Professional Documents
Culture Documents
Health Monitoring of Wind Turbine Blades Through Vibration Signal Using Advanced Signal Processing Techniques
Health Monitoring of Wind Turbine Blades Through Vibration Signal Using Advanced Signal Processing Techniques
Abstract— Wind turbines are used for the transformation of to be an important part of the wind turbine because of its
wind energy into electrical energy with the help of rotating huge structure and has a major role in the energy conversion
blades connected to the generator. The blades of the wind mechanism [3]. The price of blades in the wind turbine can
turbine play an important role since these are used for the reach up to 15–20% of the whole setup. The blades are
conversion of kinetic energy into electrical energy. The blades exposed to the environment therefore blades are subjected to
of the wind turbine experience severe vibrations due to adverse various faults due to different environmental conditions and
environmental conditions, huge size, variation in wind speeds, huge structures. The vibration caused in blades due to
and continuous operation throughout. These vibrations lead to different wind speed, interaction with foreign particles and
serious damage to the turbine resulting in the reduction of its
various environmental changes (rain, snowfall, etc.) results in
productivity and may lead to failure in the future. The faults
slow rotation or even failure of the turbine which affects the
must be recognized at the early stage so that energy conversion
is not affected. Therefore, effective health monitoring and fault overall production and may cause a shutdown. The vibration
diagnosis are important for evading severe damage to the wind of the blades is tough to evaluate on-line due to the huge
turbines. The identification of faults in any system requires structure, various operating conditions and the remote
physical knowledge of the parts which are not readily available location of the wind turbines [4, 5].
every time. Hence data-driven models are used for Every rotating element carries its vibration and acoustic
classification and diagnosis of faults. The present work aims to signatures in its healthy condition. But when a defect occurs
diagnose the faults in a wind turbine blade using vibration
in it then a variation appears in its vibration and acoustic
signals as the measured signal which is acquired from the
hardware setup and classify the faults using different machine
signatures. This variation can be observed in time and
learning techniques and then the performance of the classifiers frequency analysis [6, 7]. Vibration and acoustic signal
are compared. The experimental work is carried out using a processing are some of the vividly used ways to detect and
wind turbine set-up by introducing the different conditions of diagnose the different machine faults. In some literature, it is
the blades i.e. healthy and defective blades. To find the found the use of wavelet transform or time-frequency
suitability of the proposed method, the signal acquisition is analysis and adaptive wavelet transform to diagnose the
done at three different speeds using a suitable instrumentation bearing fault [8, 9]. Some literature introduces the signal
system. The required statistical parameters are extracted from strength improvement techniques using an active noise
the measured vibration signals. Then three different machine cancellation (ANC) technique [10]. Andrew Kusiak et al.
learning classifiers are applied for the classification of faults. shows blade pitch fault monitoring in wind turbines [11].
The performances of the classifiers are evaluated in terms of
the percentage of accuracy from the confusion matrix. The In current scenario, machine learning techniques are
proposed work shows that the machine learning technique is a widely used to diagnose different types of machine faults.
good approach for wind turbine fault classification and the Therefore, a proper signature of faults is essential which
vibration signal is a good choice as a measuring signal for the allows us to move forward in the fault classification, results
detection and diagnosis of the faults in wind turbine blades. in a better and accurate fault diagnosis of the machine
elements [12, 13].
Keywords—Wind Turbine, Health Monitoring, Fault
Diagnosis, Machine Learning, Statistical Features, Confusion In the proposed work, one healthy blade and three
Matrix. defective blades are used for the experimental work and
vibration signal is used as a measuring parameter for the
I. INTRODUCTION diagnosis of faults in wind turbine blades. The
In the modern era, the development of wind turbines had instrumentation setup for the experiment is developed. Then
been very fast due to advancements in technology and at the from the instrumentation setup the measured vibration signals
same time, it has given a few obstacles. It is essential to are acquired. The required statistical features are extracted
sustain the healthful state of the operating turbine to maintain from those vibration signals. The machine learning classifiers
its efficiency. There is a continuous requirement to lessen the are trained by this evaluated statistical features. The machine
operation and maintenance costs [1, 2]. Hence, identification learning classifiers used in this work for the fault
of defects/malfunctions at an early stage is important to classification are k-nearest neighbour (KNN), support vector
decrease the shutdown and productivity can also be machine (SVM) and decision tree. The performance
increased. Wind turbines are used for the transformation of comparisons of different classifiers are presented in terms of
wind energy into electrical energy with the help of rotating accuracy depicted from the confusion matrix. The proposed
blades connected to the generator. The blades are considered work aims to determine whether the blades of the wind
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.
turbine are in a good mode of operation or in a defective
condition and then diagnose the faults in blades if it is
defective. The work flow of the present work is shown in
figure 1.
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.
IV. STATISTICAL ANALYSIS TABLE II. STATISTICAL PARAMETER COMPARISON FOR VIBRATION
SIGNAL AT 600 RPM
The statistical parameters are computed from vibration
Sl. Static Healthy Blade Blade Blade
signals acquired from the healthy and defective blades by No Parameters Blade Bend Crack Erosion
using the following mathematical formulas: . (Type- 0) (Type-1) (Type-2) Type- 3)
1. Mean 2.0347 8.4520 1.1318 6.8783
Mean 𝜇 = ∑ 𝑋
2. Standard 2.6399 3.0260 3.2048 4.1151
deviation
3. Root mean 2.6476 3.1747 3.3986 4.1720
Standard Deviation = ∑ (𝑋 − 𝜇) square
4. Sum .02035 .8452 1.1318 .6878
Minimum = smallest value 11. Variance 6.9690 9.1569 1.0271 1.6934
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.
Fig. 8. Acquired vibration signal from different blades at 600rpm
V. RESULTS
Fig. 11. Accuracy opted using Decision Tree Classifier
TABLE IV. OF PERCENTAGE MODEL ACCURACY OFFERED BY THE Fig. 12. Accuracy opted using SVM Classifier
DIFFERENT CLASSIFIER ALGORITHM (500 RPM)
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.
TABLE V. COMPARISON OF PERCENTAGE MODEL ACCURACY
OFFERED BY THE DIFFERENT CLASSIFIER ALGORITHM (600 RPM)
Sl. No Classifier % Model
Algorithm Used Accuracy
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.
adopted for the health monitoring of a wind turbine blade to
decrease the turbine shutdown and to increase the overall
productivity.
REFERENCES
Authorized licensed use limited to: Tsinghua University. Downloaded on April 29,2021 at 02:40:08 UTC from IEEE Xplore. Restrictions apply.