A hybrid prognosis approach for robust lifetime control of commercial wind turbines

Abstract

Dynamic fluctuations in the wind field to which a wind turbine (WT) is exposed to are responsible for fatigue loads on its components. To reduce structural loads in WTs, advanced control schemes have been proposed. In recent years, prognosis-based lifetime control of WTs has become increasingly important. In this approach, the prognostic controller gains are adapted based on the stateof-health (SOH) of the WT component to achieve the desired lifetime. However, stochastic wind dynamics complicates estimation of the SOH of a WT. More recently, robust controllers have been combined with real-time damage evaluation models to meet prognosis objectives. Most rely on model-based online load cycle counting algorithms to determine fatigue damage, with analytical models providing the degradation estimate. However, most use load measurements that are either unreliable or unavailable in commercial WTs, limiting their practicality. In this contribution, a hybrid prognosis scheme combining data-driven load prediction and model-based damage estimation models for robust lifetime control of commercial WTs is proposed. A data-driven support vector machine (SVM) regression model is trained using loading data obtained from dynamic simulations using a μ-synthesis robust disturbance accommodating controller (RDAC). The regression model uses available WT measurements to predict tower load. Based on this prediction, an online rain-flow counting (RFC) damage evaluation model estimates the damage level and lifetime of the tower. The RDAC controller gains are dynamically adapted to achieve a predefined damage limit and lifetime. The proposed approach is evaluated on a 5 MW reference WT and its performance is compared with a model-based prognosis scheme using ideal WT tower measurement. Results demonstrate the efficacy of the proposed approach to control the fatigue lifetime in WT components.