New mathematical tools for modelling uncertainties in offshore turbine operation and maintenance

The UK is planning to make massive investments in offshore wind farms which will result in several fleets of similar wind turbines being installed around the UK coastline. The economic case for these wind turbines assumes a very high technical availability, which means simply that the turbines have to be working and ready to generate electricity for nearly all of the time. Not achieving this availability could well result in large economic losses. Unfortunately there is relatively little operational experience of offshore systems on which to base the estimates used. The systems may turn out to behave in unexpected ways by failing earlier than expected, or by proving more difficult to maintain. Even well-known systems can behave differently when used in new environments, which is why reliability databases often indicate ranges of failure behaviour rather than single number estimates. Availability is difficult to model because, in addition to the unknown impact of different environments, there is often a period of adjustment in which operators and manufacturers adapt their processes and systems to the new situation, leading to the potential for availability growth. However, with a new fleet of turbines there is also an aging process as they all grow older together which could lead to lower availability. The economic case for offshore systems depends a lot on whether high enough availability can be achieved, particularly in the early years of operation which are important for paying back the investment costs. This project looks at the degree of uncertainty there is in availability estimates for offshore wind turbines. This uncertainty is not one that averages out when there are a large number of turbines, because it has a systematic affect across all the turbines in a wind farm and therefore leads to corresponding uncertainty in the overall availability across the wind farm. This type of uncertainty is often called state-of-knowledge uncertainty and only gets reduced by collecting data over the longer term. Even if we are not yet able to collect operational data, we can still gain an understanding of the sources of state-of-knowledge uncertainty. Mathematical models can help us understand how different sources of uncertainty affect the uncertainty about availability, and to find out which ones we should be most concerned about. That, in turn, will help researchers to focus their energies on resolving the issues that ultimately have the biggest impact. In this project, operations researchers will work together with engineers and other researchers in the renewables sector, in order to build credible mathematical models to help answer these questions. Doing that requires the development of new mathematics, particularly in the way we represent how uncertainties are affected by different environmental and engineering aspects. It requires us to find better ways of getting information from experts into a form that we can use in the mathematical models, and it also requires us to find new ways of running the models on a computer.

"The key outcomes of the award are a set of models and processes that can be used to support decision making for stakeholders in offshore wind farms - operators, investors, OEMs in particular. The models have been designed so that they take into account the degree of uncertainty there is in different aspects of the system and of its environment. This uncertainty is assessed by expert groups and is based on engineering judgements and engineering inputs. The uncertainty is then propagated through a simulation model that is designed to look at the way different types of subsystems within the offshore windfarm are affected by these uncertainties. Realisations of different types of uncertainty affect the overall system in different ways - for example, uncertainties in design affect all units, while uncertainties in manufacture and construction have a systematic impact but not at 100%. The model captures many features that are observed in practice - in particular the impact of early systemic failures, the need for back-fitting, and the longer-term growth of availability through learning about the system and environment. The model is quantified using a combination of historical data (which shows typical reliability behaviour of mature systems), and expert assessment of uncertainties. The overall model can be used to make judgements about the value of information arising from, for example, extra testing or improved environmental assessment. Within the project we have demonstrated how the main simulation model can be approximated by a Gaussian Process simulator to carry out such calculations.
The modelling approach has been developed through extensive discussions and workshops with industry representatives. There has been excellent collaboration between Management Science and Electrical and Electronic Engineering at Strathclyde, which has led to another collaboration around risk in electrical network planning, and also to the provision of a course in the DTC Wind Energy."