Propeller cavitation can produce strong noise, and serious cavitation will also lead to the damage of propeller blades. Therefore, it is necessary to conduct real-time detection of propeller cavitation. Acoustic method is the main research orientation of propeller cavitation detection. In this study, an experiment is carried out in cavitation tunnel to obtain noise of a propeller model under differ-ent cavitation states. A number of typical parameters related to energy intensity, pulse character-istics, spectrum structure of the noise signal are calculated and their changes with the develop-ment of cavitation are analyzed. The effective parameters are chosen to comprise feature vectors, and a classify based on support vector machine (SVM) is designed to identify the propeller cavita-tion state. The result shows that the method is effective, and the recognition rate of the classifier reaches 95.8% using 240 test samples.
