Application of Multivariable Data Analysis in Mineral Processing

Abstract

Data analysis and application of machine learning (ML) have demonstrated successful performance in various data rich industrial applications. Mineral processing and metallurgical operations are considered suitable for implementation of novel ML-based algorithms. The key operating performance and product outputs are usually obtained from the lab measurements and analyses that can be expensive, complex, and time consuming. Therefore, the development and application of a soft sensor and/or a state observer is a useful option to be considered due to their ability to provide the distribution of desired outputs in a continuous manner. In addition, the motivation to apply a soft sensor (a data-based model) is to provide guidance and/or information feedback to the operator in charge of making operational decisions. The soft sensor was developed at Sherritt's Metal Plant in Fort Saskatchewan as a nonlinear neural network model and it was based on two years of plant historical data. The model was also validated based on historical data, live testing, and additional sampling of process streams during simultaneous sampling campaign.