Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'Advanced rotor technologies' taking place on Tuesday, 11 March 2014 at 11:15-12:45. The meet-the-authors will take place in the poster area.

Jesus Jimenez (1) P Estefania Artigao (2) F Tasos Chatziloukas (3) Lars Schubert (4) Slim Soua (1) Tat-Hean Gan (2)
(1) TWI, Cambridge, United Kingdom (2) BRUNEL UNIVERSITY, London, United Kingdom (3) INNORA, Athens, Greece (4) FRAUNHOFER, Dressden, Germany

Printer friendly version: printer.gif Print




Current wind turbine condition monitoring process can be time-consuming and costly. These systems failed to achieve the reliability and operational efficiency required by the industry. Existing reports on the monitoring of wind turbine data are restricted to a few historical cases. For certain sensors, like Acoustic Emission (AE), is yet scarcer [1]. In order to overcome this issue, the development of a baseline data analysis method is proposed in this paper. For that, three techniques will be integrated: Acoustic Emission, Operational Modal Analysis and Motor Current Signature Analysis.


Three different techniques have been used for the development of a baseline that indicates the behaviour under normal operation of three different components: MCSA on the generator, OMA on the drive train (including gearbox) and AE on the high speed shaft. For that, Current, Vibrational and Acoustic Emission data have been collected for a period of time and this initial data has served as a training process. For each technique, different signal processing methods have been used to extract the wind turbine features. The features extracted during this initial period have been used to create a baseline (or signature) that defines the behaviour of the generator, gearbox and high speed shaft of the wind turbine under normal operation. A range of maximum and minimum values for these features have been calculated based on trends (average ± standard deviation).
In the future, the drive train monitoring process can be performed by comparing each new set of data acquired to the original baseline created during the initial stage.
The necessary hardware for the signal acquisition of the three techniques has been tailor made for the project needs. The system architecture uses multiple sensor modules (MCSA, OMA, and AE) that sample the analogue sensors at rates of up to 2MSPS. Each module samples the sensor(s) for a specified time window, stores the samples temporarily and transmits them to the server via Ethernet. Special development was required for the AE Module, in this case the sensors are attached to the rotating high speed shaft; therefore wireless transfer of the data, energy autonomy, and small size were mandatory requisites. After acquisition, the signals were processed by each module and the baseline strategy applied.
Fig. 1: Work distribution within CMSWind’s Research Organisations
The tailor made hardware development allows for a cost reduction of the fully integrated system.

Main body of abstract

It is presented in this paper the signature development for the three different techniques applied to the three different components under study: generator, drivetrain (including gearbox) and high speed shaft of the WINDMASTER300.
Fig. 2: Acquiring from WINDMASTER300
MCSA is the procedure of acquiring the motor current signal, performing signal conditioning and analysing the derived signal to identify faults [2]. The principle behind MCSA is that, in a theoretical situation with no harmonics, only one peak in the frequency spectrum would be seen. Therefore by locating and analysing fault related harmonics one can assess the generator currents’ health. The different frequency components associated to different faults in induction machines are well known, some are shown in Fig. 3, these can be used for fault detection and diagnosis.
Fig. 3: Fault-related frequencies
Short intervals defined around those fault-related frequency components will be monitored. Relevant features of the extracted sub-waveforms will be trended for the signature development during the training stage and for monitoring at the later stage. In this way, this technique has been upgraded from being used for off-line or incidental measurements to a continuous monitoring subsystem.
Modal Analysis is used for extracting data needed to understand and describe the dynamic behaviour of a given structure. In order to describe the behaviour of a dynamic system, the use of Modal Analysis is regarded as an established method to extract required key parameters; extracted for each individual mode, and consist of the natural frequency, the deflection of the structure corresponding to the frequency and the modal damping. The modal properties are not only dependent on the material parameters like mass, stiffness and damping, but also on boundary conditions such as bearings, clamping and attenuators. If these properties change, the mode’s frequency, mode shape and damping, may vary as well. Inversely, this can be used to identify structural change by structural health monitoring techniques. To identify a system description and derive the needed modal parameters, multiple approaches are available for OMA: Frequency-Domain based methods (Frequency Domain Decomposition, Enhanced Frequency-Domain Decomposition, Curve-Fit Frequency Domain Decomposition), and Time-Domain based methods (Stochastic Subspace Identification) [3]. After identifying spectral peaks using one of these methods, one has to divide the spectral peaks into frequencies coming from noise and periodic distortion (both unwanted) or structural frequencies (wanted).
Fig. 4: Measurement concept: Cluster analysis of modal parameters frequency and damping
By extracting the significant features related to each cluster’s mode, a signature for the drivetrain can be defined during the initial training stage. At the monitoring stage, such features are used for monitoring by comparison with the stated limits.
AE refers to the generation of transient elastic waves produced by a sudden redistribution of stress in a material. When a structure is subjected to an external stimulus (change in pressure, load, or temperature), localised sources trigger the release of energy, in the form of stress waves, which propagate to the surface and are recorded by sensors. Studying Acoustic Emission does not require supplying energy to the object under examination; AE simply listens for the energy released by the object [4]. The level of AE activity during multiple load cycles will form the signature development for the high speed shaft during the training process. Such activity (determined by feature extraction as well as in the previous methods) will be monitored at the later stage to determine the appearance of faults.
The signals acquired from the three different components using the three mentioned techniques will be classified in bins as a function of the wind turbine power output. Binning example for a 300kWatt wind turbine shown in Fig. 5.
Fig. 5: Bin allocation per power output for a 300 kWatt wind turbine
For each bin, the relevant parameters of each signal will be extracted (RMS, peak-to-valley, Crest factor, etc.). During the training process the different features and its changes because of environmental influences under normal operation will be obtained and trended. After sufficient data is recorded the training process stops and the monitoring process starts. The new signals acquired during monitoring that do not fall within the specified limits (average ± standard deviation) will indicate an abnormal behaviour and thus the possible appearance of a defect.


Since monitoring data for wind turbines is scarce, an approach consisting on the development of a baseline is presented in this paper. For that, a training process will take place, for an initial period of time, where the necessary data to develop the baseline will be gathered. After acquiring from the three different techniques, each data set is allocated in its correspondent bin, depending on the wind turbine power output. Such data will be trended, which will define the normal operation of the wind turbine. Once the baseline has been created the monitoring stage will start, where the new sets of data will be compared against the baseline. If the new data set does not fall within the established limits, the system will generate alarms indicating the abnormal operation of the relevant part.
For that, three different techniques have been combined: MCSA to monitor the generator, OMA the drivetrain, and AE the high speed shaft. Novelties for every method have also been applied:
 The MCSA technique has been upgraded from being used for off-line or incidental measurements to a continuous monitoring subsystem, therefore elevating the method from a fault assessment technique to a condition monitoring one.
 Contrary to Experimental Modal Analysis (EMA), OMA does not rely on specific excitation; the method rather assumes a white noise spectrum as input. The loads applied directly to the structural system cannot be measured and are not known, thus it is not possible to separate the dynamic influence of the environment and the structure itself. This represents a great advantage of the method, since modal parameters are extracted for structures under actual operational conditions.
 Thanks to the tailor-made hardware developed specifically for the project, AE, which is a novel technique in wind turbines, can be applied on the high speed shaft, i.e. a rotating component. Its wireless technology has proven successful to wirelessly transfer the acquired signals even at the very demanding sampling rate needed for AE (2MSPS).
The present work is part of the CMSWind FP7 Project, partly funded by the EC (Grant Agreement no. 286854) [5].

Learning objectives
- MCSA used for condition monitoring of the wind turbine generator.
- OMA used for condition monitoring of the wind turbine gearbox.
- AE used for condition monitoring of the wind turbine high speed shaft.
- Bespoke hardware development, including wireless data transmission under demanding sampling rates.
- Baseline development as a monitoring strategy for wind turbines.

[1] S. Soua, B. Bridge, L. Cebulski, T-H. Gan, “Statistical analysis of accelerometer data in the online monitoring of a power slip ring in a wind turbine”. IOP Publishing, February 2012.
[2] Mohamed El Hachemi Benbouzid, “A Review of Induction Motors Signature Analysis as a Medium for Faults Detection”. IEEE Transactions on Industrial Electronics, Vol.47, No.5, October 2000.
[3] T. Jacob, M. Zundel, “Comparison between classic experimental modal analysis and operational modal analysis using the example of a wind turbine gear box”, VDI-Berichte Nr. 2168, 2012.
[5] Description of Work (DoW) of the CMSWind Project, 1-June-2012, Grant Agreement Number 286854.