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 operation & maintenance' taking place on Thursday, 13 March 2014 at 11:15-12:45. The meet-the-authors will take place in the poster area.

Carole Murray Helitune Ltd., United Kingdom
Chong Ng (1) P Peter Morrish (2) F Carole Murray (2) Michael Mulroy (1) Nick Lieven (1) Matthew Asher (3)
(1) Narec, Blyth, United Kingdom (2) Helitune, Torrrington, (3) University of Bristol, Bristol, United Kingdom

Printer friendly version: printer.gif Print

Presenter's biography

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

Carole Murray is a Project Engineer at Helitune Ltd. specialising in condition monitoring for the wind turbine industry. She holds a BEng in Electroacoustics from the University of Salford and is a Chartered Engineer. Carole previously worked for QinetiQ (formerly DERA) where she led acoustic research programmes and was involved in the development of a radar based runway debris monitoring system.


Novel health and usage monitoring solution for the offshore wind turbine industry


Cost-optimised maintenance is becoming increasingly important due to remote locations and access limitations of offshore wind farms. Maintenance and repair costs increase significantly due to the specialist vessels required and delays caused by unfavourable weather conditions.

The key innovation described in this paper is the transfer of condition monitoring technologies and maintenance models from the aerospace industry to the offshore wind turbine industry. A new health and usage monitoring system is proposed, enabling predictive maintenance and extending component life through the use of advanced maintenance protocols. This will minimise risk for insurers and reduce the levelised cost of energy.


The primary technical innovation of this project is based on the investigation of novel feature extraction algorithms that continuously monitor varying structural dynamic loads, resulting from transient wind conditions and then quantifies their impact on remaining turbine life. It has long been established that structural degradation is associated non-linearly with load [1].

The key attribute which makes this challenging is the non-deterministic nature of the outcome i.e. shock loading though gusts have a greater effect on the structural performance than an equivalent static load. The very nature of wind turbines subject to gust loading means that the dynamic loading properties have a significant impact on remaining life.

The approach adopted here is to isolate the peak responses in the time domain via a discrete wavelet algorithm, thus identifying the resonant properties under combined static and dynamic loads. The wavelet decomposition is then mapped against failure modes and associated with a pass/fail condition using a recursive neural network to establish the hyperplane which indicates pass/fail in real time. Thus - as in the aerospace industry [2] - the outcome is then associated with the maintenance schedule seek either maintenance credit or more simply an estimate of remaining life.

Accelerometers have been installed on the main bearing, gearbox and generator of a 950kW wind turbine. Vibration data has been recorded over a period of several months encompassing a variety of weather conditions. Meteorological data from the nearby mast was used to identify features in the vibration data that can be attributed to gusts and yaw misalignment. The recorded data was then used in the algorithm development.

The paper will demonstrate the ability and performance of the new algorithms to identify real-time features through testing at Narec’s facility at Blyth. Baseline tests will be performed to establish the vibration profile of the drive train operating in a condition similar to the wind turbine. Known faults will be introduced into various components such as bearings and gearboxes to provide a suitable comparison.

Main body of abstract

The specifications of wind turbine condition monitoring systems with current or pending approval by Germanisher Lloyd [3] have been evaluated and compared with helicopter health and usage monitoring systems and other techniques used in the aerospace industry. A gap analysis of operations and maintenance strategies within the two industries was also undertaken to identify capabilities within the aerospace industry that could be used to improve predictive maintenance and asset management for offshore wind farms.

The experience of the aerospace industry has shown that remaining life in structural components is governed by extreme transient events. Helicopter operators can elect not to fly in harsh conditions and return them to the comfort of their hangars, but offshore wind turbines are continually exposed to these elements. Both are impacted by dynamic loads, particularly when subjected to rapid changes in wind speed and direction. In addition, wind turbines are affected by yaw misalignment and wind shear across the blade diameter, both of which can introduce unsteady loads on the blades that may excite vibrations in critical components.

In general, vibration based wind turbine condition monitoring systems tend to either sample events at a relatively low sampling rate or to perform local averaging functions to avoid transmitting large data files to Operations and Maintenance personnel. As a consequence, transient events are averaged out and only their long term effects can be monitored.

The proposed wind turbine health and usage monitoring solution will utilise techniques developed by Helitune, for use in the helicopter industry, to measure the impact of rapid fluctuations in wind speed and direction and assess the impact on the remaining life of key components. The University of Bristol will adapt algorithms used in the aerospace and medical industries for use with wind turbines to enable the early detection of incipient faults.

Specifically the approach adopted by this paper is a two stage approach, firstly to filter the time domain measured signal via a discrete Wavelet deconvolution [4], as described by equation (1) the outcome of which is shown in (Figure 1).

where is the shape function of the wavelet transform and a and b are the shape and scale coefficients for the individual wavelets.

The second stage is to use a recursive Neural Network to classify the wavelet decomposition against pass/fail criteria. This is achieved by forming co-terminus hyperplanes allowing for multiple failure modes. Note that the term “failure” in this paper is the definition adopted by the aerospace industry which deems failure as a condition whereby a component or system does not meet a specified performance requirement – it is not imply structural failure (which is too late for the application we are considering here). An example of co-terminus hyperplanes is shown in Figure 2

The early detection of developing faults with the correct diagnosis can allow maintenance to be scheduled to replace/repair the damaged component before it fails preventing secondary damage or catastrophic failure. Significant cost reductions can be achieved by making better use of condition monitoring information to make more efficient and effective maintenance decisions. Replacement parts and specialist access equipment can be ordered in advance, reducing downtime. Work can be scheduled for times when the wind speed is predicted to be low, minimising production losses.

Through continuous monitoring and identification of peak loading transients, a maintenance scheduling strategy has been developed for the helicopter industry which we believe would be directly transferable to offshore wind turbines. The resultant through-life history will enable financial risk mitigation, allowing Capital Markets and Insurance Institutions to adapt their strategies and premiums, as appropriate.

In this paper, the outcome of the investigation into the condition monitoring techniques from both helicopter and wind turbine industries will be presented. The investigation will correlate the experience and technologies from both industries and identify the best practice to be implemented with the methodology mentioned above.


This paper presents the findings of a study into the feasibility of transferring existing successful condition monitoring techniques and maintenance models from the aerospace industry into the offshore wind turbine industry.
Novel feature extraction algorithms are being adapted from the aerospace industry with the goal to continuously monitor varying structural dynamic loads, resulting from transient wind conditions and to detect incipient faults in gears and bearings. This information can then be used to assess the impact on the remaining life of key components to inform a condition based maintenance system.

Initial studies have demonstrated encouraging indications of the ability and performance of the new algorithms to identify real-time features and their applicability to the Offshore Wind industry. Further analysis, algorithm development and subsequent testing will confirm this and inform the generation of a design specification for a new wind turbine health and usage monitoring system.

Significant cost reductions can be achieved by making better use of condition monitoring information schedule maintenance more efficiently and effectively. The decision to replace a component can be based upon its current condition and predicted remaining useful life. This will minimise the cost of parts and maintenance effort whilst improving the availability of the wind turbine and preventing secondary damage or catastrophic failure. This is particularly important for offshore wind farms due to the remote location and access limitations.

The proposed health and usage monitoring system for offshore wind turbines will enable predictive maintenance and the extension of component life through the use of advanced maintenance protocols. This will minimise risk for insurers and ultimately reduce the levelised cost of energy.

Learning objectives
The contribution to the industry knowledge is the adoption of time domain structural health monitoring and classification methods linked directly to a predictive maintenance schedule. The technologies which are deployed to achieve this goal are time domain data compression using discrete wavelet transforms coupled with self directed learning (pattern recognition) to establish pass/fail hyperplanes. The outcome is linked to a maintenance schedule to assist operators to make timely interventions to extend safe operations.

[1] Miner, M.A. ‘Cumulative Damage in Fatigue’, Journal of Applied Mechanics, Vol. 67, P. A159-A164, 1945.
[2] Wood, A. ‘An Optimal Component Maintenance Policy for Rotorcraft Components’, Engineering Doctorate, University of Bristol, 2010.
[3] Germanisher Lloyd. “Guideline for Certification of Condition Monitoring Systems for Wind Turbines”, 2013.
[4] Lieven N.A.J., Clayton J., Hoyle “Identification of Damaged Hip Prostheses”, Proc SEM-IMAC XIX, pp1099-1105, 2001