Back to the programme printer.gif Print



Thursday, 13 March 2014
11:15 - 12:45 Advanced operation & maintenance
Science & Research  


Room: Llevant
Session description

The session covers the entire area of wind farm and wind turbine operation and maintenance, e.g. how to access, repair and organise operation and maintenance logistics onshore and offshore. In order to keep in hand the current health of turbines and farms, failure detection, identification and prognosis methods are also presented. Maintenance operations are also addressed from the viewpoints of required activities and efficiency. In order to cover management aspects, operation and lifetime cost calculation methodologies are also introduced. Experts from various European countries share their results during the session.

Learning objectives

  • Advanced operation and maintenance
  • Fault detection methods
  • Reliability calculation techniques
  • Monitoring on the field
  • Statistical and artificial intelligence-based solutions for diagnostics data- and model-based solutions for fault detection
Lead Session Chair:
Zsolt Viharos, Hungarian Academy of Sciences, Hungary

Co-chair(s):
Christopher J. Crabtree , University of Durham, United Kingdom
Zhenyou Zhang Kongsberg Maritime AS, Norway
Co-authors:
Zhenyou Zhang (1) F P
(1) Kongsberg Maritime AS, Trondheim, Norway

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

Zhenyou Zhang is a senior engineer in wind turbine condition monitoring at the Department of Wind Park Management of Kongsberg Maritime AS. He graduated in Mechanical Engineering at Shanghai University as Master Student in 2009 and then started his PhD work of Condition-based Maintenance (CBM) in Norwegian University of Science and Technology (NTNU). He is currently responsible for data analysis and algorithm development of fault diagnosis and prognosis for wind turbine condition monitoring system.

Abstract

Comparison of data-driven and model-based methodologies of wind turbine fault detection with SCADA data

Introduction

Wind energy is the most developed renewable energy technologies worldwide with more than 282.587 GW installed capacity at the end of 2012 [1] and there is a great demand to reduce maintenance cost. The goal can be reached by detecting and identifying the fault of wind turbines in early stage which gives the operator sufficient time to make more informed maintenance decision. Most wind farms have installed Supervisory Control and Data Acquisition (SCADA) system but not used in effective way. This paper proposes and compares data-driven and model-based fault detection methods for wind turbines based on existing SCADA data。

Approach

All faults of wind turbines have corresponding indicators which are the performance parameters of wind turbine. The approach develops the Artificial Neural Network (ANN) model and Mathematical model of the fault indicator for components of wind turbines in normal condition based on SCADA data which are so-called normal behavior model. Once the models are established, the theoretical value of the indicator can be estimated and further to compare with the real value. Through the difference of estimated and real value of indicator, the fault can be found in early stage.

Fig. 1 shows the procedure of fault detection based on history SCADA data using both ANN method and mathematical model methods. There are mainly three steps to build and apply the methods to wind turbine fault detection. The first step is to establish normal behavior ANN/mathematical model for components of wind turbines in normal condition. The second step is to compare the difference of estimated values and real values of indicator stored in SCADA data, and set the different thresholds for early warning and close alarm in different level. The final step is to apply the models and thresholds to online system which can detect abnormal behavior of wind turbine component in very early stage so that the operator can make best maintenance plan accordingly.


Fig. 1 Procedure of Fault Detection based on SCADA Data

The paper takes the main shaft rear bearing of wind turbine as an example and the bearing temperature is selected as fault indicator. Based on the results of the example, the methods of ANN model based and mathematical model based are compared.

Main body of abstract

1) SCADA Data and Parameter Selection

Typical parameters recorded by SCADA on wind turbines are 10 minute average values which could be broadly categorized into four types, i.e. parameters of wind, performance, vibration and temperature, which could be used in fault detection and prediction activity [2].

There is no gearbox in direct-driven wind turbine and thus avoid to faults of comprehensive gearbox. The main bearings are the key components and thus it is main monitoring object in this paper. The parameters may affect or inflect the temperature of bearing contain: active power output, nacelle temperature, turbine speed and cooling fan status. Unfortunately, the cooling fan status is not available in current SCADA data and thus the parameters selected to establish model for the parameter of main shaft rear bearing temperature can be chosen as seen in Table 1.



2) ANN Approach

ANN is a model that emulates a biological neural network [3] which can deal with multi-input, multi-output, complex system with very good abilities of data fusion, self-adaptation and parallel processing. It’s very suitable for fault detection and prediction of wind turbine. The training data of ANN model should be as varied as possible in normal condition. Accordingly, the data shown in Fig. 2 and Fig. 3 are chosen to be training data and test input data for ANN model. The test result is shown in Fig. 4 in which the average difference between actual and estimated value is 0.09, and the root mean square error is 0.29, which is considered to be an acceptable level for diagnosis requirement for successful fault detection.


Fig. 2 Training Data of ANN Model for Turbine Rear Bearing Temperature


Fig. 3 Rear Bearing Model Testing Input Data


Fig. 4 ANN and Mathematical Model Outputs in Normal Condition (with Three Month Training Data)

3) Model-based Approach

After analyzing the history data trend and discussing with the experienced staff who worked in the field of wind farms, the mathematical model of main shaft rear bearing temperature can be written as Eq. (1)


The same data are used as ANN approach as seen in Fig. 2 and Fig. 3 to train and test the mathematical model respectively. Fig. 4 also shows the result of mathematical model approach. The average of difference between actual and estimated value is 0.44, and the root mean square error is 0.78 which is also an acceptable level for fault detection requirement.

4) Detection of Rear Bearing Fault

Once the models are trained and tested in normal behavior, the models can be used to detect the fault of wind turbine. Fig. 5(a) shows the evolution of rear bearing temperature from the period of July 2010 to March 2011 which contains eight months where it eventually fails. Fig. 5(b) shows the difference trend between the estimated and actual temperature of rear bearing in this period. The first important deviation from the model estimates occurred from the start of October 2010, i.e. point ①. From point ②, the deviation from the model estimates increased to 4 and lasted to point ③ where the turbine was stopped because of overheating. Then, the operator of wind farm tried to solve the problem two times in point ③ and point ④, but not successful and finally the turbine was completely stopped. From this figure, the warning can be given as early as three months in point ① before the failure happens. With the evolution of the failure, the deviation from model estimation increases and the alarm can be given to operator when the deviation reaches the level of point ②.


Fig. 5 Fault Detection Results of Rear Bearing

5) Comparison
Fig. 4 also shows a comparison of two models with huge history data in normal condition is available for training models. In this case, both methods are good enough at the estimation of the theoretical temperature. The ANN model is a little better than mathematical model as its estimated values are closer to the real values.

Fig. 6 shows a comparison of two models when only a small amount of history data available (10 days). In this case, the mathematical model-based approach is apparently better than data-driven approach. The mean errors are 1.0298 and 2.8405 respectively and the values of root mean square are 1.25 and 3.63 respectively. The mathematical model is still reliable but the ANN model is not in this case.


Fig. 6 ANN and Mathematical Model Outputs in Normal Condition (with 10 Days Training Data)



Conclusion

ANN is a method to establish the relational model between inputs and output by training using history data with high adaptability, robustness and flexibility. It is very suitable to be applied in the field of fault detection and identification for wind turbines. However, the accuracy of ANN model highly depends on the quality of training data. Fig. 6 shows that the ANN model is not accurate enough when the training data is few or in bad quality.

When the mathematical model is available for wind turbines, mathematical model-based are more accurate than ANN model to estimate the indicators and thus it can detect fault more precisely. Once the mathematical model is established, it can detect all type of faults even the faults are not presented in the training data.

Therefore, when the mathematical model or physical model is available, it is better to choose as a method for fault detection and identification. However, the mathematical model or physical model is not always available because of increasingly complexity of wind turbines. In this case, if there is huge history data available to train the data-driven model, the data-driven approach is a good option to detect fault of wind turbines. Most wind farms have installed SCADA system and collect lots of history data which can be used to established ANN model and thus the data-driven should be chosen in this case.

One suggesting future work is to combine the data-driven and model-based approaches to improve the precise of fault detection, identification and prediction.



Learning objectives
1. Learning methodologies to detect the fault of wind turbines.
2. Knowing the advantages and disadvantages of data-driven and model-based approach so that they can be chosen correctly.



References
[1] L. Fried, S. Sawyer, S. Shukla, and L. Qiao, “Global Wind Report Annual market update 2012,” 2012.
[2] A. Verma and A. Kusiak, “Fault Monitoring of Wind Turbine Generator Brushes: A Data-Mining Approach,” J. Sol. Energy Eng., vol. 134, no. 2, p. 021001, 2012.
[3] K. Wang, Applied Computational Intelligence in Intelligent Manufacturing Systems. Australia: Advanced Knowledge International Pty Ltd, 2005.