Monitoring and Control of Hydrocyclones by Use of Convolutional Neural Networks and Deep Reinforcement Learning

Abstract

The use of convolutional neural networks for monitoring hydrocyclones from underflow images was investigated. Proof-of-concept and applied industrial considerations for hydrocyclone state detection and underflow particle size inference sensors were demonstrated. The behaviour and practical considerations of model-free reinforcement learning, incorporating the additional information provided by the sensors developed, was also discussed in a mineral processing context.