(409c) A Revolutionary, Evolutionary Approach to the Characterisation of Particulate Media: From Bulk Measurements to Micromechanical Properties

There exists a wide array of commercial powder characterisation tools capable of measuring a diverse assortment of bulk powder properties, such as flowability or cohesive index. Many of these properties are, however, purely heuristic and offer little physical insight into the micromechanical properties and behaviours of the materials characterised. This issue is troublesome not only in that it inhibits our understanding of particulate systems, but also means that numerical models thereof (e.g. discrete element method simulations) which require detailed knowledge of powders' microscopic (particle-level) properties cannot be calibrated using commercially-available devices. While in academia researchers have created custom systems for the measurement of particle properties, these ad hoc approaches are not well-validated, nor are they viable for industry.

A lack of knowledge regarding the micromechanical properties of particulate media not only inhibits our fundamental understanding thereof, but also prevents the meaningful application of numerical simulations such as the discrete element method (DEM), which may otherwise prove transformative to a variety of particle-handling industries.

In this talk, we present a novel approach to powder characterisation, using evolutionary algorithms coupled to DEM digital twins [link] of widely-used powder characterisation devices to autonomously determine the microscopic properties of powders from conventional bulk measurements made using said tools. The tool, Autonomous Characterisation and Calibration using Evolutionary Simulation (ACCES) [link] operates, in broad terms, as follows:

(i) A bulk measurement of the material of interest is performed using one or more chosen powder characterisation devices. This may be, for example, the free surface shape formed by a dynamic angle of repose tester (see figure).

(ii) A simulation of said characterisation device(s) is set up.

(iii) Known particle and system properties are set as ‘fixed’, unknown properties as ‘variable’.

(iv) An initial ‘population’ of simulations is set up by applying random ‘mutations’ to variable

(v) DEM simulations are run for each ‘phenotype’ (i.e. parameter set) created, modelling the real-life procedure of the characterisation device in full.

(vi) The ‘fitness’ of each phenotype is assessed based on a given ‘cost function’, i.e. its ability to reproduce the experimental measurement(s). This may be, for example, the integral difference between an experimentally-measured and DEM-simulated free surface (see figure).

(vii) The fittest phenotypes are further mutated, to produce a new ‘generation’ of simulations.

(viii) Steps (v) to (vii) are repeated until a suitable parameter set, representing the true values of the particles’ micromechanical properties, is found.

Unlike conventional methods for the determination of particle-level properties from bulk measurements, which are highly labour-intensive, the ACCES process is entirely autonomous, thus removing arguably the most significant bottleneck from the characterisation process.

As well as acting as a micromechanical characterisation tool, the same methodology may also be applied directly to a piece of industrial process equipment, allowing the fully-autonomous calibration of a DEM simulation thereof.

In this talk, we provide an overview of the ACCES methodology, alongside examples of its application to real industrial problems and materials.