Monitoring of Froth Flotation with Transfer Learning and Principal Component Models*

Abstract

Froth flotation is widely used in mineral processing to separate valuable mineral ores from gangue or waste material. As such, improved monitoring and control of flotation systems can have a significant impact on mineral processing efficiency. To this end, videographic monitoring of flotation cells is well established commercially to enable decision support in plant operations, but its application in automated monitoring and control is still emerging. In this paper, the incorporation of transfer learning with deep convolutional neural networks in traditional multivariate process monitoring is considered. It is shown that despite their high dimensionality, froth image features extracted with AlexNet provides better performance than achievable with traditional multivariate image methods.