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.

Graeme Hawker University of Strathclyde, United Kingdom
Co-authors:
Graeme Hawker (1) F P David Macmillan (1) Athena Zitrou (1)
(1) University of Strathclyde, Glasgow, 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

With 9 years experience in wind energy, Graeme is a Chartered Scientist working as a Researcher in Future Energy Systems at the Institute for Energy and Environment at the University of Strathclyde, Glasgow.

Abstract

Modelling the effect of maintenance strategies and reliability for long-term wind yield assessment

Introduction

In order to replace the broad assumptions on availability that are typically used in wind plant yield assessments, a framework is proposed for modelling the specific maintenance and warehousing strategy to be used on a site, in the context of a subsystem reliability model, so as to quantitatively derive the impact of availability losses and turbine performance due to technical faults on the long-term yield of a wind farm.

Approach

Separate models are used to represent, respectively, the failure rates and modes of a set of turbines on a wind farm, modelled at a subsystem level, alongside external outage causes; the location and resources of a maintenance centre responding to outages on a number of wind farms, with a set despatchable resource of technicians, and a spare parts warehousing strategy; and a wind resource model capturing the geographic and temporal variability of power across a group of turbines and wind plant locations, which may further contribute to turbine failures. A statistical combination of these models is used within a Monte Carlo simulation to derive the likely losses, uncertainty and sensitivity in resultant energy yield for a wind farm.

It is understood, however, that detailed modelling for a wind farm during the energy yield analysis process may not be possible, and that instead a set of heuristics connected with parameters describing the site in question would be preferable for use in financial models. This would further permit the financial decision-making process to easily involve the comparison of different site configurations, such as if different maintenance companies were under consideration, or for spare part warehousing decisions, such as the number of gearboxes to have available for replacements.

This analysis does not consider the off-shore maintenance scenario, where additional complexity exists due to weather access windows, additional environmental influences and greater operational uncertainties – however, this is considered as a possible extension of the maintenance optimisation model used in this paper. The increasing size of turbines, especially in the offshore environment, is considered in terms of additional risks and uncertainties.


Main body of abstract

In assessing the long-term energy yield of a wind farm, typically the gross yield is derived, representing the amount of energy the site will generate at the turbines assuming perfect performance after losses due to wind flow, shear and turbulence. The losses due to availability (among other secondary losses) are then applied in a pro-rata fashion to this figure to give a net yield, and this availability loss is usually based on industry-wide statistics of historical availability levels. This raises 3 issues; the question of whether the past availability of turbines adequately represents future performance; that the modelling assumes a ‘typical’ availability rather than one tuned to the specifics of the site in question; and that the time-based availability, defined in [1], of the site is equal to the energy-based availability.

Additionally, it is frequently assumed that the distribution of annual yields for a wind farm corresponds to a normal distribution, due to the underlying stochastic processes of the wind resource, which ignores the asymmetrical nature of losses such as availability. For example, a turbine which achieves on average a time-based availability of 97% may only achieve 100% as an upper cap, but has no lower cap besides 0%, and hence a symmetrical distribution for availability losses does not adequately describe the potential loss in yield and associated risk for financial decisions.

In order to formulate a framework for providing a site-specific estimate of energy-based availability losses, we instead combine a series of sub-models as illustrated in fig. 1.


Fig. 1: Relationship between sub-models, showing dependencies

This includes:

1. A subsystem failure model, consisting of a Markov Chain representation of a turbine, as described in [2], generating a time series of failures and modes consistent with models of component failures as presented in work such as [3] and [4]. This allows for the reliability curves presented in work such as [5], including considerations of both ‘infant mortality’ and wear-out to construct a lifetime reliability curve.

2. A centralised maintenance dispatch model, representing a turbine company independent of the wind farm owner, contracted to perform maintenance on the turbines according to a specific framework of response times, liabilities, incentives and penalties. This is based on a survey of typical contractual terms found within turbine maintenance contracts, and used to construct a discrete event simulation of responses to turbine outages across multiple sites. Optimisation of this strategy follows work in [6].

3. A warehousing model, with lead times given for each major component optimised for cost per component and expected failure rates, similar to the approach taken in [7].

4. A wind resource model, describing the variance in power output diurnally and seasonally, with an interdependence with failure rates and availability as described in [8].

This combined model is run for a number of different site configurations, with a Monte Carlo approach taken to extraneous model parameters in order to investigate sensitivities. The asymmetric nature of the effect of these losses on an energy yield distribution is also considered, highlighting that the frequent assumption that a normal distribution is a suitable approximation to annual energy yields may inadequately address the underlying uncertainties.


Conclusion

This work demonstrates that the use of flat-rate availability assumptions in yield assessment disguises many of the site-specific factors which may affect the actual yield of an operational site in the long-term. The underlying assumption that turbine availability is independent of wind speed in particular may lead to over-prediction of yields. It is recommended that in constructing loss estimates for yield analysis, a methodology which takes into account the maintenance strategy proposed, as well as spare parts provision and the specific model/scale of turbine should be used, with heuristics suggested to replace the detailed modelling used in this paper, based on the results of the sensitivity analysis.

It is also suggested that rather than allocating a single yield distribution per operational year, analyses of projected yield should contain a prediction of how this yield distribution will change over the operational life of the site, reflecting the change in mechanical reliability which occur, as well as the performance efficiency reductions over time seen from component wear and degradation, reflecting the lifetime reliability curve constructed in other analyses.

Furthermore, it is a known weakness of these forms of projections that turbine technology is still in a period of maturation and, especially as turbines increase in size and rated power, the assumption that historical performance is a valid indicator of future turbine models’ reliability may itself carry an uncertainty which should be accounted for, and that performance data from the early life of new turbine models is particularly valuable to yield analysis of future wind farms. Recommendations are made for further datasets which could be used to extend this analysis.



Learning objectives
To introduce a framework for the use of maintenance models for yield analysis; the application of reliability data to site development and financing; understanding the relevance of interdependency between wind resource and turbine availability to yield analysis.


References
[1] International Electrotechnical Commission, “Standard 61400-26: Time based availability for wind turbines,” 2010.
[2] F. Besnard and L. Bertling, “An approach for condition-based maintenance optimization applied to wind turbine blades,” … Energy, IEEE Trans., vol. 1, no. 2, pp. 77–83, 2010.
[3] F. Spinato, P. J. Tavner, G. J. W. van Bussel, and E. Koutoulakos, “Reliability of wind turbine subassemblies,” IET Renew. Power Gener., vol. 3, no. 4, p. 387, 2009.
[4] R. Project, “Report on Wind Turbine Reliability Profiles,” 2008.
[5] H. Stiesdal and P. H. Madsen, “Design for reliability,” in Proc. Copenhagen Offshore Wind. International Conference, Copenhagen, Denmark, 2005.
[6] M. Marseguerra and E. Zio, “Optimizing maintenance and repair policies via a combination of genetic algorithms and Monte Carlo simulation,” Reliab. Eng. Syst. Saf., vol. 68, no. 1, pp. 69–83, Apr. 2000.
[7] R. Sarker and A. Haque, “Optimization of maintenance and spare provisioning policy using simulation,” Appl. Math. Model., vol. 24, no. 10, pp. 751–760, Aug. 2000.
[8] F. C. Sayas and R. N. Allan, “Generation availability assessment of wind farms,” Gener. Transm. …, vol. 143, no. 5, p. 507, 1996.