Nonlinear Transient Heat Transfer Analysis Blew Up at Beginning of Analysis

[dstuebgen]

I recently migrated to Autodesk Inventor 2023 software from 2021 this past weekend. While conducting a nonlinear transient heat transfer analysis on this model before I migrated to 2023, I had convection loads (vertical plate, horizontal hot face up plate , and horizontal hot face down plate assumptions) which included radiation and this worked very well for the kiln cycle I was modeling.

When I migrated, I saw that 2023 was updated to be able to have convection and radiation loads on the same faces without any issue. So, I adjusted my convection loads and added a radiation load separately. When I ran the analysis with the new loads, it immediately blew up and failed. The temperature gradient went from 90°F to 57000°F, and then proceeded to blow up further, which is astronomically out of the range I was expecting. Can someone help out?

[John_Holtz]

Hi @dstuebgen 

I tried a test model and see a similar behavior.

I suspect the problem is the conversion from F or C to absolute temperature R or K. For example, the normal method of applying a time varying load is to apply a magnitude of 1 and enter the "real" temperature in the transient table. My guess is the software converts 1 F or 1 C to R or K, and then multiplies it by the scale factor from the transient table. I will device a test model to see if this is true or not.

If it is correct, the solution would be to set the Units (the branch in the model tree underneath the analysis name) to SI. Enter the radiation temperature as 1 K and set the transient table to the desired temperature in K.

John

[John_Holtz]

Hi,

I have confirmed that the issue with the radiation temperature is related to the conversion from °F or °C to the absolute temperature (R or K) for the radiation heat transfer calculation. The expectation when using a time dependence table (also known as "Transient Tabular Data" or "Table Data") is the scale factor from the table multiplies the assigned load. This is how it works for other loads. In the example from Figure 1, the absolute temperature is expected to be as follows where Tconv is the conversion to absolute temperature scale:

T x LSF + Tconv = 1°F x 70 + 460 = 530 R = 70°F.

 Figure 1: Typical input when using a table. The Load Scale Factor typically multiplies the applied load: T x LSF.

The calculation that Nastran uses for the absolute temperature for radiation is as follows:

(T + Tconv) x LSF = (1°F + 460) x 70 = 32270 R.

The solution for radiation loads is one of the following:

• Assign a temperature of zero C or zero F and calculate what Load Scale Factor is needed to give the desired temperature in absolute temperature units (R or K). See Figure 2.
• Or change the unit system to SI where the temperature input is K and Tconv = 0. Then the input temperature can be 1 K and the Load Scale Factor in the table is the desired temperature (in K).

Figure 2: Correct input to obtain an ambient temperature of 70°F. For the ambient temperature, enter 0°F. For the Load Scale Factor, enter (desired temp+Tconv)/Tconv . In this example (0°F+460)*1.1522=530 R = 70 F. When using Celsius, enter an ambient temperature of 0°C and calculate the Load Scale Factor based on Tconv=273.

Note that Figures 1 and 2 are using a constant temperature throughout the analysis. Since the Table Data is extrapolated when the time is beyond the range of the table, the final time can be any value. A horizontal line from 0 to 1 is the same as a horizontal line from 0 to any time value! When the ambient temperature changes with time, the real time needs to be entered

John

[dstuebgen]

Hi John,

I tried using this method for my model and the results still exploded to the same range as when I made the original post. I'd be more than glad to discuss this via email if you are ok with that.

For the time being, using a combined convection coefficient, which approximates the amount of radiation based on the formula h_combined = h_conv + ε*σ*(T_surface + T_ambient)*((T_surface)^2 + (T_ambient)^2) is a nice workaround for this issue. In the equation, epsilon (ε) is the emissivity of the material and sigma (σ) is the Stefan-Boltzmann constant. T_suraface and T_ambient must be in absolute units (Kelvin/Rankine).

I hope in the next software update for Inventor and Inventor Nastran 2023, the radiation issue is resolved so the radiation can be applied correctly.

I am eager to hear from you soon!

-Dru

[John_Holtz]

Hi Dru,

I suggest you create a support case, www.autodesk.com/gethelp, and provide your model. It should be relatively easy to find out what is occurring in your model.

I know there will not be a change in the "next" release. The next release is due to be out in a few weeks, and any changes to the release have already been finalized. Also,

1. I have not had a chance to submit this yet. 😞
2. It is possible that there is a reason the solver was designed this way, whether good or not.
3. A more important issue that the developers will need to figure out is how to handle existing model. Having the results suddenly change is not good.

John

[dstuebgen]

Hi John,

I tried the link you posted in the reply and an error 404 occurred. But regardless of that, I will create a help ticket to hopefully get this issue figured out.

-Dru

[dstuebgen]

Update @John_Holtz

I just submitted a case ticket with a zip file of the parts and assembly. Hopefully I get a swift reply.

Thanks,
-Dru

[John_Holtz · Accepted solution]

Hi,

Sorry, I was so focused on the radiation load that I did not test the combination of radiation and convection in my model. I would have expected the time curve for the convection to use T x LSF and the time curve for radiation to use (T + Tconv) x LSF. This is how you have your model setup. (Dru provided the model through a support case.)

It looks like the time curve for the convection also needs to be based on (T + Tconv) x LSF, where T is the desired temperature in deg F at any time, Tconv = 460 to convert from deg F to R, and LSF is the load scale factor to control the desired absolute temperature as a function of time.

The easiest way to do this without going crazy is to enter the ambient temperature as 0 deg F for the radiation and convection load. Then calculate what LSF is needed to give the real desired temperature in deg R as a function of time. In other words, (0+460) x LSF = (T+460). Therefore LSF = (T+460)/460. For the convection load on the inside of your furnace, this works out to be the following load scale factors:

deg F	deg R	Time	LSF
90	550	0	1.195652
90	550	720	1.195652
300	760	64800	1.652174
997	1457	126000	3.167391
2130	2590	205200	5.630435
2330	2790	241200	6.065217
2330	2790	259200	6.065217
1550	2010	302400	4.369565
1354	1814	316800	3.943478
1354	1814	327600	3.943478
904	1364	417600	2.965217
716	1176	432000	2.556522
199	659	471600	1.432609
70	530	471708	1.152174
70	530	500400	1.152174

The third and first columns are the "normal" input of time and temperature that you would enter in Nastran. The second and fourth columns are calculations. When arranged in this order, the third and fourth columns can be copied from the spreadsheet and pasted into the time curve in Nastran. The entered ambient temperature would be 0 deg F.

For the convection loads on the outside of the furnace, change the ambient temperature from 1 deg F to 0 deg F. The time curve uses a similar calculation above, which you already did for the radiation load. You can use the same time curve table for the outside convection and radiation since the ambient temperature for convection and radiation are the same. (I will send the revised model through the support case.)

John

[dstuebgen]

Hi everyone,

After doing some research from my heat transfer book and some other stuff I kept from college and having discussions with John via the support case I opened up as well as some of my old professors, the way John described how to keep the radiation and convection loads is correct. The way 2023's version was programmed makes sense to convert the radiation ambient temperature into absolute temperature, but the way the program does it is I believe now how it should not be executed by changing both convection AND the radiation with a load scale factor vs time table.

John also made a note that he will have discussions with the programmers to see why they programmed it that way, and hopefully try to resolve it for future updates.

Hope this helps if anyone else has this issue.

-Dru