A weak fault identification method of micro-turbine blade based on sound pressure signal with LSTM networks

Abstract

The working condition of micro-turbine blades is very harsh, which is easy to cause blade faults such as fouling, cracks, and fractures. Therefore, it is of great significance to accurately and timely identify the early faults of turbine blades. In this paper, an early weak fault identification method of micro-turbine blade based on sound pressure signal and long short-term memory (LSTM) networks is proposed. Firstly, the sound pressure signals generated by different turbine structures are analyzed by frequency spectrum. Then, the wavelet packet decomposition method is adopted to extract wavelet packet coefficient energy as the fault feature vector, and the weak fault identification model of the turbine blade is established based on LSTM networks. Finally, different faults of turbine blades under different working conditions are identified and tested by using the sound pressure test platform of the micro-turbine. The results show that it is feasible to identify turbine blade faults based on sound pressure signals. Through multiple test results, it can be seen that the identification accuracy of the proposed method for turbine blade faults reaches more than 98%, indicating that it is effective for weak fault identification of turbine blades under complex working conditions, which can provide reference and guidance for the identification and warning of early weak faults of micro turbine blades.