Application of machine vision to analyze grinding media kinematics in grinding chambers of vibratory mills

A fundamentally novel approach to determining the kinematic parameters of individual grinding bodies in the grinding chamber of a vibratory mill is presented based on application of machine vision algorithms. The YOLO model, version 23, was trained to determine the kinematic parameters of the balls in the vibratory mill grinding chamber. The model is able to track objects within a video stream, facilitating both object identification in a single frame and determination of the inter-frame displacements. A convolutional neural network, configured for object detection, was developed to identify objects within the frame. Balls were employed as the target objects (grinding bodies). Training was conducted on a manually annotated dataset comprising video frames captured during the operation of a vibratory mill loaded with four marker balls. A total of 108 frames were annotated, of which 90 constituted the training set and 18 served as the validation set. Annotation was performed using the Label Studio software. Automatic data acquisition on the ball paths was performed using the trained model, and visualization of the trajectories was provided. A dependencies of the average kinetic energy of the balls on the fill factor of the vibratory mill grinding chamber were obtained. It was established that the optimal fill factor is not dependent of the presence of the core and lies within the range of 0.8–0.9. The results of the conducted experiments provide a basis for determining future research directions aimed at investigating the energy intensity of the grinding process.