General Question: Error of simulation

[xuxxx679]

Hi,

I was asked a general question: how much error between simulation results and experimental data can be considered as accurate? I'm really not sure how to answer this question. My area is thermal analysis. Does anyone working on thermal have some thoughts on that?

Thanks,
Danny

 

[L_K]

Many times I hear the number 10% thrown around. Maybe for a thermal simulation (temperature) the number is a bit lower
than for structural (deformation) simulation.

 

[hendersdc]

I would say it depends on what you used as input for the simulation. If you were just going with a quick approximation to figure out where you are on the scale between "definitely a problem", "Maybe a problem, investigate more" and "No chance of a problem" an order of magnitude may be accurate enough for you.

You are also routinely making worst case assumptions or analyzing worst case conditions. What are the odds that your experimental setup is representative of that?

 

[xuxxx679]

thank you guys very much. The purpose of this simulation is to match the experimental results. The setup is very simple, it's mainly just heating up a metal bar. The issue i'm facing that the error i got for a transient analysis is between 0.5 to 2 degrees(temperature increase is 30 to 80), which I think for such an easy model, it seems to large. But I know a lot of factors can cause the error, such as inaccurate material properties, contact thermal resistance, constant heat transfer coefficient (should be temperature and location dependent).I'm just not sure what is usually considered as the 'acceptable error' for all the factors i listed above.

 

[hendersdc]

A couple of things to consider that may be throwing off your results:

- How is the bar being supported? Supports can be a path for heat flow through conduction, if it is in an oven that may heat it up faster, if it is being resistively heated the supports may act like a heat sink, lowering the temperature.

- How is the bar being heated? Resistive heating is relatively easy to model, a heating coil will create non uniform temperature distribution so moving your thermocouple around will change the experimental result. An oven would also create a non uniform temperature through the thickness of the bar. If the bar is out in the open are there any air currents?

- Are you taking natural convection into account in you analysis?

- What orientation is the bar? If vertical the top will be hotter than the bottom by a couple degrees at least.

In general though, I would consider any result from CFD that was accurate to 0.5 - 2C pretty darn good. In my line of work that would be more than sufficient since we would typically want a much greater margin than that to determine that the performance of the design is acceptable.

 

[L_K]

Simple model where heat is entering the bar from one surface and leaving through natural (and forced?) convection and radiation from others. If the error
is 2/80 = 2.5% that seems very acceptable to me. Did you model the bar in Ansys Mechanical or CFX/Fluent?