Recognition of rotor rubbing fault types based on BP neural networks

For the aim of recognition of rotor rubbing fault types, this paper presents a new intelligent method - BP neural networks to identify three different conditions: normal state, coupling faults of crack collision friction and coupling faults of misalignment respectively, which is based on the experimental data of a single-span double-disc rotor experimental stage obtained in three conditions mentioned above. A BP neural network is constructed and trained to be classifier, in addition, the process of learning and recognition is also descripted in the paper. The results show that the recognition rate of rubbing fault signal is increased to 85 percent via the classification method with a BP neural network.