Development of deep learning-based detection technology for vortex-induced vibration of a ship's propeller

Abstract

Vortex-induced vibration (VIV) resulting lock-in phenomenon which is the vibration frequency of the structure immersed in a fluid is locked in its resonance frequencies within a flow speed range is a potential cause of vibration fatigue and/or singing of the propellers of large merchant ships. A deep learning-based indirect VIV detection algorithm that works by vibration measurement data of a hull structure instead of a propeller during the sea trial phase is proposed in this study. RPM-frequency representations of the vibration signal by stacking the vibration frequency spectrum relative to various propeller RPM's, so-called waterfall chart of 2-dimensional data like an image, are measured and fed into the proposed convolutional neural network (CNN) architecture for VIV detection i.e. VIV frequency and RPM range. To generate the large data set, artificial data looks real, for learning, a method based on the modal superposition method instead of time-consuming fluid-structure interaction analysis is proposed. 10k set data for training, 1k set for hyper-parameter tuning, and 1k set for the test were generated. After the training, it showed a diagnosis success rate of 82% for the test set. To test the proposed VIV detection algorithm, investigations were carried out in a laboratory on a small-scale ship propulsion system designed in such a way that the vortex shedding frequency and the underwater natural frequency match each other. The proposed VIV detection algorithm was tested using the structural vibration signals measured at stationary structures in the air instead of the rotating propeller immersed in water. Lastly, the validity of the proposed algorithm was tested using the structural vibration signal measured at the hull structure of a crude oil carrier in which a propeller singing occurred during her sea trial.