Development of a methodology for prediction and calibration of parameters of DEM simulations based on machine learning

This research work aims to develop a robust methodology for the global estimation of interaction parameters based on machine learning, applicable to the discrete element method (DEM) in the study of granular materials. The specific objectives include establishing a theoretical framework that relates the interaction micro-parameters with the macro-parameters and the physical behaviour of the material; performing DEM simulations for different material parameters; developing a machine learning-based model for global parameter estimation; and consolidating the methodology for obtaining micro-parameters for a given material and behaviour. The methodology will be applied specifically to the case of dry copper ore, evaluating its limitations and the possibility of extension to materials with other characteristics. This approach does not consider direct experimental tests, but focuses on the characterisation of the relationship between the input parameters of the material and its response through simulations, validating the response and sensitivity of the model in its different stages. The methodology is expected to allow the systematic estimation of interaction properties for a DEM model, considering micro-parameter duplicities and their global selection, aspects little addressed in the literature. The final verification includes mechanisms and key questions that facilitate future modifications and improvements, allowing its application to materials of different characteristics beyond the specific case study.