Deep-Learning Based Identification of Inducer Cavitation Instability

Abstract

A deep-learning based method is introduced to detect and identify the inducer cavitation instability. To identify alternate blade cavitation, which is a common cavitation instability occurs at two-bladed inducer, synthetic unsteady pressure data under equal length cavitation and alternate blade cavitation have been generated and used as training data sets. The neural network is trained to categorize the unknown unsteady pressure signal into with or without cavitation instability. In the present research, the network shows good performance in capturing the key features of the instability and robustness against random noise compared with the previous mode analysis technique