Data acquisition in a simplified turbine model for prediction of unsteady vortex phenomena

Abstract

Abstract The utilization of machine learning in finding decisions of engineering problems is the optimal way. This study presents a new tool that applies machine learning algorithms, to predict the frequency response of an unsteady vortex phenomenon known as the precessing vortex core (PVC) that appears in a conical draft tube behind a runner. The basic values involved in Linear Support Vector Classification model training are the two components of the time-averaged velocity profile at the cone diffuser inlet and cone angle which should be accurately measured. The paper introduces the approach to accumulating an experimental database and conducting primary analysis of the implemented regimes of swirling flow in a simplified hydraulic turbine model. It was obtained that it is necessary to clearly identify the zone of recirculation flow. The presence of this zone is a necessary, but not sufficient condition for the formation of the PVC in the flow. Injection of an axial jet in a situation with moderate swirl flow allows to shift the PVC frequency about by 10% relative to the PVC frequency without an additional jet.