This is one of the problems associated with finite element method. I have the same problems as you with FlexPDE because this software is also based on finite element method. If you change your method and use finite volume (Fluent), it may be solve your problem.

thank you for your answer. But it take me a long time to redefining all equations and change the base software. Besides all the new methods and solvers or even software take times to learn and solving errors. I prefer finding out the best mesh and solver.

It would help if you acknowledge the limitations of the approach at step 0 as one of my professors used to tell me. All in all, appreciate the problem at hand for what it is and not what you want it to be. For your case, I highly recommend getting involved in online CFD forums (https://www.cfd-online.com/ did help me a lot).

If you are trying to get the help you need to provide the reader with sufficient details like but not limited to:

Dimensions (helps to pinpoint the scales you interested in) from the complexity of the system I would assume that your reactor exists in a lab-scale setup but I might be wrong.

Mesh (grid resolution) you haven't mentioned what grid sizes you simulated. In my experience, if you are not sure about the mesh it is not worth bothering with a fully equipped model. That is something you need to do!

I am not getting what are you referring to by "technique" in the following sentence: The two studies run in totally different techniques.

Can you elaborate a bit?

What type of simulation are you interested in steady-state or a transient one? in either of those cases, there are protocols for the stability of the solution which might help.

From your description, you must be interested in a photoreaction in a mixing tank!

If such modules exist in COMSOL it totally makes sense for the simulation timer to increase, right? more equations---->more calculation!

and on a side note your tank should be aerated right? I would assume you need to get the dissolved oxygen to a minimum level to produce enough radicals. If so, your case is also multiphase!

To wrap up:

1- it is dangerous to sacrifice grid resolution for computation time.

2- list the phenomena (or physics in COMSOL) and find the relevant scales this gives you an idea of how fine of a grid is needed!

3- Steady-state or not? that might be the problem!

4- do not try to have the PERFECT model! Try to use simplifying assumptions based on your expertise in the field.

Generally, no two case studies are similar in CFD but what you might consider is to come up with a systematic way to see where your problem lies.

For example, if you say, all your models are working in a stand-alone mode then maybe you can monitor the computation resources and see which of them has the highest demand and if you find it then maybe you can go to the literature to find an alternative model to simplify it.

thank you for your attention and time Christopher Sarkizi Shams Hajian

I have a photo- reactor lab scale with a chemical reaction that work in ambient temperature and pressure. I used three physics include: mixer, radiation and transported in diluted species. yes you right, I should have multiphase module but I first decided to model a simple one, and then add other physics.

I run mixer physics in a time dependent study but it didn't run in stationary study. My two other physics run in a different time dependent study with different time range and work in stationary study too. but the three of them didn't run together and in a unique time range or in stationary study.

I create a 5 step mesh as below:

1-size>fluid dynamics>predefined>Extra fined

2-size2>all Domians>fluid dynamics>predefined> finer

3-size3>boundaries(all)>general physics > predefined > extra fine

4-corner refinement

5- Boundary layers> all domains>Number of boundary layers

I used fully coupled and segregated(with 4step for parameter calculation) method for stationary(frozen rotor) and time dependent study. I change many details in my study but it still don't work.

Hello Maedeh Asgharian ,

I agree, in general, with the comments of Christopher Sarkizi Shams Hajian . I also agree with your desire to stay with the software you are using, Comsol, rather than jumping to different software with its attendant steep learning curve. I, too, am not exactly sure what you mean by the term 'technique', but I assumed it was a different DE (Differential Equation) solver. I am not familiar with Comsol, but in Matlab, for example, there are different DE solvers the user can invoke depending on the type of problem.

It has been my experience that people who pose questions on RG rarely give detailed descriptions of their problem, so you are not alone in this regard. Sometimes this is deliberate due to the fact that their work is being funded by a business and the information is proprieary, or there could be some other valid reason for not revealling more information.

You have not mentioned whether or not any of your equations are nonlinear. This information would also be useful for the RG readers to know, and, of course, the presence of a nonlinearity would be relevant to the choice of which DE solver you use, e.g., in Matlab/Simulink, a nonlinear DE or coupled (simultaneous) DE's would usually be solved using the Simulink portion of the software. Finally, have you verified that the equations are indeed correct and that they have been correctly inputted into the software, i.e., are you sure there are no outright errors caused by you? Have you asked someone else to verify your work? I know this sounds trite, but it is also essential.

Regards,

Tom Cuff

thanks a lot for your answer Thomas Cuff

I add more details in above answer. I should note that I get some reasonable results of my model when I used a short time, and I got in the trouble when matched them together of increase time in the study. The first and second study include stationary and time dependent solution, respectively.

my details in study are as below:

stationary study: all segregated steps used MUMPS and PARDISO solution.

time dependent study: all segregated study used MUMPS solution and constant(Newton) as nonlinear method.

Hello Maedeh Asgharian ,

Thank you for the two screen shots of your Comsol console. As I said, I am not familiar with Comsol, but it appears you are solving coupled, nonlinear DE's. I don't know if you can do this, but is it possible to linearize all the DE's and see if Comsol is able to solve the coupled, linear DE's? Then add the nonlinear terms back, one at a time, and see if Comsol can still solve the systems of DE's. In other words, suppose there are three nonlinear terms in the system of DE's, then you add back the first nonlinear term, try a solution, remove that nonlinear term, add back the second nonlinear term, try a solution, etc. If each individual nonlinear terms allows a solution, then try them in pairs, etc. Unless you know which nonlinear term or terms is/are causing the problem, then you are wandering around in the dark.

Regards,

Tom Cuff

Hello dear Thomas Cuff

I'm not very professional in comsol too, but this software have defined equations and physics, which allowed you choose your model base on them. If you want solve them alone or together, you could do it, but it's not so easy to define new equations or solver in comsol. In this tool, you should have just right choose for your physics and use the suitable boundary conditions in the right place. After them, you should choose a proper mesh and solver which have a base line to choose but some times as my case, they need some more experience and knowledge. I need help for selecting solver and mesh in my model.

Maedeh Asgharian

Of course, it would make sense that each phenomenon is taking place on different timescales as you mentioned already. Frozen rotor approaches should be handled with care.

First off, I do not see your grid but from the qualitative explanation, I would assume that you used an automatic meshing procedure, correct?

It should be clear if you are using a single reference frame (SRF) or a multiple reference frame (MRF). You can answer this based on your design and if it includes baffles or not. (hint: visit cfd-online or comsol manual it does have an extensive one!)

I do like the approach Thomas Cuff proposed but I do not know if it would fit in your thesis but nevertheless that would be the best way for you to find the bottleneck.

As an alternative, I do suggest getting yourself familiar with the software of interest so that you would know what module adds which equations.

Although the software itself is a tool used in many disciplines it is almost inevitable for one to just use it without knowing the equations governing the phenomena of interest.

I also notice that your machine has less than 2 GB of RAM dedicated to your computation. Although hardware limitations should not cause divergence, it will have a significant effect on wall time required.

From a chemical engineering point of view, you need to specify (quantitatively) what is the smallest resolutions required. for a lab-scale reactor, I would assume the volume of the reactor should fall somewhere in 1-10 liters right? you should see what parameter of interest is the most important one for you! is it the power input (pneumatic + mechanical) or is it the gas-flow regime? or maybe average velocity magnitude? Also bear in mind that hardware limitation might pose a serious flaw in your simulation (you might never get to the pseudo-steady-state).

Software packages do make it look like a plug and play magic but the truth is that unless you are familiar with the underlying code, at least to a minimal amount, you are risking using models based on assumptions other than your system.

Of course including new functions in software is not easy, but nevertheless it is one of the crucial parts of academic research (If somebody did it before what would be the novelty of the work, right?)

In a nutshell, CFD is not as easy as it looks so you need to build some foundation one way or another.

I wish you the best with your project.

Regards,

Chris

P.S. if you are new to COMSOL then try the tutorials on youtube and other social media to speed up the learning phase.