Brazil, January, 16th, 2008.

Subject: CAE program for heat conducting

Which is the better CAE program to simulate the heat conducting in a block? I tried using FEMLAB 3.0a (3.0.0.228) from COMSOL, but the results had a big deviation! It happened with the CFX-v5.6 program too. The softwares are listed below:

FLUENT, ABAQUS v6.7, ANSYS CFX-v11.0, Flow3D or some another one. Thanks!

Details: I wanna simulate a sample (=cemented carbide tool) like that:

Total dimension of the sample (x, y, z): 0.0127 x 0.0127 x 0.0047 (m) = 1.27 x 1.27 x 0.47 (cm) = 12.7 x 12.7 x 4.7 (mm)

Dimension of the cavity with air (Pr = 0.70) (x, y, z) (for example, 10 times the dimension of the sample): 127.0 x 127.0 x 47.0 (mm)

Interval of acquisition of data = 0.222000 seconds

Initial temperature of the sample = 29.2 (ºC) = 302.35 K

Number of thermocouples = 4

Co-ordinated (s) of the (s) thermocouple (you are):

x= 0.0095 (m) y= 0.0035 (m) z= 0.0047 (m) x = 9.5 (mm) y= 3.5 (mm) z= 4.7 (mm)

x= 0.0043 (m) y= 0.0035 (m) z= 0.0047 (m) x = 4.3 (mm) y= 3.5 (mm) z= 4.7 (mm)

x= 0.0035 (m) y= 0.0089 (m) z= 0.0047 (m) x = 3.5 (mm) y= 8.9 (mm) z= 4.7 (mm)

x= 0.0065 (m) y= 0.0059 (m) z= 0.0047 (m) x = 6.5 (mm) y= 5.9 (mm) z= 4.7 (mm)

Area (s) (s) for the (s) heat flow (s) (the heat flux [W m^-2] is in the center of the square for the coordinate z=0.0 ):

xo= 0 x=0.0104

yo= 0 y=0.0104

zo= 0.0 z=0.0

About the sample: The values of thermal properties used to calculate these temperatures are 43.1 W/m.K and 14.8 x 10^-06

m²/s.

As boundary conditions, it was considered that all the faces were submitted to a constant convection heat transfer

coefficient (h = 20 W/m²K).

Rogério Fernandes Brito - Mechanical Engineering

Hi,

All the simulation codes you mention can accurately solve your model. The error is not caused by error in the simulation codes but the way the simulation was set up. You need to properly verify and validate your simulation (no matter what software you use) to get accurate answers.

Glenn Horrocks

How do you know that the programs has good results? The programs, are which will be, need validation for each studied problem!

Hi,

This is a very basic simulation you have described. Getting accurate answers for the boundary conditions you describe is also quite easy as no fluid flow is involved, just heat transfer and that is a linear equation and therefore far easier to get accurate answers. You can be sure any simulation software which can do heat transfer will be able to do it quite accurately.

The issue is how accurate do the boundary conditions describe the real condition? I think you will find this is the main problem.

Glenn Horrocks

I´m making a comparison with experimental data and it had a big deviation, around 29%.

In your original post, you said that

"As boundary conditions, it was considered that all the faces were submitted to a constant convection heat transfer coefficient (h = 20 W/m²K)."

Perhaps it is time to check this assumption? Certainly vertical sides, top and bottom sides would all be expected to have *different* heat transfer coefficients, due to the different orientation of the surfaces and buoyancy forces. So already it seems your model is not a refined as perhaps it should be.

I think Glenn is also suggesting that the error is more likely to lie with this type of assumption for boundary conditions that have been selected than a fundamental errror in the simulation software...

So, perhaps you should research available heat transfer correlations in more detail? [Or perhaps do a full CFD calculation of the surrounding air - but I would probably not advise that straight away unless you have experience of CFD.]

How would I proceed in your position? Well, if I find disagreement between simulation and experiments suggests my first action would be to carefully check my modelling assumputions, such as the details of the boundary conditions. My second action would be to double check them, and to discuss them with an experienced colleague. Then I would check that my measurements are correctly calibrated and correctly interpreted. Only after a *lot* of testing would I come to the conclusion that a widely respected code might have a bug...

So: *If* you can generate a detailed model with a *fine* mesh using *exactly* the same material properties, boundary conditions, and tight convergence criteria on several different codes and they still disagee with each other after detailed checking, then you might have some evidence of a problem with one of the codes.

Good luck, andy

I´m using CFX-v5.6 and making a comparison with experimental data and also with a numerical data (method elemnt method) from Solidonio* reseacher (*from Federal University of Uberlandia, in Brazil - www.ufu.br). Solidonio has been used this kind of boundary condition (h=20 and T=302.35 [K]). In my study, i had to put the sample (metal) within a bigger cavity with air (laminar flow) with isothermal wall at T = 302.35 [K]. Solidonio did a simulation using only the sample, only a heat conducting simulation. On the sample, i´m using MC1 = 0.8 and this sample has 12.7 x 12.7 x 4.7 [mm]. Maybe, i would improve the mesh, but i don´t have a good PC, onlye a AMD Athon with 512 MB of memory RAM. It´s good to remember that my Biot number on this sample is much lower than 0.1, so my results show that, in all the sample, i have the same temperatura, around 338 [K] for t = 48 [s] and the experimental data gives around T = 49 + 273.15 [K]. A big deviation.

Thanks for your attention for this moment.

Best regards.

I said: "...It´s good to remember that my Biot number on this sample is much lower than 0.1, so my results show that, in all the sample, i have the same temperature...". I mean that i don´t know if the results will get better, if i make another mesh wit more elements on the sample. On the fluid (air), i´m using a coarse mesh! I don´t want to study the flow around the sample. On CFX-v5.6, i can´t make only a heat conducting simulation in a unique solid. I must make two solids, one is the fluid (air at 25 ºC) e another is the solid.

The method used by Solidonio* is the Finite Element Method (FEM), it´s not a commercial software. He* had developed this 3D software. The sample has the following thermal parameters. Density "ro": 14,900.0 [kg/m³]; Specific Heat Capacity Cp: 195.85 [J/(kg K)]; Thermal Conductivity K: 43.1 [W/(m K)]; Thermal Diffusivity Coefficient "alfa": 14.8E-06 [m²/s].

Total time of simulation: 110 [s] Timestep: 0.22 [s]

Hi,

My comment remains the same, and andy2o has reinforced it. The error is with your assumption of simple boundary conditions, not CFX. The heat transfer model of CFX has been extensively validated and for simple cases like you describe I would be extremely surprised (more like flabbergasted) if you found an error in the heat transfer model. Likewise for the heat transfer model of any other respected CFD code, it is such a basic part of the code.

To improve accuracy against experiment you need to correctly model the air flow around the part the get the true convective heat transfer. Depending on surroundings conduction and radiation may also be significant.

Obviously you will have to model the flow accurately - this will require a fine enough mesh and everything else required for an accurate simulation.

Glenn Horrocks

The deviation is around 30% when i´m comparing with experimental data! I don´t know what´s happening! The sample {12.7 x 12.7 x 4.7 [mm]} is too smal with a small Biot number (much lower than 0.1). So, the gradient of the temperature is the same through the sample in a z direction. When I improve the mesh (with small volumes), there is no advance on the results! I put 0.38 as MC1 on sample mesh and 20.0 (MC2) on the cavity walls (mesh). Next week, I´ll change the g (9.81 [m s^-2] direction. Today, I´m simulating with K = 100 and not with K = 43.1 (thermal conductivity). I´m making a simulation on CFX-v5.6. I´ll use ANSYS CFX-11.0 to see if I will have a different results.

Thanks for your contribution!

Talk with a friend in another news group:

It seems like you're describing a basic transient thermal analysis of a solid with linear properties, a heat flux load, and a convection load.

Answer (myself): Yes, it´s a basic simulation ! From t = 0.0 [s] up to 110 [s].

All of the codes you mentioned should be capable of performing this type of analysis, and all should you yield essentially the same results.

Answer (myself): I hope so! I´ve got to validate this code to keep on my investigation!

If your predictions aren't correlating with the experimental measurements, then perhaps your material properties or loads are sufficiently accurate (e.g., the loads may be varying with time, the material properties may be temperature- dependent, there are convection cells or some other complicating factor).

Answer (myself): I wanna ask to my co-or- dinator about if the material properties could be temperature- dependent.

Also, make sure you understand whether the code is applying the heat flux and convection loads instantaneously rather than ramping them up over the load step.

(myself): The Heat flux [W m^-2] varies with the time!

(myself): For the time t=0 up to 63.27 [s]:

Flux_a (B*(19453.29701*(t/A)^3.07678)/(4.73947^3.07678+(t/A)^3.07678))

A 1.0 [s] B 1.0 [W m^-2]

For the time t=63.27 up to 110.27 [s]:

Flux_b (B*(-310.90531+42173.66886*exp(-((t/A)-60.93432)/1.77107)+16687.48239*exp(-((t/A)-60.93432)/1.7707)+9118.37502*exp(-((t/A)-60.93432)/6.52949)))

A 1.0 [s] B 1.0 [W m^-2]

I will verify about the material properties, but I think that they are all constants along with the time! This is an assumption, of course, on making the experiments, but Solidônio´s researcher had made it on his simulation, when he had used the his finite element code (not a commercial code).

Thanks!

The Initial temperature of the sample is 29.2 (ºC) = 302.35 K.

When I put initial temperature on CFX-v5.6, with value 0.0 [ºC] = 273.15 [K], i have good results.

My question: even i have a small sample (12.7 x 12.7 x 4.7 [mm]), how can i put, on experimental measurements, a 29.2 [ºC] = 302.35 [K] as the initial temperature for all the interior of the sample?

Glenn Horrocks,

please give a look on this paper from Mr. Sandro. The subject is about my studied problem.

http://rogeriofernandesbrito.googlep...Sandro2006.pdf

Abstract: The temperature fields generated in the cutting processes are subject of extensive research. The studies of these thermal fields in machining are very important for the development of new technologies aiming to increase the tool lives and to reduce production costs. Since the direct temperature measurements at the chipâ€"tool interface are very complex this work proposes the estimation of the temperature and the heat flux at the chipâ€"tool interface using the inverse heat conduction problem technique. The thermal model is obtained by a numerical solution of the transient three-dimensional heat diffusion equation that considers both the tool and the tool holder assembly. To determine the solution equation the finite volume method is used. Changing in the thermal properties with the temperature and heat losses by convection are also considered. Several cutting tests using cemented carbide tools were performed in order to check the model and to verify the influence of the cutting parameters on the temperature field.