2D CNN을 이용한 풍력발전기용 베어링의 결함 검출 연구

Abstract

This study simulated the defects that could occur in bearings when installed in wind turbines, conducted experiments, and detected them using a 2D CNN. Tapered roller bearings, which have similar characteristics to those of wind turbine bearings, were selected to overcome the limitations of accessing actual bearing data. In particular, four cases were simulated, including a normal bearing and three defective bearings corresponding to outer raceway, roller, and combined outer raceway defects. Furthermore, vibration data were collected by setting the load and rotational speed as variables to simulate the environmental changes caused by wind acting on the wind turbine. These collected data were analyzed using a 2D CNN-based model, and the model reliability was verified through cross-validation. The results were compared with those of SVM, KNN, and 1D CNN to verify the model performance. The results demonstrate that the proposed 2D CNN model can successfully detect bearing defects and is expected to be useful for diagnosing defects in bearings installed in actual wind turbines.