Data optimised operations & maintenance for offshore wind farms
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Offshore wind farm developments continue to grow globally at a rapid pace, with growth estimates of 630 GW by 2050. Developers of wind farms are increasingly struggling with increased costs and unfavourable project economics. Offshore wind farm operations and maintenance costs are estimated at approximately 30% of lifetime costs, reducing this expenditure is key for facilitating further deployment of offshore wind developments. This thesis describes a data-driven framework that can be applied to facilitate operations and maintenance cost reductions at offshore energy assets. The framework can be applied to any operational wind farm that has been fully commissioned. The three-part framework uses data-analysis to establish components that are key to wind farm revenue losses from downtimes due to failure. It uses machine learning, trained on the SCADA data to develop failure prediction models for these components to try and reduce turbine downtimes and therefore revenue losses, as well as avoid unnecessary replacements. Finally, it uses these predictions as inputs to a maintenance schedule optimisation problem, to provide maintenance plans and vessel routes to minimise revenue losses and maintenance costs. Within the work, this has been applied to a case-study of an operational offshore wind farm in the UK. Stage 1 of the framework involves a data analysis of maintenance logs and monitoring data, for a windfarm chosen by the operator, with the purpose of identifying turbine failure rates, availabilities, and losses and costs from maintenance and failures. When applied to the casestudy wind farm the analysis has revealed that maintenance actions can be characterised into two brackets, cost of maintenance led and cost of downtime led. Cost of maintenance led actions are ones where the cost is driven by the actual costs of performing the maintenance, i.e. expensive spare parts and specialist vessel hire. Cost of downtime driven actions are ones where the cost is primarily driven by the cost of the revenue losses associated with the downtime from component failures. These types of actions of have different characteristics and as such they should be approached differently in the context of failure predictions. It is also revealed that electrical components are critical to the failure rate and energy losses due to maintenance at the OWF. In the windfarm used as the case-study and for the period of time considered, failures in electrical components account for approximately 28% of all failures. Furthermore, the revenue losses associated with their failure accounts for nearly 40% of all the wind farm revenue losses due to the downtime from failed components. The power converters drive most electrical failures and are of key commercial interest to the farm. As a result, the power converters should be the target for future predictive model development and are considered in stage 2 of the presented framework. In stage 2 of the framework, a method is presented for developing failure prediction models for wind turbine components. First, the component’s failure mechanisms and symptoms of failure must be understood. Based on the data which contain symptoms of upcoming failure, a data collection scheme is designed. This data is then used to train the predictive modes. The models are evaluated using a novel scoring function, which assesses the impact on maintenance costs of utilising the trained predictive models in practice at a site. Several machine learning and deep learning models are trained on the wind farm SCADA data to predict the power converter failures. Results suggest that implementing an artificial neural network failure prediction model offers approximately 40% reduction in power converter maintenance costs compared to business as usual. Further improvements to these models could entail the acquisition of high frequency monitoring data specific to the power electronics in the power converters with signals showing the symptoms of failures identified in the root cause failure analysis. Collecting more data in volume (e.g. from a bigger fleet of turbine, or for a longer period of time) would also be a route to explore to improve model performance. Beyond specific components or predictive models, it is learned that failures can be anticipated via data analysis and fed into the maintenance decision optimisation addressed in stage 3. Finally, in the third stage a novel maintenance scheduling model designed to both minimise the costs of performing maintenance and revenue losses due to maintenance downtime, is presented. This model facilitates the planning on both tactical and operational timescales. The tactical timescale is addressed by solving a long-term job allocation problem, formulated as a mixed integer linear program, that allocates jobs to time periods within a year, e.g. individual weeks. As new maintenance jobs arise and forecasts are updated, the long-term job allocation problem is re-solved and the maintenance teams choose a new tactical plan. A stochastic maintenance routing and scheduling problem is then solved, e.g. for the following week, for the jobs assigned to the relevant time period by the tactical model, producing a schedule and set of routes that the maintenance teams can use to plan their operational activities. A novel formulation of the model as a mixed-integer linear program is presented and a stochastic solution method based on simulation optimisation and an adaptive large neighbourhood search process is developed for solving the problem. Computational experiments verify that the proposed metaheuristic method is able to obtain rapid solutions on the order of minutes and offer up to 69% improvements in maintenance costs over current practices in deterministic conditions. Under uncertainty in weather, job duration and power price, the simulation optimisation method is able to produce solutions on the order of hours, which is acceptable, and leads to improvements over the current practices of up to 60% whilst producing schedules that are robust to changing weather conditions. Further computational experiments reveal that the combined solving of the tactical and operational timescales, applied to a full year of historic maintenance jobs, could lead to an improvement of 66% over the historical baseline. The data-driven techniques in this framework offer a promising route to reducing operations and maintenance costs, further facilitating the deployment of offshore renewable energy assets.
Fields of Study & Subjects
Reliability and Maintenance Optimization
Wind Energy Research and Development
Power System Reliability and Maintenance
Publication details
- Journal / source
- ERA
- Publication date
- 30 March 2026
- Language
- EN
- Document type
- Thesis