Cylindrical Nozzle Cavitation Water Jet Characteristics And Neural Network Reynolds Stress Prediction

Cavitation water jet technology has wide application prospects in deep well drilling,cleaning and rust removal,metal surface strengthening and other fields due to its advantages of strong impact energy,low working pressure and green environmental protection.The flow field study of cavitated water jet is of great significance for understanding the cavitation mechanism and effectively using cavitation water jet technology.At present,the industry mostly uses the Reynolds time-average N-S equation(RANS)to solve the turbulent flow field.However,because the vortex-viscosity model used in the RANS equation is often based on the Boussinesq hypothesis to satisfy the linear relationship between anisotropic Reynolds stress and average strain rate,it is difficult to fully capture the anisotropic characteristics existing in the flow field,which leads to great differences between simulated data and real data.In this paper,the Cavitation Jet Flow Field of cylindrical nozzles with inlet pressures of 5MPa,7MPa and 12 MPa is numerically simulated by using the Mixture multiphase flow model,ZGB cavitation model and WALE model in the large eddy model.The results show that cylindrical nozzle cavitation clouds have periodic evolution of primary,developing,shedding and collapsing.As the pressure increases,the shedding cycle of the cavitation cloud and the length of the cavitation cloud continue to increase.The high-speed camera shooting results are compared with the simulated cavitation flow field,and it is found that the simulated cavitation cloud evolution period and cavitation cloud length are in good agreement with the high-speed camera shooting experimental results,which verifies the accuracy of the LES method simulation.Using the LES simulation result data as the input and output features of the neural network,taking the RANS model simulation result data as the baseline data,and based on the DNN deep neural network,the dimensionless average strain rate tensor and the average rotational strain rate tensor were selected and then two tensor invariant primary terms,dimensionless mixed density and dimensionless average velocity gradient were used as the input features of the neural network,and the anisotropic Reynolds stress tensor was used as the output features of the neural network.The 5MPa and 12 MPa large eddy simulation data were used for DNN training,and the cavitation jet flow field of RANS model with inlet pressure of 7MPa was predicted,and the DNN prediction results were compared with the corresponding LES simulation results,and the results showed that the error between DNN prediction results and LES numerical simulation results was about 10%,which could better reveal the development law of anisotropic Reynolds stress in cavitation jet flow field compared with RANS model DNN.It is shown that DNN has good prediction ability for anisotropic Reynolds stress tensor of cavitation jet flow field of cylindrical nozzle,which provides a basis for modeling turbulence in cavitated water jet.