A hybrid workflow for investigating wide DEM parameter spaces

• Published in: Powder technology Vol. 404; p. 117440
• Main Authors: Forgber, T., Khinast, J.G., Fink, E.
• Format: Journal Article
• Language: English
• Published: Lausanne Elsevier B.V 01-05-2022; Elsevier BV
• Subjects: Algorithms; Artificial neural network; Bulk characterization; Calibration; Computer applications; DEM; Optimization; Parameter identification; Regression; Regression models; Surrogate model; Workflow; Regression; Bulk characterization; Calibration; Artificial neural network; Surrogate model; DEM
• ISSN: 0032-5910; 1873-328X
• DOI: 10.1016/j.powtec.2022.117440

Calibration of contact parameters for the DEM approach remains one of the critical obstacles for an accurate description of powder flows. Ideally, such a calibration approach relies on various macroscopic responses to identify an acceptable set of contact parameters. A significant challenge arises since the parameter space for models contains at least 10 degrees of freedom. The delicate task is to develop a framework that addresses the above-mentioned problems. In this paper, a flexible framework is presented that tackles these challenges by combining DEM simulations with regression methods. A surrogate model is trained, making it possible to identify parameter combinations in a fast and effective manner. The applicability was proven for a test powder, and multiple varieties of DEM parameters were determined. Due to the analytical structure of the surrogate model, it becomes computational feasibility to combine it with any optimization algorithm.
[Display omitted]
•Framework for DEM parameter calibration is presented.•Application of decision-tree-based regression methods (XGBoost).•The influence of DEM contact parameters on bulk tests is quantified.•The surrogate model (ANN) relates DEM parameters to powder characterization tests.•Identification of acceptable DEM parameter space for model powder.