Process model of vibrating screen based on DEM and physics-informed machine learning

Abstract

Vibrating screens are widely used in industry but it is difficult to consider the enormous controlling variables and complicated particle-particle interactions in current screen process models. This study develops a new process model for vibrating screens by the combination of discrete element method simulation and physics-informed machine learning. First, based on the assumption that the particle passing in a part of a screen is dependent on the local screen and flow conditions, the relationship between the passing flow rate and local conditions is established through machine learning of data generated by a series of controlled DEM simulations. In particular, a universal local passing function is developed to predict the passing flow rates of different size particles on a screen segment, according to the segment's vibration conditions, inclination angle, and the inlet flow. Then the process model for the whole screen is developed by connecting different segments based on mass continuity, which can predict overflow partition curves under various conditions in a much more efficient way than DEM. Moreover, the process model can be applied not only to the original simulated incline vibrating screen, but also to modified screens with varied inclination angles or vibration conditions in different segments. The model would be helpful to the smart design and control of industrial screens and other similar particle classification processes. • Combined DEM and physics-informed machine learning for process model of screen. • Local passing function established by machine learning of DEM data. • Local flow in segment predicted by ML and linked by mass conservation. • Process model accurately predicts partition curve but much more efficient than DEM. • Process model is versatile to banana screens and varied vibration screens.