Abstract: At present, incipient cavitation in a model turbine is identified as before using the manual method whose shortcomings are a long period of data acquisition, strong subjectivity, low accuracy, and low efficiency. To improve incipient cavitation identification, this paper develops an intelligent identification method of turbine cavitation－namely the intelligent recognition method of multimodal bubble sound PSVFR based on feature extraction of cannon sound spectra and special pulsation spectra. In this method, turbine cavitation noise data are processed using MTCSPC, a multistate algorithm independently-developed by the authors. By collecting the feature vectors of incipient cavitation tones, a matrix model is constructed; then calculation and judgment are made through feature comparison with the qualitative matrix in the sample database, so as to help the machine complete the learning and recognition of model turbine cavitation noise. Compared with previous technologies, this method improves the accuracy and efficiency of machine identification of turbine incipient cavitation, with a recognition efficiency reaching up to 80%.

Key words: model turbine cavitation, recognition of cavitation, bubble sound intelligent recognition, sound state feature vector, characteristic spectrum