Transfer learning-enhanced fault diagnosis for wind turbine main bearings under diverse wind farm conditions

The performance of wind turbines relies heavily on the stability and durability of their main bearings, which endure the unpredictable forces inherent in wind energy fluctuations. Given the heightened risk of faults stemming from this variability, ensuring the reliability of each turbine???s main bearing under dynamic conditions within wind farms is paramount. To address this challenge, an innovative diagnostic framework, aiming to detect faults in main bearings from their vibration signals and overcome the diverse operating conditions and bearing types encountered in wind farms, is proposed in this paper. At its core, the framework harnesses the power of pre-trained convolutional neural networks (CNNs) to extract weight parameters, thereby enhancing the efficiency of feature extraction through transfer learning (TL). To assess the efficacy of this diagnostic approach, diverse bearing datasets have been meticulously established. These datasets provide a comprehensive evaluation of the framework’s diagnostic accuracy across various scenarios, including bearings of different types and under diverse operating conditions. Furthermore, experimental validation has confirmed the adaptability and effectiveness of transfer learning strategies across different CNN architectures. This underscores the versatility of the proposed framework in addressing a wide array of diagnostic challenges encountered in wind turbine systems. In summary, the implementation of this advanced diagnostic framework represents a significant advancement in facilitating intelligent maintenance practices for wind turbines. By promptly identifying and addressing main bearing faults, this approach contributes to the efficient and sustainable operation of wind energy systems.