See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/270966930
Wind turbine blade shear web disbond detection using rotor blade operational
sensing and data analysis
Article in Philosophical Transactions of The Royal Society A Mathematical Physical and Engineering Sciences · February 2015
DOI: 10.1098/rsta.2014.0345 · Source: PubMed
CITATIONS
READS
3
1,809
3 authors:
Noah Myrent
Douglas Adams
Romax Technology, Boulder, Colorado
Vanderbilt University
7 PUBLICATIONS 55 CITATIONS
238 PUBLICATIONS 2,724 CITATIONS
SEE PROFILE
D. Todd Griffith
University of Texas at Dallas
118 PUBLICATIONS 1,197 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
machine learning View project
Segmented Ultralight Morphing Rotor View project
All content following this page was uploaded by D. Todd Griffith on 18 December 2015.
The user has requested enhancement of the downloaded file.
SEE PROFILE
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
rsta.royalsocietypublishing.org
Wind turbine blade shear web
disbond detection using rotor
blade operational sensing
and data analysis
Research
Noah Myrent1 , Douglas E. Adams1
Cite this article: Myrent N, Adams DE,
Griffith DT. 2015 Wind turbine blade shear web
disbond detection using rotor blade
operational sensing and data analysis. Phil.
Trans. R. Soc. A 373: 20140345.
http://dx.doi.org/10.1098/rsta.2014.0345
One contribution of 17 to a theme issue
‘New perspectives in offshore wind energy’.
Subject Areas:
energy, mechanical engineering,
environmental engineering, civil engineering,
pattern recognition, computer modelling and
simulation
Keywords:
wind energy, blade damage, condition
monitoring, shear web, disbond,
sensitivity analysis
Author for correspondence:
Douglas E. Adams
e-mail: douglas.adams@vanderbilt.edu
and D. Todd Griffith2
1 Laboratory for Systems Integrity and Reliability, Vanderbilt
University, 566 Mainstream Drive, Nashville, TN 37228, USA
2 Wind and Water Power Technologies, Sandia National Laboratories,
1515 Eubank SE, Albuquerque, NM 87123, USA
A wind turbine blade’s structural dynamic response
is simulated and analysed with the goal of
characterizing the presence and severity of a shear
web disbond. Computer models of a 5 MW offshore
utility-scale wind turbine were created to develop
effective algorithms for detecting such damage.
Through data analysis and with the use of blade
measurements, a shear web disbond was quantified
according to its length. An aerodynamic sensitivity
study was conducted to ensure robustness of the
detection algorithms. In all analyses, the blade’s flapwise acceleration and root-pitching moment were
the clearest indicators of the presence and severity
of a shear web disbond. A combination of blade and
non-blade measurements was formulated into a final
algorithm for the detection and quantification of the
disbond. The probability of detection was 100% for
the optimized wind speed ranges in laminar, 30%
horizontal shear and 60% horizontal shear conditions.
1. Introduction
Wind energy is the second largest energy resource being
added to the grid (behind natural gas), making up
32% of US electric generating capacity additions in
2011 [1]. However, wind operations and maintenance
(O&M) costs are estimated to be more than double
that of natural gas powered generation [2]. Although
lowering the cost of wind energy is challenging, these
costs can be reduced as part of an overall condition-based
2015 The Author(s) Published by the Royal Society. All rights reserved.
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
2
.........................................................
maintenance strategy. Current wind turbine condition monitoring systems analyse vibration, oil,
strain, temperature and generator power to assess the structure’s health [3]. Owing to a wind
turbine’s unique operating conditions primarily being influenced by the stochastic aerodynamic
loading, the current condition based monitoring systems have had limited effectiveness,
especially for the application of blade damage detection [4]. Wind turbine blades are currently
certified by material testing and full-scale blade testing of a single blade. However, blades
encounter several failure modes which cannot be detected by coupon or blade testing. These
failures occur mainly due to manufacturing defects. In addition, the mechanical properties of
blade bond lines are very sensitive to design and material processing technologies [5]. Stresses
in the shear web bond lines which may be caused by stress concentrations or manufacturing
variability are not directly measured in the full-scale blade tests [6]. As a result, large safety factors
are used to minimize and mask risks in the structural bond lines.
Poor bond lines are known to be relatively common in the wind blade industry, primarily due
to the manufacturing process [7,8]. Common adhesive bond defects include bond thicknesses
out of tolerance and voids due to missing adhesive. In 2007, several Gamesa blades exhibited
complete debonding of the shells at the trailing edge and internal spars. A faulty adhesive
applicator would later be publicized as the cause of the defect due to inadequate quantities of
adhesive being applied [9]. Figure 1 shows one of the Gamesa blade debonding failures.
If a bond void is already present between the shear web and the blade shell before the turbine
is even commissioned, the shear web disbond will likely increase in length due to shear stresses
and fatigue. Out-of-plane deformations of the panels above and below the shear web result in
peeling stresses in the bond lines which may be the cause of fatigue and failure in shear web
adhesive joints [10]. A blade becomes more susceptible to buckling with the presence of shear
web bond voids and the risk rises as the extent of the void increases [7]. Essentially, the void
creates a concentration of lap shear. If the stress concentration at the edge of the void reaches
the lap shear strength of the adhesive, the failure will rapidly propagate in both directions,
causing the bond line to unzip. Wind turbine blade shear webs have been experimentally and
analytically shown to possess post-buckling load carrying capability [11] and unzipping of the
shear web bond line is a major concern if a disbond of 1 m or greater is present. If increasing
buckling loads are introduced to the wind blade, then the spar and blade panels are susceptible to
buckling and potentially cracking, especially during extreme wind events. Therefore, the proposal
in this paper is to use low cost blade and nacelle mounted inertial sensors in combination
with blade root strain gauges to detect a shear web disbond on one of the blades and to
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
Figure 1. Debonding failure of Gamesa 43 m blades [7]. (Online version in colour.)
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
In order to evaluate the feasibility of detecting a shear web disbond on utility-scale wind
turbine blades, computer simulations were carried out on a 5 MW offshore wind turbine model
developed by the National Renewable Energy Laboratory (NREL), with blade models developed
by Sandia National Laboratories (SNL). Simulation modelling was done using NREL’s Fatigue,
Aerodynamics, Structures and Turbulence (FAST) code, which is a comprehensive aeroelastic
simulator for two- and three-bladed horizontal axis wind turbines (HAWTs). FAST includes a
variety of input parameters including turbine control settings, environmental conditions, blade
and tower models, and many others. There are also several hundred different outputs, including
generator power and blade inertial measurements.
FAST uses AeroDyn to predict the aerodynamics of HAWTs. AeroDyn is an aerodynamics
software library which uses several subroutines in conjunction with aeroelastic code for wind
turbine applications, including the blade element momentum theory, the generalized dynamicwake theory and dynamic stall based on the semi-empirical Beddoes–Leishman model. FAST also
uses the stochastic inflow turbulence tool TurbSim to provide a numerical simulation of a full-field
flow that contains coherent turbulence structures that reflect known spectral models. The FAST
model combines these aerodynamic parameters with a multibody dynamics formulation and it
performs a time-marching analysis of the nonlinear equations of motion. For a more detailed
description of the working principles of the FAST code, see the FAST user’s guide [12].
For this initial investigation, damaged blade models with shear web disbonds were developed.
All of the disbonds were assumed to have initiated at the max chord of the blade (at the 14.35 m
span location) and propagated outwards toward the tip of the blade. This location was chosen
because buckling of the surface of the blade at the maximum chord section is one known type
of field failure [13]. This section includes a variety of different analyses that were conducted at
various stages throughout the modelling and simulation processes.
(a) Five megawatt turbine model description
The simulations in this paper were performed as part of an ongoing structural health and
prognostics management project for offshore wind turbines in conjunction with SNL and
Kusnick [14] using a representative utility-scale wind turbine model. The NREL offshore 5 MW
baseline wind turbine model was developed to aid research aimed at evaluating offshore wind
energy technology [15]. The wind turbine model represents a three-bladed, upwind, variable
speed, variable blade-pitch turbine and was created using design information from documents
published by turbine manufacturers, with a heavy emphasis on the REpower 5 MW machine.
Basic specifications of the model are listed in table 1. A new blade model was developed in
Sandia’s Numerical Manufacturing and Design (NuMAD) tool to be used with the NREL 5 MW
turbine model, which is the same model used in the initial studies [16]. A detailed blade model
did not exist and was needed so that damage could be introduced into the blade structure
within the multi-scale modelling and simulation framework (as described above). The detailed
blade model was developed by SNL using blade geometry data from the Dutch Offshore Wind
Energy Converter Project (DOWEC) and composite layup information from the European Union’s
UpWind programme. The distribution of material layers along the blade span is illustrated
in figure 2.
Two-thirds of the blade span uses the TU-Delft family of aerofoils, and the last third of the
blade span uses the National Advisory Committee for Aeronautics (NACA) 64-series aerofoils.
Mediate aerofoil shapes were developed which preserve the blending of camber lines as well as
provide a fluent blade thickness profile. Figure 3 shows the finite-element model of the blade in
.........................................................
2. Methodology
3
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
determine its severity. The ability to characterize the shear web disbond will allow early and
effective maintenance actions to be performed which can lower the overall O&M costs for a
wind turbine.
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
120
4
GELCOAT
TRIAX-1/3
TRIAX-2
UDCarbon
UD_TE
SKINFOAM
60
40
20
0
0
10
20
30
40
blade span (m)
50
60
70
Figure 2. Model of the distribution of material layers along the span of the blade [16]. (Online version in colour.)
Table 1. Gross properties of the NREL 5 MW baseline wind turbine [15].
property
value
rating
5 MW
rotor orientation, configuration
upwind, 3 blades
control
variable speed, collective pitch
drivetrain
high speed, multiple-state gearbox
rotor, hub diameter
126 m, 3 m
hub height
90 m
cut-in, rated, cut-out wind speed
3 m s−1 , 11.4 m s−1 , 25 m s−1
cut-in, rated rotor speed
6.9 r.p.m., 12.1 r.p.m.
rated tip speed
80 m s−1
overhang, shaft tilt, precone
5 m, 5, 2.5
rotor mass, nacelle mass, tower mass
110 000 kg, 240 000 kg, 347 460 kg
water depth
20 m
wave model
JONSWAP/Pierson–Moskowitz spectrum
significant wave height
6m
platform
fixed-bottom monopile
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
Figure 3. ANSYS finite-element mesh for the 5 MW blade model. (Online version in colour.)
ANSYS with the coloured sections representing different composite materials. This high degreeof-freedom model was reduced into a model consisting of several beam elements using SNL’s
blade property extraction (BPE) tool. The BPE code applies loads in each of the 6 d.f. at the tip of
the blade model in ANSYS, then calculates the resulting displacements at selected nodes along
.........................................................
80
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
no. layers
100
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
(a)
(b)
zn
yn
xn
xb,i
Figure 4. Blade coordinate system (a) and nacelle/yaw coordinate system (b) [12]. (Online version in colour.)
the blade to generate the 6 × 6 Timoshenko stiffness matrices for the beam. This reduced degreeof-freedom model is subsequently used to define the blade properties in FAST [16].
(b) Fast simulation turbine coordinate systems
FAST makes use of different coordinate systems for input and output parameters which will be
referenced throughout this paper and were extracted from the FAST user’s guide [12]. Although
FAST’s coordinate systems use a downwind turbine configuration, the same coordinate systems
apply for the upwind turbine used in this study. The blade coordinate system is expressed
as xb,i (or xb ) in the flap-wise direction, yb,i (or yb ) in the edge-wise direction and zb,i (or zb )
in the span-wise direction. The nacelle inertial measurements make use of the nacelle/yaw
coordinate system with xn in the axial direction, yn in the transverse direction and zn in
the vertical direction for nacelle inertial measurements. Both coordinate systems are shown
in figure 4.
(c) Shear web disbond damage modelling methodology and simulation methods
To model the presence of a shear web disbond on a wind turbine blade, the NuMAD blade model
was modified so that shear web nodes were split into two different nodes. This effectively split the
blade model at the shear web in a similar way to how the blade is physically constructed through
bonding the high pressure clam shell to the shear webs. To simulate a healthy bond across the
blade, the top and bottom shear web nodes were connected using constraint equations in all 6 d.f.
In the area of the blade in which the shear web disbond existed, the constraints were removed so
that there was no connection between the top of the blade and the shear web. A similar approach
was used by Griffith et al. [16] to simulate a trailing edge disbond on the same blade model.
While this modelling disbond methodology is effective in modelling a disbond in which the blade
and shear web do not come into contact, it fails to take into account the possible interaction of
the top and bottom surfaces of the disbond. For large cracks in which interaction between the
top of the blade and the shear web may have a significant influence, the relative decrease in
stiffness due to the disbond is likely overestimated because the added stiffness due to the disbond
face interaction was not taken into account. Modelling the interaction between the two surfaces
could be achieved using nonlinear surface contact constraints between the top of the blade and
the shear web but this was not accomplished during this initial investigation and remains as
future work.
FAST simulations were performed for several wind profiles and turbine blade conditions.
Among the wind profiles used in the FAST simulations were constant wind speed and direction,
IEC Kaimal model with A turbulence, IEC Kaimal model with B turbulence and the NREL
.........................................................
yb,i
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
zb,i
5
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
Table 2. FAST simulation matrix for each blade damage type.
6
2m
disbond
3m
disbond
4m
disbond
5m
disbond
10 m
disbond
wind speed (3–25 m s−1 )
101
101
101
101
101
101
101
horizontal shear
303
303
303
303
303
303
303
..........................................................................................................................................................................................................
(30%, 60%, 90%)
..........................................................................................................................................................................................................
turbulence (A, B, KHTEST)
303
303
303
303
303
303
303
..........................................................................................................................................................................................................
National Wind Technology Center (NWTC) wind model with a Kelvin–Helmholtz billow
(KHTEST). For the constant wind profile, the wind speed was set to 11.4 m s−1 , with a 1/7 power
law vertical shear profile. The IEC Kaimal model is defined in IEC 61400-1 [17] and assumes
neutral atmospheric stability.
A mean wind speed of 13 m s−1 was used. The spectra K = u, v, w, where u, v and w are the
three orthogonal wind components, are given by
SK ( f ) =
4σK2 LK /ūhub
(1 + 6fLK /ūhub )5/3
,
(2.1)
where f is the cyclic frequency and LK is an integral scale parameter. More information can
be found in IEC 61400-1 [17] or the TurbSim user’s guide [18]. The NREL model (NWTCUP)
represents turbulent inflow characteristics at the NWTC, downwind of a major mountain range.
A mean wind speed of 13 m s−1 was used. For neutral and stable flows, the NWTCUP spectra are
defined by adding scaled versions of the SMOOTH-model spectra
SK ( f ) =
NumPeaks
K
pi,K SK,SMOOTH (Fi,K f ),
(2.2)
i=1
where NumPeaksK = 2 for all wind components K = u, v, w and the function SK,SMOOTH is
defined within the SMOOTH model. More information can be found in the TurbSim user’s
guide [18]. The sample time spacing was 0.01 s, corresponding to a sample rate of 100 Hz.
As the per-revolution harmonics were mainly of interest and the maximum rotor speed was
12.1 r.p.m., or 0.2 Hz, this sample rate was sufficient. Simulations were conducted for a healthy
turbine in addition to turbines with one of the three blades containing a shear web disbond
of 1, 2, 3, 4, 5 or 10 m in length. Two hundred output variables were recorded from the
simulations, including generator power, blade root moments, tri-axial blade accelerations along
the span, nacelle accelerations and many others. The first 30 s of simulations were discarded
when analysing the data to allow any start-up transients to damp out. The total simulation
time for each test, eliminating the first 30 s, was 1 h, allowing sufficient data with which to
perform averaging.
(d) Sensitivity analysis methods and parameters
A sensitivity analysis was performed by varying the extent of damage and inflow conditions
for the turbine. The NREL offshore 5 MW baseline wind turbine model and FAST were used
to simulate the varying parameters. Table 2 shows the matrix of FAST simulations performed
for the sensitivity analysis. Operational measurements were analysed for a healthy turbine in
addition to turbines with one of the three blades containing a shear web disbond of 1, 2, 3, 4, 5 or
10 m in length. Mean wind speed, horizontal shear and turbulence were among the aerodynamic
parameters used in this study. For all of the wind profiles, a 1/7 power law vertical shear profile
was applied. The wind speed was varied from 3 to 25 m s−1 in 0.22 m s−1 increments (totalling
101 simulations per turbine damage type). Horizontal shear parameters of 0.3, 0.6 and 0.9 (or
.........................................................
1m
disbond
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
healthy
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
35
per cent decrease (%)
10
de
sta 15
tio
n
25
20
15
10
5
bla
0
20
5 10 )
2 3 4 gth (m
1
25 0 d len
on
disb
–5
0
5
10
15
blade station
20
25
Figure 5. The per cent decrease of the torsional stiffness values for varying length disbonds for beam model segments spaced
along the length of the blade. (Online version in colour.)
30%, 60% and 90% horizontal shear) were used while varying the wind speed from 3 to 25 m s−1
in 0.22 m s−1 increments (totalling 303 simulations per turbine damage type). The horizontal
wind shear parameter is expressed as a linear spectrum of wind speed across the rotor disc. The
horizontal wind shear parameter ranged between −1 and 1, and it represents the wind speed
at the blade tip on one side of the rotor minus the wind speed at the blade tip on the opposite
side of the rotor, divided by the hub-height wind speed. The horizontal shear is measured in
the direction perpendicular to the normally prevailing wind vector. The turbulence models used
include the IEC Kaimal model with A turbulence, the IEC Kaimal model with B turbulence
and the NREL NWTC wind model with a KHTEST intense disturbance. See §2c for more
information on these turbulence models. For all turbulent wind profiles, the mean wind speed
was varied between 3 and 25 m s−1 in 0.22 m s−1 increments (totalling 303 simulations per turbine
damage type).
3. Shear web disbond simulation analysis and inflow sensitivity study
(a) Shear web disbond sensitivity and structural effects
The sensitivity of the extracted blade beam model stiffness values to the shear web disbond
was determined by calculating the per cent decrease in each of the stiffness values for all of the
sections in the reduced order model. All four stiffness parameters (axial, flap-wise, edge-wise and
torsional) decreased at the damage location as the disbond length was increased. The shear web
disbond also had the largest effect on torsional stiffness, indicating that measurements which are
sensitive to the blade’s torsional response will be good indicators that a shear web disbond is
present. Figure 5 shows the per cent decrease in torsional stiffness.
(b) Analysis of shear web disbond without blade sensors
Algorithms were first developed for detecting the shear web disbond using only the outputs from
FAST that would not require blade-mounted sensors to see if it was possible to construct a shear
web disbond detection method without blade sensors. Of the 200 variables that are provided
.........................................................
30
5
7
no disbond
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
per cent decrease (%)
40
40
35
30
25
20
15
10
5
0
–5
0
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
0.07
8
0.06
per cent change
0.04
0.03
0.02
0.01
0
–0.01
–0.02
–0.03
4
6
8
10
12
14
16
18
20
22
24
wind speed (m s–1)
Figure 6. RMS per cent change of power output for shear web disbond in varying wind speeds. (Online version in colour.)
as outputs from the FAST simulation, those which displayed significant percentage changes in
their RMS value or frequency response magnitude at the operating speed given a shear web
disbond were identified as key measurement channels. The rotor azimuth position output from
FAST was used as the reference signal for time synchronous averaging. The rotational resampling
was performed such that the azimuth signal was converted to radians, unwrapped and then the
measurement signal was interpolated so that each revolution contained the same number of data
samples with each sample corresponding to the same azimuth position of the rotor rotation. Data
blocks containing three revolutions were averaged together. By using more than one revolution
in the block size, the frequency resolution of the discrete Fourier transform of the time-averaged
signal was increased. The shear web disbond detection algorithms for non-blade sensors were all
based on detecting changes from baseline measurements either in the RMS response or 1p power
spectral density magnitude.
(i) Generator power
Overall, the generator power did not change significantly in the presence of a shear web disbond
when varying the wind speed, horizontal shear and turbulence wind profiles. The power output
experienced a few transients between the cut-in and rated speeds (3 and 11.4 m s−1 , respectively)
during the turbulent simulations, although all of the power output changes after the turbine
reached the rated speed were negligible. Figure 6 shows the RMS per cent change in power output
for the laminar wind profile in the presence of a shear web disbond.
(ii) Nacelle inertial measurements
For all wind profiles and damage cases, the RMS value of the nacelle acceleration in all three
directions increased at the turbine’s rated wind speed (11.4 m s−1 ) or higher. The transverse
nacelle acceleration showed a clear RMS increase for all cases between the rated speed and
approximately 20 m s−1 (shown in figure 7). In addition, the nacelle accelerations increased as
the shear web disbond length was increased. The 1p rotor order domain response magnitude was
analysed as well, but the trends of an increasing magnitude were not apparent for all of the wind
loading cases. Because these measurements were made at the nacelle hub, it is not possible to
determine the blade which has the shear web disbond. However, these measurements can be used
.........................................................
0.05
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
25
9
per cent change
15
10
5
0
–5
–10
–15
4
6
8
10
12
14
16
wind speed (m s–1)
18
20
22
24
Figure 7. RMS per cent change of transverse nacelle acceleration (yn ) for shear web disbond in 60% horizontal shear. (Online
version in colour.)
1.5
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
per cent change
1.0
0.5
0
–0.5
–1.0
–1.5
4
6
8
10
12
14
16
18
20
22
24
wind speed (m s–1)
Figure 8. RMS per cent change of edge-wise blade tip acceleration (yb ) for shear web disbond in laminar flow. (Online version
in colour.)
to indicate that a shear web disbond is present and then trigger more sophisticated measurements
to determine which blade has the disbond and the severity of the damage.
(c) Analysis of shear web disbond with blade sensors
Section 3b illustrated that some non-blade measurements are sensitive to the presence of a shear
web disbond in one of the three blades, but they lacked the ability to determine which blade(s)
contains the disbond. The next sections will investigate outputs from the FAST simulations that
would depend on blade-mounted sensors in an operating turbine.
.........................................................
20
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
40
10
per cent change
30
25
20
15
10
5
0
4
6
8
10
12
14
16
18
20
22
24
wind speed (m s–1)
Figure 9. 1p magnitude per cent change of span-wise blade tip acceleration (zb ) for shear web disbond in laminar flow. (Online
version in colour.)
2
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
0
–2
per cent change
–4
–6
–8
–10
–12
–14
–16
–18
4
6
8
10
16
12
14
wind speed (m s–1)
18
20
22
24
Figure 10. RMS per cent change of flap-wise blade tip acceleration (xb ) for shear web disbond in 90% horizontal shear. (Online
version in colour.)
(i) Blade Tip acceleration response
The per cent change in the RMS response magnitude of the edge-wise blade tip acceleration for
shear web disbond at different wind speeds in laminar flow is shown in figure 8. Although the
edge-wise blade tip acceleration was affected by the presence of a shear web disbond in all wind
conditions, these algorithms did not present a trend that could be correlated to an increase in
disbond length.
The span-wise blade tip acceleration 1p response differences in laminar flow are shown in
figure 9. When a shear web disbond was present, the 1p power spectrum response difference
was always positive up to 18 m s−1 for all inflow cases (and for all wind speeds in laminar
.........................................................
35
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
25
11
per cent change
15
10
5
0
–5
4
6
8
10
12
14
16
18
20
22
24
wind speed (m s–1)
Figure 11. 1p magnitude change of blade root-pitching moment for shear web disbond in varying wind speeds. (Online version
in colour.)
flow). Although there does not appear to be a trend that shows the severity of the damage, this
measurement can serve as a good indicator that a shear web disbond is present.
The flap-wise blade tip acceleration RMS response differences in the presence of 90%
horizontal shear are shown in figure 10. For all wind cases, there was a clear decrease in the
RMS response at the turbine’s rated speed (11.4 m s−1 ) for shear web disbond lengths of 2 m or
greater. The trend of a decreased flap-wise blade tip acceleration RMS response was apparent at
rated speed for all of the FAST simulations conducted in this study. In addition, the RMS response
decreased as the shear web disbond length was increased. Therefore, this measurement can serve
as a feature to indicate shear web disbond severity.
(ii) Blade root-pitching moments
Figure 11 shows the blade root-pitching moment 1p response differences for the laminar wind
loading case. For all of the wind cases up to a wind speed of 16 m s−1 , the 1p response increased
for a 4, 5 and 10 m shear web disbond. This measurement can be used as another indicator that a
severe shear web disbond is present in one of the blades. The blade root-pitching moment can be
measured with strain gauges located at the root of each blade.
(d) Summary of shear web disbond detection strategy
The results of the sensitivity analysis and key measurements have been used to develop a shear
web disbond detection strategy flowchart, as shown in figure 12. This strategy employs both blade
and non-blade sensor measurements. Specifically, non-blade sensor measurements are used as the
first indicator that a shear web disbond may be present and the blade sensors are used to confirm
that the damage is present and its level of severity. Using a single sensor measurement to first
identify potential damage will drastically reduce the necessary amount of processing and data
flow in situ. The strategy is summarized as follows:
1. Detect if a shear web disbond exists in one of the blades.
2. Determine the severity of the shear web disbond.
.........................................................
20
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
healthy
1 m disbond
2 m disbond
3 m disbond
4 m disbond
5 m disbond
10 m disbond
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
transverse
nacelle
accelerations
yes
span-wise tip
accelerations
>5% increase in
1p response
magnitude?
no
no disbond
yes
shear web disbond
blade tip
accelerations
apply data analysis
methods to compute RMS and
1p power spectrum response
determine shear
web disbond
severity
blade root
pitching
moments
Figure 12. Shear web disbond detection flow chart. (Online version in colour.)
3. Notify turbine operator of the disbond and severity so that a repair can be scheduled or
coordinated with other maintenance.
The results of the sensitivity analysis and key measurements have been used to refine the
shear web disbond detection strategy. This strategy employs both blade and non-blade sensor
measurements. Specifically, non-blade sensor measurements are used as the indicator for a shear
web disbond and the blade sensors (strain gauges at the blade root) are used to detect the
problematic blade and assess the level of severity. Each shear web disbond has been assigned
thresholds corresponding to the severity of the damage, as shown in table 3.
(e) Shear web disbond probability of detection analysis
Probability of detection (POD) values were calculated for detecting the presence of a shear web
disbond. In addition, probability of classification (POC) values were calculated for correctly
categorizing the three different shear web disbond damage states which vary by severity. See
table 3 for the damage state classifications of shear web disbond. State 2 refers to a 1–2 m
disbond, state 3 is a 3–5 m disbond and state 4 is a disbond of 10 m or more. These damage state
classifications were used for each FAST simulation and inflow condition. If the measurement
at a given wind speed, profile and damage state met the criteria described in table 3, then
it was deemed a success. Otherwise, it was deemed a failure. For example, the blade rootpitching moment is extracted from the simulation for the 3.88 m s−1 laminar wind profile and
for a turbine with a blade which has a 4 m shear web disbond. If there is an increase in the
.........................................................
no disbond
1%
increase in
RMS
response?
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
no
12
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
Table 3. Shear web disbond damage state and corresponding feature used for classification.
13
..........................................................................................................................................................................................................
..........................................................................................................................................................................................................
state 2 (1, 2 m disbond)
1% increase in measured RMS transverse nacelle acceleration, less than
0.5% increase in 1p blade root-pitching moment
..........................................................................................................................................................................................................
state 3 (3, 4, 5 m disbond)
greater than 0.5% and less than 5% increase in 1p blade root-pitching
moment
..........................................................................................................................................................................................................
state 4 (10 m disbond or longer)
greater than 5% increase in 1p blade root-pitching moment
..........................................................................................................................................................................................................
blade root-pitching moment 1p and that increase is greater than 0.5% the healthy response and
less than 5% greater than the healthy response, then the detection is a success and given a ‘1’
value at that data point. If it does not meet the criteria, it is given a ‘0’ value. The number of
successes are then added up and that total is divided by the total number of simulations in that
wind profile (101 simulations). The resultant percentage is the POD or POC for that damage
state and wind profile. For the overall shear web disbond POD calculation, the measurements
needed to meet the minimum damage state criteria to be deemed detectable, i.e. a 1% increase in
measured RMS transverse nacelle acceleration versus expected healthy RMS transverse nacelle
acceleration. Table 4 shows the POD values for detecting the presence of a disbond as well as
the POC values categorizing the damage into each damage case, respectively. The POD and
POC values were calculated over the entire wind speed range in addition to an optimized wind
speed range for maximizing the resulting POD/POC value for accurate damage detection for
all wind loading cases. The optimized wind speed range and corresponding POD/POC values
are highlighted in bold in the table. In addition, each POD and POC value was weighted by the
Weibull distribution to incorporate the probability of occurrence of each hub-height wind speed
used within the analysed range, i.e. a uniform distribution was applied for the wind speed
bins used for the raw probability and a Weibull distribution was applied for the wind speed bins
used for the Weibull weighted probability. The POD results show that the developed algorithms
are 100% successful for all of the laminar, 30% horizontal shear and 60% horizontal shear wind
profiles. The POD and POC values are also approximately 75% or greater for all wind profiles
except the 90% horizontal shear simulations. There is a large decrease in that POD because
the aerodynamic loading greatly influences the transverse nacelle acceleration response and
this feature becomes the dominating feature at that measurement location rather than a shear
web disbond in one of the three blades. In the real world, however, a 90% horizontal shear
wind profile does not occur nearly as often as the laminar and less extreme horizontal shear
wind profiles.
4. Conclusion
Aeroelastic simulations and aerodynamic sensitivity analyses were conducted on a 5 MW offshore
wind turbine to demonstrate the effectiveness of rotor blade sensing and data analysis methods
for monitoring the structural dynamic response of wind energy systems. The analysis provided
an understanding of the loading experienced by the turbine rotor as well as the influence this
loading had on the reliability of the wind turbine. Time synchronous averaging, RMS vibration
analysis and order domain vibration analysis were the primary data analysis methods applied.
Several sensing and analysis schemes were evaluated for shear web disbond detection in a
simulated 5 MW offshore wind turbine. The shear web disbond was modelled in NuMAD at
several different lengths. The stiffness properties were extracted from the blade models and
a large reduction in torsional stiffness was observed as the shear web disbond length was
.........................................................
1% increase in measured RMS transverse nacelle acceleration and
>5% increase in measured 1p span-wise blade tip acceleration
response
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
state 1 (shear web disbond is present)
8.5–17.08 m s−1 (%)
3–25 m s−1 (%)
3–25 m s−1 (%)
3–25 m s−1 (%)
8.5–17.08 m s−1 (%)
state 3 (3–5 m disbond)
3–25 m s−1 (%)
8.5–17.08 m s−1 (%)
state 4 (≥10 m disbond)
84.16
77.24
Weibull weighted
100.00
84.16
100.00
83.33
100.00
84.16
100.00
77.24
100.00
77.16
100.00
77.24
100.00
74.26
78.07
Weibull weighted
100.00
77.79
100.00
69.11
100.00
72.79
100.00
77.95
100.00
77.55
100.00
77.95
100.00
44.55
60.18
Weibull weighted
100.00
35.73
100.00
36.61
100.00
36.17
100.00
58.30
100.00
58.61
100.00
58.46
100.00
15.84
25.19
Weibull weighted
32.50
10.82
32.50
11.29
32.50
10.98
32.50
23.21
39.18
23.58
39.18
23.34
39.18
37.62
50.49
Weibull weighted
75.00
34.27
75.00
29.80
75.00
27.94
75.00
50.03
74.99
48.31
74.99
47.91
74.99
45.54
62.86
Weibull weighted
85.00
42.39
85.00
36.07
85.00
33.82
85.00
61.73
84.98
59.38
84.98
59.65
84.98
45.54
68.43
Weibull weighted
80.00
33.82
80.00
34.72
80.00
28.41
76.00
64.93
88.06
65.41
88.06
60.51
85.76
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
.........................................................
.............................................................................................................................................................................................................................................................................................................................................
88.06
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
KHTEST turbulence
.............................................................................................................................................................................................................................................................................................................................................
84.98
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
B turbulence
.............................................................................................................................................................................................................................................................................................................................................
74.99
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
A turbulence
.............................................................................................................................................................................................................................................................................................................................................
39.18
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
90% shear
.............................................................................................................................................................................................................................................................................................................................................
100.00
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
60% shear
.............................................................................................................................................................................................................................................................................................................................................
100.00
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
30% shear
.............................................................................................................................................................................................................................................................................................................................................
100.00
.............................................................................................................................................................................................................................................................................................................................................
raw
.............................................................................................................................................................................................................................................................................................................................................
Laminar
.............................................................................................................................................................................................................................................................................................................................................
8.5–17.08 m s−1 (%)
state 2 (1–2 m disbond)
presence of damage
Table 4. POD and POC for shear web disbond. The POD values are shown under presence of damage and the POC values are shown in the other columns.
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
14
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
under the title ‘Wind turbine blade shear web disbond detection using rotor blade operational sensing and
data analysis’.
Funding statement. The authors acknowledge Sandia National Laboratories (contract agreement 1067259) and
the US Department of Energy for their support of this paper and their continuing support of the wind energy
research efforts being performed at Vanderbilt University.
References
1. Wiser R, Bolinger M. 2012 2011 Wind technologies market report. Springfield, VA: US
Department of Commerce.
2. Greco A, Sheng S, Keller J, Erdemir A. 2013 Material wear and fatigue in wind turbine systems.
Wear 302(Suppl. 1), 1583–1591. (doi:10.1016/j.wear.2013.01.060)
3. García Márquez FP, Tobias AM, Pinar Pérez JM, Papaelias M. 2012 Condition
monitoring of wind turbines: techniques and methods. Renew. Energy 46, 169–178.
(doi:10.1016/j.renene.2012.03.003)
4. Yang W, Tavner PJ, Crabtree CJ, Wilkinson M. 2010 Cost-effective condition monitoring
for wind turbines. IEEE Trans. Ind. Electron. 57(Suppl. 1), 263–271. (doi:10.1109/tie.2009.
2032202)
5. Sayer F, Antoniou A, van Wingerde A. 2012 Investigation of structural bond lines in wind
turbine blades by sub-component tests. Int. J. Adhesion Adhes. 37, 129–135. (doi:10.1016/
j.ijadhadh.2012.01.021)
6. Sayer F, Kleiner F. 2012 Rotor blade reliability. Wind Systems, March 2012, pp. 30–35. See
http://www.windsystemsmag.com/article/detail/344/rotor-blade-reliability.
7. Wetzel K. 2009 Defect-tolerant structural design of wind turbine blades. In 50th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conf., Palm
Springs, CA. Reston, VA: American Institute of Aeronautics and Astronautics.
8. Tippmann JD, Manohar A, Lanza di Scalea F. 2012 Wind turbine inspection tests at UCSD.
Proc. SPIE 8345, 83451Q. (doi:10.1117/12.915323)
9. Mellott K. 2007 Gamesa pinpoints turbine-blade flaw. The Tribune-Democrat, 11 May 2007.
10. Jensen FM, Kling A, Sørensen JD. 2012 Scale-up of wind turbine blades—changes in failure
type. In Proc. European Wind Energy Conf. and Exhibition (EWEC 2012), Copenhagen, Denmark,
16–19 April 2012.
11. Hermann TM, Mamarthupatti D, Locke JE. 2005 Post-buckling analysis of a wind
turbine blade substructure. J. Solar Energy Eng. 127(Suppl. 4), 544–552. (doi:10.1115/1.
2037093)
.........................................................
Data accessibility. Datasets that support the results in this article can be accessed on Dryad (http://datadryad.org)
15
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
increased. Shear web disbond lengths of 1, 2, 3, 4, 5 and 10 m were included and the wind
profile was varied by mean wind speed, horizontal shear and turbulence characteristics. The
transverse nacelle acceleration clearly increased for all aerodynamic cases between the rated wind
speed of the wind turbine (11.4 m s−1 ) and approximately 20 m s−1 . Although measurements
made at the nacelle hub cannot indicate the problematic blade, this measurement can serve
as a good indicator that a shear web disbond is present on one of the three blades. The RMS
value and 1p magnitude of the power spectrum of the blade tip accelerations were sensitive to
the presence of a shear web disbond. In particular, the flap-wise RMS differences were a good
indicator of shear web disbond severity. The measurements of the blade root-pitching moment 1p
power spectrum magnitude were effective in detecting a shear web disbond. This measurement
was applied to the detection strategy as a secondary feature for indicating the severity
of damage.
On the basis of results of the sensitivity study, the detection strategy was developed to
accommodate several wind loading cases and thus increased the robustness of the detection
algorithms. Robust probabilities of detection were derived as an algorithm success metric. The
POD was 100% for the optimized wind speed ranges including the laminar, 30% horizontal shear
and 60% horizontal shear conditions. These POD values validate the detection strategy which
incorporates multiple measurements to detect and quantify a shear web disbond in order to
optimize turbine maintenance protocols and minimize associated O&M costs.
Downloaded from http://rsta.royalsocietypublishing.org/ on October 29, 2015
.........................................................
View publication stats
16
rsta.royalsocietypublishing.org Phil. Trans. R. Soc. A 373: 20140345
12. Jonkman JM, Buhl ML. 2005 FAST user’s guide. NREL/TP-500-38230. Golden, CO: National
Renewable Energy Laboratory.
13. Ghoshal A, Sundaresan MJ, Schulz MJ, Pai PF. 2000 Structural health monitoring
techniques for wind turbine blades. J. Wind Eng. Ind. Aerodyn. 85(Suppl. 3), 309–324.
(doi:10.1016/s0167-6105(99)00132-4)
14. Kusnick J. 2012 Rotor blade operational data analysis methods and applications for condition
monitoring of vertical and horizontal axis wind turbines. West Lafayette, IN: Purdue
University.
15. Jonkman J, Butterfield S, Musial W, Scott G. 2009 Definition of a 5-MW reference wind turbine
for offshore system development. Golden, CO: National Renewable Energy Laboratory.
16. Griffith DT, Yoder N, Resor B, White J, Paquette J. 2012 Structural health and prognostics
management for offshore wind turbines: an initial roadmap. SAND2012-10109. Albuquerque,
NM: Sandia National Laboratories.
17. International Electrotechnical Commission et al. 1999 IEC 61400-1: wind turbines—
part 11: acoustic noise measurement techniques. IEC 61400-11:1998. (doi:10.3403/
01559962)
18. Jonkman BJ, Buhl ML. 2006 TurbSim user’s guide. NREL/TP-500-39797. Golden, CO: National
Renewable Energy Laboratory.