Monitoring and control of hydrocyclones by use of convolutional neural networks and deep reinforcement learning

The Challenge

Hydrocyclones are complex apparatus controlling material flow in many modern mineral processing systems. Their efficient operation can deliver significant efficiency and cost benefits, but hydrocyclones are typically monitored only at a cluster level, and operated with low-level control.

Key Findings

Image-based hydrocyclone monitoring using convolutional neural networks and model-free learning delivered significant performance benefits.
Trained systems were able to distinguish between correct and fault states under industrial conditions
This performance provides the basis for automated monitoring and control of hydrocyclone circuits, identifying obstruction or suboptimal performance and modifying operating conditions accordingly.

Benefits to WA

Keith’s research paves the way to automation of hydrocyclone operation in mineral processing and related material industries. Remote monitoring and control of this critical infrastructure could deliver important performance improvements in terms of both efficiency and safety to the WA mineral processing industry.