摘要：AbstractHydrocyclones are classifiers that are widely used in closed-circuit grinding operations and they play a pivotal role in the production and energy efficiency of grinding circuits. Online performance monitoring of these units is not yet established, resulting in mostly manual control based on visual inspection by process operators. Previous studies have indicated that computer vision systems show promise for online particle size estimation using convolutional neural networks. In this paper, further validation of these results is provided by conducting explainability analyses of model predictions. A model was trained using GoogLeNet in transfer learning mode to classify the particle size distribution of the underflow of a pilot-scale hydrocyclone, using images of the underflow. The model could reliably separate the images into distinct classes. Occlusion sensitivity and Grad-CAM maps were used to visualize the internal reasoning of the models. These results indicated that the convolutional neural network was making predictions based on sensible characteristics of the underflow, such as the spray profile and slurry textures.
