Following an extended period of consulting and working closely with the internal team and gaining a deep understanding of their requirements, we were engaged for the design, development and migration of the client’s predictive model.

Advanced Blending Model For Feed Management In Prediction Of The Metallurgical Performance.

Following an extended period of consulting and working closely with the internal team and gaining a deep understanding of their requirements, we were engaged for the design, development and migration of the client’s predictive model.

The main objective of the project was to develop a new dynamic model (Recovery/Grade Model) to predict the recovery with a stable blending (Inputs) of different deposits based on the blending% mass as inputs.

The Challenge

• Currently the client is in need to change the metallurgical model of the  processing plant which is no longer working due to the change in mineralogy and mode of operation related to the blending of multiple deposits (Variability).
• The seven (7) deposits are blended to maintain a constant head grade feeding the concentrator.

• Initially each deposit had a mathematical formula established following studies of laboratory flotation tests on cores samples more than ten years ago during the development and design of the concentrator. As per Klimpel, these models are based on the first kinetics flotation model. The models contain formulas for predicting the recoveries of Ni, Cu, Co, Pd, Pt, and Au, as well as Ni concentrate grade and Cu concentrate grade etc…
• Subsequently, the actual historical performances are used in the utility to correct the models by manually adding a factor (on the formula or the Belding %) which makes the curve “drift” upwards or downwards and fit the model.
• The blend is used in the concentrator for a month or two while the deposit lasts in the stockpile. When the stockpile changes, or the % mass per deposit changes, the formula is again manually manipulated to get the same outcome. Which is a big problem and tedious work for metallurgists.

The Solution

Model developed using AI-Machine learning and time series analysis.

Variables for model development:

• Mass (dmt)
• Previous Blending (%)s
• Head Ni&Cu % (Feed)
• Concentrate Ni&Cu %.(Product)
• Ni/Cu % Recoveries

The Results

The new predictive and dynamic model was successfully developed for the client as a tool for predicting blending ratios and respective expected Cu/Ni recoveries.