(419a) An Industry-Focussed, AI-Driven Approach to the Calibration, Validation, and Optimisation of DEM and CFD-DEM Simulations

Numerical methods such as the discrete element method (DEM), computational fluid dynamics (CFD), and multiphase particle-in-cell (MP-PIC) provide potentially powerful tools for the investigation and optimisation of diverse particulate and multiphase systems. However, without rigorous calibration and validation, their outputs may be inaccurate, or even entirely unphysical. The process of calibration can, however, prove highly time- and labour-intensive, and both calibration and validation are sorely lacking a clearly-defined "Best Practice". In this work, we present the results of a 5-year project funded by the International Fine Particle Research Institute, in which we worked with 12 industrial users of DEM in order to analyse the shortcomings of current methodologies and, based on our findings, develop a Best Practice. In this work, we detail the Best Practice determined, and how it may be integrated into a wider framework for the calibration, validation, and AI-driven optimisation of DEM.

The calibration methodology developed, termed Autonomous Characterisation and Calibration using Evolutionary Simulation (ACCES), uses evolutionary optimisation strategies coupled to DEM simulations to autonomously “evolve” toward a suitable set of DEM parameters to match a chosen set of characterisation data and/or experimental results. The outputs of ACCES are then validated against wider experimental data from the system of interest, here produced using positron emission particle tracking, a powerful technique facilitating high-resolution 3D imaging of particulate systems. A similar evolutionary strategy can then be used to optimise the simulated system according to a specific set of goals determined by the industrial end-user. These may be to increase throughput (and thus revenue) or, particularly important in the current age of climate crisis, improve sustainability through e.g. increased energy-efficiency or waste reduction.

Our work demonstrates, using industry-relevant case studies, a complete workflow for the calibration, validation and optimisation of diverse industrial and scientific systems, using a set of open-source, user-friendly tools.