Discussion Forum

Hello fellow Comsol-Users,

as part of my masters thesis, I'm currently working on a simulation model to analyze the thermal conduction of support structures produced by additive manufacturing. I'm using a bulk macro-scale approach, i.e. adding a whole layer at a set temperature every few seconds. I'm planning to take at least conduction and convection into account and maybe adding radiation at a later date.

The problem is, I am struggling to build a usable, functioning model with comsol itself.

The paths I've already tried are time-dependent material additions and using methods to create multiple studies with added geometries. The time dependent addition is using the events interface (sort of a birth-and-death workaround), but the results are way off the measured data. Furthermore, using this approach, the model is only able to take conduction into account and hinders me from adding convection into the simulation (since only outermost layers can be used for convection within comsol).

After writing with support, I was advised to try the methods within the application builder - and to create multiple studies or study steps to simulate the addition of material. The good news is, the results using single layers (and manually adding a second or third one) are quite a lot better than the method using the events interface. On the other hand (as far as I can see it for now), the automation using methods is flawed. Since I'm adding new geometries to the model, I cannot use the "previous solution" choice for subsequent study steps, since comsol doesn't take initial value nodes for new geometries and hence, initializing the added layer at 0°K. The workaround for that would be using two initial value nodes, using one for the initial material deposition temperature and the other for a temperature at the end of the first simulation step.

All in all this seems to become a cluttered mess if I'd simulate close to 500 layers of material, not even regarding the combination of previous solutions into one another to get a complete set of data. So while it might be possible to simulate with this, the usability of the model seems to be rather limited.

My question to more experienced users is if you know a simpler method for this problem or can offer any advice. I'd be very grateful for you help.

Best Regards,

Max Schöler