Automatic Fault Detection for Wind Turbines Using Single-Class Machine Learning Methods
This work adopts data related to the rotor efficiency of wind turbine to estimate the performance of wind turbine. To achieve this goal, two novel machine learning methods are adopted to build models for wind-turbine fault detection: one is the support vector data description (SVDD) and the other is the kernel principal component analysis (KPCA). The data collected from a normally-operating wind turbine are used to train models. In addition, we also build a health index using the KPCA reconstruction error, which can be used to predict the performance of a wind turbine when it operates online. The data used in our experiments were collected from a real wind turbine in Taiwan. Experiments results show that the model based on KPCA performs better than the one based on SVDD. The highest fault detection rate for KPCA model is higher than 98%. The results also indicate the validity of using rotor efficacy to predict the overall performance of a wind turbine.
Y. H. Liu et al., "Automatic Fault Detection for Wind Turbines Using Single-Class Machine Learning Methods", Applied Mechanics and Materials, Vols. 58-60, pp. 2602-2607, 2011