ACCES: Non-Invasive Simulation Calibration via Optimisation using Evolutionary Algorithms and Metaprogramming

Abstract

Calibration is a vital step in the development of rigorous digital models of diverse physical and chemical processes, yet one which is highly time- and labour-intensive. In this paper, we introduce a novel tool, Autonomous Calibration and Characterisation via Evolutionary Simulation (ACCES), which uses evolutionary optimisation to efficiently, autonomously determine the calibration parameters of a numerical simulation using only simple experimental measurements. Modelling calibration as an optimisation problem, it can use any combination of sensibly-chosen experimental measurements as calibration targets for virtually any numerical modelling technique, e.g. computational fluid dynamics (CFD), finite element analysis (FEA), discrete element method (DEM). We provide a case study demonstrating the application of the method to DEM modelling and comprehensively validate the calibrated DEM particle properties using positron emission particle tracking (PEPT) experiments in conjunction with DEM digital models of systems spanning orders-of-magnitude differences in scale and covering both collisional and frictional operating regimes, showing excellent agreement. Key results show that i) even simple, well-chosen targets can achieve highly-accurate calibration, ii) though the calibration process only used images corresponding exclusively to a free-flowing surface shape, it was able to reproduce completely different quantities as well, such as velocities and occupancies, with quantitative precision, and thus iii) DEM can quantitatively predict complex particulate dynamics even outside the calibration system. ACCES is a portable, tested, documented and officially-registered Python package available under the GNU General Public License v3.0.