[Dr Youssef Hafez]

Dear ANSYS developers and users,

Having been working on CFD developments for over 30 years, one could not un-notice the presence of the very popular CFD code such as ANSYS. Being in particular interested in turbulent flows in wall bounded regions (ducts) and in open channels, made me think or at least curious to know how ANSYS is handling these cases. My interest is in how ANSYS and in general other commercial CFD codes are treating the relationship between the turbulent stresses and the strain rates (The constitutive equations or stress-strain rate relation) and are there any differences between wall- bounded (ducts or closed channel flows) and open channel flows?.
To my surprise and in spite of the vast scientific CFD publications using ANSYS, I could not find clear answers to my questions. Most articles using ANSYS cite a linear stress-strain relationship which gives the impression that the community using ANSYS are either not paying attention to this point or thinking it is a linear relationship. No information is also available about differences between closed and open channel flows in ANSYS. These two points became like a puzzle to me that I wanted to solve.
It was until recently that I discovered on ANSYS website a report titled” Best Practice: Generalized k- (GEKO) Two-Equation Turbulence Modelling in Ansys CFD Version 1.01”. This report gave me an answer to the stress- strain relation through the GEKO model and allowed me to compare between GEKO and my nonlinear k-ε code Ph.D. code as follows.
In brief a comparison between the GEKO and my nonlinear k-ε code is as follows:
1- For wall bounded flows especially at corners (such as at an airplane's body to wing), my code succeeded in an excellent manner in simulating the main velocity, secondary velocities and turbulence structure (turbulent viscosity, K, and ε). In addition, distributions of the non-gravitational pressure and turbulent stresses are well predicted too. Along the corner bisector the maximum secondary velocity divided by the average shear velocity is calculated as 0.31 which is in agreement with experimental data of Brundrett & Baines (1964) of about 0.32. For the wall bisector a value of 0.21 is predicted versus a measured value of 0.20 by Gessner and Emery 1980. Bulging of the velocity contours toward the corner is well predicted which is important in reducing separation due to adverse longitudinal pressure gradients. In my code no up-winding or (over/under) relaxations are used. Non-linear Newton–Raphson that has quadratic convergence is used for dealing with the system of nonlinear equations. The boundary shear stress is calculated and is shown to be affected by the secondary velocities.

On the other hand it looked to me that the works by the group of ANSYS users or developers are less in simulating wall-bounded or duct flows. In Matyushenko and Garbaruk (2017), doi:10.1088/1742-6596/929/1/012102), a non-linear correction for the k-ω SST model was developed and the corrected model (SST-NL) was examined. Along the corner bisector they obtained the maximum lateral velocity divided by the bulk velocity, V/U, as 0.010 but at 0.1 H (H is the duct height) while DNS data have corresponding value of V/U as 0.011 but at 0.33 H. The Ph.D. FE code predicted V/U as 0.010 at distance of 0.34 H. GEKO Turbulence Modeling in Ansys CFD (it is on ANSYS website) uses generalized k-ω, predicted V/U of 0.08 at 0.15 H. This confirms that for wall bounded flows especially at corners my nonlinear K-ε performs better that the k-w NL model.

2- For open channel flows: a very distinct feature in my nonlinear k-ε model is the use of an-isotropic turbulent viscosity in which the turbulent viscosity in the vertical direction differs from that in the lateral direction, a feature not existing in any existing CFD code to my knowledge. The model succeeded in predicting the depression in the main velocity maximum to be at 0.6 from the bed in a channel with aspect ratio of 1:2. The secondary velocity structure is also well predicted. Another important feature is prediction of the cellular secondary cells due to periodic roughness changes along the bed or walls. This helps in investigation of turbulent flow over ribbed surfaces.


An important point is that as CFD codes are all about predictions, one cannot be sure of their results no matter how many validation tests by any code are made. This is because models and codes are tested and validated under a certain set of conditions of constant, parameters and boundary conditions which are often different from the conditions under which the intended problem simulation or prediction is required. This is true no matter you have the best available software in the market (e.g. ANSYS, OPENFOAM , FLOW3D, COMSOL Multiphysics, ABAQUS, …etc.) as any single code still represents one point of view as far as mathematical treatment of the problem at hand.

Therefore, using only one code to investigate a certain phenomenon might not be the best approach as there is no assurance that this single code will produce reliable results. Overcoming this issue is through use of a different model or even using several models in parallel to tackle a new problem. This will allow comparisons of the simulation results and if all the codes’ results are within a narrow and accepted range then the results would have more confidence than when using only one code.

In summary, two added advantages could be provided to ANSYS which are (1) simulating accurately wall-bounded corner flows in ducts or at body-wing corners and (2) providing an-isotropic turbulent viscosity in open channels. Of course there are many other distinct modelling details that were not discussed to keep this message as short and brief as possible. CFD development big companies should always be interested in different technologies and look for new codes or software tools that give different functionality to open up new opportunities as an Engineering Consultancy or clients.