soman et al-2015-wind energy

8182019 Soman Et Al-2015-Wind Energy

httpslidepdfcomreaderfullsoman-et-al-2015-wind-energy 112

RESEARCH ARTICLE

Bi-axial neutral axis tracking for damage detection in

wind-turbine towersRohan N Soman

1 Pawel H Malinowski

1and Wieslaw M Ostachowicz

12

1 Institute of Fluid-Flow Machinery Polish Academy of Sciences 14 Fiszera Street Gdansk 80-231 Poland2 Faculty of Automotive and Construction Machinery Engineering Warsaw University of Technology Narbutta 84 02-524 Warsaw

Poland

ABSTRACT

This work concentrates on Structural Health Monitoring (SHM) of a wind turbine tower A decision level data fusion based

on bi-axial tracking of change in neutral axis (NA) position is proposed A discrete Kalman Filter (KF) is employed for the

estimation of the NA in the presence of measurement noise from the strain sensors The KF allows data fusion from the

strain sensors and the yaw mechanism for the accurate estimation of the NA Any change in the NA position may be used

as an indicator for the presence and location of the damage The ratio of the change in the NA along two perpendicular axes

is taken and used for the localization The study has been carried out on the simulated 1047297nite element (FE) model of the wind

turbine tower and indicates that bi-axial NA tracking based on data fusion is indeed necessary and at the same time is

sensitive to damage The sensitivity studies carried out indicate that the metric is robust enough to overcome the effects

of measurement noise and yawing of the nacelle Copyright copy 2015 John Wiley amp Sons Ltd

KEYWORDS

structural health monitoring (SHM) damage detection wind turbine tower neutral axis (NA) kalman 1047297 lter (KF) data fusion strain

Correspondence

Rohan N Soman Institute of Fluid-Flow Machinery Polish Academy of Sciences 14 Fiszera Street Gdansk 80-231 Poland

E-mail rsomanimpgdapl

Received 28 November 2014 Revised 23 February 2015 Accepted 5 April 2015

1 INTRODUCTION

Wind energy is seen as one of the most promising solutions to manrsquos ever increasing demands of a clean source of energy

But the main drawback of the wind energy is the high initial cost for setting up and maintenance This makes the energy

more expensive than the conventional energy sources like fossil fuels and nuclear and hence it has not been widely

accepted Thus there is a concerted effort to reduce the cost of energy production This can be achieved by increasing

the lifetime of the wind turbines reducing maintenance costs and ensuring low downtime of the turbine The lifetime

may be increased by ensuring a more robust design while the maintenance cost and the downtime of the equipment

may be lowered through the use of condition monitoring and structural health monitoring (SHM) 1 SHM allows early

detection of damage and allows maintenance planning which reduces the cost 2 and can allow us to avoid unnecessary

downtime hence increasing the availability of the system

The SHM needs to be low cost and suitable for continuous monitoring The SHM techniques are based on the concept

that the change in mechanical properties of the structure will be captured by a change in its dynamic characteristics 3

Vibration-based damage indicators have been traditionally used because of their low cost implementation Many vibra-

tional parameters like changes in natural frequency 4 mode shapes 5 mode shape derivatives 6 and modal 1047298exibility indices7 have been suggested in the literature These methods are global level damage detection techniques and are sensitive to

large scale damage only and may not detect local level damage Thus there is an increasing trend to use strain sensors

for local level damage detection Strain sensors are local level sensors and hence more sensitive to smaller levels of damage8 Many strain-based damage detection techniques have been proposed in literature and their performance has even been

compared with displacement-based methods9ndash11

WIND ENERGY

Wind Energ (2015)

Published online in Wiley Online Library (wileyonlinelibrarycom) DOI 101002we1856

Copyright copy 2015 John Wiley amp Sons Ltd



The SHM methods are still not up to the desired standards and as such there is a need for a robust damage detection tech-

nique The discrepancy between the expected and the measured results of the different SHM techniques is mainly attributed to

the uncertainty in the measurement environment with respect to noise temperature and excitation mechanism of the structure

Thus thesearch for an SHM system that will be able to detect small levels of damage butat the sametimebe robustenough to

overcome the ambient noise and temperature changes and detect damage under operational conditions is ongoing

The use of SHM techniques in wind turbines is gaining prominence especially with the use of offshore wind turbines

where costs of unscheduled maintenance are high and downtime results in major loss of revenue Ciang et al 12 and Hyers

et al 13 provide reviews on condition-monitoring strategies for all wind turbine subsystems and components Tower dam-

age is the third most common damages in wind turbines 12 As a result the research in the area of SHM of towers is now

becoming more signi1047297cant Furthermore as seen in Figure 1 the downtime for minor damages in towers (support and hous-

ing) is far less signi1047297cant (014 days) than in the case of major damage (2801 days) So there is a huge economic incentive

for early detection of minor failures through the use of proper SHM techniques for towers

The most commonly used sensors for SHM of towers are accelerometers and displacement sensors15ndash17 These sensors

are suitable only for global damage identi1047297cation On the other hand strain sensors allow a more local damage detection

approach Bas et al and Bang et al make use of 1047297bre Bragg grating (FBG)-based strain sensors for SHM of tower1819

Bas et al 18 have studied the structural response of the tower to various stopping events which may be used as baseline

for detecting deterioration of the tower over a period of time Bang et al 19 have studied the stability of the FBG sensors

over time and their ability for dynamic strain assessment on a real structure Benedetti et al 20 make use of conventional

strain gauges for the fatigue monitoring near welds of the tower The results obtained are quite encouraging but the use of

this method is restricted to known hot spots of potential damage

Thus in this paper a method for damage detection of towers is proposed which is able to detect lower levels of damagethroughout the structure without a priori knowledge of the structural behaviour The methodology proposed is a customiza-

tion of neutral axis (NA) as a damage-sensitive feature for tower structures The NA has been proposed for damage detection

in bridges21ndash23 The application of the methodology directly to tower structures is not possible because of the yawing of the

turbine and the resulting mass and stress redistribution Thus a robust methodology has been proposed to overcome these

application-speci1047297c challenges and further the effectiveness of the methodology for more precise damage isolation

The present research builds on the research work presented by the authors in 24 and 25 In this paper the methodology for

damage detection using NA as a metric is extended to bi-axial neutral tracking using decision level data fusion It was

shown in 24 that the position of NA of the tower structure may be used as a robust damage indicator This is possible as

the NA is the property of the cross-section of the tower independent of the bulk temperature effects and the ambient wind

loading The position of the NA can be assessed by measuring the strains on opposite surfaces of the tower in bending The

NA of the tower subject to unknown loading both in magnitude and direction was estimated using the discrete Kalman

Filter (KF)2627 The authors then extended the methodology to include the effect of yaw of the nacelle on the observability

of the NA in25 But even then only the determination of the altitude of the damage was possible In this study through the

use of bi-axial tracking and decision level data fusion a better resolution of the damage location is possible The study is

undertaken on a 1047297nite element (FE) model of the 10 MW wind turbine tower which is seen as the future of offshore wind

Figure 1 Reliability characteristics of several sub-assemblies14

Bi-axial NA tracking for SHM in towers R N Soman P H Malinowski and W M Ostachowicz

Wind Energ (2015) copy 2015 John Wiley amp Sons Ltd

DOI 101002we



turbine industry28 The study also includes some sensitivity studies in order to establish the robustness of NA tracking as a

damage indicator under different damage locations damage extents and presence of measurement noise

The paper is organized as follows 1047297rstly the concept of NA and the theoretical formulation of the use of NA as a damage

indicator and how it can be estimated by measuring the strain at diametrically opposite locations is explained The section also

explains the theoretical background and the implementation of the KF The next section explains the FE modelling approach

used for the Technical University of Denmark (DTU) Reference Wind Turbine (RWT) 28 and the loading scenarios for the

simulations The section on numerical simulation 1047297rst establishes the need for yaw tracking followed by the need of KF for data

fusion The section on bi-axial NA tracking introduces the decision level data fusion and in turn allows us to locate the damage

with betterresolution andas such is the main contribution of the paper Some sensitivity studies of the indicator to measurement

noise and severity of damage have also been carried out in this section The last section then draws some conclusions based on

the observations made in the simulated scenarios and also points at the area of future research

2 THEORETICAL BACKGROUND

21 Neutral axis

The primary function of the tower structure is to support the hub and the nacelle of the wind turbine The nacelle and the

hub are axial loads that are eccentrically loaded on the tower This eccentric loading gives rise to axial compressive loads as

well as bending loads as shown in Figure 2The axial compression is uniform over the entire cross-section while the bending loads will be tensile at one end and

compressive at the other Furthermore the tower experiences wind loads which result in bending strains in the tower

The axial strains are given by Equation (1) while Equation (2) gives the bending strains

ε axial frac14 F

EA(1)

Figure 2 Flexural strain distribution over the beam cross-section subject to eccentric loading

Bi-axial NA tracking for SHM in towersR N Soman P H Malinowski and W M Ostachowicz


DOI 101002we



where ε axial is the axial strain F is the net axial force due to the nacelle hub and other wind turbine components E is the

Youngrsquos modulus and A is the area of the cross-section

ε bending frac14 plusmn M b y

EI (2)

where ε bending is the longitudinal strain in bending M b is the net bending moment at the cross-section due to wind loading

and eccentricity and I is the area moment of inertia and y is the distance from the NA to the location of the sensor23

Thus one surface of the tower experiences a combination of two axial compressions (right side in Figure 2) while the

other end experiences a combination of compressive load because of the weight and tensile load due to the bending (left

side in Figure 2)

The net strains on the right and left surface in Figure 2 are given by Equation (3) while

ε r =l frac14 ε axial plusmn ε bending (3)

If the line connecting the two strain levels is extended there will be a point where the strain experience will be zero

which is identi1047297ed as the NA point The NA of the section is a function of the 1047298exural rigidity of the structure and does

not depend on the applied bending loads thus by measuring the strains at the opposite edges of the beam the NA can

be located which in turn may be used as an indicator of the damage Figure 2 explains the abbreviations used and the con-

cept The NA can thus be estimated based on the strain measurements

This NA location ( L ) may be found by linear extrapolation and is given by equation

L frac14 εbending εaxial

w

2εbending

frac14 εl w

2 εl ε r eth THORN (4)

It should be noted that the dimensions of the NA location are that of distance so this metric may be represented as a

dimensionless quantity NA estimate (NAE) given by equation

NAE frac142 L

w(5)

The NAE can be calculated at each time instant based on measured strains at that instance using Equation (5) This

approach is termed as direct estimation in further sections or they can be estimated based on the previous estimates and

updated at each time instant using the new measurements using KF

22 Kalman Filter

The KF is a set of mathematical equations that provides an ef 1047297cient computational (recursive) solution of the least-squares

method26 Theoretically KF combines a system rsquos dynamic model (physical laws of motion) and measurements (sensor

readings) to form an estimate of the systems varying quantities (system state) that is better than the estimate of the system

obtained by measurement alone26

The KF allows estimation of the state variable x of a discrete time controlled process governed by the linear stochastic

difference equation In the absence of the control scalar it can be given by Equation (6) 26

x k thorn1 frac14 Ak x k thorn wk (6)

where A is the state transition matrix and wk is the white process noise and k indicates the time step

The measurement equation is given by Equation (7)

zk frac14 Hx k thorn vk (7)

where H is the measurement matrix that relates to the observation zk and vk is the measurement noise

The goal of using KF is to 1047297nd an equation that computes a posteriori state x k k as a linear combination of an a priori

estimate x k k -1 and a weighted difference between an actual measurement zk and a measurement prediction H x k k -1 as shown

in Equation (8)

x k =k frac14 x k =k 1 thorn K zk Hx k =k 1

(8)



DOI 101002we



where K is the Kalman gain The value of K is chosen to minimize the error covariance There are several forms of the K

matrix discussed in literature2930 One form of this matrix used for the gain computation is given by Equation (9)

K k frac14 Pk =k 1 H T k H k Pk =k 1 H T

k thorn Rk

1(9)

where P is the error covariance matrix and Rk is the measurement noise covariance matrix

Figure 3 concisely explains the implementation of the KF

In the present application the state estimate variable is X k = [NAE 1 θ ]T NAE in undamaged condition should remainconstant independent of the applied loads and the second variable tracked is the constant value 1 This constant is incor-

porated to ensure a correct relation between the state estimation matrix the observation matrix and the measurement matrix

The added bene1047297t of the constant is it makes the measurement matrix square which allows faster computations The third

component of the vector is the variable θ for the yaw angle It is a linear estimate of the measurement from the sensor The

input for the KF algorithm is essentially the state transition matrix ( A) which relates the state estimate variable in time In

this case A is a unity matrix of dimension 3 as the state estimates are random and not co-related in time The other input is

the measurement matrix ( H ) which links the state estimation variable X k and the measurement variable ( Z k ) at each time

step (measurement from the sensors) The H matrix takes into consideration the observability of the NA based on the lo-

cations of the sensors and is designed for accurate system modelling while maintaining the linearity of the measurement

step In the present case Z k = [ϵl ϵr θ ]T

vector consists of the strain measurements from the left (ε l ) and the right side

(ε r ) of the tower and the yaw angle measurement (θ ) of the nacelle

23 Damage sensitive feature

As mentioned in earlier section the NA location is independent of the loading conditions and depends only on the condi-

tion of the structure Thus the change in the NAE is taken as the damage sensitive feature and is given by Equation (10)

ΔNAE frac14NAEhealthy NAEmonitored

NAEhealthy

100 (10)

The NAEhealthy is developed at the time of the installation of sensors when the structure is known to be in healthy

condition NAEmonitored is the estimate at every time step If the change exceeds a certain threshold an alarm is raised in-

dicating damage This threshold is based on engineering judgement

3 FINITE ELEMENT MODELLING

The proposed methodology was veri1047297ed on a simulated FE model of the DTU 10 MW RWT28 The tower is a 115630 m

tall hollow steel structure The outer diameter varies linearly from 83 m at the base to 55 m at the top of the tower The

tower is divided into 10 sections where the wall thickness is constant in each section but gradually decreasing from the

bottom to the top (Figure 4) The tower is encastred at the bottom The tower is made from steel S355 with a Youngrsquos mod-

ulus of 210 GPa Poissonrsquos ratio 03 and the density 8500 kgm 3 (8 increase of the density to account for the secondary

structural components) The model was simulated in commercial FE modelling software ABAQUS31 using shell elements

based on the design data in the reference28

Figure 3 Flow chart for the implementation of the KF X is the estimate of the state A is the state transition matrix P is the state var-

iance matrix K is the Kalman gain H is the measurement matrix z is the measurement variable and the lsquok rsquo indicates the time step k



DOI 101002we



The nacelle and hub loads were applied as point loads at speci1047297ed eccentricity and height indicated from the design spec-

i1047297cations A random wind pressure was simulated with the peak wind pressure based on the Eurocodes32

The wind pressure was applied on the surface area facing the wind in order to compute the force The force increases

according to the wind pro1047297le power law along the height of the tower 33 and the wind loads were varied accordingly on

the structure The applied wind loads were similar to those experienced by the tower in actual operation The dynamic loads

on the tower were simulated for a period of 30 s The blades however were assumed to be pitched into a full aerodynamic

brake position to ensure minimal rotor motion and consequent change in mass distribution which may affect the NA34

The dynamic direct analysis was carried out in ABAQUS which uses the modal superposition for estimating the displace-

ments and the strain so in order to achieve accurate results and limit the computational load the number of extracted mode

shapes was 50 The mesh size of the element was then chosen in order to achieve stable and smooth mode shapes for the

extracted modes

4 NUMERICAL SIMULATIONS

Numerical simulations were carried out on FE model described in the previous section

The numerical simulations were carried out

bull to establish the need for yaw tracking for accurate estimation of NA

bull to exhibit the robustness and performance of the KF-based NA tracking methodology as compared with the direct es-

timation method

bull to demonstrate the use of NAE as a damage sensitive feature in presence of noise and

bull to show the effectiveness of bi-axial NA tracking for accurate damage isolation

41 Effect of Yaw on NA Location

As shown in Equation (4) the NA location is directly proportional to the axial strain and inversely proportional to the bend-

ing strains In ideal conditions the axial load experienced will not change with the yaw angle On the other hand the

Figure 4 Finite element modelling details of the tower



DOI 101002we



bending moment along the axis changes with the yaw angle This is because of the changing perpendicular distance be-

tween the point of loading and the axis about which the strain is measured This distance is the cosine component of the

yaw angle along x axis and the sine component of the yaw angle along y axis As a result the NA location changes with

the yaw angle The NA location change to the yaw angle is plotted in Figure 5 The NA being inversely proportional to

the cosine component plots a curve similar to the secant function of the yaw angle

As can be clearly seen in Figure 5 the location of the NA undergoes signi1047297cant changes and hence as such needs to be

monitored quantity for robust damage detection

42 KF for NA estimation

The KF is a powerful tool for the estimation of the state variables especially in the presence of measurement noise So the

use of KF will improve the estimation Figure 6 shows the qualitative relative performance of the KF for the estimation NA

compared with the direct estimation method Five different cases varying the measurement noise for the strain and yaw an-

gle measurement are considered in order to check the robustness of the KF-based estimation Table I also shows the quan-

titative superiority of the KF-based methods

As seen in Table I the standard deviation of the KF-based estimation is orders of magnitude lower than the direct esti-

mation method especially in the presence of measurement noise It should be noted that because of the presence of

Figure 5 Plot of NA location against Yaw angle along the x axis

Figure 6 Comparison between direct and KF estimation methods (a) 0 noise for strain and yaw angle (b) 5 noise for strain and

0 noise for yaw angle (c) 10 noise for strain and 0 noise for yaw angle (d) 0 noise for strain and 5 noise for yaw angle and (e)

5 noise for strain and 5 noise for yaw angle



DOI 101002we



measurement noise the mean of the direct prediction changes appreciably this in turn directly affects the accuracy of the

damage detection methodology Hence the use of KF estimator is necessary

Kalman Filter is indeed a very robust tool in addition to the sensitivity studies in presence of measurement noise

more studies were performed for different values of initial guess of NA location and different process noise esti-

mates These studies indicate that the initialization of the KF does not affect the 1047297nal NAE estimate An improper initialization just affects the time required for achieving the convergence to the true value of the NAE 35 This robust-

ness to initialization allows easy implementation to real strain data from the wind turbine The KF-based NA estima-

tion has been applied to strain data available from the Nordtank NTK50041 wind turbine and may be found in 36

The implementation of the KF was carried out based on engineering judgement alone and yield very promising

results

43 Bi-axial NA tracking for damage detection

The sensor pairs are located perpendicular to each other as indicated in Figure 4 and at the centre of each element As such

if the damage is at any location not in the vicinity the observability of the damage is in the form of the sine and the cosine

component Furthermore because of the non-linearity the damage may be detected but the isolation of the damage may be

a problem Thus the bi-axial NA tracking data should be combined in order to get more realistic damage isolation The

intuitive way of combination is by taking a ratio of the observed NA along the two axes Although this may yield an ap-proximate estimate of the location it cannot be applied directly It should be kept in mind that the periodicity for sine and

cosine is 2π while that for tan ratio is π thus leading to loss of directionality Thus a decision level data fusion is necessary

where the change in the directionality is overcome by proper study of the change in the NA location along both the sensor

axes The decision level data fusion refers to the use of the individual signs of the change in the NA location in order to

make an assessment of the damage location

The damage detection strategy is based on the principle of change in the NA location in healthy and damaged state The

damage will be indicated when the relative change in the location is more than a certain threshold This threshold needs to

be determined based on engineering judgement taking into consideration the probability of positive false detection and

negative false detection and risk to the entire structure Positive false detection occurs when a signi1047297cant damage goes un-

detected while the negative false detection occurs when the damage alarm is raised when there is no damage in the struc-

ture In order to quantify the threshold based on probability several damage scenarios were studied under different

simulated noise levels The threshold was set at 1 change in NA location when the yaw angle measurement and bi-axial

NA tracking are carried out

Equation (11) gives the location of the damage in terms of the angle without taking the signs of the individual change of NA locations into consideration (without fusion)

DL frac14 tan1 ΔNAEB

ΔNAEA

(11)

where DL is the damage location in degrees (deg) ∆NAEB is the percent relative change in the NAE at B while

∆NAEA is the percent relative change in the NAE at A

The notations B and A are indicated in Figure 7 which also shows the different damage scenarios

For accurate detection of location the difference in the periodicity of the tangent function and sine and cosine functions

should be noted

Table I Statistical performance of estimators (NAE)

Scenario

KF estimation Direct estimation

Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we



Table II shows quantitatively that the approach is valid at the same time decision level data fusion is necessary for

accurate isolation of damage

The damage scenarios I and II indicate the validity of the decision level data fusion and taking the ratio of the changes inNA is a valid technique for fusion The damage isolation for cases I and II is valid without fusion as the damage lies in the

90ndash90deg range The damage scenario III is a point diametrically opposite to II and it shows the signi1047297cance of decision

level data fusion If the signs corresponding to the change of the location are not taken into account the isolation gives a

diametrically opposite point that may be termed as false isolation of the damage (shown in grey background) and hence

the decision level data fusion is necessary The damage scenario IV indicates that when the damage is at locations perpendic-

ular to one of theaxes theresults obtained for the isolation are notaccurate but this in essence proves that bi-axial tracking is

indeed necessary Also change in the location of NA in scenario IV in the estimate at A is signi1047297cant and hence false

detection of the damage is highly unlikely The highchangeexceeding10 also indicates very close proximity of the damage

to the strain sensors and gives further idea about the location of the damage Furthermore the decision level data fusion will

recognize that the threshold for the damage detection has not been exceeded (shown in grey background) and as such the data

should not be used thus making the decision level data fusion-based strategy more robust against false localization

44 NA-based damage detection

As explained in the earlier section the NA of the cross-section of the tower is the property of the condition of the structure

and may be used as a damage indicator

In order to validate the use of NA as damage indicator arti1047297cial damage was introduced in one element of the tower by

reducing the 1047298exural rigidity of that particular element by 20 Reduction of 1047298exural rigidity is a valid damage simulation

strategy as indicated by 24 It may be treated equivalent to loss of material thickness because of corrosion or cracking and is a

commonly used strategy for global level damage simulation in bridge structures 1 The simulated damage was detected by com-

paring the NAE of the damage and the undamaged element The relative change in the location of the NA is given in Table III

The damage is detected if the change in the NA estimation of the damaged and undamaged states is more than a speci1047297ed

threshold which is determined on engineering judgement As can be clearly observed even in the presence of measurement

noise there is a signi1047297cant difference in the change of the NAE of the damage element and the others so the chances for a

false detection are quite minimal and as such a lower threshold may be possible 1 in the case where yaw angle is being

Figure 7 Damage scenarios indicating the need for bi-axial NA tracking

Table II Bi-axial NA tracking for damage detection



DOI 101002we



tracked as well The use of yaw tracking allows this higher con1047297dence and as such is an advantage for detecting lower

levels of damage

45 Sensitivity to Severity of Damage

Ideally the damage metric should be able to detect even minor changes in the system but in actual practice these changes

are often masked by changes in ambient condition changes and measurement noise Hence the sensitivity to damage is

investigated

Table IV indicates the percent change in the NAE with change in the damage severity

It can be seen that damage above 15 severity of damage can be easily detected through the tracking of NA and as such

the methodology promises to be better than the conventional vibration-based damage detection methods which are unableto detect such low levels of damage in the presence of same levels of measurement noise 24

5 CONCLUSIONS

The paper proposes bi-axial tracking and decision level data fusion for a more accurate damage localization The method-

ology is based on tracking of NA along two perpendicular axes using KF-based estimator The study 1047297rst establishes the

effect of yaw on the measured strain and in turn the tracked NA Then it establishes the merits for the use of KF for NA

tracking estimation and data fusion of yaw angle and measured strain This KF-based NA estimation is then used to detect

damage in the simulated tower structure of the 10 MW DTU RWT

Table III NA-based damage detection in presence of noise

Table IV Performance of NA with changing severity of damage

Damage extent Undamaged NAE Damaged NAE NAE ()

5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we



The study indicates that the NA is a property of the condition of the structure and remains relatively unaffected by the

measurement noise Furthermore the robustness of the metric has been studied in the presence of measurement noise From

the results obtained it can be seen that the bi-axial NA tracking is a promising SHM methodology for improved damage

isolation

The authors believe that the methodology proposed is robust to the effects of ambient temperature The tower being a

metallic structure will undergo only bulk temperature effects The temperature gradient if any will be extremely low

and in such conditions the NA position is insensitive to temperature changes thus making the methodology insensitive

to ambient condition changes

The FBG-based sensors are durable have a long life and are relatively easy to install The instrumentation similar to the

one assumed in the paper has been carried out in1933 Once the strain data are available the setting up of KF needs low

expertise as the NAE estimate is robust even if the initialization of the KF is incorrect 35 and has been successfully imple-

mented in36

The present study aims at giving a proof of concept and the validity of the use of data fusion for NA tracking for damage

detection in tower structures in the presence of yawing The authors acknowledge that the actual loading conditions in-

service and the pitching and the rotation of the blades may increase the complexity for the use of the metric The rotation

of the blades will change the load distribution in bending which in turn will affect the strains measured these effects may

be compensated for by fusing the rotation speed of the wind turbine in the estimation process Similarly pitch angle too

affects the strain response of the tower as shown in18 Thus a more inclusive fusion of data from all the different sensors

available is necessary to compensate for these effects and as such is identi1047297ed as the next step of the research In addition

more realistic damage scenarios like fatigue-induced cracks need to be simulated and the sensitivity of the method needs

to be validated in these scenarios

ACKNOWLEDGEMENTS

The authors would like to acknowledge the European Commission for their research grant under the project FP7-PEOPLE-

2012 ITN 309395 ldquoMARE-WINTrdquo (new Materials and REliablity in offshore WINd Turbines technology) The authors

would also like to thank DTU Wind Energy for providing valuable information for the modeling of the ten MW RWT

towers for the purpose of this study The authors are also grateful to TASK-CI for allowing the use of their computational

resources The opinions expressed in this paper do not necessarily re1047298ect those of the sponsors

REFERENCES

1 Jang S Jo H Cho S et al Structural health monitoring of a cable-stayed bridge using smart sensor technology deploy-

ment and evaluation Smart Structures and Systems 2010 6(5ndash6) 439ndash459

2 Doebling SW Farrar CR Prime MB A summary review of vibration based damage identi 1047297cation techniques Shock

and Vibration Digest 1998 30(2) 91ndash105

3 Abedwuyi A Wu Z Serker NHKM Assessment of vibration-based damage identi1047297cation methods using displacement

and distributed strain measurement Structural Health Monitoring 2009 8(6) 443ndash461

4 Cawley P Adams RD The location of defects in structures from measurements of natural frequencies J Strain Anal

1979 14 49ndash57

5 Hunt DL Application of an enhanced coordinate modal assurance criterion In Proceedings of the 10th International

Modal Analysis Conference San Diego CA 1992 1 66ndash71

6 Pandey AK Biswas M Samman MM Damage detection from changes in curvature mode shapes Journal of Sound

and Vibration 1997 145(2) 321ndash332

7 Pandey AK Biswas M Damage detection in structures using changes in 1047298exibility Journal of Sound and Vibration1994 169(1) 3ndash17

8 Chakraborty S DeWolf JT Development and implementation of a continuous strain monitoring system on a multi-

girder composite steel bridge Journal of Bridge Engineering 2006 11(6) 753ndash762

9 Zonta D Bernal D Strain-based approaches to damage localization in civil structures In Proceedings of XXIV

international modal analysis conference Saint Louis 2006

10 Benedetti M Fontanari V Zonta D Structural health monitoring of wind towers remote damage detection using strain

sensors Smart Materials and Structures 2011 20(5) 055009

11 Adewuyi AP Wu ZS Modal macro-strain 1047298exibility methods for damage localization in 1047298exural structures using long-

gage FBG sensors Structural Control and Health Monitoring 2011 18(3) 341ndash360



DOI 101002we



12 Ciang C Lee J Bang H Structural health monitoring for a wind turbine system a review of damage detection methods

Measurement Science and Technology 2008 19 122001

13 Hyers RW McGowan JG Sullivan KL Manwell JF Syrett BC ldquoCondition monitoring and prognosis of utility scale

wind turbinesrdquo Energy Materials 2006 1(3) 187ndash203

14 Faulstich S Hahn B Tavner PJ Wind turbine downtime and its importance for offshore deployment Wind Energy

2011 14(3) 327ndash337

15 Swartz RA Lynch JP Zerbst S Sweetman B Rolfes R ldquo

Structural monitoring of wind turbines using wireless sensor networksrdquo Smart structures and systems 6 no 3 (2010) 183ndash196

16 Smarsly K Hartmann D Law KH ldquoAn integrated monitoring system for life-cycle management of wind turbinesrdquo

International Journal of Smart Structures and Systems 2013 12 2

17 Lu KC Peng HC Kuo Y-S Structural health monitoring of the support structure of wind turbine using wireless sensing

system Le Cam Vincent and Mevel Laurent and Schoefs Franck EWSHM-7th European Workshop on Structural

Health Monitoring Nantes France 2014

18 Bas J Smith J Carriveau R Cheng S Ting D Newson T ldquoStructural response of a commercial wind turbine to various

stopping eventsrdquo Wind Engineering 2012 36(5) 553ndash570

19 Bang H Jang M Shin H ldquoStructural health monitoring of wind turbines using 1047297ber Bragg grating based sensing

systemrdquo In SPIE Smart Structures and Materials + Nondestructive Evaluation and Health Monitoring International

Society for Optics and Photonics 2011

20 Benedetti M Fontanari V Battisti L ldquoStructural health monitoring of wind towers residual fatigue life estimationrdquo

Smart Materials and Structures 2013 22(4) 045017

21 Sigurdardottir D Glisic B ldquoDetecting minute damage in beam-like structures using the neutral axis locationrdquo Smart

Materials and Structures 2014 23(12) 125042

22 Sigurdardottir D Glisic B ldquoNeutral axis as damage sensitive featurerdquo Smart Materials and Structures 2013 22(7)

075030

23 Xia HW Ni YQ Ye XW ldquoNeutral-axis position based damage detection of bridge deck using strain measurement

formulation of a Kalman 1047297lter estimatorrdquo In Proceedings of the 6th European Workshop on Structural Health

Monitoring Dresden Germany 2012

24 Soman R Malinowski P Ostachowicz W Neutral axis tracking for damage detection in wind turbine towers In

Proceedings of the EWEA 2014 Barcelona Spain 2014

25 Soman R Malinowski PH Ostachowicz W Kalman-1047297lter based data fusion for neutral axis tracking for damage

detection in wind-turbine towers Le Cam Vincent and Mevel Laurent and Schoefs Franck EWSHM - 7th European

Workshop on Structural Health Monitoring Nantes France 201426 Welch G Bishop G An introduction to the Kalman 1047297lter 1995 [accessed online on 6-Nov-14 at httpclubsens-

cachanfrkrobotolddatapositionnementkalmanpdf ]

27 Brown RG Hwang PYC Introduction to Random Signals and Applied Kalman Filtering 3rd edn John Wiley amp Sons

New York 1997

28 DTU Wind Energy Report-I-0092 Description of the DTU 10 MW Reference Wind Turbine 2013

29 Maybeck PS Stochastic Models Estimation and Control Academic press 1982 3

30 Sorenson HW ldquoLeast-squares estimation from Gauss to KalmanrdquoSpectrum IEEE 1970 7(7) 63ndash68

31 ABAQUS Analysis User rsquos Manual Version 6 12-3 edn 2013

32 Eurocode NS-EN 1991-1-4 ldquoGeneral actionsmdashwind actionsrdquo Standards Nor-way 2005+NA 2009

33 Şen Z Altunkaynak A Erdik T ldquoWind velocity vertical extrapolation by extended power lawrdquo Advances in Meteorol-

ogy 2012 2012 6 Article ID 178623 DOI 1011552012178623

34 Bas J Carriveau R Cheng S Newson T Strain response of a wind turbine tower as a function of nacelle orientation InBIONATURE 2012 The Third International Conference on Bioenvironment Biodiversity and Renewable Energies

2012 12ndash18

35 Soman R Malinowski P Ostachowicz W ldquoThreshold determination for neutral axis tracking based damage detection

in wind turbine towersrdquo In Proceedings of the EWEA Offshore2015 Copenhagen Denmark 2015 (submitted)

36 Soman R Malinowski P Ostachowicz W Paulsen U ldquoKalman 1047297lter based data fusion for neutral axis tracking in wind

turbine towersrdquo In Proceedings of the SPIE smart Structures NDE San DiegoUS 2015 (submitted)



DOI 101002we



The SHM methods are still not up to the desired standards and as such there is a need for a robust damage detection tech-

nique The discrepancy between the expected and the measured results of the different SHM techniques is mainly attributed to

the uncertainty in the measurement environment with respect to noise temperature and excitation mechanism of the structure

Thus thesearch for an SHM system that will be able to detect small levels of damage butat the sametimebe robustenough to

overcome the ambient noise and temperature changes and detect damage under operational conditions is ongoing

The use of SHM techniques in wind turbines is gaining prominence especially with the use of offshore wind turbines

where costs of unscheduled maintenance are high and downtime results in major loss of revenue Ciang et al 12 and Hyers

et al 13 provide reviews on condition-monitoring strategies for all wind turbine subsystems and components Tower dam-

age is the third most common damages in wind turbines 12 As a result the research in the area of SHM of towers is now

becoming more signi1047297cant Furthermore as seen in Figure 1 the downtime for minor damages in towers (support and hous-

ing) is far less signi1047297cant (014 days) than in the case of major damage (2801 days) So there is a huge economic incentive

for early detection of minor failures through the use of proper SHM techniques for towers

The most commonly used sensors for SHM of towers are accelerometers and displacement sensors15ndash17 These sensors

are suitable only for global damage identi1047297cation On the other hand strain sensors allow a more local damage detection

approach Bas et al and Bang et al make use of 1047297bre Bragg grating (FBG)-based strain sensors for SHM of tower1819

Bas et al 18 have studied the structural response of the tower to various stopping events which may be used as baseline

for detecting deterioration of the tower over a period of time Bang et al 19 have studied the stability of the FBG sensors

over time and their ability for dynamic strain assessment on a real structure Benedetti et al 20 make use of conventional

strain gauges for the fatigue monitoring near welds of the tower The results obtained are quite encouraging but the use of

this method is restricted to known hot spots of potential damage

Thus in this paper a method for damage detection of towers is proposed which is able to detect lower levels of damagethroughout the structure without a priori knowledge of the structural behaviour The methodology proposed is a customiza-

tion of neutral axis (NA) as a damage-sensitive feature for tower structures The NA has been proposed for damage detection

in bridges21ndash23 The application of the methodology directly to tower structures is not possible because of the yawing of the

turbine and the resulting mass and stress redistribution Thus a robust methodology has been proposed to overcome these

application-speci1047297c challenges and further the effectiveness of the methodology for more precise damage isolation

The present research builds on the research work presented by the authors in 24 and 25 In this paper the methodology for

damage detection using NA as a metric is extended to bi-axial neutral tracking using decision level data fusion It was

shown in 24 that the position of NA of the tower structure may be used as a robust damage indicator This is possible as

the NA is the property of the cross-section of the tower independent of the bulk temperature effects and the ambient wind

loading The position of the NA can be assessed by measuring the strains on opposite surfaces of the tower in bending The

NA of the tower subject to unknown loading both in magnitude and direction was estimated using the discrete Kalman

Filter (KF)2627 The authors then extended the methodology to include the effect of yaw of the nacelle on the observability

of the NA in25 But even then only the determination of the altitude of the damage was possible In this study through the

use of bi-axial tracking and decision level data fusion a better resolution of the damage location is possible The study is

undertaken on a 1047297nite element (FE) model of the 10 MW wind turbine tower which is seen as the future of offshore wind

Figure 1 Reliability characteristics of several sub-assemblies14



DOI 101002we















21 Neutral axis






ε axial frac14 F

EA(1)




DOI 101002we






EI (2)





side in Figure 2)










w

2εbending

frac14 εl w




NAE frac142 L

w(5)




22 Kalman Filter














in Equation (8)


(8)



DOI 101002we






k thorn Rk

1(9)



















NAEhealthy

100 (10)
















DOI 101002we













extracted modes






timation method









DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we















21 Neutral axis






ε axial frac14 F

EA(1)




DOI 101002we






EI (2)





side in Figure 2)










w

2εbending

frac14 εl w




NAE frac142 L

w(5)




22 Kalman Filter














in Equation (8)


(8)



DOI 101002we






k thorn Rk

1(9)



















NAEhealthy

100 (10)
















DOI 101002we













extracted modes






timation method









DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we






EI (2)





side in Figure 2)










w

2εbending

frac14 εl w




NAE frac142 L

w(5)




22 Kalman Filter














in Equation (8)


(8)



DOI 101002we






k thorn Rk

1(9)



















NAEhealthy

100 (10)
















DOI 101002we













extracted modes






timation method









DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we






k thorn Rk

1(9)



















NAEhealthy

100 (10)
















DOI 101002we













extracted modes






timation method









DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we













extracted modes






timation method









DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we
























DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we











results





















ΔNAEA

(11)





should be noted


Scenario


Standard deviation

(mean)

Standard deviation

(mean)

(a) 00444 01064

(1633) (1637)

(b) 00474 01862(1633) (1622)

(c) 00507 04527

(1635) (1666)

(d) 00044 01168

(1633) (1641)

(e) 00499 02111

(1633) (1686)



DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we
































DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we




levels of damage




investigated




5 CONCLUSIONS









5 7008 7160 2169

10 7008 7276 3825

15 7008 7374 5215

20 7008 7479 6717

25 7008 7593 8340



DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we






isolation








mented in36










ACKNOWLEDGEMENTS






REFERENCES








1979 14 49ndash57
















DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we








2011 14(3) 327ndash337


















075030











New York 1997









2012 12ndash18







DOI 101002we

soman et al-2015-wind energy

Documents