Standard Article

You have free access to this content

Wind Turbines

Other Applications

  1. Goutham R. Kirikera1,
  2. Mannur Sundaresan2,
  3. Francis Nkrumah2,
  4. Gangadhararao Grandhi2,
  5. Bashir Ali2,
  6. Sai L. Mullapudi3,
  7. Vesselin Shanov4,
  8. Mark Schulz3

Published Online: 15 SEP 2009

DOI: 10.1002/9780470061626.shm116

Encyclopedia of Structural Health Monitoring

Encyclopedia of Structural Health Monitoring

How to Cite

Kirikera, G. R., Sundaresan, M., Nkrumah, F., Grandhi, G., Ali, B., Mullapudi, S. L., Shanov, V. and Schulz, M. 2009. Wind Turbines. Encyclopedia of Structural Health Monitoring. .

Author Information

  1. 1

    Northwestern University, Center for Quality Engineering and Failure Prevention, Evanston, IL, USA

  2. 2

    North Carolina A&T State University, Department of Mechanical and Chemical Engineering, Greensboro, NC, USA

  3. 3

    University of Cincinnati, Department of Mechanical Engineering, Cincinnati, OH, USA

  4. 4

    University of Cincinnati, Department of Chemical and Materials Engineering, Cincinnati, OH, USA

Publication History

  1. Published Online: 15 SEP 2009

1 Introduction

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

With increasing costs associated with fossil-based energy, there is large interest in deploying small and large wind turbines (WTs) for generating distributed power for residential and rural areas [1]. It is estimated that a small but significant part of the energy needs of the United States could be obtained through large-scale deployment of WTs [2]. WTs are deployed in rural areas where access to these structures is sometimes not feasible because of the elevated positions. Damage to the WT could cause large financial losses including loss of energy tapped during this outage; see Migliore et al. [1]. Structural health monitoring (SHM) of WTs is needed to ensure safety and to avoid overdesign of components. Overdesign will not allow maximum power from the wind to be captured. This article is a review of the current literature describing techniques for health monitoring of WTs. The main focus of the article is on monitoring the turbine blades, which may be the most critical component of the turbine. Continuous health monitoring of the drive train (gearbox and bearings) and the wind turbine blade (WTB) will ensure proper performance of the WT. Bently Nevada Inc. has developed a commercially available module for continuous monitoring of the drive train [3] of WTs. Monitoring bearings and other components may be done using accelerometers. Since the frequencies of vibration and operation are low, the data-acquisition problem is simplified as a consequence of which allows a larger number of accelerometers or other sensors to be used. Similarly, continuous health monitoring of the WTB is needed. The design of the WTB is a critical factor in the performance (power production) and reliability of WT systems. The trend in blade design is toward improving the aerodynamic efficiency of the blade thus necessitating higher strength-to-weight-ratio materials such as carbon composites owing to their thinner airfoil profiles. A low-cost continuous health monitoring system could provide critical information about the location and propagation of damage in WTBs, and provide predictive maintenance information prior to a blade becoming unsafe. If a blade fails, the rotor can become unbalanced and cause the blade to impact the turbine tower, which can severely damage the turbine drive train. It is common for blades made of composites to have sudden audible acoustic emissions (AEs) during damage growth. AE is defined as the class of phenomena whereby transient elastic waves are generated by a rapid release of energy from a localized source or sources of damage; see Wells et al. [4]. Acoustic emission techniques (AETs) are often used to locate damage and to detect the growth of cracks during qualification testing of WTBs. Use of AE and other techniques for SHM of blades is discussed in this article.

Several researchers have attempted in the past to monitor WTBs for damage in a laboratory setting using various techniques. The simplest technique is to use strain gauges to detect high strains that could indicate damage. In general, too many strain gauges would be needed to have a high probability to detect small damage before failure could occur. Another approach is to use AE sensors. These are typically heavy barrel-type sensors that would be difficult to use in practice owing to the size and requirement that a large number of sensors and channels of high-rate analog-to-digital (A/D) data conversion would be needed to monitor the blade. Typically triangulation is used to locate damage similar to how the epicenter of an earthquake is located. Strain gauges and AE methods are passive methods wherein no artificial excitation of the structure is used, and the structure must be operating for damage to be detected. Active methods are another approach for damage detection. In active methods, a diagnostic waveform or some other form of artificial excitation is used to probe the structure for damage. Sundaresan et al. [5] used an actuator to pulse the WTB intermittently during quasi-static loading and store the received waveforms. The current data was compared with baseline data to identify damage. Migliore et al. [1], Dutton et al. [6], Joosse et al. [7], and other researchers performed testing on WTBs using AETs. AE “threshold-based” arrival systems can be used to approximately locate damage. An advantage of AE methods is that knowledge of the wave speed is not required to detect damage. Wave speed is a function of material properties and the geometry of the structure. WTBs consist of multiple anisotropic materials and have complex geometric features. An advantage of AE methods is that they can detect damage on complex anisotropic structures like the WTB shown in Figure 1. On the other hand, the following disadvantages are inherent in existing conventional AE-based systems: (i) each sensor requires an electrical circuit containing a preamplifier that increases the overall mass of the sensor; (ii) each sensor's output must be converted into a digital format using a high sampling rate A/D converter; (iii) a large amount of data is obtained, which requires rapid real-time data collection, storage, and processing; and (iv) the cost of the overall health monitoring system increases because of the above factors.

thumbnail image

Figure 1. Wind turbines in action and cross section of a typical blade. [Reproduced from Ref. 8. © Derek Berry, 2004.]

With the improvement in the field of microelectromechanical systems (MEMSs), the mass of the sensor and associated electronics can be mitigated. Recently, a structural neural system (SNS) has been developed to overcome some of the limitations of current AETs. The SNS is a signal-processing system that emulates how the human body processes signals from large numbers of highly distributed neurons and receptors (sensors). The SNS uses piezoelectric sensors bonded onto the blade. The surface-mounted sensors are easy to install, and can be repaired, retrofitted, or updated, and the sensor cannot degrade the integrity of the blade. These sensors are highly sensitive and can detect AE stress waves caused by small damage propagating in the blade, as shown by Joosse et al. [7]. The SNS is based on receiving AE signals and is a passive health monitoring method since no artificial means of exciting the structure is used. The frequency of the AE signal that is sensed is on the order of hundreds of kilohertz depending on the blade material and geometry. The SNS architecture was installed on a 9-m-long WTB that was tested to failure during quasi-static loading. Results of detecting and locating growing damage are summarized in this article. Information on testing of blades using conventional AE methods can be found at the US National Renewable Energy Laboratory (NREL) web site [2]. Information on testing different sensor types on WTBs for health monitoring and control is at present available through Sandia National Labs [8-11]. Nondestructive evaluation (NDE) and SHM of WTs is discussed in [9-36]. WT failures that have occurred over the past 30 years and safety concerns are briefly discussed next to help designers decide how practical SHM techniques should be developed.

1.1 Wind Turbine Failures and Safety

To make WTs widely accepted, the public must be assured that WTs are reliable, safe, and benign [37-40]. This section provides an overview of WT failures to give designers an idea of how SHM systems might improve reliability and safety of WTs. WT failures are not well documented and data is not available for new designs. An unofficial and not comprehensive summary of WT accidents from 1975 to November 30, 2007 is given in [39]. The total number of accidents was 403. These resulted in 49 fatalities, of which about 35 involved wind industry workers. Fourteen accidents were public fatalities. The most common cause of fatalities was falls from turbines. Human injury occurred in a further 18 accidents. Blade failure is the most common failure mode on turbines. A total of 118 separate incidences of blade failure were reported in the study. Blade failure may result in whole blades or pieces of blade being thrown from the turbine. Fire is the second most common accident cause. Fire can arise from a number of sources including electrical malfunction and lightning strikes. Turbine fires cannot usually be stopped because of the turbine height, unless there is an onboard fire suppression system. Burning debris from a turbine may be scattered causing a wider-area fire risk. Structural failure is the third most common accident cause as per the data in [12]. Structural failure is assumed to be a major component failure usually owing to high wind gust exposure. Ice falling or throwing is another mode of danger. Transport accidents involve turbine sections falling from transporters including at sea. Driver distraction by turbines, thrown ice, and blade pieces landing on the road have also caused accidents. Some cases of environmental damage including bird and bat deaths have been reported. Other types of damage and accidents are owing to component malfunction, hail, and lightning strikes. Some of the possible indirect problems caused by WTs include interference with TV or microwave reception, depreciating property values, WT noise, increased traffic, road damage, rotating shadows from the blades, aesthetics, concerns about electrical danger, and increased lightening strikes.

Failures of WT components [36] on a percent basis are electrical control 13%; gearbox 12%; yaw system 8%; entire turbine 7%; generator 5%; hydraulic 5%; grid 5%; blades 5%; brakes 3%; entire nacelle 1%; mechanical control 2%; air brake 2%; axle/bearing 1%; other components 30%; and the tower, foundation, hub and coupling <1%. Large WTs are equipped with a number of safety devices to ensure safe operation during their lifetime. Existing safety devices/approaches include vibration analysis, oil analysis, component temperature measurement, thermographics, shaft alignment, strain measurement, acoustic analysis, photo/thermo elastic analysis, electrical effects, overspeed protection, aerodynamic braking systems including tip brakes and mechanical braking systems, and visual and aural inspection. Difficulties that are to be overcome in order to install detailed SHM systems on WTs include the following: the requirement that they should last 30 years without causing false positives for failure, the cost to monitor data from the turbine, and the difficulty in servicing the SHM system. Research in SHM of WTs is being conducted in national labs [2, 11, 17, 33, 35-38], industry [3, 8, 36, 39], and universities [4, 7, 13, 21-28, 31, 34]. In particular, the Danish WT industry is a leader in commercialization. Twenty per cent of Danish domestic electricity production comes from wind. Based on the above data, SHM systems could be applied to the major areas where failures occur, such as blades and system components, to increase the safety and reliability of turbines. The SHM system would identify degraded parts in time for replacement to prevent failure, and this could prevent injuries that occur owing to failure of components or repair of failed turbines. Approaches for SHM of WTBs are discussed next.

2 Methods for Damage Assessment and Prognosis on Wind Turbines

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

This section discusses possible new methods of SHM of WTs [12, 13, 40]. Several different methods are needed to monitor the entire turbine because of the variety of mechanical components that are present in the turbine.

2.1 General Methods for SHM of Wind Turbines

There are many methods of SHM being developed for a large number of applications. General characteristics of these methods are presented in order to help developers consider candidate techniques for specific WT applications. Monitoring the blades in a rotating system presents the problem of data transfer from the rotating frame to the fixed frame for all the methods discussed. A summary of the main methods is given in Table 1. It is anticipated that a medley of sensors and methods will be the best approach to monitor WTs. A fully integrated SHM system that can use different damage detection techniques and types of sensors would be the ultimate goal. Recent papers in the area of SHM (2002-up to the present) are published in Structural Health Monitoring: An International Journal [14]. This journal has a fairly comprehensive list of techniques for SHM. Other references on SHM are given in [9-11, 15-35] and can be found in [41-54].

Table 1. General Methods for SHM
MethodSensor/actuator typeDescription of the method
VibrationAccelerometer, piezo, or MEMSThe natural frequencies of the blade can be monitored for changes indicating damage. Small damage is difficult to detect
StrainFoil strain gauge or fiber-optic cableStrain can be monitored at critical points in the blade and other components. It is difficult to measure strain inside the composite and to have enough gauges to detect small damage. Fiber-optic Bragg gratings can provide a large number of low-bandwidth strain measurements. Passive method [33]
Ultrasonic wave propagationPiezoelectric waferGood for monitoring uniform sections or hot spots. Need predamage reference data that may vary owing to environmental changes or sensor aging. Can detect damage when the blade is operating or not operating. Commercial systems are well along in development for several applications. Active method [12, 34]
Smart paintPiezoelectric or fluorescent particlesPaint changes color when damaged. Visual technique that is low cost but not automated for remote applications. Passive method
Acoustic emission conventionalAE wideband barrel sensorCan detect damage in complex structures. Fretting can cause false indications of damage and the turbine must be operating. Sensors are large and many are required. Fast multiplexing is simplifying the hardware requirements. MEMS AE sensors are reducing size and weight of the system. Passive method
Structural neural system (SNS)Piezoelectric wafers or other sensor typesOvercomes problems of conventional AE by using piezoelectric wafer sensors and biomimetic highly distributed massively parallel signal processing. Multistate sensors (nanotube thread, pressure, temperature, etc.) can also be used. Passive method. Active SNS has also been tested and uses a simple method of neuron firing to detect damage in the passive and active systems [13]
ImpedancePiezoelectric waferHigh-frequency method that can detect local damage. Active method that uses an impedance analyzer and diagnostic signal; Inman [34]
Laser vibrometryScanning laser Doppler vibrometerDifficult to measure the rotating blade which is also changing orientation, and the cost is high. Good for characterizing the blade in the lab. Active or passive
Impedance tomographyCarbon nanotube or other conductive fillerThis method uses carbon nanotubes or other conductive particles to make the blade material electrically conductive. The impedance between arrays of electrodes is monitored to detect damage. The size of damage the method can detect depends on the size of the electrode patterns, and the method is passive and simple and does not require the structure to be operating.
ThermographyInfrared cameraThis method uses an infrared camera to map the temperature of the structure. Damage can produce a local temperature rise owing to crack and delamination breathing. The method is expensive and difficult to use on a rotating system and it is difficult to detect interior damage
Laser ultrasoundLaserA laser beam pulse excites the structure and also measures the response of the structure. Like the NDE ultrasound technique but noncontact. Difficult and expensive to use in the field and minor damage to the surface of a composite may occur by the laser excitation. Good for automated NDE of metals.
NanosensorsElectronic particleThis future approach uses small particles embedded in the blade during the fabrication process. The sensors are transceivers that reflect an RF signal which changes owing to very local damage [13]. See Nanoengineering of Sensory Materials for further information
Nonlinear dyn., condition monitoring, etc.Accelerometers, strain gauges, othersA wide variety of specialized methods for SHM are described in [14] by Todd, Adams, Zimmerman, Chang, Farrar, Pai, Peairs and Inman, and many others
Buckling health monitoring (BHM)PZ T patchesBHM has received little attention in the literature but is important for wind turbine blades and other structures such as radomes, civil infrastructure, and tower structures. Some work has been done by Frank Pai using the nonlinear finite element code Geometrically Exact Structural Analysis (GESA) and by Sundaresan using wave propagation

Most of the SHM techniques and sensor types listed in Table 1 are discussed in other articles of this encyclopedia. However, there is not much information on application of these techniques to SHM of WTs. Several methods that have been applied to WT SHM include a scanning laser vibrometer [32], the fiber-optic method [33], vibration and other methods [34], and the impedance method [35]. The SNS, in particular, was developed for SHM of WTs and is not discussed elsewhere in the encyclopedia. Recent results are also available using the SNS to monitor damage on a WTB in a proof test. Thus, an overview of the SNS method and its testing are given in this article.

2.2 Introduction to the Acoustic Emission Technique as an SHM Tool

The AET has been used as a nondestructive testing (NDT) technique for several decades and is fairly well developed with commercially available instrumentation and standardized testing procedures [15]. The AET relies on the dynamic release of elastic strain energy as damage grows within materials under stress. The released elastic energy propagates through the structure in the form of guided waves. AE sensors suitably located on the structure can detect these signals, as shown in Figure 2. The requirement of loading the structure and the need for the growth of damage for the AET to evaluate the structure separates this technique from other NDE or SHM techniques, which frequently do not need the damage to be propagating under load. This very same requirement also provides the AET the potential to directly quantify the damage severity possibly in terms of crack growth rate or the remaining fatigue life, and makes it suitable for health monitoring of structures in the field. However, one of the major shortcomings of the traditional AET is its susceptibility to false positives. False positives can be triggered by a large number of extraneous signals including mechanically induced noise signals such as friction of mating surfaces (fretting) and hydraulic noise, as well as radio frequency electrical noise. With the availability of new types of sensors including MEMS devices, miniaturized electronics, advanced signal-processing techniques, and pattern recognition, it is likely that false positives will be reduced significantly.

thumbnail image

Figure 2. Conventional acoustic emission parameters extracted from a waveform.

Health monitoring of WTBs is currently practiced only during laboratory tests and certification tests [7, 16-18]. Such tests have been monitored using multichannel AE monitoring systems during repeated static loading to meet the certification requirements as well as during fatigue loading. Weak blades have been successfully identified by the AET during repeated ramp loading to the normal operational load of the blade, or a slightly higher level, followed by constant load hold periods and subsequent unloading. Composite materials that make up the wind turbine blades are known to be profuse sources of AE events and the difficulty of evaluating the blade lies in separating AE signals that are related to the critical damage events from the noncritical damage events. During these blade tests, it is assumed that good quality blades would emit AE signals during the load ramp-up and remain quiet during the hold period. When AE signals continue during the hold period, it is assumed to be a sign of active damage growth when none is expected and hence is an indication of a weak blade.

In addition to the AET, other techniques that have been considered for the SHM of WTBs include vibration monitoring using accelerometers and fiber-optic sensors [18]. Since the critical damage sizes in WTBs are relatively large and may introduce measurable changes in the frequency and mode shapes, accelerometers are considered a viable option. Microbend-type fiber-optic sensors are suggested for monitoring bond lines in WTBs. Remote monitoring of rotor blades in large offshore WTBs [18, 33] is considered to be both economical and technically feasible, and it is estimated that the cost of the additional expense related to the SHM instrumentation will be recouped within 3–8 years of the operation of the turbine. The large WTs (with capacities approaching 5 MW for a single turbine) that are being designed now are among the largest composite structures ever built. The initial investment and the reliability requirement of these blades are sufficiently high to justify the development and incorporation of SHM capability in these blades.

The transition of the techniques that have been used to monitor structural tests in the laboratory to actual SHM on operating turbines will require considerable effort. Embeddable sensors, instrumentation, and information transfer are needed. The design of the SHM system will have to consider the unique requirements of the current and next generation WTs including fabrication techniques (often on-site), material failure behavior [18, 34], the load spectrum, environmental conditions, and economics. Figure 3 shows a schematic of a WTB embedded with sensors and electronics that can potentially monitor AE signals, vibration characteristics, and blade deformation. Moreover, there is recent interest in using sensors on blades for control purposes. Thus, sensor systems on turbines may serve dual uses, for SHM and also for feedback control of the turbine.

thumbnail image

Figure 3. Schematic illustration of a blade with sensors for SHM under field conditions.

2.3 Damage Assessment and Prognosis using the AE Method

Newer WTBs are made from several different materials and are tailored to harvest the maximum energy from the wind through bending-induced twist coupling and other means. Simultaneous achievement of good dynamic properties, light weight, and structural integrity is a challenge. Both gradual as well as sudden changes in the material properties are anticipated during the service life. Gradual reduction of properties could be owing to normal fatigue, ultraviolet (UV) radiation, hygrothermal effects, erosion, etc., while the sudden changes could be owing to lightning strike or gust loads. In addition, blade failure could result from local buckling, which is one of the most common modes of failure. An SHM technique should be able to recognize and address the interaction between these different causes of damage evolution. In addition, the sensors integrated into the blade could be utilized to assess the initial condition of the blade after the fabrication and establish the baseline responses useful for tracking the evolution of damage in the blade. A combination of sensors capable of addressing different aspects of damage evolution is likely to be successful. Embedded wireless resistance strain gauges and accelerometers will be inexpensive approaches to track the gradual degradation of elastic moduli and the blade's natural frequencies. Power harvesting from vibration of the blade is an approach to power the wireless sensors.

The AET offers a real-time SHM capability, which is particularly useful to identify sudden increases in the damage growth and hence is likely to be particularly useful in the prevention of catastrophic failure. AE-based SHM has the advantage of directly assessing the damage growth rate unlike other techniques that attempt to measure the damage size. Recently, bondable “continuous sensors” with individual nodes having wideband characteristics and sensitivities comparable to commercial resonant frequency AE sensors have been developed. This continuous sensor and the array sensor with multiple sensor nodes interconnected and with one signal output offer a means of simple and inexpensive monitoring of large WTBs with minimal hardware and signal processing.

The continuous sensors were shown to perform well in the fatigue life prediction of individual composite specimens and for their life extension. This approach of predicting the fatigue life of individual components has been tested on four different groups of specimens with different combinations of geometry, material properties, and loading conditions. Figure 4 shows a schematic representation of the service life of a structural component experiencing a spectrum loading, gradual aging, and encountering adverse events such as impact or lightning strike. While it is not possible to prevent damage to the blade owing to such unexpected events, it is possible to assess the combined effect of these factors and prevent further steep degradation in structural integrity by limiting the operational envelope of the turbine.

thumbnail image

Figure 4. Schematic representation of load, adverse events, and aging of wind turbine blades.

Figure 5 shows a portion of the life of the turbine simulated using laboratory specimens. A group of 30 woven fabric composite tensile specimens were subjected to varying levels of impact damage followed by simulated spectrum loading with load spikes. For these specimens, AE signals were collected only during the load increment. The specimens were subjected to fatigue load with peak stress at 50% of the average static strength of these specimens. As indicated by the circles near the left dashed line in Figure 5, the cumulative AE energy during the incremental load correlated well with the subsequent fatigue durability of these specimens. A subset of this group was selected to verify if this prognosis could be used to identify and separate specimens that are bound to fail prematurely so that they could be used in less demanding services. The predicted lives of these specimens, at the load amplitude employed for the rest of the group, are shown as black triangles in Figure 5. To simulate a less demanding service life, the fatigue load was reduced from 50 to 40% of the average static strength. The distribution of fatigue lives at this reduced load amplitude is shown as circles near the right dashed line in Figure 5. The reasonable correlation between AE data and extended fatigue life for these specimens appears to support both the validity of AE-based prognosis and the opportunity for optimal utilization of available WTBs.

thumbnail image

Figure 5. Correlation between AE prediction and actual fatigue life of woven fabric composite specimens subjected to impact load. The left dashed line represent the 50% static load case, the right dashed line represents the 40% static load case.

2.4 The Structural Neural System

Composite structures deteriorate gradually owing to operational effects and aging, and owing to unpredictable events such as impact and lightning strikes, and contamination from the environment. It is important to monitor the condition of large systems to prevent catastrophic failure owing to the combination of all the deterioration effects. Because multiple physical states exist in systems, the use of different types of sensors simultaneously may be a more accurate way to identify degradation, contamination, and damage before it reaches a critical size or level. From a practical standpoint, conventional methods of sensing and signal processing are often too expensive, heavy, and complex for comprehensive in situ monitoring of large systems, such as WTs, where degradation or chemical contamination (e.g., owing to rain erosion, acid rain, water ingestion, UV degradation, and volcanic ash in the air) can occur anywhere in the system.

Approaches for structural damage detection that use a large number of individually wired sensors, or storing large sets of predamage data, or using amplifiers to generate diagnostic waves, or performing complex signal processing may not be feasible for WTs. Future sensor architectures ideally would produce information based on ambient conditions, rather than using complex analytical models and predamage data to diagnose degradation and damage. To develop an efficient health monitoring system, we can gain inspiration from the human neural system [19, 20]. The human body is composed of complex anisotropic heterogeneous materials. To monitor its health, millions of receptors and excitatory and inhibitory neurons are distributed throughout the body, which simultaneously process signals in an efficient hierarchical way. This section describes how the functional capability of the human neural system can be mimicked using electronic components. Continuous sensors attached to a signal processor form a neuron, which is analogous to the biological cell body or soma. The processor performs thresholding, firing, and inhibition to measure waves or strains that are caused by growing or breathing damage in a structure, or the change in impedance of a chemical sensor, or by measuring other physical states of the system. The receptor neurons can be designed to sense many types of physical response thus opening the door for many important applications on WTs and other structures.

Description of the SNS

The SNS is a biomimetic signal-processing system, which uses a multiple state continuous sensor network for health monitoring of large composite and metallic structures and other large systems. Passive sensing based on AE monitoring is used to detect damage such as fiber breaking and delamination in composites. Parallel signal processing inspired by the biological neural system is used to combine continuous sensors in a grid pattern into four channels of data acquisition to locate damage and capture the sensor signal. The SNS is a generic biomimetic signal-processing architecture designed to simplify health monitoring of large structures. In Figure 6(a), each vertical line indicates a column neuron and the horizontal line represents a row neuron. The neurons are not connected to each other. Figure 6(b) shows the general architecture of the SNS. The neurons are a continuous sensor connected to an analog processor at the end [21-28]. A continuous sensor also called a neuron is a series connection of many individual sensors with only one output signal per neuron. The output of the neurons is represented by V1 through V20 in Figure 6. V1–V10 are column neurons and V11–V20 are row neurons. Thus, a large sensor grid is formed wherein a reduction in the required number of channels of data acquisition is achieved by processing the outputs of only the neurons that produce anomalous signals, and keeping a track of which neurons are firing using the structural neural system analog processor (SNSAP). The output of the first 10 channels (column neurons V1–V10) is the input to the SNSAP 1, which reduces the required number of data-acquisition channels to only two using firing of the neurons. One of the channels (C1) predicts the location of damage based on firing of the neurons. An algorithm is used to decode the firing signals and tell which neurons are firing. The other channel (T1) is the actual time response of the neuron and is used to qualitatively predict the severity of the damage. The same procedure follows for the row neurons (V11–V20) using the SNSAP 2 processor and the outputs C2 and T2. Thus, with an arbitrary number of sensor nodes, only four channels of data acquisition are required to predict the location and estimate the severity of the damage within a grid of sensors. Since each neuron can have 10 (or more) piezoelectric sensor nodes, and there are 20 neurons in this example, signals from 200 sensors are monitored using four channels of data acquisition.

thumbnail image

Figure 6. The SNS for WTBs: (a) detail of sensor connections and (b) sensor array.

The configuration of the electronic processor used to mimic the biological neural system is described in [21-26]. A brief explanation of how the neurons fire is as follows. When a neuron receives an AE signal, the signal is evaluated by the analog processor. If the signal meets set frequency and amplitude characteristics related to damage, the neuron “fires” a unique location signal and also passes the acoustic signal with those of any other neurons that are firing. A combined AE response analog signal and a second analog signal containing information on the locations of damages are sent to a central PC that performs A/D conversion and locates the damage and approximates the magnitude of the damage. Because of the high coverage of continuous sensors (neurons) on the structure, a neuron can be close to any damage site. This will allow the damage signal to be detected before it is attenuated and distorted and this makes signal processing and filtering easier when compared to using conventional AE sensor systems.

An important application of SHM is detecting crack initiation in areas of high feature density such as at joints. It is difficult to place strain gauges or to propagate waves to detect small damage in joints. The SNS is a simple onboard real-time NDE approach that listens for AE signals from damage in joints. There is no other sensor system that enables the simultaneous monitoring of tens or more long continuous sensors using as few as four channels of data acquisition. In comparison to other techniques, the SNS does not need an amplifier, diagnostic waves, or individual wires for large numbers of discrete sensors/actuators, and there is no need for storage of predamage data. The SNS provides in situ simultaneous sensing and intelligence at the sensor level that reduces hundreds of signals into easily interpretable damage information. The SNS is applicable for onboard real-time monitoring of any type of large sensor system in which anomalous events must be detected. The SNS is accurate, simple, miniaturized, lightweight, interior or exterior surface mounted, repairable, redundant, passive, requires low power, and is safe. This meets the objectives of reliable, accurate, and cost-effective operation. The SNS characterizes the severity of the damage right from damage initiation and tells if the damage is serious while the structure is operating.

Damage Location Algorithm

The neurons that are firing are used to locate damage. A combinatorial algorithm in MATLAB is used to decode which neurons are firing at any time. Because of the limited voltage range of the electronics and the limited number of unique combinations of voltages that can be decoded, this approach of identifying which neurons are firing is limited to about 10 neurons. Thus, a digital approach to locate the firing neurons is suggested for future work to allow up to hundreds of neurons to operate on a digital data bus. Four digital data buses and one power supply wire from a computer are expected to be able to transmit signals from hundreds of neurons in the SNS. This approach can make NDE using the AET practical and can allow monitoring of tens of row and column neurons and hundreds of miniature sensors on a large structure.

Developing the SNS

Initially a two-neuron first prototype of the SNS was designed and built. Results of testing the passive SNS two-neuron prototype are given in [24]. Subsequently, an SNSAP was designed with much faster electronic components and also with better electronic shielding. Also, to prove that the system is practical on a larger scale, the two-neuron prototype was extended to a four-neuron prototype. Results of testing the four-neuron prototype on a thin composite plate in a laboratory-controlled environment were reported by Kirikera et al. [25]. Kirikera et al. [26] implemented the SNS on a WTB for identifying the location of propagating cracks during the quasi-static testing. Also, to understand the propagation of Lamb waves on a thin plate, a wave simulation algorithm was developed [25, 27]. The wave simulation algorithm was derived using the modal superposition method and is based on classical thin plate theory. The wave simulation algorithm was extended to excite a composite structure using an actuator at a specific frequency. Long continuous piezoelectric sensors were recently modeled in the wave simulation algorithm to study the ability of continuous sensors, which detect acoustic wave propagation [28]. Testing of the SNS is described in the next section.

3 Monitoring a Wind Turbine Blade During Proof Testing

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

The SNS was used to identify damage initiation and propagation on a 9-m-long WTB during a quasi-static proof test to failure at the NREL test facility in Golden, CO. The spar caps of the blade were constructed using a constant thickness variable width quasi-unidirectional unitary 3WEAVE carbon/Glass hybrid material [55, 56] developed by 3TEX Inc. The shear web is constructed from fiberglass and balsa wood. Balsa wood is also used in the leading and trailing edge and serves as a panel stiffener. A brief summary of identifying damage on the WTB is discussed in this section. For a detailed explanation the reader is referred to the article by Kirikera et al. [22, 26].

WTBs are composite structures with complex geometry and sections that are built of different materials. The 3D structure, large size, anisotropic material properties, and the potential for damage to occur anywhere on the blade make damage identification a significant challenge. During this test, 12 piezoelectric sensors were bonded onto the surface of the WTB and were connected to form four continuous sensors, which were used in the SNS to identify damage. Although 12 sensors monitored the WTB, the SNS produced only two output signals; the first signal identified and located damage and the second comprised combined AE waveforms. Figure 7(a) shows the top view of the outline of the blade with the locations of three load saddles. The load saddles are used for quasi-static testing. Figure 7(b) is an enlarged view of the sensor area on the blade, which is represented by a circle in Figure 7(a). The circled portion has surface-bonded continuous sensors (V1–V4). Twelve piezoelectric wafer sensors are connected in series to form four continuous sensors. Signals from these four continuous sensors are sent as analog inputs to the SNSAP. These signals are further reduced to two channels (T1 and C1). A load profile was applied to the blade using the saddles as shown in Figure 7(c). It is noted that this blade failed at a load above the design load.

thumbnail image

Figure 7. Experimental setup for the blade and SNS sensors: (a) geometry of the blade; (b) sensor locations on the blade; and (c) test setup with three load saddle. (Figure 7c is courtesy of 3TEX Inc. and the National Renewable Energy Laboratory).

Figure 8(a) shows the load profile applied to the WTB and also the response of SNSAP (the T1 and C1 channels in Figure 6). Each continuous sensor was assigned a unique voltage by the response of the C1 channel of SNSAP. A software was used to convert the unique voltages into corresponding neuron numbers, as shown on the ordinate of Figure 8(b). Figure 8(b) indicates the times at which each neuron received the AE signal.

thumbnail image

Figure 8. Test data from the SNSAP: (a) T1 channel of the SNSAP superimposed on the load profile and (b) response of the C1 channel of SNSAP used to locate the damages.

Based on the neuron that receives the AE, multiple damage zones were manually mapped out as seen in Figure 9(a). At the end of the test, multiple damage zones were identified by the SNS (Figure 9a). After the predictions of the damage zones were made based on the SNS results, the blade was cut into sections by the NREL engineers to determine the actual locations of damage. The predicted damage zones were compared with three observed damage locations (Figure 9b) based on cutting the blade into sections. Zones 1A and 1B correspond to damage 1, zones 5, 1A, 1B, and 2 correspond to damage 3, and zones 3A and 3B correspond to damage 2. Zone 4 as predicted by the SNS was not identified in the visual analysis. On subsequent discussions with NREL engineers, it was concluded that damage indeed existed at zone 4 and was missed during visual postfailure observations.

thumbnail image

Figure 9. Comparison of the predicted and actual damage locations on the wind turbine blade: (a) damage locations predicted by the SNS before sectioning of the blade and (b) damage locations observed by the NREL engineers after the postfailure sectioning of the blade.

Figure 10 shows the number of AE hits received from each zone. The number of zones is shown in Figure 9(a). On the basis of the number of AE hits, it is concluded that zone 1A has a large portion of the damage compared to zone 1B. Both zones 1A and 1B correspond to the same damage (damage 1 in Figure 9b). Similarly zone 3B has a larger damage than zone 3A as concluded from the visual postfailure analysis of Figure 9(b).

thumbnail image

Figure 10. The number of AE hits produced from each zone.

Damages 1 and 2 were visible on the surface of the WTB, damage 3 was not visible on the surface of the WTB. The blade was cut open to understand the cause of failure. The primary mode of failure was panel buckling [29]. The buckling could have started by either the leading edge of the WTB buckling inward or the trailing edge buckling outward leading to a rotation of the spar cap about its axis effectively peeling the spar cap off from the shear web section of the WTB. The rotation of the spar cap was likely a secondary mode of failure [29]. The out-of-plane movement of the panel led to catastrophic damage on the surface of the WTB, Figure 11. The separation of the spar cap from the shear web caused a bond line failure running from 700 to 1650 mm on the WTB, and is shown in Figure 12. Also, a dye penetrant was used to detect brittle gel coat cracks present on the surface of the WTB. The dye pattern did not correlate with the damage locations. The spar cap remained intact indicating that it had performed its function of being the primary load carrying member, without sustaining catastrophic damage. Also, strain gauges were used on the WTB. The strain data did not indicate the damage progression, but detected the damage just at the onset of buckling failure. Strain gauge locations were not sufficiently close to the failure region to adequately capture the damage.

thumbnail image

Figure 11. Postfailure sectioning showing visible damage: (a) “damage 1” on the trailing edge of the WTB and (b) “damage 2” on the leading edge of the WTB. [Pictures courtesy of NREL. Reproduced with permission.]

thumbnail image

Figure 12. Cut sections of the WTB (a) indicating bond line failure between the spar cap and shear web and (b) showing the interior geometry of the blade. [Pictures courtesy of NREL. Reproduced with permission.]

The SNS indicated the general area where the damage started and how the damage progressed, which is valuable information for verifying and improving the blade design and for verifying the manufacturing procedure. In the future, a grid pattern of sensors (neurons) could be attached inside the blade along the shear web to detect damage inside the blade with greater sensitivity. A major outcome of this testing was to provide confidence that SHM of large anisotropic composite structures that have complex geometry and multiple materials is practical using a simple, low-cost SNS. This test also showed that the SNS can detect cracking that precedes buckling and may be useful for buckling health monitoring (BHM) of large structures.

4 Buckling Health Monitoring Techniques

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

WTBs are hollow tubes designed to have bending twisting coupling and are prone to local buckling. Uncontrolled local buckling can lead to catastrophic failure of individual blades and subsequently the whole turbine. Since buckling initiates as an elastic deformation, the subsequent damage to the structure could be entirely prevented if this tendency is identified in real time and countermeasures are taken. Two different approaches have been examined for identifying the local buckling in WTBs.

The first approach is to propagate low-frequency Lamb waves across the region that has the tendency to undergo local buckling, regions A and B in Figure 13(a), as was used during static testing of a WTB [5]. During this test, a 5-kHz Lamb wave signal was able to indicate the initiation of buckling deformation at 40% length of the blade from the root end. The blade failure at about 4500 lbs (2040.82 kg) load, Figure 13(b), was premature and was caused by this buckling. The initiation of buckling is indicated by the reduction in amplitude of the received Lamb wave signals around 4000 lbs (1814.34 kg) as shown in Figure 13(c).

thumbnail image

Figure 13. Early detection of local buckling in a WTB using low-frequency Lamb wave signals.

The second approach is the monitoring of the natural frequency of the region that has a potential for buckling. This condition was simulated in a 4-in.-wide, 49-in.-long, and 0.125-in.-thick glass fiber epoxy composite strip. Finite element analysis of the vibration characteristics of this bar at various stages of buckling indicated that easily recognizable changes in frequency and mode shapes result even in the early stages of buckling. Experimental measurement of the frequency and mode shapes confirmed the feasibility of this approach. In this experiment, the bar was excited at the third natural frequency corresponding to the unbuckled configuration using a five-cycle windowed sine wave. The oscillation in the unbuckled bar lasted longer than 1 s. The duration of oscillation dropped significantly even in the early stages of buckling when the deformation was less than 0.25 in., providing a clear indication of the instability. Figure 14(a) shows the experimental configuration and Figure 14(b) shows the different time responses of the beam for different stages of prebuckling.

thumbnail image

Figure 14. Detection of buckling through changes in natural frequency.

5 Multistate Continuous Sensors

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

This section describes recent advances in continuous sensor technology that allows different types of distributed sensors to be utilized inexpensively, reliably, and with low weight for monitoring complex structures such as WTs. The continuous sensor architecture is for use in an SNS described above. SNS is an electronic sensor network developed using highly distributed interconnected continuous sensors coupled to a parallel signal-processing system to enable in situ real-time monitoring of WTBs. The generic SNS can use almost any type of passive sensor. Examples include piezoelectric ceramic wafers to monitor AEs due to crack propagation, and carbon nanotube (CNT) neurons (threads or film) to monitor strains, detect large cracks, and to detect electrolytes related to early stage corrosion and hygrothermal degradation. In structural applications, an SNS can detect damage such as composite fiber breaking and tell when impact damage is propagating. SNS is low cost and simple because no predamage data, actuation, or multiplexing is needed. Also, it is relatively insensitive to drift in properties of the sensors owing to environmental effects because reference data is not used. From an operational viewpoint, SNS is more than a monitoring system—it is an artificial central nervous system that monitors the condition of the structure without interruption and identifies when anomalous events and early degradation occur. It also defines the operational performance envelope and provides confidence in terms of safety and reliability of the system. In structures, degradation often occurs at complex built-up or joined sections where the load and acoustical paths are complex. SNS is excellent for use in these areas of high feature density where other monitoring techniques are impractical to use. Multistate continuous sensors extend the usefulness of SNS.

This section describes ongoing development of a suite of multistate continuous sensors, including MEMSs continuous accelerometers, CNT crack and corrosion sensors, pressure, temperature, and other types of sensors for use in an SNS. It reduces signals from many sensors to a small amount of essential information that can be equated to degradation of the system. Different types of sensors can be used with an SNS, but the design of the continuous sensor might differ for these different types of sensors. Moreover, the SNSAP parameters change depending on the type of continuous sensor. Details of filtering and thresholding also will, in general, be different for every sensor type. We are presently developing a suite of multistate continuous sensors for use in an SNS. The goal is to achieve integration of multiple sensor types into a single signal-processing architecture.

Dual use application of the SHM techniques for WT monitoring and other applications should be considered to reduce development cost and to take maximum advantage of technology. Besides mechanical applications such as monitoring WTBs, sensor networks are becoming increasingly important to safeguard the environment and homeland, and for medical applications. SNS has potential application in many types of sensor systems to simplify the hardware by reducing the number of wires and A/D convert boards. SNS is designed to sense anomalous values of a particular state variable over a large area using a continuous sensor. Figure 15 shows some of the individual sensors and sensor materials that can be used to form continuous sensors. The design of different multistate sensors for an SNS is discussed in the following text. It is also important to model the transfer function of the continuous sensor and SNS for new types of sensors. In particular, individual sensor nodes must be integrated properly to form continuous sensors—they cannot be arbitrarily connected. The rows overlap the column neurons and are electrically insulated. The size of the cells in the grid can be small to detect small damage. A description of the different sensor types is as follows.

  1. Carbon Nanotube Continuous Sensor

    This sensor is to detect cracks, delamination, and corrosion damage based on electrical impedance. A prototype CNT-based sensor has been built and cracking and electrolyte representing corrosion have been detected based on the impedance of the sensor. CNT film sensors, highly distributed on the surface or embedded within a composite, are used to form a continuous sensor for crack and corrosion monitoring. An example of nanotube neurons on a simple panel is explained. A highly distributed network of nanotube continuous sensors, which sense along their entire length and have a biomimetic architecture, allows coverage of large structures. An example of the change in resistance as a crack passes through a single neuron is shown in Figure 16(a). There is a small change in resistance as the crack begins propagating through the neuron. The change becomes larger and approaches infinity as the crack passes through the neuron. An example of the change in capacitance as electrolyte is put on a single neuron is shown in Figure 16(b). Up to a factor of 50 increase in capacitance occurs because of the double layer supercapacitance property of nanotubes. The change in resistance owing to the electrolyte is only 6% as shown in Figure 16(b). Therefore, crack and corrosion sensing are mostly decoupled and can be measured using the same neuron. Signal processing using the nanotube neurons is greatly simplified because the electrical properties of the neurons, rather than structural waves, are used to characterize damage.

    A prognostic method for damage modeling can be developed based on the changes in electrical parameters of the CNT neurons. The modeling of the neuron is based on the Randal Warburg circuit. The CNT is a strain sensor (piezoresistive effect) and a highly sensitive corrosion sensor (electrochemical impedance spectroscopy (EIS) effect). A model should be developed to predict the crack length based on the EIS spectra of the nanotube composite. An electrode configuration can also be used on a nanocomposite plate where the nanotubes are dispersed in the polymer matrix. Low-cost carbon nanofibers can replace high-cost CNT for large applications of an SNS.

    An important application of the CNT neuron in WTB is health management of polymer matrix composites (PMCs). The CNT neuron can be developed for service life monitoring of PMCs used WTB structural applications. Two of the primary life-limiting mechanisms in polymer composites are hygrothermal degradation and oxidative degradation. Hygrothermal and oxidative degradation can lead to chemical changes in the resin system causing cracking and embrittlement in the surface layers of the composites. Within the oxidized layer of the composite, the tensile strength, strain to failure, flexural strength, density, and toughness decrease while the modulus increases. Surface cracks provide pathways for the transport of moisture and oxidants to the fiber/matrix interfaces that act as high-diffusion paths thereby increasing the degradation rate. The CNT neuron is a passive continuous sensor that can monitor the electrical impedance of composite materials to indicate degradation of PMCs used in WT structures. Future work should investigate the scientific, technical, and commercial feasibility of the CNT neuron for PMC health management including monitoring oxidative degradation for a neat resin aged in low temperature and oxidizing environments.

  2. MEMS Capacitive Accelerometer

    Accelerometer evaluation boards from Analog Devices Inc. were used to build a four-element continuous sensor. The accelerometers are two axis each and measure in-plane acceleration. The four accelerometers were used to form a continuous accelerometer with four inputs and one output in the x axis and a second continuous accelerometer with four inputs and one output in the y axis. These accelerometers work based on the capacitance of a MEMS cantilever. The accelerometers required a modified circuit to form the continuous sensor to sum and amplify signals and to prevent low-frequency signal feed-through. Initial testing was successful, opening up the use of MEMS continuous accelerometers for vibration measurement, impact detection, and condition monitoring of bearings, unbalance, and damage.

  3. Thermistor Continuous Temperature Sensor

    This sensor is to monitor operating temperatures based on resistance of the sensor. Certain types of damage may cause an increase in temperature. The continuous temperature sensor may indicate hot spots in a structure.

  4. Piezoelectric Lead Zirconium Titanate (PZT) wafers

    These are being tested with a brass backing to sense AEs owing to cracks and delamination in metallic and composite structures. Prototypes were tested in the lab with good results. This sensor is more rugged than PZT wafers without a backing. Studies using the transfer function voltage output of the sensor divided by strain input (V0/S) show that a large number of piezoelectric sensors will not reduce the quality of the data obtained for AE detection. Transfer functions must be determined for other sensor types in the project.

  5. Continuous Strain Gauge

    This sensor can be formed using miniature strain gauges in series or parallel. Equivalent electric circuits can be used to develop a transfer function. The strain gauge continuous sensor will be tested in the project to monitor large strain simultaneously at multiple gauges.

  6. Continuous Pressure Sensor

    This sensor can be built using a piezoresistive or capacitive film sensor typically with applications in biomechanics. The piezoresistive and capacitive pressure sensors are easy to incorporate as a continuous sensor. This sensor has many types of commercial applications.

    Other sensor types include an inclinometer, humidity sensors, proximity sensor, magnetic sensor, flow sensor, and liquid level sensors. Another class of sensors is nanowire sensors. A nickel nanowire neuron continuous sensor is expected to be useful to detect cracks, delamination, and corrosion damage using electrical impedance or eddy current. The Ni nanowires are being produced at University of Cincinnati. Use of multiple sensor types and fusing data is possible using SNS and is an area of future work.

thumbnail image

Figure 15. Sensor elements that can be used to form continuous sensors: (a) nanotube impedance film sensors to detect corrosion, cracking, and delamination for an SNS; (b) MEMS accelerometer on a circuit board; (c) thermistor that can from a continuous temperature sensor; (d) PZT commercial wafers with brass backing used to detect acoustic emissions; (e) strain gauge sensor; and (f) pressure mapping sensor.

thumbnail image

Figure 16. Change in capacitance and resistance of a carbon nanofiber neuron owing to simulated damage: (a) crack propagation damage, the resistance changed from 9.3 kΩ to infinity, the capacitance changed from 44 pF to zero; (b) electrolyte corrosion damage, the resistance changed from 117 to 123 kΩ, which is a 5% increase, the capacitance changed from 19.1 to 1109 pF, a 60 times increase.

5.1 Manufacturing an SNS

Features of an SNS include the following: (i) it is a passive system; (ii) detects anomalous events; (iii) continuous sensors combine 10 or more sensor nodes into a one wire sensor; (iv) highly distributed continuous sensors called neurons can detect small damage on large structures with complex geometry and high feature density; (v) bioinspired parallel signal processing reduces the number of channels of data acquisition from tens or hundreds to four; (vi) sensor modules can be developed for different types of sensors (piezoelectric, piezoresistive, thermistor, strain, magnetic, etc.); (vii) multistate sensing is possible in one system architecture; (viii) sensing AEs to detect cracking is possible using high-bandwidth data-acquisition boards (up to 1 MHz); (ix) sensing vibration using accelerometers is possible using moderate-bandwidth data-acquisition boards (up to 50 kHz); and (x) sensing strain, pressure, temperature is possible using low-bandwidth data-acquisition boards (up to 1 kHz). Overall, SNS is simple, low cost, redundant, and reliable. Other types of SHM systems may also have many of the attributes mentioned above. It is suggested that some type of SHM test bed be developed for WTBs to practically evaluate the different techniques for SHM.

A plan for manufacturing SNS should be developed such that the SNS is available for prototype testing on operating WTs. To a first-order approximation, the cost of an SNS can be quite reasonable (∼$20 000, including the analog electronics, computer, A/D boards, and sensors), and the system can be small and lightweight. Dual use commercialization of SNS should be possible. In military, the technology can be applied to aircraft for structural and engine applications, and for exhaust wash structures to monitor high-temperature PMCs. For commercial applications, the technology can be applied to civilian and military aircraft and nonaerospace applications such as bridges, buildings, rocket casings, and any structure with complex geometry that must be monitored for degradation in real time.

6 Wireless MEMS Accelerometers for SHM in Rotating Systems

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

Existing accelerometers and the data-acquisition systems used for structural analysis and monitoring studies are often cumbersome to use. Therefore, a wireless system using small MEMS accelerometers is being developed. The first stage of this development has been performed successfully at the University of Cincinnati. The accelerometer system was built by integrating commercial MEMS accelerometers (ADXL 278, Analog Devices Inc.) with wireless sensor technology from MicroStrain Inc. The wireless measurement system can be installed on a rotating WTB without being tethered to any external devices or computers. The current prototype system has eight MEMS accelerometers wired to an A/D converter and wireless transmitter (MicroStrain Inc.). The cost is about $3000. This battery-powered system (Figure 17) is in initial testing and continuously streams acceleration data wirelessly to a laptop computer with a bay station. This wireless system cannot store much data locally, but seems suitable for WT use. Battery life is a limitation and power harvesting from blade vibration may be used to charge the batteries. SNS could be adapted to the wireless system.

thumbnail image

Figure 17. Wireless data-acquisition system with eight MEMS accelerometers: (a) MEMS accelerometer/board, battery-powered wireless module, laptop computer with bay station and (b) one channel response due to a shock input. This system was assembled at the University of Cincinnati using commercial components. Continuous accelerometers can be used to allow 32 or more individual MEMS accelerometers to be used with this system.

6.1 Damage Detection and Fatigue Testing of Complex Joints

Detecting and locating cracks in WT structural components that have high feature densities is a very challenging problem and is not discussed much in the literature. In general, few SHM techniques have been applied to the monitoring of joints and complex structural geometries. However, reliable low-cost assessment of joints is crucial to maintain operational availability and productivity, reduce maintenance cost, and prevent catastrophic failure of WTs. SNS and other SHM systems should be deployed on components or sections of composite structures and tested together to understand the capabilities of the different approaches for SHM. SNS could become a standard signal-processing architecture for general SHM systems, where reducing the number of data-acquisition channels and hardware is of critical importance.

6.2 Resources for SHM

An increasing number of publishers, companies, and research organizations are developing and providing information, sensors, and measurement systems that may be used for SHM including for WTs. A partial list of resources for SHM components and systems is given in [41-54].

7 Summary and Conclusions

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

Different techniques for monitoring wind turbines and particularly WTBs are discussed. Overall, we can say that monitoring degradation in WT structures is a complex problem that might be solved using multiple types of sensors and different damage detection techniques. New signal-processing architectures are also needed that simplify the hardware and reduce the cost of the overall health monitoring system. An SNS was developed as a simple approach for health monitoring of WTs. The neural system listens for AEs to detect damage in a WTB. The SNS was tested during proof testing of WTB and multiple damage zones were located. Subsequent postfailure sectioning of the failed blade showed close correlation between the predicted and actual damage locations. An SNS with multistate continuous sensors is also being developed to measure other variables such as pressure, acceleration, and electrical impedance to identify different types of damage. Power harvesting and wireless data transmission from the rotating blades to the fixed frame are other areas where research is needed. It is suggested that a test bed for WTBs be developed to practically evaluate the different techniques for SHM.

Acknowledgments

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References

The work related to the SNS was supported by the NREL under subcontract number XCX-2-31214-01. Mr Alan Laxson was the technical monitor of this project. Much of the help in making the prototype of the SNSAP was provided by engineers from Texas Instruments and Analog Devices. Also Mr Henry Westheider, Instrumentation specialist, and Mr Doug Hurd, Machinist, in the Department of Mechanical Engineering at the University of Cincinnati provided valuable help in building the SNS prototype. The static test of the 3-TEX-100 blade was performed at NREL test facility in Golden, CO. The LABVIEW expertise provided by Dr Vahan Gevorgian of NREL during the on-site testing of the SNSAP on the WTB is gratefully appreciated. Mr Derek Berry and Dr Mansour Mohamed are the principal technical leads from TPI Composites and 3 TEX, Inc., respectively. Mr Scott Hughes and Mr Jeroen van Dam are the NREL engineers who assisted during the WTB testing.

References

  1. Top of page
  2. Introduction
  3. Methods for Damage Assessment and Prognosis on Wind Turbines
  4. Monitoring a Wind Turbine Blade During Proof Testing
  5. Buckling Health Monitoring Techniques
  6. Multistate Continuous Sensors
  7. Wireless MEMS Accelerometers for SHM in Rotating Systems
  8. Summary and Conclusions
  9. Acknowledgments
  10. References