The condition monitoring data of the wind turbogenerator has the characteristics of large quantity

multiple sources

heterogeneity

and complex and rapid growth. Existing fault diagnosis and early warning methods are hardly able to deal with such issues quickly while ensuring accuracy under the big data. A model of wind turbines on-line fault diagnosis and early warning is put forward by referring to real-time streaming data processing technology Storm and memory batch processing technology Spark. And the gearbox fault diagnosis and early warning are taken as an example to explain the model. Firstly Storm is introduced to deal with the state monitoring data flow and the topology structure of flow data processing is designed as well. Secondly a resilient distributed dataset(RDD)programming model is employed to realize a naive Bayes algorithm and back propagation(BP)algorithm

which are used for fault diagnosis and prediction based on device status information. Finally the experimental results show the satisfactory speed-up ratio with ensured precision and the correctness and validity of the fault diagnosis and early warning model. This work is supported by National Natural Science Foundation of China(No. 61300040)and Scientific Research Project of Hebei Province(No. Z2012077).