Temperature (colour coded on the left) is diffused using the SPH approach. Per particle user defined conductivity values allow for a controlled melting process while an ambient temperature and conductivity is cooling the particles down. The viscosity of the particles is directly influenced by the temperature. Hotter particles receive a lower viscosity, i.e. they flow more freely and quickly, while colder particles move with a higher viscosity. When the temperature drops below a certain threshold, the particle is no longer advected by SPH but either stops, becomes a DEM particle or forms a rigid body with its neighbouring particles in case their temperature also dropped below the threshold. The neighbourhood data can again be collected utilising e.g. a uniform grid.

On the right, the particles are meshed using a marching cubes implementation. The particles’ position, colour and velocity information is sampled on a field for further refinement. A 3D smoothing kernel can be applied to distinct channels as can the ISO value, weight and kernel types be adjusted to generate a tight mesh representation of the particles or field respectively. By also sampling the velocity channel we can support motion blur. The current implementation only supports Blinn, Wyvill and a simpler Metaball kernel, but averaging approaches for smoother surfaces as introduced by Bridson et al. are likely to be added later on.