Posters

Abstract

The efficiency of wind turbine restart processes following fault events or maintenance interventions has a direct impact on operational availability and the overall cost of wind farm operation. This work presents a method for calculating a Restart Success Probability (RSP) index, developed through the integration of operational, condition monitoring, and maintenance data, combined with supervised machine learning techniques. The proposed approach introduces a unified and operationally grounded framework for quantifying restart success, integrating heterogeneous data sources that are typically analysed in isolation. These sources include high-frequency operational variables, component health indices, fault event records, maintenance work order information, and reset events from the turbine control system. The data are consolidated through a dedicated ingestion, processing, and storage architecture, enabling consistent statistical analysis, feature engineering, and predictive modelling across different turbines and operating contexts. The problem is formulated as a probabilistic classification task, in which the target represents restart success or failure, defined based on the stability of the turbine operational state after the completion of a maintenance work order, considering different post-restart time windows. This formulation allows restart performance to be evaluated beyond immediate recovery, capturing short- and medium-term operational stability. Explanatory variables include maintenance severity indicators, recent fault history, aggregated component health status, pre-restart operational state, as well as physical, environmental, and turbine-specific historical variables. Tree-based and ensemble learning models proved particularly effective in modelling restart success, demonstrating strong capability to capture nonlinear relationships, complex interactions among variables, and asset-level heterogeneity inherent to wind turbine fleets. The probabilistic outputs generated by these models showed consistent alignment with the observed post-restart behaviour of the turbines, enabling reliable discrimination between operating conditions associated with higher and lower restart risk. The application of the RSP index enabled the identification of turbines and operational contexts with increased likelihood of post-restart failure, providing actionable insights for maintenance prioritization, operational adjustments, and decision support. By transforming raw operational and maintenance data into an interpretable, asset-specific risk metric, the proposed approach enhances operational reliability and supports more informed decision-making in wind farm management. It is concluded that the RSP index represents a novel and practical contribution to wind turbine operation and maintenance analytics, highlighting the value of integrated industrial data and machine learning techniques for improving availability and reducing operational risk.