Flotation Plant Silica Soft Sensor Using Machine Learning

Can machine learning measure the silica impurity of an iron-ore flotation concentrate every hour, between the laboratory samples? This project builds soft sensors on the real data of a Brazilian iron-ore flotation plant - persistence of the last lab value, an AR(3) model of the lab results, PLS and ridge regression, gradient boosting, LightGBM, LSTM and GRU networks - and tests them honestly on later months with a time-ordered split. It explains why published notebooks report R-squared above 0.9 (random splits, leaking iron values), shows what the models learn about amine and starch dosing, identifies an ARX model and simulates an amine-dosing loop with lab or soft-sensor feedback. A web application runs the soft sensors and what-if dosing on new data; GNU Octave and python-control cross-check the results and a complete MATLAB / Simulink version is included.

Category: Mining Engineering

Mining Engineering Project MN02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.