EDEM API Contact + Body Force Model | Problem with custom heat conduction model

MQ

Hi Everyone,

I am trying implement a custom heat conduction model based on Batchelor and O’Brien's formulation, by modifying the given EDEM's  used model (Chaudhari et al.). I have a certain contact model that does not use Hertz-Mindlin, so EDEM's default model cannot be used unfortunately.

I cannot seem to figure out the issue with the resultant change in temperatures. Can anyone please advise on resolving this issue?

Result

Initial temperatures: Particle 1 = 1000 K | Particle 2 = 500 K

Expected: Particle at higher initial temperature losing heat up to the point of average temperature (750 K), and vice versa.

Current: Continuous Sin/Cos wave pattern: Particle 1 drops to 500 K and then climbs back to 1000 K. Vice-versa for Particle 2.

The code is roughly as following:

Contact model (heat flux):

configforTimeStep:
particleManager->resetCustomProperty("myptc", HEAT_FLUX.c_str(), 0.0);

calculateForce:
if (element2.isSphere != 0)
{
double contactRadius = sqrt(2 * equivRadius(element1, element2) * contact.normalContactOverlap); // Coble geometric model
double heatCoefficient = 2 * contactRadius * resistivityFactor * thermalConductivity; // Batchelor and O’Brien'
double change = heatCoefficient * (*elem2Temp - *elem1Temp);
*elem1Flux += change;
*elem2Flux -= change;
}

Body force (temperature update):

externalForce:
double heatFlux = heatFluxVals[0] / (particle.mass * heatCapacity);
*temperatureChangeVals += heatFlux * timeStepData.timeStep;

Thank you.

Best regards,

MQ

 

Stephen Cole

Hi MQ,

 

How quickly do the particles change temperature? I've seen similar behaviour if the inputs (e.g. heat capacity or thermal conductivity) are not within expected ranges, for example having a very small heat capacity or super high conductivity can give you huge heat flux values.  Also if you are running at a very high time-step this can make the heat transfer unstable.

 

Otherwise from first viewing the equations look OK to me, could there be any external factors or additional physics models influencing the result?

Regards

Stephen

 

 

MQ

Stephen Cole_21117 said:

Hi MQ,

 

How quickly do the particles change temperature? I've seen similar behaviour if the inputs (e.g. heat capacity or thermal conductivity) are not within expected ranges, for example having a very small heat capacity or super high conductivity can give you huge heat flux values.  Also if you are running at a very high time-step this can make the heat transfer unstable.

 

Otherwise from first viewing the equations look OK to me, could there be any external factors or additional physics models influencing the result?

Regards

Stephen

 

 

Hi Stephen,

Thanks for your response.

- The temperature gradient appears to be extremely steep, compared to the EDEM's default model, for the same heat capacity and thermal conductivity.

- They appear to interchange the temperatures (P1 1000 K to 500 K, P2 500 K to 1000K) in a cyclic manner, rather than coming to rest at the average 750 K. However, they stay within this bound.

- I bring the particles down to complete rest after a brief contact is made, i.e., by disabling gravity and HM contact, and applying a stopping force (using a custom API). And then I apply this thermal conductivity model.

- I have set the time step (TS) size within 20% of Reyliegh TS, which corresponds to Hertz-Mindlin equations. Although I am not sure how small a TS should be used for this model's stability.

Best regards,

MQ

 

MQ

Stephen Cole_21117 said:

Hi MQ,

 

How quickly do the particles change temperature? I've seen similar behaviour if the inputs (e.g. heat capacity or thermal conductivity) are not within expected ranges, for example having a very small heat capacity or super high conductivity can give you huge heat flux values.  Also if you are running at a very high time-step this can make the heat transfer unstable.

 

Otherwise from first viewing the equations look OK to me, could there be any external factors or additional physics models influencing the result?

Regards

Stephen