Abstract

Particulate materials are ubiquitous in nature (soils, ores, grains) and constitute a large portion of industrial feedstock (chemicals, plastics, pharmaceuticals). The Discrete Element Method (DEM), in which individual particles and their interactions are modelled, is a high-fidelity approach for modelling particulate materials in a wide range of applications in the mining[1,2], metallurgy[3], agriculture[4], additive manufacturing [5], batteries [6], pharmaceuticals [7], chemicals [8] and defence industries [9]. Producing fit for purpose DEM models requires the determination of appropriate values for the micro-mechanical parameters that describe particle interactions. The physical measurement of these parameters is challenging, and an indirect method of determination, whereby the model parameters are optimised to reproduce a bulk measurement such as the angle of repose of a granular pile, is commonly adopted [10]. This model optimisation process can be time consuming, resource intensive and may require DEM modelling expertise. Material model databases can significantly minimise these requirements and therefore have considerable utility. However, their generation using traditional methods is expensive and time consuming. In this work, machine learning methods are investigated as a means of significantly improving the efficiency of generation of a DEM material model database. Artificial Neural Network (ANN) based methods and regression-based methods are evaluated and compared in terms of their accuracy for predicting the commonly used static angle of repose and bulk density calibration responses. A dataset of thousands of material models in the commercial DEM code EDEM is used for this purpose. It is found that machine learning methods have good predictive accuracy for the angle of repose and bulk density responses and a systematic approach for the rapid generation of a material model database using these methods is proposed. [1] Gröger, T., & Katterfeld, A. (2007). Application of the discrete element method in materials handling. Bulk Solids Handling, 27(1), 17–22. [2] Rodriguez, V. A., de Carvalho, R. M., & Tavares, L. M. (2018). Insights into advanced ball mill modelling through discrete element simulations. Minerals Engineering, 127(May), 48–60. [3] Mio, H., Kadowaki, M., Matsuzaki, S., & Kunitomo, K. (2012). Development of particle flow simulator in charging process of blast furnace by discrete element method. Minerals Engineering, 33, 27–33. [4] Horabik, J., & Molenda, M. (2016). Parameters and contact models for DEM simulations of agricultural granular materials: A review. Biosystems Engineering, 147, 206–225. [5] Fouda, Y. M., & Bayly, A. E. (2020). A DEM study of powder spreading in additive layer manufacturing. Granular Matter, 22(1) [6] Schreiner, D., Klinger, A., & Reinhart, G. (2020). Modeling of the calendering process for lithium-ion batteries with DEM simulation. Procedia CIRP, 93, 149–155. https://doi.org/10.1016/j.procir.2020.05.158 [7] Yeom, S. Bin, Ha, E., Kim, M., Jeong, S. H., Hwang, S. J., & Choi, D. H. (2019). Application of the discrete element method for manufacturing process simulation in the pharmaceutical industry. Pharmaceutics, 11(8). [8] Guo, Y., & Curtis, J. S. (2015). Discrete element method simulations for complex granular flows. Annual Review of Fluid Mechanics, 47, 21–46. [9] Wasfy, T. M., Mechergui, D., & Jayakumar, P. (2019). Understanding the Effects of a Discrete Element Soil Model’s Parameters on Ground Vehicle Mobility. Journal of Computational and Nonlinear Dynamics, 14(7). https://doi.org/10.1115/1.4043084 [10] Coetzee, C. J. (2017). Review: Calibration of the discrete element method. Powder Technology, 310, 104–142. https://doi.org/10.1016/j.powtec.2017.01.015