Transcript PowerPoint-presentatie

Master Thesis Presentation
Blade Load Estimations by a Load
Database for an Implementation in
SCADA Systems
17-7-2015
Carlos Ochoa A.
TUD idnr. 4145658
TU/e idnr. 0756832
October 23Th, 2012
Delft
University of
Technology
Challenge the future
CONTENTS
1. Introduction
2. Objective
3. OWEZ Data
4. Method
5. Load Comparison Between Turbines
6. Load Database Construction
7. Database Estimators Validation
8. Conclusions
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
2
1. Introduction
• Real Wind Conditions
Z
Different inflow parameters affect the

FT(V,u,z)
turbine behavior, factors as:

• Wind Speed
• Wind Shear
Occurrences
MY(Ω)
FG
• Turbulence
• Atmospheric stability
• etc.
All these parameters have an impact over
the forces and moments of the turbine.
Y
Ω
Q(V)
Turbulence
Wind Speed
X
FC
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
3
1. Introduction
• Real Wind Conditions
• Loads and Fatigue
The cyclic loads affects the fatigue in
the materials, this limits the lifetime
of a wind turbine.
In a wind turbine, the blades are
structural
components that have the largest provability
of failure after determinate period.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
4
1. Introduction
• Real Wind Conditions
• Fatigue
• SCADA
Collect, monitor & storage of turbine behavior through
the Standards Signals:
• Generator rotational speed and acceleration
• Electrical power output.
• Pitch angle.
• Lateral and longitudinal tower top acceleration.
• Wind Speed and wind direction.
Only the main Statistics of the selected variables
are computed.
• Min, max, average & standard deviation.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
5
2. Objectives
Develop a method to estimate the blade load behavior by retrieving information from a measurement
database depending on the standard signals of the wind turbine, which are usually stored by the SCADA
system.
How accurate are the fatigue damages and the cumulative fatigue estimations when comparing them against
other load estimation methods results?
Neural Networks
Regression Techniques
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
6
3. OWEZ Data
High frequency measurement data (32Hz) from two turbines were obtained trough a measuring campaign at
OWEZ. 41 different signals were measured for each different turbine for several months.
Key Signals Measured (32Hz):
•Stain signals from the root of the blade
• Edgewise
• Flapwise
•Other 70 signals
• Standard signals
Standard Reconstruction of SCADA data
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
7
4. Method
The data was classified depending on the turbine, mean winds speed and turbulence intensity. Under each
wind inflow condition different load behavior is produced. From these, Rainflow counting matrixes and load
amplitudes histograms are obtained.
From the load amplitude histograms,
Load time Series
Site Inflow Condition Characterization
load estimators can be derived. The
database.
To
perform
elements
of
a
the
load
estimation,
the
database can be
retrieved by the use of the SCADA
Turbulence Intensity
groups of estimators are storage on a
2
4
6
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
8 10 12 14 16 18 20 22 24
30
Rainflow Counting
Matrixes
28
26
24
22
0
20 Histograms
Load Amplitude
18
16
14
12
Load Distribution
10 Functions
2940
8
6
4
2
2
4
6
8 10 12 14 16 18 20 22 24
Load Estimators
Mean Wind Speed
m s
data.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
8
4. Method
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
9
4. Method
To convert the Rainflow cycle matrixes to load histograms certain material characteristics were assumed.
The geometry of the blade root (thickness and chord) was estimated.
Stress Load Amplitude S [MPa]
S-N Curve for the Assumed Material
300
250
A linear Goodman diagram was obtained from the use of the
200
150
assumed blade characteristics. By its use, load cycle histograms
Red. Chi-S
R^2
qr
1359.72611
0.54569
Equation
100
S = a N ^b
Coefficient
a
b
Value
433.668
Std Err
or
9.6486
-0.09242
0.00247
were obtained.
50
2
10
3
10
4
10
5
10
6
10
7
10
Cycles to Failure N
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
10
5. Load Comparison Between Turbines
From the OWEZ data, the load patterns from both turbines were compared. From all the wind conditions, the
comparison results shown a remarkable similitude between loads.
Histogram for the Wind Bin at 11
13 m s and 15
13
TI Case with Bins of 5 KNm.
Edgewise Distribution
Flapwise Distribution
Ocurrences
Ocurrences NNFlap
Flap
Ocurrences
Ocurrences NNFlap
Flap
100
100
10
10
1
1
0.1
0.1
0.01
0.01
0.001
00
200
400
600
800
200
400
600
800
Cycle Amplitude
Amplitude
KNm
Cycle
KNm
100
100
10
10
1
1
0.1
0.1
0.01
0.001
0.01
0
0
200
400
600
800
200
400
600
800
Cycle Amplitude
KNm
Cycle Amplitude
KNm
• Turbine 8
• Turbine 7
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
11
6. Load Database Construction
All the inflow condition measured were processed
to obtain the load database. Interesting patterns
came up when analyzing the changes of
the
load behavior trough the wind speed.
Especially in the edgewise direction.
Edgewise Mean Peak Load Variation with the Wind Speed
Mean Loading (KNm)
500
400
300
200
100
0
4
8
12
16
20
24
Wind Speed (m/s)
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
12
6. Load Database Construction
In contrast, other patterns came up when analyzing the load behavior changes trough the turbulence intensity.
Edgewise
Site Inflow Condition Characterization
4
6
8 10 12 14 16 18 20 22 24
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
2
4
6
0
Site Inflow Condition Characterization
2940
2
8 10 12 14 16 18 20 22 24
Mean Wind Speed
m s
Mean Wind Speed 7m/s.
Turbulence Intensity:
• 9%
• 11%
• 13%
• 15%
• 17%
Turbulence Intensity
Turbulence Intensity
2
4
6
8 10 12 14 16 18 20 22 24
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
2
4
6
0
2940
8 10 12 14 16 18 20 22 24
Mean Wind Speed
m s
Flapwise
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
13
6. Load Database Construction
From all the load histograms generated, load distributions functions were constructed; all these were
normalized to 10-min. All the load distribution functions were made by piecewise functions, for the edgewise
case three polynomials were used. For the flapwise functions only two functions were used.
Site Inflow Condition Characterization
Turbulence Intensity
2
4
6
8 10 12 14 16 18 20 22 24
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
2
4
6
0
2940
8 10 12 14 16 18 20 22 24
Mean Wind Speed
m s
To fit better the tail behavior, a moving average with a ratio of 1:5 was used . The tails were fitted with a linear or a
quadratic function in the logarithmic scale.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
14
6. Load Database Construction
Respect to the idling condition, it was characterized only for all the speeds lower the cut-in wind speed. It was
interesting to note the apparent gravity peak pattern seen in the flapwise direction.
The same gravity peak appear at
power production cases with low
winds speeds. It is caused by the
high pitching angles of the idling
conditions.
In the edgewise direction, it
causes
the
appearance
of
a
double peak.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
15
6. Load Database Construction
From all the load distribution functions load estimators can be derived; they can take form as equivalent loads,
fatigue damages or even maximum load values were obtained. The next are examples from the fatigue
damages normalized for 10-min.
Linear fatigue damage increase with the turbulence intensity for the edgewise direction, exponential for flapwise.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
16
7. Database Estimators Validation
When comparing a single random 10-min. load sequence with the load distributions from the database, it was
observed they does not match well. Scatter appears especially at the tail of the edgewise distribution.
Site Inflow Condition Characterization
Turbulence Intensity
2
4
6
8 10 12 14 16 18 20 22 24
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
30
28
26
24
22
20
18
16
14
12
10
8
6
4
2
2
4
6
0
2940
8 10 12 14 16 18 20 22 24
Mean Wind Speed
m s
Furthermore, it was noticed the histogram data points show spaces between bin counts. Not every 5KNm in
the cycle load amplitude axis has a count.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
17
7. Database Estimators Validation
From the database constructed is possible to estimate the cumulative fatigue of such turbine. It can be estimated
with the database information and compared with the sum of all the 10-min. calculated fatigue damages.
From:
200-300 KNm
11/20 counts
From:
650 -700 KNm
7/10 counts
The error range from 31.4% and 41%. They can be attributed to the
scatter and the missed counts trough each single load histogram.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
18
7. Database Estimators Validation
It was possible to improve the cumulative fatigue
estimation by the use of a multiplication constant.
The main idea was not to fix the final value of the
estimation with the calculation result, but to make
the slope of this line as similar as possible to the
calculation line. The multiplication constant
obtained was 0.835.
With this, the errors diminished to 10.7% and 15%.
Using the database from the turbine 7 data and its
correction, the cumulative fatigue of the turbine 8 was
estimated and its errors range from 9.44 to 10.3%
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
19
7. Database Estimators Validation
From the database made with the turbine 7 another turbine cumulative fatigue was estimated.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
20
7. Database Estimators Validation
For the previous results, all the single fatigue estimation were retrieved from the load database by means
of the reconstructed SCADA data. For this, the pitching angle information is extremely useful to identify the
turbine status. The main statistical values of the wind speed where used as well.
Wind Direction
In real life applications, other variables from the
SCADA data, as the electrical power output or the
generator
Power Production
Pitch Angle: 0-25°
Idling
Pitch Angle: 25-40°
speed,
could
be
used
to
corroborate the turbine status.
The load estimators do not necessarily have to
be retrieved from the database each 10-min.
Start Up
Pitch Angle: ～ 45°
This period can be fixed by the frequency the
SCADA system update its variables.
Pause, Stop & E. Stop
Pitch Angle: ～ 90°
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
21
8. Conclusions
• It was possible to create a load estimation method based on previous turbine measurements and on SCADA data
information.
• The fatigue accumulation estimations from both turbines give back smaller errors than other methodologies.
The errors range from 9 to 15%.
• Estimations by neural networks produce errors ranging from 12 to 22% depending on the number of nodes used in
the network.
• Regression techniques have errors ranging from 2 to 23%.
Nevertheless, the methodology proposed in this report still needs to be validated by more turbines.
• Given the similar load patterns obtained from different turbines under the same wind conditions, the method
developed could be applied to other couple of turbines.
• Thanks to the cumulative loading estimation of the turbine blades, would help to determine wheatear or not to
extend the turbine service lifetime or modify the turbine maintenance program, this could mean to be a
significant monetary advantage.
Blade Load Estimations
by –Database
for SCADA
SET MSc
Wind Energy
22
Thanks for the Attention
Questions…?
New
York–Long
340MW
BladeThe
Load
Estimations
byIsland
for Project
SCADA
SETSupport
MSc
–Database
Wind
Energy
Esbjerg,
Structure
Design 23