Detection and Level Estimation of Cavitation in Hydraulic Turbines with Convolutional Neural Networks

Abstract

In this paper a method for detecting and furthermore estimating the intensity of cavitation occurrences in hydraulic turbines is presented. The method relies on analyzing high frequency signals with a convolutional neural network (CNN). The CNN is trained in an adversarial manner in order to get more robust results. After successful training the obtained network is modified in such a way, that it is possible to obtain estimations of the intensity. For evaluation purposes a separate dataset is investigated.