The control over characteristics of the iron ore blend delivered is one of the most important aspects of the mining business. The iron ore price varies according to its chemical composition and physical properties, based on the assessment of the Value-in-use differential made by Platts. Hence, sales penalties are applied to the prices in the case of out-of-specification deliveries.
For example, if the iron content in the delivered product is lower than established in the contract, a penalty is applied to the price, reducing the revenue. The content of impurities, such as alumina (Al2O3), silica (SiO2), and phosphorus (P), is also monitored.
In a standard iron mining company, the production chain is composed of mining, ore beneficiation, and client supply. These penalties are usually the result of uncoordinated decision-making between the mine planning and the client supply, the first and final stages. When the decisions are made on a "local level", the intermediate stage – the beneficiation plant – struggles to deliver the best blend possible.
Most of the time, the way the ROM is processed at the plant is not coordinated with the product delivery. Furthermore, the ore is usually blended in a unique homogenization pile. This results in the production of iron ore with an uncertain quality due to the typical variability in the ROM characteristics. However, at the end of the process, the contracts define very strict specifications. This aspect, combined with the fact that the volumes and qualities defined in the contracts are always at the limit of production capacity, generates out-of-specification deliveries and the loss of revenue.
As an example, imagine a generic iron mine with a beneficiation plant equipped with magnetic separators. The plant generates three types of iron ores sub-products, named here as Nonmagnetic (NMAG), Magnetic (MAG), and Concentrate (CON). They can be sold separately or blended in different proportions to meet different quality specifications. The production rate and composition of each sub-product vary in function of the lithology of ROM being fed to the plant, as described below:
We can arbitrarily consider that the weekly consumption of ROM follows the same distribution as the monthly availability of each lithology. The operational planning is divided into 21 sequential time slots, where the first 12 represent weeks (W1 to W12) and the last 9 represent months (M4 to M12). Each week has 28 shifts of 6 hours each and one of those is exclusive for maintenance.
The plant supplies three clients, classified as High Quality (HQ: min 63.5% Fe and max 5.5% SiO2), Medium Quality (MQ: min 62.5% Fe and max 6% SiO2), and Low Quality (LQ: min 59.5% Fe and max 8.8% SiO2), with custom products to be delivered in specific amounts every month. The sub-products stocked at the train terminal are blended and loaded to meet both the chemical specifications and the requested mass. The delivery of a product out of the contract specification is a permanent challenge.
Considering those guidelines and the availability of ROM, the plant can schedule its trains to better supply its clients. However, even with flexibility in the delivery, some trains are loaded with products out of specification. In this specific case, at month 1, one out of the three trains delivered to client MQ (2.8% of overall supply) exceeded the SiO2 content by 0.22%. Also, at month 3, four out of the eleven trains delivered to HQ client (3% of overall supply) exceeded the SiO2 content by 0.31%.
To avoid this, end-to-end integrated planning can be used. It is important to use the variability as an opportunity instead of a problem. To achieve this, as a first step, a combination of clustering and logistic classification techniques is applied to predict the results of the beneficiation process. The clustering is first used to define groups of ROMs that yield similar composition for a given product. Then, a logistic regression model is used to predict the group of a given ROM for each type of product based on its lithology.
Applying this to our example, we can segregate the three sub-products produced with high variability by the ROM pile consumed (here we have three types of ROM: A, B, and C):
The second step is to use the variables defined to create a mathematical programming model. In this model, the production of each sub-product per ROM at each period, the blending of different sub-product qualities to meet the contracts' specifications, and the train/ship loading plan can be optimized. In our example, this results in an operation planning that ensures all deliveries within clients' specifications:
The main driver for this solution is the better utilization of ROM. The plant has benefited from the possibility of choosing which ROM to treat each week, reducing the overall inflow mass to the plant. The figure below shows the comparison of ROM type consumption proportion per period between the optimal planning and the contextualization case, where we arbitrarily considered a uniform consumption of ROM A, B, and C:
Applying this combination of clustering, classification, and mathematical programming, we can see that end-to-end integrated planning can turn the difficulties into opportunities and minimize the penalties in the deliveries of out-of-specification products, increasing the profits of a mining company.
If you are interested in more benefits that can be achieved by this kind of model, check out our Insights for more content with optimization opportunities!
Author: Cassiano Lima - Senior Consultant at Cassotis Consulting
Co-authors: Fabio Silva - Senior Manager at Cassotis Consulting
Emmanuel Marchal - Managing Partner at Cassotis Consulting