Synthetic Data Generation for AI-Driven Fault Detection in Wind Turbines

The wind energy sector is projected to triple its installed capacity by 2030, driven by a favorable investment climate. However, the development of fault diagnosis technologies, behavior prediction models, and lifetime estimation methods is hindered by the limited availability of open-access datasets for the scientific community. In this context, digital twins emerge as a key tool to address this limitation. This work presents a comprehensive digital twin framework for wind turbines, designed to generate synthetic, properly labeled data adaptable to various analysis scenarios. The digital twin integrates physical modeling through OpenFAST (an open-source tool for Fatigue, Aerodynamics, Structures, and Turbulence simulation) and electrical and control system modeling via MATLAB/Simulink, allowing configuration according to the specific needs of the end user. The proposed framework was validated using the manufacturer’s power curve data from a 2.35 MW wind turbine. The goal is to provide a versatile platform for synthetic data generation that supports the development of advanced learning models for fault detection tasks.