Automatic classification of cavitation states using hydroacoustic measurements (vor Ort)
* Presenting author
Abstract:
Due to future restrictions on the underwater sound emission of merchant ships, the design of low-noisepropellers is becoming increasingly important. Since cavitation is the dominant sound source on propellers,there is an increased demand for automated detection methods that deliver detailed information aboutcavitation type, inception, area and intensity in order to adapt the propeller design. Considering that eventhe smallest cavitation bubbles contribute to sound emission, the classical optical cavitation observation is notsufficient for a reliable analysis. In this paper, an automatic method for classification of propeller cavitationbased on acoustical hydrophone measurements is presented. It is shown that by using a Random Forestclassifier the cavitation states can be distinguished with high accuracy. A feature selection algorithm wasused to identify the relevant features that indicate the statistical quantity and frequency range in which thecavitation states differ most. By a reduction of the feature set it could be shown that besides differences alsosimilarities between the cavitation states exist. These results offer great potential for the investigation of theunderlying physical processes of cavitation.