International Journal of Recent Technology and Engineering (IJRTE)

ISSN: 2277-3878, Volume-8, Issue-2S11, September 2019

Logistic Model Tree Classifier for Condition

Monitoring of Wind Turbine Blades
A. Joshuva, G. Deenadayalan, S. Sivakumar,
R. Sathishkumar, R. Vishnuvardhan 

Abstract--- Wind energy is one of the essential renewable the key segments of the wind turbine like gearbox,
energy resources because of its consistency due to the mechanical brake, electrical generator, control frameworks,
development of the technology and relative cost affordability. The yaw drive, and so forth [4].
wind energy is converted into electrical energy using rotating
The main objective of condition monitoring method is to
blades which are connected to the generator. Due to
environmental conditions and large construction, the blades are create the power yield as calculated. When there is an
subjected to various faults and cause the lack of productivity. The awareness of the time and categories of defects,
downtime can be reduced when they are diagnosed periodically maintenance activities can be arranged ahead of time [5].
using condition monitoring technique. These are considered as a Thus, condition monitoring of blade fault is a critical
machine learning problem which consists of three phases, activity. There are two types of approaches which are
namely feature extraction, feature selection and fault
carried out for condition monitoring: traditional approach
classification. In this study, statistical features are extracted from
vibration signals, feature selection are carried out using J48 and machine learning approach. The traditional approach is
algorithm and the fault classification was carried out using mainly used for frequency component does not change with
logistic model tree algorithm.” respect to time.
Keywords--- Fault Diagnosis; Condition Monitoring; Rotating machines produce non-stationary signals. Since
Statistical Features; J48 Algorithm; Logistic Model Tree (LMT) the frequency components change due to wear and tear, fault
Algorithm. discrimination is very difficult using an automated system in
the traditional approach. Hence, not preferred. In machine
I. INTRODUCTION learning approach, algorithms have the capability to learn
In recent years, wind energy becomes the most demanded continuously and adapt themselves to the varying situations.
resource in power generation. Because of worldwide Researchers often resort to machine learning approach for
ecological contamination due to other resources, the fault diagnosis of mechanical systems.”
sustainable energy and natural resources like wind energy Many studies were carried out using machine learning
have risen [1]. “Wind energy is one of the efficient studies and simulation studies to name a few, Godwin and
renewable sources and an alternative option for commonly Peter Matthews [6] have done classification and detection of
used sources. The objective of the wind turbine is to boost wind turbine pitch faults through SCADA data analysis and
ideals by extreme energy extraction and in this manner, the RIPPER algorithm which yield them 87.05% classification
performance of the wind turbine frequently differs with accuracy in pitch angle fault. “N. Dervilis et al., [7] carried
wavering winds. To make wind energy more focused from out a research on damage diagnosis for a wind turbine blade
different sources of energy, like performance, accessibility, using pattern recognition such as principal component
reliability, effectiveness, the life of turbines must be analysis (PCA), nonlinear principal component analysis
improved [2]. (NLPCA), artificial neural network (ANN), auto-associative
Generally, horizontal axis wind turbine (HAWT) is used neural network (AANN) and Radial Basis Functions (RBF)
for power generation since it is more effective than the using vibration signals.
vertical axis wind turbine [3]. Most HAWT are with three Mark Mollineaux et al., [8] carried out a work on damage
rotor blades typically placed upwind of the tower and the detection methods of wind turbine blade testing with wired
nacelle. The nacelle is typically furnished with anemometers and wireless accelerometer sensors using benchmark data
and a wind vane to calculate the wind rate and direction of and autoregressive moving average (ARMA) and
the wind. The nacelle also contains aviation light signal and Continuous Wavelet Transform (CWT) used as modeling
techniques.
Mahmood Shafiee et. al [9] have done work on an
Manuscript received September 16, 2019.
A. Joshuva*, Centre for Automation and Robotics (ANRO), opportunistic condition based maintenance policy for
Department of Mechanical Engineering, Hindustan Institute of Technology offshore wind turbine blades subjected to stress corrosion
and Science, Old Mahabalipuram Road, Padur, Kelambakam, Chennai, cracking and environmental shocks and simulated using
T.N, India. (e-mail: [email protected])
G. Deenadayalan, Centre for Automation and Robotics (ANRO), MATLAB.
Department of Mechanical Engineering, Hindustan Institute of Technology H. M. Slot et al., [10] have made a review of coating life
and Science, Old Mahabalipuram Road, Padur, Kelambakam, Chennai, models on the leading edge erosion of coated wind turbine
T.N, India.
S. Sivakumar, Department of Mechanical Engineering, Hindustan
blades and he suggested the material and prediction of the
Institute of Technology and Science, Old Mahabalipuram Road, Padur, lifetime of the blade.
Kelambakam, Chennai, T.N, India..
R. Sathishkumar, Department of Automobile Engineering, Hindustan
Institute of Technology and Science, Old Mahabalipuram Road, Padur,
Kelambakam, Chennai, T.N, India.
R. Vishnuvardhan, Department of Mechatronics Engineering, Sri
Krishna College of Engineering and Technology, Coimbatore, T.N, India.

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 202 & Sciences Publication
 LOGISTIC MODEL TREE CLASSIFIER FOR CONDITION MONITORING OF WIND TURBINE BLADES

Shizhong Zhang et al., [11] carried out a work on design Wind Turbine with
Accelerometer
and analysis of jet based laboratory equipment for
performance evaluation on the erosion of wind turbine blade Data Acquisition (Vibration signal)
coatings.
Xiang Li et al., [12] have carried out a research on crack Feature Extraction (Statistical)

growth, sparse pursuit for wind turbine blade using vibration

data and lamb wave propagation. Wei Xie et al., [13] have Feature Selection using J48 Algorithm
made a novel folding blade of wind turbine rotor for
effective power control using MATLAB and achieved
maximum reduction of 51.1% in pitch control (Optimum Training Data Set Test Data Set

fold angles of 550 and 900).

Simon Hoell and Piotr Omenzetter [14] have carried out a Training model
structural damage detection in wind turbine blades based on
time series representations of dynamic responses using
Trained Model
vibration data and cross-correlations, principal component
analysis (PCA), genetic programming (GP) as the diagnostic
algorithms.” Output

Wang Yongzhi et al., [15] have carried out a study on a

composite wind turbine blade aerodynamic and structurally Fault Detection in Blade

integrated design optimization based on radial basis function Figure 1: Methodology

(RBF) meta-model, blade element momentum (BEM)
theory, finite element method (FEM) and multi-island II. EXPERIMENTAL STUDIES
genetic algorithm (MIGA).
“This paper aims to reduce the mass of blade under some The main aim of this study is to classify whether the
constraints, including the power and deflection at the rated blades are in good condition or in a defective state. “If it is
wind speed, and the strength and deflection under the defective, then the objective is to identify the type of fault.
ultimate case. P.U. Haselbach et al., [16] studied the effect The experimental setup and experimental procedure are
of delamination on local buckling in wind turbine blades described in the following subsections.
using FEM analysis. Experimental Setup
They simulated by bending the blade in flap-wise
The experiment was carried out on a 50W, 12V variable
direction and conducted the experiment. Although numerous
speed wind turbine (MX-POWER, model: FP-50W-12V)
works were carried out using simulation studies, only a very
[18]. The technical parameters of a wind turbine are given in
few of the experimental studies for condition monitoring on
Table 1. The wind turbine was mounted on a fixed steel
wind turbine blades were carried out.
stand in front of the open circuit wind tunnel outlet. The
Some researchers carried out fault diagnosis system using
wind tunnel speed ranges from 5m/s to 15 m/s and acts as a
machine learning approach and they considered very few
wind source to start the wind turbine. The wind speed was
faults in their study.
varied continuously in order to simulate the environmental
Only selected algorithms were used to perform the fault
wind condition. The experimental setup is shown in Figure
prediction on wind turbine blades and many algorithms are
2. Piezoelectric type accelerometer was used as a transducer
yet to be studied [17]. Figure 1 shows the methodology of
for acquiring vibration signals. It has high sensitivity for
the work done.”
detecting faults. Hence, accelerometers are widely used in
The contribution of the present study
condition monitoring. In this case, an uniaxial accelerometer
1. This study considers more than two faults for fault
of 500g range, 100 mV/g sensitivity, and resonant frequency
diagnosis of the wind turbine (blade crack, erosion,
around 40 Hz was used.
hub-blade loose connection, pitch angle twist and
Table 1: Technical Parameters of Wind Turbine
blade bend).
Model FP-50W-12V
2. Vibration signals are taken for the blade using data Rated Power 50 W
acquisition. Rated Voltage 12 V
3. Statistical analysis is used for feature extraction. Maximum Current 4A
Rated Rotating Rate 850 rpm
4. The J48 algorithm is used for feature selection.
Start-up Wind Speed 2.5 m/s
5. Logistic model tree (LMT) algorithm is used for Cut-in Wind Speed 3.5 m/s
feature classification. Cut-out Wind Speed 15 m/s
The rest of the paper is organized as follows. Section 2 Security Wind Speed 40 m/s
presents the experimental setup and experimental procedure. Rated Wind Speed 12.5 m/s
Engine Three-phase permanent magnet generator
In section 3, feature extraction is explained, followed by
Rotor Diameter 1050 mm
feature selection in section 4. The classifiers used in this Blade Material Carbon fiber reinforced plastics
study are explained in section 5. The classification accuracy
of the LMT model was discussed in section 6. Conclusions
are presented in the final section (section 7).

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 203 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-2S11, September 2019

The piezoelectric accelerometer (DYTRAN 3055B1) was Experimental Procedure

mounted on the nacelle near to the wind turbine hub to In the present study, three-blade variable horizontal axis
record the vibration signals using an adhesive mounting wind turbine (HAWT) was used. Initially, the wind turbine
technique. It was connected to the DAQ system through a was considered to be in good condition (free from defects,
cable. The data acquisition system (DAQ) used was NI USB new setup) and the signals were recorded using an
4432 model. The DAQ card has five analogue input accelerometer. These signals were recorded with the
channels with a sampling rate of 102.4-kilo samples per following specifications.
second with 24-bit resolution. The accelerometer is coupled 1. Sample length: The sample length was chosen
to a signal conditioning unit which consists of an inbuilt long enough to ensure data consistency; and also
charge amplifier and an analogue-to-digital converter the following points were considered. ARMA
(ADC). From the ADC, the vibration signal was taken. measures are more meaningful when the number of
These vibration signals were used to extract features through samples is sufficiently large. On the other hand, as
feature extraction technique. One end of the cable is plugged the number of samples increases the computation
to the accelerometer and the other end to the AIO port of time increases. To strike a balance, a sample length
DAQ system. NI – LabVIEW was used to interface the of 10000 was chosen.
transducer signal and the system (PC). 2. Sampling Frequency: The sampling frequency
should be at least twice the highest frequency
contained in the signal as per Nyquist sampling
theorem. By using this theorem sampling
frequency was calculated as 12 kHz (12000Hz).
3. Number of samples: Minimum of 100 (hundred)
samples were taken for each condition of the wind
turbine blade and the vibration signals were stored
in data files.

Figure 2: Wind Turbine Setup

Good condition blade Blade with crack

Blade with pitch angle twist Blade with erosion

Hub-blade loose connection Blade with bend (Top View)

Figure 3: Various Blade Fault Conditions
The following faults were simulated one at a time while
all other components remain in good condition and the

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 204 & Sciences Publication
 LOGISTIC MODEL TREE CLASSIFIER FOR CONDITION MONITORING OF WIND TURBINE BLADES

corresponding vibration signals were acquired. Figure 3 wind. The blade was made to flap wise bend with
shows the different blade fault conditions which are 10 º angle.
simulated on the blade. b) Blade crack (BC-2): This occurs due to foreign
a) Blade bend (BB): This fault occurs due to the object damage on the blade while it is in operating
high-speed wind and complex forces caused by the condition. On the blade, 15mm crack was made.

Figure 4: Vibration Signal Plot

c) Blade erosion (BE): This fault is due to the heavy vibration to the framework. To attain this
erosion of the top layer of the blade by the high- fault, blade pitch was twisted about 12º angle with
speed wind. The smooth surface of the blade was respect to the normal blade condition.
eroded using emery sheet (320Cw) to provide an In Figure 4, the vibration signals (sample number vs.
erosion effect on the blade. amplitude) are shown which were taken for different
d) Hub-blade loose contact: This fault generally conditions of the wind turbine blade (good condition blade,
occurs on a wind turbine blade due to an excessive blade bend, blade erosion, hub-blade loose connection,
runtime or usage time. The bolt connecting the hub blade crack and pitch angle twist).
and blade was made loose to obtain this fault.
e) Blade pitch angle twist (PAT): This fault occurs
due to the stress on the blade caused by high-speed
wind. This makes the pitch get twisted, creating a

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 205 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-2S11, September 2019

III. FEATURE EXTRACTION – STATISTICAL value v. Note the first term in the equation for gain is just
ANALYSIS the entropy of the original collection S and the second term
The vibration signals were acquired for good and other is the expected value of the entropy after S is partitioned
faulty conditions of the blade. If the time domain sampled using feature A. The expected entropy described by the
signals are given directly as inputs to a classifier, then the second term is simply the sum of the entropies of each
number of samples should be constant. “The number of subset Sv, weighted by the fraction of samples |Sv|/|S| that
samples obtained which are the functions of rotation of the belong to Sv. Gain (S, A) is, therefore, the expected
blade speed. Hence it varies with speed. Also, the number of reduction in entropy caused by knowing the value of feature
digitized data points in a signal is too large; generally the A. Entropy is a measure of homogeneity of the set of
work of classifiers may not handle it efficiently. Therefore, examples and it is given by
a few features must be extracted before the classification (2)
process. Descriptive statistical parameters [19] such as sum, where, c is the number of classes, Pi is the proportion of S
mean, median, mode, minimum, maximum, range, belonging to class „i’.
Skewness, kurtosis, standard error, standard deviation and The J48 decision tree algorithm has been applied to the
sample variance were computed to aid as features in the problem of feature selection. The input to the algorithm is
feature extraction process. the set of statistical features described above and output of
the decision tree shown in Figure 5. It is clearly shown that
IV. FEATURE SELECTION - J48 ALGORITHM the top node is the best node for classification. The other
features in the nodes of decision tree perform in descending
J48 decision tree algorithm is adapted from the C4.5
order of significance. It is to be mentioned here that only
algorithm in WEKA [20]. “It consists of a number of
features that contribute to the classification appear in the
branches, one root, a number of nodes, and a number of
decision tree and other features do not contribute much. The
leaves. One branch is a chain of nodes from the root to a
features which have the less discriminating capability can be
leaf, and each node involves one attribute. The occurrence
consciously discarded by deciding on the threshold. This
of an attribute in a tree provides information about the
concept is made use for selecting good features. The
importance of the associated attribute [21]. A decision tree
algorithm identifies the good features for the purpose of
is a tree based knowledge representation methodology used
classification of the given training data set, and thus reduces
to represent classification rules. J48 decision tree algorithm
the domain knowledge required to select good features for
is a widely used one to construct decision trees [22]. The
pattern classification problem [24-26]. Referring from
procedure of forming the decision tree and exploiting the
Figure 5, one can identify the most dominating features to
same for feature selection is characterized by the following:
represent the blade conditions are the sum, range, standard
1. The set of features available at hand forms the
deviation, and kurtosis.
input to the algorithm; the output is the decision
tree.
2. The decision tree has leaf nodes, which represent
class labels, and other nodes associated with the
classes being classified.
3. The branches of the tree represent each possible
value of the feature node from which they
originate.
4. The decision tree can be used to classify feature
vectors by starting at the root of the tree and
moving through it until a leaf node, which provides
a classification of the instance, is identified.
5. At each decision node in the decision tree, one can
select the most useful feature for classification
Figure 5: J48 Tree Classification for Feature Selection
using appropriate estimation criteria. The criterion
used to identify the best feature invokes the
V. FEATURE CLASSIFICATION - LOGISTIC
concepts of entropy reduction and information
MODEL TREE (LMT) ALGORITHM
gain.
Information gain measures how well a given attribute A logistic model tree essentially comprises a normal
separates the training examples according to their target decision tree arrangement with logistic regression tasks at
classification. The measure is used to select the candidate the leaves [27]. “As in normal decision trees, an assessment
features at each step while growing the tree [23]. on one of the qualities is connected with each internal hub.
Information gain is the expected reduction in entropy caused For an identified property with k values, the hub has k child
by portioning the samples according to this feature. hubs, and illustrations are sorted down one of the k branches
Information gain (S, A) of a feature A relative to a relying upon their estimation of the feature. For numeric
collection of examples S, is defined as: features, the hub has two child hubs and the test comprises
(1)
where Value (A) is the set of all possible values for
attribute A, and Sv is the subset of S for which feature A has

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 206 & Sciences Publication
 LOGISTIC MODEL TREE CLASSIFIER FOR CONDITION MONITORING OF WIND TURBINE BLADES

of contrasting the characteristic significance to the From logistic model tree (LMT), the kappa statistic was
threshold. Generally, a logistic model tree comprises of a found to be 0.884. It is used to measure the arrangement of
tree arrangement that is comprised of an arrangement of likelihood with the true class. The mean absolute error was
internal or non-terminal hubs N and an arrangement of found to be 0.0496. It is a measure used to measure how
leaves or terminal hubs T. Let S indicate the entire close forecasts or prediction are with the ultimate result
occurrence in space, spread over by all characteristics that [30]. The root mean square error was found to be 0.1621. It
are available in the information [28]. At that point, the tree is a quadratic scoring rule which processes the average size
structure gives a separate section of S into areas St, and each of the error. The detailed class wise accuracy is shown in
area is characterized by a leaf in the tree. Table 3. The class wise accuracy is expressed in terms of the
(3) true positive rate (TP), false positive rate (FP), precision,
Not like all decision trees, the leaves t ∈ T has a related recall, F-Measure, receiver operating characteristics (ROC)
logistic regression function ft rather than only a class name. area [31-34].
The regression function ft considers a subset Vt⊆V of all TP is used to predict the ratio of positives which are
characteristics existing in the data and models the class correctly classified as faults. FP is commonly described as a
relationship possibilities as false alarm in which the result that shows a given fault
condition has been achieved when it really has not been
(4) achieved. The true positive (TP) rate should be close to 1
and the false positive (FP) rate should be close to 0 for a
Where better classifier [35-41]. One can observe from table 3, the
TP rate of most of the classes are close to 1 and FP rate were
If =0 for vk ≠ Vt. The model denoted by the whole close to 0. This reassures that the result presented by
logistic model tree is given by, confusion matrix in Table 2.
(5) Table 2: Confusion Matrix of LMT
Where I(x∈ St) is 1 if x∈ St and 0 otherwise. Blade
Goo Ben Crac Loos
Pitc
Erosio
condition h
d d k e n
VI. RESULTS AND DISCUSSION s twist
Good 86 0 0 14 0 0
The vibration signals were acquired for good condition Bend 0 92 4 0 0 4
and faulty blade conditions using DAQ (total 600 samples Crack 0 4 92 3 0 1
were collected; 100 samples for each condition). From Loose 13 0 4 82 1 0
vibration signals, twelve descriptive statistical features were Pitch
extracted. Out of them, four best contributing features were 0 0 0 0 95 5
twist
selected using J48 decision tree algorithm. They are the Erosion 0 3 0 0 5 92
sum, range, standard deviation, and kurtosis. From Figure 5, Table 3: Class Wise Accuracy of LMT
the feature „sum‟ is the most contributing features when TP F-
compared to other features. The other contributing features FP Precisio Recal ROC
Class Rat Measur
are a range, standard deviation, and kurtosis. Rate n l area
e e
The minimum number of instances per leaf and the 0.8 0.02 0.97
number of data used for reduced-error pruning was kept at Good 0.878 0.86 0.869
6 4 4
50 for selecting 4 dominating feature in J48 decision tree 0.9 0.01 0.99
algorithm. The rest of the features like mean, median, mode, Bend 0.93 0.93 0.93
3 4 4
minimum, maximum, skewness, sample variance and 0.9 0.98
Crack 0.01 0.948 0.91 0.929
standard error were eliminated as they contribute very less 1 7
in fault classification. With the selected features, the fault 0.8 0.97
Loose 0.04 0.81 0.85 0.829
classification process was carried out using logistic model 5 1
tree. The classification accuracy of LMT algorithm using Pitch 0.9 0.98
0.01 0.949 0.94 0.945
WEKA software package was found to be 90.33% with the twist 4 8
minimum number of instances to be 20 and the logitboost Erosio 0.9 0.01 0.98
0.912 0.93 0.921
iteration to be -2. The confusion matrix of the logistic model n 3 8 4
tree algorithm is shown in Table 2. In confusion matrix, the From 600 samples, 542 samples are correctly classified
diagonal elements represent the correctly classified (90.33%) and remaining 58 are misclassified (9.67%). The
instances and the others are misclassified instances [29]. time taken to build the model is about 0.23 seconds. This
Also one can observe more misclassifications between good can be used in real time for the condition monitoring of the
and loose conditions. For the loose condition, the bolts wind turbine due to low computational time.
between the hub and the blade were made loose (please note The logistic model tree algorithm is a better classifier that
that the blade was in good condition). However, at high can be used in multiclass fault prediction on wind turbine
wind speed, the blade can stick to the hub and behave like a blades.”
good condition during operation. Because of this, the
signature of the loose condition sometimes resembles good
condition and the classifier finds difficult to distinguish
between them; hence, more misclassifications.

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 207 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-8, Issue-2S11, September 2019

12. Li X, Yang Z, Zhang H, Du Z, Chen X. Crack growth

VII. CONCLUSION sparse pursuit for wind turbine blade. Smart materials
and structures. 2014 Nov 18; 24(1):015002.
The wind turbine is very important in the production of 13. Xie W, Zeng P, Lei L. A novel folding blade of wind
wind energy in the day to day life. This paper presents an turbine rotor for effective power control. Energy
algorithmic based clarification of vibration signals for the Conversion and Management. 2015 Sep 1; 101:52-65.
assessment of wind turbine blade conditions. “From the 14. Hoell S, Omenzetter P. Structural damage detection in
wind turbine blades based on time series representations
acquired vibration data, a model was developed using data
of dynamic responses. In Smart Materials and Non-
modeling technique. LMT algorithm was used to learn and destructive Evaluation for Energy Systems 2015 2015
classify the different conditions of the blade. The model is Mar 27 (Vol. 9439, p. 94390B). International Society for
tested under 10-fold cross validation and correctly classified Optics and Photonics.
instances were found to be 90.33%. The error rate is 15. Wang YZ, Li F, Zhang X, Zhang WM. Composite wind
relatively less and may be considered for the blade fault turbine blade aerodynamic and structural integrated
design optimization based on RBF Meta-Model. In
diagnosis. Hence the logistic model tree can be practically
Materials Science Forum 2015 (Vol. 813, pp. 10-18).
used for the condition monitoring of wind turbine blade to Trans Tech Publications.
reduce the downtime and to provide uninterrupted wind 16. Haselbach PU, Bitsche RD, Branner K. The effect of
energy. delaminations on local buckling in wind turbine blades.
Renewable Energy. 2016 Jan 1; 85:295-305.
REFERENCES 17. Joshuva A, Sugumaran V. Fault diagnostic methods for
wind turbine: A review. ARPN Journal of Engineering
1. Bellarmine GT, Urquhart J. Wind energy for the 1990s and Applied Sciences. 2016 Apr 11; 11(7):4654-4668.
and beyond. Energy Conversion and Management. 1996 18. Joshuva A, Sugumaran V. A data driven approach for
Dec 1; 37(12):1741-1752. condition monitoring of wind turbine blade using
2. Márquez FP, Tobias AM, Pérez JM, Papaelias M. vibration signals through best-first tree algorithm and
Condition monitoring of wind turbines: Techniques and functional trees algorithm: A comparative study. ISA
methods. Renewable Energy. 2012 Oct 1; 46:169-178. transactions. 2017 Mar 1; 67:160-172.
3. Veers PS, Ashwill TD, Sutherland HJ, Laird DL, Lobitz 19. Joshuva A, Sugumaran V. Wind turbine blade fault
DW, Griffin DA, Mandell JF, Musial WD, Jackson K, diagnosis using vibration signals through decision tree
Zuteck M, Miravete A. Trends in the design, algorithm. Indian Journal of Science and Technology.
manufacture and evaluation of wind turbine blades. Wind 2016 Dec 29; 9 (48):1-7.
Energy: An International Journal for Progress and 20. Joshuva A, Sugumaran V, Amarnath M. Selecting kernel
Applications in Wind Power Conversion Technology. function of Support Vector Machine for fault diagnosis
2003 Jul; 6(3):245-259. of roller bearings using sound signals through histogram
4. Liu WY. The vibration analysis of wind turbine blade– features. International Journal of Applied Engineering
cabin–tower coupling system. Engineering structures. Research. 2015; 10(68):482-487.
2013 Nov 1; 56:954-957. 21. Joshuva A, Sugumaran V. A Comparative Study for
5. Ciang CC, Lee JR, Bang HJ. Structural health Condition Monitoring on Wind Turbine Blade using
monitoring for a wind turbine system: a review of Vibration Signals through Statistical Features: A Lazy
damage detection methods. Measurement science and Learning Approach. International Journal of
technology. 2008 Oct 13; 19(12):122001. Engineering and Technology. 2018; 7(4.10):190-196.
6. Godwin JL, Matthews P. Classification and detection of 22. Joshuva A, Sugumaran V. A Machine Learning
wind turbine pitch faults through SCADA data analysis. Approach For Condition Monitoring Of Wind Turbine
IJPHM Special Issue on Wind Turbine PHM. 2013 Jul: Blade Using Autoregressive Moving Average (ARMA)
90. Features Through Vibration Signals: A Comparative
7. Dervilis N, Choi M, Taylor SG, Barthorpe RJ, Park G, Study. Progress in Industrial Ecology–An International
Farrar CR, Worden K. On damage diagnosis for a wind Journal. 2018; 12(1/2):14-34.
turbine blade using pattern recognition. Journal of sound 23. Joshuva A, Sugumaran V. Classification of various wind
and vibration. 2014 Mar 17; 333(6):1833-1850. turbine blade faults through vibration signals using
8. Mollineaux M, Balafas K, Branner K, Nielsen P, Tesauro hyperpipes and voting feature intervals algorithm.
A, Kiremidjian A, Rajagopal R. Damage detection International Journal of Performability Engineering.
methods on wind turbine blade testing with wired and 2017; 13(3):247-258.
wireless accelerometer sensors. In EWSHM-7th 24. Joshuva A, Sugumaran V. Failure Analysis on Wind
European Workshop on Structural Health Monitoring Blade Using Vibration Signals and Classifying the
2014 Jul 8. Failures Using Logit Boost Algorithm. International
9. Shafiee M, Finkelstein M, Bérenguer C. An opportunistic Journal of Control Theory and Applications. 2016;
condition-based maintenance policy for offshore wind 9(52):225-234.
turbine blades subjected to degradation and 25. Joshuva A, Sugumaran V. A comparative study of Bayes
environmental shocks. Reliability Engineering & System classifiers for blade fault diagnosis in wind turbines
Safety. 2015 Oct 1; 142:463-471. through vibration signals. Struct Durab Health Monit
10. Slot HM, Gelinck ER, Rentrop C, van der Heide E. (SDHM). 2017; 12(1):69-90.
Leading edge erosion of coated wind turbine blades: 26. More Vasudha, Panditrao Harshal, Joshuva A,
Review of coating life models. Renewable Energy. 2015 Sugumaran V. Effect of Sampling Frequency and Sample
Aug 1; 80:837-848. Length on Fault Diagnosis of Wind Turbine Blade.
11. Zhang S, Dam-Johansen K, Nørkjær S, Bernad Jr PL, Pakistan Journal of Biotechnology. 2018; 15(Special
Kiil S. Erosion of wind turbine blade coatings–Design
and analysis of jet-based laboratory equipment for
performance evaluation. Progress in Organic Coatings.
2015 Jan 1; 78:103-115.

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 208 & Sciences Publication
 LOGISTIC MODEL TREE CLASSIFIER FOR CONDITION MONITORING OF WIND TURBINE BLADES

Issue ICRAME 17):14-17. 41. Joshuva A, Sugumaran V. Fault Diagnosis and

27. Shoombuatong W, Hongjaisee S, Barin F, Localization of Wind Turbine Blade [dissertation],
Chaijaruwanich J, Samleerat T. HIV-1 CRF01_AE Vellore Institute of Technology, Chennai Campus. 2018.
coreceptor usage prediction using kernel methods based
logistic model trees. Computers in biology and medicine.
2012 Sep 1; 42(9):885-889.
28. Joshuva A, Sugumaran V. Fault Diagnosis of Wind
Turbine Blade Using Vibration Signals through J48
Decision Tree Algorithm and Random Tree Classifier.
International Journal of Control Theory and
Applications. 2016; 9(52):249-258.
29. Joshuva A, Sugumaran V. Multiclass Classifier
Approach for Fault Diagnosis of Wind Turbine Blade
Using Vibration Signals through Statistical Analysis.
International Journal of Control Theory and
Applications. 2016; 9(52):235-247.
30. Joshuva A, Sugumaran V, Amarnath M, Lee SK.
Remaining Life-Time Assessment of Gear Box Using
Regression Model. Indian Journal of Science and
Technology. 2016 Dec 28; 9(47):1-8.
31. Joshuva A, Sugumaran V. A Study of Various Blade
Fault Conditions on a Wind Turbine Using Vibration
Signals through Histogram Features. Journal of
Engineering Science and Technology. 2018 Jan;
13(1):102-121.
32. Manju BR, Joshuva A, Sugumaran V. A Data Mining
Study for Condition Monitoring on Wind Turbine Blades
Using Hoeffding Tree Algorithm Through Statistical and
Histogram Features. International Journal of Mechanical
Engineering and Technology. 2018; 9(1):1061-1079.
33. Joshuva A, Sugumaran V. Wind Turbine Blade Fault
Diagnosis Using Vibration Signals and Statistical
Features through Nested Dichotomy Classifiers.
International Journal of Pharmacy and Technology.
2017 Apr; 9(1): 29147-29164.
34. Joshuva A, Sugumaran V. Fault Diagnosis for Wind
Turbine Blade through Vibration Signals Using
Statistical Features and Random Forest Algorithm.
International Journal of Pharmacy and Technology.
2017 Apr; 9(1): 28684-28696.
35. Deenadayalan, Sivakumar, S., Vishnuvardhan R., Sathish
Kumar R. Fabrication and Characterisation of B-H-G
Fiber with Teak Wood Particles Reinforced Hybrid
Composite. International Journal of Engineering and
Technology. 2018; 7(2.31): 208-211.
36. Sivakumar S, Dhanalakshmi V, Vishuvardhan R.
Extraction of Subtractive Features of Prismatic Parts
from STEP File for CAD/CAM Integration. Journal of
Science and Technology, Malaysia. 2019 Jan; 27 (1):
343-356.
37. Vishnuvardhan R, Pooranam N, Sivakumar S, Vignesh
T. Producing Electrical Energy from Light Intensity,
Design Traking System by Dual Axis Method.
International Journal of Mechanical and Production
Engineering Research and Development (IJMPERD).
2019; Jan; 9: 503-509.
38. Joshuva A, Sugumaran V. Improvement in wind energy
production through condition monitoring of wind turbine
blades using vibration signatures and ARMA features: a
data-driven approach. Progress in Industrial Ecology, an
International Journal. 2019 Jun 21; 13(3):207-31.
39. Joshuva A, Sugumaran V. Selection of a meta classifier-
data model for classifying wind turbine blade fault
conditions using histogram features and vibration
signals: a data-mining study. Progress in Industrial
Ecology, an International Journal. 2019 Jun 21;
13(3):232-51.
40. Joshuva A, Sugumaran V. Crack Detection and
Localization on Wind Turbine Blade Using Machine
Learning Algorithms: A Data Mining Approach. Struct
Durab Health Monit (SDHM). 2019; 13(2):181-203.

Published By:
Retrieval Number: B10330982S1119/2019©BEIESP Blue Eyes Intelligence Engineering
DOI: 10.35940/ijrte.B1033.0982S1119 209 & Sciences Publication