Wang, Vazirizadeh, Rosario
Published in Proceedings of the IMPC2020 Congress, SAIMM
Advances in data science, machine learning, and artificial intelligence are transforming mining and mineral processing, and making them more algorithm intensive. The paradigm is shifting from one of detection and control to one of prediction and optimisation. To build up reliable prediction and optimisation models these steps should be passed: collect, aggregate, cleanse, and process large amounts of complex, structured, and unstructured data as a first stage. Next, the right tools must be used to extract meaningful features from the operational data for the build-up of a prediction model.
A grinding circuit is one of the challenging areas in terms of determining its overall efficiency and the prediction of product quality. In a processing plant, some of the KPIs are typically unmeasurable or infrequently measured for real-time application, e.g. ore grindability and grind size. However, these variables are required to calculate the grinding circuit efficiency and to model the operation performance.
In this paper, advanced analytic tools have been used to extract one selected KPI (cyclone overall P80or final grind size) and its relationship with measured variables for an industrial dataset. The results illustrate that the performance of the grinding circuit can be predicted for real-time monitoring by means of measured data.
Grinding circuit, data-driven, machine learning, process modelling, hybrid modelling
F. Wanga*, A. Vazirizadehb,and P. Rosarioa
aHatch MMP Group, 1066 West Hastings Street, Vancouver, Canada V6E 3X2
bHatch Digital Group, 2800 Speakman Drive, Mississauga, Ontario, Canada L5K 2R7