SCADA Data Based Wind Turbine Fault Diagnosis


                          Christian Tutivén, Yolanda Vidal and Francesc Pozo
      Control, Modeling, Identification and Applications (CoDAlab), Department of Mathematics,
                         Universitat Politècnica de Catalunya, Barcelona, Spain
          email: christian.tutiven@upc.edu, yolanda.vidal@upc.edu, francesc.pozo@upc.edu



                        Abstract                                Key words
The reliability requirements of wind turbine (WT) compo-        Machine learning, support vector machines, fault diagno-
nents have increased significantly in recent years in the       sis, health monitoring, wind turbine.
search for a lower impact on the cost of energy. In addi-
tion, the future of the wind energy industry passes through
the use of larger and more flexible wind turbines in remote
locations, which are increasingly offshore to benefit           Acknowledgments
stronger and more uniform wind conditions. The future of        This work has been partially funded by the Spanish Minis-
the wind energy industry passes through the use of larger       try of Economy and Competitiveness through the research
and more flexible wind turbines in remote locations, which      projects DPI2014-58427-C2-1-R, DPI2017-82930-C2-1-R,
are increasingly offshore to benefit stronger and more uni-     and by the Generalitat de Catalunya through the research
form wind conditions. The cost of operation and mainte-         project 2017 SGR 388.
nance of offshore wind turbines is approximately 15–35%
of the total cost. Of this, 80% goes towards unplanned
maintenance issues due to different faults in the wind tur-
bine components, therefore, condition monitoring is cru-
cial for maximum availability.
    In this work, without using specific tailored devices for
condition monitoring but only increasing the sampling
frequency in the already available sensors of the SCADA
system, a data-driven multi-fault diagnosis strategy is con-
tributed. An advanced WT benchmark is used. That is a 5
MW modern WT simulated with the FAST (Fatigue, Aer-
odynamics, Structure and Turbulence) software and sub-
ject to various actuators and sensors faults of different
type. The measurement noise at each sensor is modeled as
a Gaussian white noise.
    First, the SCADA measurements are pre-processed and
feature transformation based on multiway principal com-
ponent analysis (MPCA) is realized. Then, 10-fold cross
validation support vector machines (SVM) based classifi-
cation is applied. In this work, SVMs were used as a first
choice for fault detection as they have proven their robust-
ness for some particular faults but never accomplished, to
the authors’ knowledge, at the same time the detection and
classification of all the proposed faults taken into account
in this work. To this end, the choice of the features as well
as the selection of data are of primary importance.
    Simulation results show that all studied faults are de-
tected and classified with an overall accuracy of 98%. Fi-
nally, it is noteworthy that the prediction speed allows this
strategy to be deployed for real-time condition monitoring
in WTs.