Multidimensional Data Analysis and Fault Diagnosis Strategy of Wind Tunnel of Hydropower Generators Based on Biological Perception Principle

Chenyang Zhu
Baihetan Hydropower Plant, China Yangtze Power Co., Ltd., Liangshan, 615000, Sichuan, China
Yuguo Zhou
Baihetan Hydropower Plant, China Yangtze Power Co., Ltd., Liangshan, 615000, Sichuan, China
Long Ma
Baihetan Hydropower Plant, China Yangtze Power Co., Ltd., Liangshan, 615000, Sichuan, China
Bo Xu
Science and Technology Research Center, China Yangtze Power Co., Ltd., Wuhan 430014, Hubei, China
Hong Cao
Tsinghua AI Plus Co., Ltd., Beijing 100000, Beijing, China

Abstract:

Based on the principle of biological perception, this paper analyzes the multidimensional data of wind tunnel of hydropower generators based on the Conformer model, accurately extracts key features, and realizes fault classification, aiming to improve the accuracy of fault diagnosis and the level of intelligent maintenance of equipment. In this study, a simulation environment was constructed on a wind tunnel experimental platform. Data on key components such as turbines, bearings, rotors, and stators were synchronously collected through various sensors. The preprocessed data was input into a Conformer model that uses a CNN (Convolutional Neural Network) and Transformer dual-body structure. This model fully simulates biological multi-sensory parallel acquisition, hierarchical feature abstraction, and dynamic attention regulation mechanisms to achieve an organic fusion of global and local information. Experimental results show that the model in this paper has an accuracy of 0.984, precision of 0.983, recall of 0.976, F1 value of 0.979 and AUC (Area Under the Curve) of 0.93 on the test set, and shows high stability in 12 months of operation and 10-fold cross-validation. The fault diagnosis strategy proposed in this paper can effectively identify key fault categories such as turbine blade wear, bearing damage, rotor imbalance, stator coil short circuit, insulation aging, abnormal excitation current, overheating fault, etc. It can also realize intelligent early warning and maintenance through real-time monitoring, online updates and feedback mechanisms, providing scientific and reliable technical support for the safe operation of hydropower generator sets.