Detecting Particles Properties From Their Kinematics in Granular Flows

Abstract

This Thesis introduces a method for detecting the physical properties of particles in granular flows. It aims to establish a proof of concept that particles kinematics contain information regarding their physical properties, and can therefore be used as a proxy to detect them. To reach this aim, we used a combination of simulated granular flows involving different particles, and machine learning classifiers to try and detect particles of differing properties based on the knowledge of their kinematics. This approach is first applied to the case of individual particles dropped on an inclined plane. It is found that monitoring particles position and velocity as they bounce and roll enables to detect properties such as density, stiffness, friction and adhesion. This means that this method could significantly enhance traditional sensor-based sorting used in mineral progressing, as these properties are difficult or impossible to detect using conventional sensors during motion. The kinematic-based classification is then applied to silo flows and flows in rotating drums involving mixtures of two particle sizes. Results show that the size of large and small particles can be effectively detected based on their kinematics. Moreover, they show that the size of the smaller particles can be detected from the kinematics of the larger particles. These findings are particularly relevant to optimising industrial processes involving comminution using ball mills; the method proposed here indicate that large crusher kinematics can be used to monitor the size of smaller particles during comminution. The conclusions of this thesis pave the way to enhancing a number of industrial processes requiring in-situ detection of particles properties. Furthermore, it is likely that kinematic-based detection of particle property would apply to a range of particulate fluids including suspensions, emulsions or foams.