Anthony Thornton (Université de Twente, Pays-Bas). Multi-scale modelling of granular avalanches.

Many flows in the natural environment or industry consist of shallow rapidly moving (segregating) granular flows (e.g. from snow avalanches, landslides, debris flows, pyroclastic flows to flows in rotating drum mixers, kilns and crushers). It is important to be able to predict the distance to which hazardous natural granular flows might travel, as this information is vital for accurate assessment of the risks posed by such events. Here we focus on a multi-scale approach where we use information from both continuum models and discrete particle simulations to create prediction parameter free models.

Continuum methods are able to simulate the bulk behaviour of such flows, but have to make averaging approximations reducing the degree of freedom of a huge number of particles to a handful of averaged parameters. Once these averaged parameters have been tuned via experimental or historical data, these models can be surprisingly accurate; but, a model tuned for one flow configuration often has no prediction power for another setup.

On the other hand discrete particle methods are a very powerful computational tool that allows the simulation of individual particles, by solving Newton's laws of motion for each particle. With the recent increase in computational power it is now possible to simulate flows containing a few million particles; however, for 1mm particles this would represent a flow of approximately 1 litre which is many orders of magnitude smaller than the flows found in industry or nature.

This talk will focus on a simplified example of dry granular flows on inclined planes. We will investigate this problem via both the continuum modelling approach, and particle simulations. We conclude by discussing how both can be combined to reveal deeper insight and demonstrate how small-scale particle simulations can be used to determine the unknown parameters in common avalanche models. Meaning the link between micro-macro parameters is known, creating a parameter free predictive model.