Case Study

Using machine learning to ensure greater metallurgical recovery

Understand how Artificial Intelligence helped define startup parameters and reduce instability in ore pile switching

Mining

From crushing to transport, solutions with great experience of the processes

Project Data

IMPROVEMENT

1% increase

in metallurgical recovery

The solution has improved recovery rates

EVOLUTION

+ 30% reduction

in the time between campaigns

Campaign changeover time has been reduced, allowing more campaigns to be produced

ACCEPTANCE

90% utilisation

by the plant team

The solution had a 90% utilization factor, which shows its approval by the team

Context and Challenges

At the customer's beneficiation plant, the different material composition of each new incoming ore pile meant that stabilisation of the process took a long time.

This impacted on flotation, where operators faced difficulties in identifying good dosing parameters, which caused reagents to be overused and led to sub-optimal metallurgical recovery.

The planning team started to use some characteristics of the material to search for similar piles in the history and see which parameters best impacted the desired KPI. However, the results were not satisfactory, as in practice the stockpiles were not always similar and there were different mineralogies.

It was necessary to explore the data and develop a system that used machine learning to go beyond a few features and, considering the stack as a whole, find really similar ones. This went through the challenges of:

  1. Deepening knowledge about instability and how each discipline contributed to it
  2. Develop a system that used processes such as classification, clustering and operational patterns to analyse the similarity between stacks
  3. To test in practice the effectiveness of the system in the metallurgical recovery of the plant

Solutions Used and Equipment Provided

We did a workshop using Lean Inception to better understand the instability in the transition between stacks and the contribution of each discipline in this process, involving all the client's teams.

We use Machine Learning algorithms programmed in Python to classify and cluster the piles, allowing us to analyse their similarity in a more intelligent and effective way.

We developed a system that, taking advantage of this similarity analysis, started to search in the history for parameters that led to a better operational KPI for similar piles, prescribing mobile operation ranges according to the raw material classification. In some cases, the solution informs some parameters for the existing advanced control application in flotation.

We have developed an interface in PI Vision that, with operational KPIs, guides the production engineering and operations teams in the best decision making.

Experts

Data Scientist

Daniele Kappes

Awards

No items found.

Whitepapers

No items found.
Connect with our team of experts in various areas of industry.
Finding experts

Using machine learning to ensure greater metallurgical recovery

Understand how Artificial Intelligence helped define startup parameters and reduce instability in ore pile switching

August 6, 2021

published by

Data Scientist

Daniele Kappes

IMPROVEMENT

1% increase

in metallurgical recovery

The solution has improved recovery rates

EVOLUTION

+ 30% reduction

in the time between campaigns

Campaign changeover time has been reduced, allowing more campaigns to be produced

ACCEPTANCE

90% utilisation

by the plant team

The solution had a 90% utilization factor, which shows its approval by the team

At the customer's beneficiation plant, the different material composition of each new incoming ore pile meant that stabilisation of the process took a long time.

This impacted on flotation, where operators faced difficulties in identifying good dosing parameters, which caused reagents to be overused and led to sub-optimal metallurgical recovery.

The planning team started to use some characteristics of the material to search for similar piles in the history and see which parameters best impacted the desired KPI. However, the results were not satisfactory, as in practice the stockpiles were not always similar and there were different mineralogies.

It was necessary to explore the data and develop a system that used machine learning to go beyond a few features and, considering the stack as a whole, find really similar ones. This went through the challenges of:

  1. Deepening knowledge about instability and how each discipline contributed to it
  2. Develop a system that used processes such as classification, clustering and operational patterns to analyse the similarity between stacks
  3. To test in practice the effectiveness of the system in the metallurgical recovery of the plant

Cases and Similar Articles

Connect with our team of experts in various areas of industry.
Finding experts