X-ray inspection, infrared thermography, ultrasonic scanning, and tap test are used to detect hidden damages in composite materials [48,50,52]. However, implementation of the NDT techniques usually requires expensive instruments.

4.8. Eelectrical Signal-Based Methods

Electrical signal-based methods are widely used to detect various faults due to their distinctive advantages [37]. For example, the magnitudes of certain harmonic components in electrical current signal can be used to detect faults at early stages. Stator and rotor currents and stator voltages are measured to monitor the health condition of the generator [53]. A stator and rotor current based data technique is proposed to identify faults within the doubly fed induction generator (DFIG) [54]. Power signals calculated from voltage and current signals is used to detect rotor electrical imbalance as reported in [55]. Stator open circuit faults of DFIGs is detected using power and current spectra [33]. A mechanical fault or structural defect usually induces vibration of the component that can modulate generator electrical signals. This modulated signal involves fault related information of the mechanical components [56]. The P-amplitude of generator electrical signal indicates the rotor imbalance due to the increased blade surface roughness or yaw misalignment [57]. Electrical power spectral density indicates the reduction of blade stiffness [58]. The feature of bearing failure can be extracted by analyzing the phase and amplitude spectra of the generator’s current signals which can be used to identify the development of bearing failures in an early stage [59]. Electrical signals are also used to detect the fault of gearbox and power electronic converter [56,60]. Compared to other signals, the electrical signal-based condition monitoring methods have significant advantages in terms of ease of implementation, less hardware complexity, less cost, more reliability and potentiality.

5. Signal Processing Methods

5.1. Hilbert Transform

Vibration of a faulty component modulates the electrical, vibration and torque signals and it is quite challenging to extract fault features from such signals to diagnose the faults. Hilbert transform is used to demodulate these signals and extract various faults features [61,62]. The Hilbert transform comprises Fourier transform and empirical mode decomposition (EMD) to generate frequency or time frequency domain spectra [63].

5.2. Envelope Analysis

The envelope of the vibration signal is analyzed to detect bearing faults of the wind turbine system [64]. This technique is able to detect both of the inner and outer bearing faults. In addition, it can detect and predict frets corrosion and assembly damage of bearings in early stages [65]. The technique is based on time domain signal that needs other signal processing methods to be further processed.

5.3. Statistical Analysis

The statistical analysis methods are quite mature techniques that are widely used in the commercial wind turbine system. Appropriate statistical features such as mean value, variance, crest factor, root-mean-square value, skewness and kurtosis are calculated from the base values stored in the healthy condition of wind turbine [66]. The deviations of these features from the reference values indicate faults within the wind turbine (WT). This method can only indicate the occurrence of a fault in spite of revealing the detailed information of the fault location or mode. Moreover, statistical analysis methods are almost inapplicable in high-noise environments and they require large data sets.

5.4. Fast Fourier Transform (FFT)

Fourier analysis is probably the most frequently applied frequency analysis technique in digital systems. The variations of certain harmonic components of the frequency spectrum indicates a specific fault [53,57]. Though the classic FFT has the ability of frequency analysis for stationary signals, it cannot indicate the change in the frequency spectra for a nonstationary signal over time [62]. Therefore, FFT cannot reveal the hidden fault information in a nonstationary signal.

5.5. Synchronous Sampling

A wind turbine usually operates with variable rotating speeds that produce nonstationary vibration and electrical signals. The nonstationary signal characteristics can be converted to constant values through several synchronous sampling algorithms [60], after which classical FFT can be applied.

5.6. Short Term Fourier Transform (STFT)

Short term Fourier transform is used to detect faults such as open circuit and short circuit faults in the generator, rotor imbalance, structure damage in blades and gearbox tooth defects in the variable speed wind turbine system [33,65]. However, the resolutions of time and frequency are limited since the STFT is based on the window method.

5.7. Wavelet Transform

The wavelet transform divides a signal into different scale components with each assigned frequency. It is applied to monitor bearing failures in the generator and the gearbox [67,68,69]. This technique can also be used to detect the rotor electrical unbalance in an induction generator and rotor mass unbalance in a synchronous generator. Similar to STFT, wavelet transform is also restricted to time and frequency resolutions.

5.8. Model Based Methods

Accurate mathematical models are constructed to simulate the dynamic behaviors of a wind turbine [63]. The methods do not need high resolution signals [49,70] which removes the need for data acquisition hardware and installing additional sensors. However, it is quite challenging to design an effective model to mimic real-world applications.

5.9. Bayesian Method

The Bayesian method is used to predict the remaining useful life as well as faults of wind turbine blade [71], bearing [72], lubrication oil [73]. The accuracy of the methods mainly depends on the size of data samples and the availability of history of previous tests. This highlights the shortcoming of the real world application of this method due to the lack of prior data samples.

5.10. Artificial Intelligence Applications

Artificial intelligence (AI) is widely used to analyze condition monitoring data of a wind turbine [74]. This includes artificial neural networks (ANNs), expert systems, space vector modulations (SVMs), and fuzzy logic systems. ANNs are used to diagnose the fault of different wind turbine components such as generators, gearbox, bearings, and power electronics. The ANN based methods are time consuming and require a large amount of data to cover all possible conditions, which makes them impracticable.

In addition to the above condition monitoring techniques, several control techniques have been adopted to enhance the overall performance of the wind energy conversion systems [75,76,77], which is beyond the scope of this paper.

A summary of WT Faults, Signals and Signal processing methods is listed in

Table 1. Summary of Faults, Signals and Signal processing methods.

Component/Subsystem	Faults Occurred	Signals Monitored	Signal Processing Methods
Rotor	Fatigue Crack Surface roughness Asymmetries Reduced stiffness Deformation	Vibration AE Strain Torque Electrical NDT	Hilbert transform Synchronous sampling FFT Wavelet transform Model based method AI
Gearbox	Gear tooth abrasion Tooth crack Tooth breakage Tooth fracturing	Vibration AE Torque Electrical Temperature Oil parameters	Hilbert transform Synchronous sampling FFT Statistical Model based method AI
Bearing	Surface roughness Fatigue, Crack, Breakage Outer/inner race Ball Cage	Vibration AE Electrical Temperature Oil parameters	Hilbert transform Synchronous sampling FFT Model based method AI
Main shaft	Corrosion Crack Misalignment, Coupling failure	Vibration Torque Electrical	Synchronous sampling Hilbert transform FFT
Hydraulic system	Oil leakage Sliding valve blockage	Pressure level	Threshold comparison
Mechanical Brake	Disc/caliper wear Disc crack Hydraulic section failure Motor failure	Vibration Temperature Electrical	Statistical FFT Model based method Threshold comparison
Tower	Corrosion Crack Structural damage	Vibration	FFT Model based method
Generator	Open/short circuit Insulation damage Imbalance Rotor bar broken Bent shaft Bearing failure Air gap eccentricity Magnet failures Rotor mass imbalance	Vibration Torque Temperature Oil parameters Electrical	Synch. sampling Hilbert transform Envelope Statistical FFT Wavelet transform Model based method Threshold comparison AI
Power Converter	Capacitor PCB Semiconductor	Temperature Electrical	Statistical Model based method Threshold comparison AI
Sensors	Sensor Data processing hardware Communication Software malfunction	All related signals	Statistical Model based method AI
Control system	Sensor Actuator Controller Communication Software malfunction	All related signals	Statistical Model based method AI

[78] while

Table 2. Summary of different signals of condition monitoring for wind turbine (WT).

Signals Monitored	Advantages	Limitations
Vibration	Mostly dominated technique Reliable Ability to detect incipient fault Standardized (ISO10816)	Expensive Wiring complexity Intrusive Difficulties of sensor’s installation Subject to sensor failures Inability to detect incipient faults
Acoustic Emission	Able to detect early-stage fault High signal-to-noise ratio Good for low-speed operation Frequency range far from load perturbation	Expensive Complex Very high sampling rate required
Strain	Ability to detect small structural changes Requirement of low sampling frequency Consistent performance over transmitting distance	Complex Intrusive Expensive
Torque and Bending moment	Direct measurement of rotor load No requirement of sensors Cost effective	Intrusive
Temperature	Cost effective Reliable Standardized (IEEE 841)	Embedded temperature detector required Inaccurate fault locations Sensors are quite fragile in a harsh environment
Lubrication Oil	Early stage fault detection Direct characterization of bearing condition	Limited to bearings with closed-loop oil Supply system Expensive for online operation
Non-Destructive Testing	Ability to detect hidden damages in composite materials	Requires expensive instruments
Electrical Signals	Widely used technique Early stage fault detection Less hardware complexity Cost effective More reliability No additional sensor needed Non-intrusive Easy to implement	Displacement based rather than force based Difficult to detect incipient faults Sometimes low signal-to-noise ratio

presents the details of different signals.

Table 3. Comparison of different signal processing methods for WT condition monitoring.

Signal Processing Method	Function	Non-Stationary Signal Processing	Early Fault Prediction	Data Complexity	Resolution
Hilbert transform	Signal conditioning	Yes	Possible	Medium	High
Envelope Analysis	Feature extraction	Possible	Yes	Low/Medium	High
Statistical Analysis	Feature extraction	Possible	Possible	Low	Rely on input
FFT	Feature extraction	No	No	Medium	High
Synchronous sampling	Signal conditioning	Yes	Possible	Low/Medium	High
STFT	Feature extraction	Yes	Possible	High	Medium
Wavelet transform	Signal conditioning	Yes	Possible	Low/Medium	Medium
Model based methods	Feature extraction	Possible	Yes	Medium/High	Rely on input
AI	Feature extraction, diagnosis, prognosis	Possible	Possible	High	Rely on input
Bayesian Methods	prognosis	Possible	No	High	Rely on input

presents a comparison of different signal processing methods for WT condition monitoring.

6. Conclusions

Due to the global trend in generating electrical energy through renewable energy resources, especially from solar and wind, the reliability of such systems has become a priority for the transmission and distribution networks’ stakeholders. As such, reliable condition monitoring and fault diagnosis techniques have to be adopted to detect any incipient faults in these systems to avoid potential consequences. This paper presents a comprehensive review on common faults, signals and signal processing methods for condition monitoring and fault diagnosis of wind turbine systems. Advantages and limitations of each technique have been highlighted. The review shows that there is still a significant research gap in developing comprehensive, cost effective, online condition monitoring techniques for wind energy conversion systems. This may be attributed to a lack of field experience and the complexity of developing such techniques for wind turbines, which are usually installed in remote locations and operate in harsh environments. With the advancement in power electronics, communication technology and sensors accuracy, reliable online condition monitoring for wind turbines will be viable in the near future.