Due to the many industrial applications of rotating drums, a wide range of operating conditions, including different particle flow regimes, are used. Knowledge of the flow regimes inside a drum is beneficial for process optimisation and control. This paper shows how the unique insights provided by a discrete element method (DEM) model of a rotating drum can be used to create soft-sensor models that detect flow regime. Impacts between particles and the drum wall are simulated, from which the feature variables are extracted. A soft-sensor model which links these feature variables to flow regime is constructed using the multivariate statistical technique of Fisher discriminant analysis (FDA). This model is able to successfully classify new testing data, which are not used in soft-sensor model training, as belonging to rolling, cascading and cataracting flow regimes.