A Comparison of Experimental-based Data Driven Models for Predicting Spontaneous Cavitation Mode Switching on a NACA 0015 Hydrofoil

Hydrodynamic cavitation on a NACA 0015 hydrofoil can experience multimodal dynamics for the same operating conditions, in which the transition from one mode to another occurs spontaneously. Predicting these transitions is crucial in naval applications to prevent cavitation erosion and performance loss. As experimental data in the form of time-resolved X-ray densitometry depicting this phenomenon is available, the present study focuses on the forecasting application and comparison of two data driven reduced-order modeling techniques: dynamic mode decomposition (DMD) and cluster based reduced order modelling (CROM). In DMD, the system is represented in a reduced sense by selecting characteristic non-orthogonal modes which are obtained from an assumed linear mapping of the snapshot set, where the modes contain system temporal behavior. Unlike modal techniques such as DMD, CROM provides a probabilistic prediction tool entirely free of underlying PDE manipulation. Dynamics are captured in a forward Markov-based model in terms of a discretized phase space, which can be used to determine the probability of future transition via a PDF transport approximation. A metric is developed to compare predictive capabilities of these two ROMs to forecast the sporadic cavitation mode transition.