Abstract. Processes involving materials in gaseous and powder states cannot be modelled without coupling interactions between the two states. XDEM (Extended Discrete Element Method) is a valid tool for tackling this issue, since it allows a coupled CFD-DEM simulation to be run. Such strength, however, mainly finds in long computational times its main drawback. This aspect is indeed critical in several applications, since a long computational time is in contrast with the increasing demand for predictive tools that can provide fast and accurate results in order to be used in new monitoring and control strategies.This paper focuses on the use of the XDEM framework as a tool for fine tuning a lumped representation of the non-isothermal decarbonation of a CaCO3 sample in powder state. The tuning of the lumped model is performed exploiting the multi-objective optimization capability of genetic algorithms.Results demonstrate that such approach makes it possible to estimate fast and accurate models to be used, for instance, in the fields of virtual sensing and predictive control.Keywords: Extended discrete element method, computational fluid dynamics, discrete element method, lumped element modelling, multi-objective model optimization, virtual sensing.