Posters

Abstract

One of the major challenges in wind farms is minimizing losses caused by factors beyond unavoidable grid limitations. However, a significant portion of losses arises from preventable or at least reducible causes, such as inefficiencies, maintenance issues, and especially component failures within wind turbines. Reducing or preventing these losses can lead to increased efficiency and guaranteed profitability in wind farm operations. To enhance efficiency, we propose a detailed analysis of the behavior of individual components in each wind turbine. The goal is to anticipate failures by detecting early-stage anomalies in component performance, serving as indicators of potential issues. This early detection enables preventive action to be taken before minor anomalies escalate into major failures. Such early detection is made possible through the implementation of Machine Learning (ML) models designed to analyze critical variables of each component, such as temperatures and vibrations. By calculating optimal performance ranges—adapted to various operating conditions like power output and ambient temperature—these models can identify how frequently a component operates outside these expected ranges. This allows for targeted preventive maintenance to mitigate potential failures. The proposed early detection system uses a Quantile Regression model with features representing parameters related to the turbine's operational load. The model establishes adaptive optimal performance ranges under different operating conditions. Being an unsupervised model, it eliminates the need for pre-labeled events, making implementation faster and more efficient. Furthermore, the model’s sensitivity can be tuned by adjusting the acceptance range through the selection of quantiles, allowing it to be precise depending on specific operational requirements. The ability to predict failures at the component level enables wind farm operators to implement more efficient preventive maintenance strategies. By incorporating external parameters that may correlate with failure detection, this approach ensures a holistic analysis. Ultimately, localized failure predictions at the component level will significantly enhance wind farm efficiency by reducing downtime and losses caused by failures