Abstract

This paper shows the comparative results of an experimental and numerical study. The numerical simulations were performed with the ANSYS/Fluent software by means of CFD applying RNG k-ε and standard k-ω turbulence models.

Experimental data were recorded of a hydrocyclone mounted on a test bench which operates with a water flow. The values extracted are the inlet pressure, outlet pressure, inlet velocity and outlet velocity. In order to investigate the pressure drop underwent by a hydrocyclone, a pressure assessment was carried out in different areas of the divice geometry.

The results of the numerical study were compared and correlated with the experimental data. The numerical values report a percentage difference of 10.45% for the RNG k-ε model and 1.57% for the standard k-ω model, with respect to the experimentally obtained output pressure. The percentage difference between experimental and numerical pressure drop is 0.04% for standard k-ω model and 0.24% for RNG k-ε model.

By studying finite volumes, the components of axial velocity, tangential velocity and turbulence intensity are calculated. These phenomena are studied simultaneously in this work. The turbulence models used to generate the corresponding analyzes in this work show similarity between both results and flow behavior. The concordance between the computational results and the experimental data is acceptable, so that the numerical simulations applied reach the purpose of this investigation.

Abstract

The finite element solution of the turbulent Navier–Stokes equations developed via (k–ϵ) turbulence models was addressed in previous publications [1–4], where a (k–L)‐predictor/(ϵ)‐corrector iterative algorithm was developed. It was shown that the developed algorithm is robust and converges for the analyses of different flows without requiring the implementation of ad hoc numerical procedures. The turbulent convection–diffusion transport equations are solved by using the velocity distributions determined from the solution of the turbulent Navier–Stokes equations. The dispersion of a die in a turbulent flow can therefore be modelled and the obtained dispersion patterns are validated via flow visualizations in water models. In the present paper, the developed analysis capability is applied to the analysis of continuous casting processes.

Abstract

In the present work, we investigate the stability of turbulence closure predictions from neural network models and highlight the role of model-data-inconsistency during inference. We quantify this inconsistency by applying the Mahalanobis distance and demonstrate that the instability of the model predictions in practical large eddy simulations (LES) correlates with a deviation of the input data between the training dataset and actual simulation data. Moreover, the method of 'stability training' is applied to increase the robustness of recurrent artificial neural networks (ANN) against small perturbations in the input, which are typically unavoidable in any practical scenario. We show that this method can increase the stability of simulations with ANN-based closure term predictions significantly. The models also achieve good accuracy on the blind testing set in comparison to the baseline model trained without stability training. The work presented here can thus be seen as a building block towards long-term stable data-driven models for dynamical systems and highlights methods to detect and counter model-data-inconsistencies.