I have a bit of a problem that I just can't seem to sort out in my head, it concerns the k-ep turbulence model. Basically I have a series of jets, with swirl exhausting into a large space. I have set the turbulence intensity to 10% on each jet. and set the length scale to 20% of the inlet diameter and run a simulation. I then set the length scale to 10% of the inlet diameter and run the simulation again. In the first case I found that the recirculation was very large, a lot larger than the second case. I can't understand this as when the length scale is large the turbulent viscosity is large, therefore I thought that this would suppress recirculation rather than promote it. Any help would be most grateful.

(1). I can't change your pictures in your mind. And I think, nobody can. (2). Swirling jet flows do not follow simple logic, otherwise, the formation of tornados would be very easy to predict. (3). My suggestion is: study in great detail about the eddy-viscosity distribution and the velocity flow field. Somewhere along the line, you may be able to pick up something you like.

When you look to k-e model, you will see that 1) the length scale ~ k**3/2/eps 2) Increasing the length scale, means that the ratio of k/eps will increase. That means that Turb. kinitic energy would increase and desipation would decrease.

Hopfully that would explain your quetions.

Tareq

You may need to look at your mesh and refine the mesh in places where you expect steep gradient in the variable to capture this change.Exageration of refinement increases the round-off errors. You may also need to try another turbulent model and compare to whatever experimental work available with you. Good Luck. Ramadan

John, after months of reading posts to do with turbulence modelling with total non comprehension i decided to take a class in the subject. we wnt thru the baldwin lomax model and i liked it. it's development is clear and it seems that a reasonable person could figure out when or when not to use it. last week though we did the derivation of k-e and it came out looking brown and smelly. i've never seen so many unfounded assumptions in my life. is it just me or does the derivation of this model just seem like a mathematical exercise and a lot of order of magnitude analysis (altogether too much for quantitative purposes)? i guess i'd could do with some reassurance that it isn't a desperate grasping at straws to expect much of these models

(1). You are at the right place and the right time. (2). Thirty years ago, researchers had to make a painful step ahead. And so, the 1-equation, 2-equation, and Reynolds stresses models were born. (3). In the mixing length theory, one can prescribe the length scale distribution and make some adjustment to fit the data. For the types of problems in those days, such as the boundary layer flows, the results were acceptable. (4). The mixing length theory is not out-dated, but it is just limited. (5). Coming from the boundary layer side, the mixing length theory becomes a problem when the flow is facing adverse pressure gradients. This type of flow has rather wide range of applications, such as the diffuser flows, inlet flows, compressure flows, etc... (6). In this area, the flow separation is a serious practical concern. Therefore, it is important to bring in the history of the turbulence into the model equations. And the general form of such equation is something like momentum equations, with convection, diffusion, and source terms in the equation. (7). The derivation of the turblence kinetic energy equation (k) is rather straight forward, without any arbitrary assumptions. (8). This is not the case for the length-scale equation. Various second equations linking to the length scale have been proposed. After many years of testing, the k-epsilon equation has been the most popular one. At the same time, its defect is also well known. It is mainly related to the epsilon equation. Dr. Rodi has done some studies in this area. I also have done some modeling to bring the model to behave more consistently. (9). Another school of model is the Wilcox's k-omega model, which also has been checked out extensively in recent years. In many cases, better results have been obtained with this model. (10). Then we have another category of problems, where the swirl motion is important. This is very important in the combustor design to stabilize the flame. It is also important in turbomachinery applications. (11). So far, each case is modelled separately, and naturally, you will be seeing a lot of model variations. (12). Then there is the practical aspect of the turbulence model research vs the real life job. I mean, most people would say the turbulence modeling work belong to the researchers in modeling, it is not his job. So, the computer will continue generating numbers using old or existing models. This is true for the in-house code development and the commercial code development as well. (13). So, very little real progress will be made this way. (14). But, if one try to bring in the turbulence modeling research, the geometry modeling research, the mesh generation research, the algorithm research, and the computer graphic research, the total effort will be big. (15). With the continuous shrinking of the research effort in most companies, trying to meet the Wall street expectations, it is hard to improve the modeling effort in any area. (16). So, it seems to me that you are at the right place and the right time to do something. The turbulence modeling is wide open, it's all yours. (if you are not happy with the current state of the art) And by tracing the history through various step of modeling effort, by learning the old but not out-dated mixing length theory, one can gradually pick up the feeling in terms of the modeling.