Multi-modal framework to model wet milling through numerical simulations and artificial intelligence (part 2)

Created by W.Langdon from gp-bibliography.bib Revision:1.7047

@Article{THON:2022:cej,

• author = "Christoph Thon and Ann-Christin Boettcher and Felix Moehlen and Minghui Yu and Arno Kwade and Carsten Schilde",
• title = "Multi-modal framework to model wet milling through numerical simulations and artificial intelligence (part 2)",
• journal = "Chemical Engineering Journal",
• volume = "450",
• pages = "137947",
• year = "2022",
• ISSN = "1385-8947",
• DOI = "doi:10.1016/j.cej.2022.137947",
• URL = "https://www.sciencedirect.com/science/article/pii/S1385894722034337",
• keywords = "genetic algorithms, genetic programming, Wet stirred media mills, Genetic reinforcement learning, CFD-DEM simulation, Predictive mill models",
• abstract = "Modelling of stirred media mills is crucial because of their broad use in various industries, ranging from mechanochemistry and mining to the production of batteries and pharmaceuticals. Stirred media mills are responsible for a considerable portion of the global energy demand. However, requirements exist regarding highly specific or uniform particle sizes, process conditions, and reduced wear or abrasion. Multi-modal modelling, which is the intelligent integration of different approaches, such as experiments, simulations, and AI, benefits from respective advantages of each approach. In the first study, results of an experiment conducted via magnetic tracking of a tracer bead was compared with those of simulations, and the inner mill mechanisms were investigated. The two-way coupled computer fluid dynamics discrete element method (CFD-DEM) simulations allowed the investigation of subsequent modelling through AI methods [1]. A novel AI training technique called {"}genetic reinforcement learning{"} (hereinafter, GRL), which combines neural nets with genetic algorithms, was demonstrated for cases with limited data. Furthermore, genetic programming was applied to derive transparent mathematical equations based on the generated data. Using these methods and experimentally validated simulation data, predictive models were trained, and mathematical equations were derived. Relative velocity distributions in the entire simulation domain as well as spatial distributions via heatmaps were predicted and evaluated for independent cases. Systematic predictions for the characteristic relative velocity values were generated instantaneously for varying tip speeds and bead diameters in a parameter space, which would have required 1-10 years through simulations. Finally, a transparent equation was generated via genetic programming",

}

Genetic Programming entries for Christoph Thon Ann-Christin Boettcher Felix Moehlen Minghui Yu Arno Kwade Carsten Schilde

Citations