Share this page on:

Conference programme 

Back to the programme printer.gif Print

Poster session

Lead Session Chair:
Stephan Barth, Managing Director, ForWind - Center for Wind Energy Research, Germany
Jan Helsen OWI-Lab/BruWind, Belgium
Jan Helsen (1) F P Wout Weijtjens (1) Pieter Jan Jordaens (2) Gert Desitter (1) Christof Devriendt (1)
(1) OWI-Lab/BruWind, Brussel, Belgium (2) Sirris, Leuven, Belgium

Printer friendly version: printer.gif Print

Download poster(0.80 MB)

Presenter's biography

Biographies are supplied directly by presenters at OFFSHORE 2015 and are published here unedited

Prof. J. Helsen has been conducting research in the field of wind turbine drivetrains for seven years. He is coordinator drivetrain monitoring in the framework of the OWI-Lab. Previous positions included a post doctoral position co-funded by ZF Wind Power and Siemens and a doctoral position in cooperation with ZF Wind Power and NREL


Integrated turbine condition and health monitoring


Health and condition monitoring (CMS) is increasingly popular for offshore wind turbines due to the need for condition based rather than fault driven maintenance strategies and insurance requirements. The traditional health and CMS approach considers each monitoring source as a separate entity, which has its own dedicated measurement system, and its data analyzed independently from the others. However the availability of an integrated dateset offers the possibility to acquire much deeper understanding of the degradation mechanisms thriving faults by investigating the complete picture.


This paper leverages this potential by suggesting an integrated monitoring approach by combining traditional vibration spectra, general turbine response calculated online using advanced physical models, loads, with advanced numerical data models for anomaly detection. The main goal of this work is to discuss the concept of the integrated monitoring approach and focus on two aspects in detail: the use of data-models for anomaly detection and dynamic load case detection for fatigue influencing events during the lifetime of the turbine. The methodology has several aspects. Essentially the goal is integration of all available data in one dataset.

Main body of abstract

Given the continuous flow of monitoring data it is unfeasible to send all data to one centralized hub without downsampling. Nonetheless information for degradation is often in signal details. Therefore, analysis should be performed on high frequency data. This paper suggests an approach for distributed data storage and computing based on the Cassandra big-data framework. One dataset contains information of several sources: loads, status parameters and vibration spectra.

Responses at critical locations for structural health monitoring are estimated using a virtual sensing approach in which accelerations and strains, measured at accessible locations, are combined with a physical model. These responses are stored in the dataset in parallel with the online measurements. Based on the combination of general turbine loads and status parameters, such as pitch angles, an automatic load case detection scheme is discussed for tracking dynamic events such as e.g. starts and stops, significantly influencing fatigue.

Moreover, loads are used in statistical models describing normal turbine operation. Wind turbine behavior is subject to great variety. Therefore typical response parameters, such as acceleration levels, strains, temperatures, etc. are significantly dependent on turbine state. A data-model taking these state parameters into account is calculated to predict values for each response parameter. Comparison between measured actual value and normal reference condition results in degradation initiation insights.


The work concludes that the one dataset based integrated scalable storage and processing data-approach has the potential for early failure detection. A more complete understanding of the abnormalities in the turbine operation is achieved by leveraging historical data in advanced statistical models on turbine level. This work showed several important aspects such as the load case and anomaly detection. Moreover, the link to the load history creates insights in the origin of the abnormality.

Learning objectives
We plan to extend our existing dataset with more directly measured high frequency data at specific locations in the turbine nacelle, drivetrain and tower. This extended information in combination with improvements in our modeling and data-analysis algorithms will give additional insights in failure propagation. Moreover, weather parameters such as wave heights, wind speeds at locations close to the park and their predictions will be added to the dataset and linked to the operational behavior.