One of the biggest losses is the increasing amount of Silica (SiO2) in the concentrate coming out of their flotation plants. Excursions in contracts and downgrades of Fe content took place over the last 12 months accounting for over $ 220M

A large concern is a delay between lab samples to measure impurity levels in the output concentrate. If they could predict SiO2 impurity levels early, their engineers would be able to take corrective actions and ensure the product quality remains high and will no longer be wasted

(Refer to the 'Flotation Process.png' image in the attached zip folder)

Refer the following link to understand more about flotation process

• https://www.911metallurgist.com/blog/froth-flotation-process
• https://www.911metallurgist.com/use-column-flotation-reduce-silica-iron-ore-pellets/

You need to support us in reducing the level of SiO2 in the final concentrate coming out of the flotation process using Advanced Analytics

Data Landscape

• (C - Controllable variable, NC - Not controllable)
• (Refer to the data structure image in the attached zip folder)

(in feed.csv) : Contains information of input feed at 20s frequency. Two outputs also exist, iron concentrate and Silica concentrate.

Variables:

• Date time
• Constituents of the raw material (NC) [% silica feed, % iron feed]
• Ore pulp flow (C)
• Amina flow and starch flow(C)
• Ore pulp pH (C)
• Ore pulp Density (C)
• %Silica concentrate (Target Variable)
• %Iron concentrate (NC)

(in flotation.csv) : Contains information of process parameters at 20s frequency. It contains the air flow and column levels of 7 flotation columns

Variables:

• Date time
• Air flow in the columns (C)
• Column levels (C)

Some process Constraints are as follows:

Constraint condition	Bound Value Minimum sum of airflow for all columns  1490 Min starch and Amina flow combined  1000 %age starch Increase in 1 hour  +(-) 5% %age Amina increase in 1 hour  +(-) 5% %age pH change  +(-) 5% Airflow bound  (220,350) Level Bound	(150,750) Amina Bound    	(250,700) Pulp Flow bound	(370,415) pH bound	(8.75,10.8) Density Bound	(1.5,1.85) 

What are we looking for?

1.Can you accurately predict SiO2 so we can have a chance to act on it before lab tests come back?

2.What are the most important variables and how do they influence the level of SiO2?

Hints: Structure the feed and flotation data set into training and test sets. Consider rolling it up to one-hour frequency data. Assess the results on the scoring data set

3. Can we operate better than we have in the past? (Formulate an optimization problem to reduce the silica content in output concentrate)

4. What values should we choose for controllable variables? (Run it over a test set in the past)

5. What is the impact we can expect? (Show a sample recommendation in the past test set and assess the impact - How much silica% could your optimizer reduce?)

What you will be evaluated on:

1.Complete notebook/ script in any language of preference

2.Presentation/ Excel report highlighting the following things –

a.Name of the algorithm used for the prediction

b.All the parameters of the algorithm used

c.Feature importance along with their description

d.Model evaluation metrics on the out of time dataset (R2, MAPE, Times series plot of prediction and actuals)

e.Interpret the model results and the variable impact on the target

You get brownie points for the following:

1.Formulation of the optimization problem

2.Presentation / Excel report highlighting the following things –

a.Constraints for the optimization problem

b.Method used for the optimization

c.Recommendation for a continuous sample showing how they vary

d.High level observation from the recommendations, if any

Please submit your output .csv file exactly as in the 'SubmissionFormat.csv' file given in the zip folder.

You are also expected to upload a PDF/PPT file along with your source code. Zip these into a folder and upload it in the 'Upload Source code' section of the platform

Evaluation Metric - RMSE