Wind Turbine State Analysis via Multi-Model Majority Voting

The analysis system is segmented into multiple independent classification models (first classification model, second classification model, third classification model), each trained on different learning data sets with specific classification criteria. This segmentation allows each model to specialize in detecting particular operating states or failure modes, improving overall reliability through diversified analysis while maintaining manageable complexity through modular architecture

The results from multiple classification models are merged through a majority criterion evaluation. The operating state is determined by aggregating the classifications from all models, where the state receiving the majority of votes is selected as the final determination. This merging approach enhances reliability by cross-validating results across multiple models while providing a systematic method to resolve the increased device complexity

Learning data from multiple wind energy installations are aggregated and classification models are trained in advance during a learning phase, before actual operational analysis is needed. The models are pre-trained with diverse learning data sets that include various operating conditions and failure modes, so that when real-time analysis is required, the pre-trained models can quickly and accurately classify operating states without requiring extensive processing of raw data at that moment

Instead of processing raw learning data from multiple installations each time analysis is needed, the system creates simplified copies in the form of trained classification models. These model copies encapsulate the patterns and knowledge learned from aggregated data, allowing rapid inference and classification during operation while avoiding the time-consuming process of re-processing the original large volumes of learning data