Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.
Matching simulation results with experimental measurements for a simple heat transfer and laminar flow problem
Posted 2012年7月12日 GMT-4 16:23 Heat Transfer & Phase Change, Computational Fluid Dynamics (CFD) 7 Replies
Please login with a confirmed email address before reporting spam
Hi,
I am trying to model a simple heat transfer problem with laminar flow. My 3-D model has a silicon block with 1W power source at the bottom. At the top of the Si-block, liquid is flowing at a certain speed. I simulated this problem in COMSOL and here are my observations:
surface-average temperature at the top of the Si-block: T1 = 0.8731 K (w.r.t fluid inlet Temp)
volume-average temperature of the fluid domain: T2 = 0.0097 K (w.r.t fluid inlet Temp)
h = P/(A*dT), where dT = T1-T2 = 0.8634 K and A = 14e-3*12e-3 m^2 and P=1W
The heat transfer coefficient obtained from the above calculation is about 6894 W/m^2.K.
However, when we perform the real experiment, and compute the heat transfer coefficient, we obtain an 'h' value of almost 10 times higher than what we get in COMSOL. I know the experimental details are not provided here, but I am pretty confident that my flow-rate, temperature, etc. measurements are quite accurate. If needed, I could provide more details about my setup.
Any general comments to match simulation results with experimental results are appreciated. I am attaching my model (build using version 4.2.0.150) herewith.
Thanks,
Kapil
I am trying to model a simple heat transfer problem with laminar flow. My 3-D model has a silicon block with 1W power source at the bottom. At the top of the Si-block, liquid is flowing at a certain speed. I simulated this problem in COMSOL and here are my observations:
surface-average temperature at the top of the Si-block: T1 = 0.8731 K (w.r.t fluid inlet Temp)
volume-average temperature of the fluid domain: T2 = 0.0097 K (w.r.t fluid inlet Temp)
h = P/(A*dT), where dT = T1-T2 = 0.8634 K and A = 14e-3*12e-3 m^2 and P=1W
The heat transfer coefficient obtained from the above calculation is about 6894 W/m^2.K.
However, when we perform the real experiment, and compute the heat transfer coefficient, we obtain an 'h' value of almost 10 times higher than what we get in COMSOL. I know the experimental details are not provided here, but I am pretty confident that my flow-rate, temperature, etc. measurements are quite accurate. If needed, I could provide more details about my setup.
Any general comments to match simulation results with experimental results are appreciated. I am attaching my model (build using version 4.2.0.150) herewith.
Thanks,
Kapil
Attachments:
7 Replies Last Post 2013年12月5日 GMT-5 01:51
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
I have a few comments on your model:
1) the mesh, you should either use the default COMSOl mesh or at least more boundary mesh along the non-slip walls, for me your mesh is too coarse for the fluid (and even the solid, as locally the gradients of T are steep)
2) check carefully the heat exchange direction for your "convecting cooling", you are using "inward" Heat Flux. These two BC are very close but are slightly different in their effets, pls check carefully on a separate model to be sure the heat flow direction is correct (yours might be OK I havent checked)
3) you have 2 solver sequences, for one physics at a time, this is good to rapidly get a convergence, but you do not really couple the two physics like that, you should use the resulting of your Study 1, for the initial conditions for a combined Study 2 solving for both physics simulataneously, then you will get other results
Personally I would have started with a 2D case X-Z cut plane, as this solves quiclker, then only a 3D model. Furthermore you could consider to use a Conjugated heat transfer Physics NITF
--
Good luck
Ivar
I have a few comments on your model:
1) the mesh, you should either use the default COMSOl mesh or at least more boundary mesh along the non-slip walls, for me your mesh is too coarse for the fluid (and even the solid, as locally the gradients of T are steep)
2) check carefully the heat exchange direction for your "convecting cooling", you are using "inward" Heat Flux. These two BC are very close but are slightly different in their effets, pls check carefully on a separate model to be sure the heat flow direction is correct (yours might be OK I havent checked)
3) you have 2 solver sequences, for one physics at a time, this is good to rapidly get a convergence, but you do not really couple the two physics like that, you should use the resulting of your Study 1, for the initial conditions for a combined Study 2 solving for both physics simulataneously, then you will get other results
Personally I would have started with a 2D case X-Z cut plane, as this solves quiclker, then only a 3D model. Furthermore you could consider to use a Conjugated heat transfer Physics NITF
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Thanks for the response Ivar!
I am trying the suggestions made by you.
As you notices, I am running the study in 2-steps to reduce the simulation time. I need to run a parametric sweep kind of thing eventually, so I want to keep the simulation time small. But for the time being, I am running all physics simultaneously.
Is there a feature available in COMSOL which automatically changes the mesh sizes according to the temperature gradients in the system? For example, we start with a normal or coarse mesh and then based on the simulated temperature-profile, COMSOL reduces the size of mesh only in those areas where the gradient is large.
Also, in two-step studies, I noticed that 'cross-mark' against a study-step means that will be simulated, and a 'tick-mark' means that study will not be simulated. Although it is not a big issue, but is somewhat non-intuitive.
Thanks,
Kapil
I am trying the suggestions made by you.
As you notices, I am running the study in 2-steps to reduce the simulation time. I need to run a parametric sweep kind of thing eventually, so I want to keep the simulation time small. But for the time being, I am running all physics simultaneously.
Is there a feature available in COMSOL which automatically changes the mesh sizes according to the temperature gradients in the system? For example, we start with a normal or coarse mesh and then based on the simulated temperature-profile, COMSOL reduces the size of mesh only in those areas where the gradient is large.
Also, in two-step studies, I noticed that 'cross-mark' against a study-step means that will be simulated, and a 'tick-mark' means that study will not be simulated. Although it is not a big issue, but is somewhat non-intuitive.
Thanks,
Kapil
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
for me a X means excluded and a tick mark means this WILL be SOLVED ! ;)
You have some adaptive mesh features in the solver settings, have you tried this ?, another way is t plot grad(T) and maually make the mesh denser in these regions, often there are only a few regions that needs retweaking, and with time you will feel them and mesh denser directly without any try and repeat sequences
--
Good luck
Ivar
for me a X means excluded and a tick mark means this WILL be SOLVED ! ;)
You have some adaptive mesh features in the solver settings, have you tried this ?, another way is t plot grad(T) and maually make the mesh denser in these regions, often there are only a few regions that needs retweaking, and with time you will feel them and mesh denser directly without any try and repeat sequences
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Yeah, adaptive mesh seems to work for simple geometries. For complicated geometries, many a times, I have to keep playing with the mesh until I make COMSOL to converge and get a solution at the first place. But, thats the way FEM tools are, I guess.
Since, you mentioned that we can manually make the mesh denser in the regions of high-gradient. I think the mesh control is only available at the level of domains, boundaries and edges. I guess we can not refine the mesh in a small area inside a domain. Also, I noticed COMSOL to refine the mesh along edges and boundaries; so it assumes that the basic size of the mesh is chosen properly in the beginning.
When I save .m file of the model, there seems to be some discrepancy in the way it treats 'X' and 'tick-marks'. I might be doing something wrong here.
Since, you mentioned that we can manually make the mesh denser in the regions of high-gradient. I think the mesh control is only available at the level of domains, boundaries and edges. I guess we can not refine the mesh in a small area inside a domain. Also, I noticed COMSOL to refine the mesh along edges and boundaries; so it assumes that the basic size of the mesh is chosen properly in the beginning.
When I save .m file of the model, there seems to be some discrepancy in the way it treats 'X' and 'tick-marks'. I might be doing something wrong here.
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi
if you see any discrepancies in the behaviour, as these things happens ;)
pls do not forget to send the info to "Support" so they can fix it for next patch release, as here on the FORUM we are only users and have no direct control on the software development.
But my experience is that COMSOL developpers are really listening to our suggestions sent to "support", many of "mine" have found their way into the new releases, I notice, that is the advantage with dealing with a "young company" with more engineers than marketers and lawers ;)
--
Good luck
Ivar
if you see any discrepancies in the behaviour, as these things happens ;)
pls do not forget to send the info to "Support" so they can fix it for next patch release, as here on the FORUM we are only users and have no direct control on the software development.
But my experience is that COMSOL developpers are really listening to our suggestions sent to "support", many of "mine" have found their way into the new releases, I notice, that is the advantage with dealing with a "young company" with more engineers than marketers and lawers ;)
--
Good luck
Ivar
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Hi Kapil,
I am curious to know whether you were able to match simulation and experimental results after all. It would be great if you could share how. Thank you.
Best,
Vishal
I am curious to know whether you were able to match simulation and experimental results after all. It would be great if you could share how. Thank you.
Best,
Vishal
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
Sorry, I just saw your message in SPAM folder.
I could not match the simulation and measurements exactly, but I could match them reasonably well. Actually, I made lot of changes over time to the model. For example, I found more accurate material properties, and I even tried simulating turbulent flow once.
Nevertheless, I will look into the model that I uploaded here and my improved model.
I will get back to you.
Best,
Kapil
I could not match the simulation and measurements exactly, but I could match them reasonably well. Actually, I made lot of changes over time to the model. For example, I found more accurate material properties, and I even tried simulating turbulent flow once.
Nevertheless, I will look into the model that I uploaded here and my improved model.
I will get back to you.
Best,
Kapil