Machine learning to improve mineral processing efficiency

Researchers from EMPA and SELFRAG have trained a computer to recognise how much a piece of rock has been broken by electric pulse fragmentation by the sound of the electrical discharge. This has the potential to allow future systems to recognise how much the rocks are breaking and automatically adjust energy input to achieve target size reduction.

Excessive energy consumption in the mining industry is a constant problem with some 95%+ of energy used in rock breakage being wasted as heat and noise, with very little being used to break rocks. Electric pulse /fragmentation disaggregation (EPF/EPD) has the potential to enhance comminution process efficiency

The method used acoustic emission sensor and advanced machine learning algorithms to monitor fragmentation of a gold-copper ore in the Pre-Weakening Test Station (PWTS) in single stone and semi-continuous process experiments, simulating an industrial environment.

In the semi-continuous experiments, an unsupervised learning method based on Laplacian support vector machine was used for the classification task to recognise whether a discharge cause none, some, or heavy breakage of a rock particle.

Results for the single stone tests showed accuracy above 90% in discriminating the three categories. For semi-continuous tests, we demonstrated that the unsupervised classification can be applied efficiently to estimate the amount of fracturing (and therefore weakening) of the treated ore, with the team being confident that the proposed method can be easily industrialised to monitor in situ and in real-time the electric discharge process within a comminution operation.

The open-acceess article is available from the Journal of Cleaner Production HERE.

For more information, contact us on the button to the right: