Improving Mining Outbound Logistics with Agent-Based Simulation Modeling


One of the largest resource companies in the world, with over $80 billion in sales, decided to enter a new market. It was planning to build a new potash mine with 90% of the resources exported. They wanted to design a reliable supply chain, with a high speed of supply replenishing, and the ability to recover from natural disasters and man-made crises benefiting from such volatility. Amalgama and Goldratt companies contracted this project to design the potash mining operations and a full supply chain of outbound logistics.

Before the contractor initiated the project, it was important to understand the bottlenecks of the simulation system, which was built for this project earlier by another company. The customer was able to get some benefits from it, however the model behaved like a black box because it produced numbers without any explanation of what was happening in the supply chain. Simulation modeling was supposed to visualize the supply chain processes and help:

The wrong decisions might lead to a hundred million dollars' profit loss over a 20-year period.


To meet the requirements, the model had to:

AnyLogic simulation tool fulfilled these requirements. It allowed engineers to create a model of the supply chain, adjustable to any inclusions. AnyLogic modeling clarified the processes inside locations (ports, hubs, etc.), and showcased how modeling process constituents work and interact.

The mining logistics process started at plants and mine storages. When the products were mined and ready to be transferred, a decision was made whether to ship the product to an export channel, or leave it for the domestic market. The products got to a hub or port storage by train, and were then shipped by vessels or transferred by trucks for local distribution.

Mining Supply Chain Simulation Modeling

In the designed agent-based model, sea ports and mines, as well as trucks, trains, and vessels, acted as stand-alone agents, interacting with each other. The model also included different sources of randomness, for example, port strikes, weather delays, disruption at production sites, customer demand variability, etc. The graphs in the model showed output statistics of the supply chain and its components.

Mining Logistics Simulation Model

While running the model, sensitivity analysis was performed to define the best policy for the supply chain – Push, Hybrid, or Pull. The analysis implied adding rail cars into the system (from 2.5 thousand up to 5.5 thousand rail cars), changing the amount of storage capacity at port and mine (from 150 thousand up to 500 thousand tons), and identifying the service level alteration. The world-class service level was predefined as 98% (green), and lower service levels were marked as red and yellow.

The graph shows that Push scenario did not give any positive results. Hybrid scenario provided the required level of performance, however it was possible to reach it with Pull policy at a combination of 4.5/ 3.5 thousand rail cars with 250/300 kilotons. The system turned out to be very sensitive in terms of storage capacity.

Mining Agent-Based Simulation Model

After defining the optimal policy, complexity and volatility factors were added to the model to see how the service level would alter. Push policy was negatively impacted by adding new products, customers, hubs, or ports, whereas with Pull strategy, high service levels were maintained regardless of any factors. Each policy was then tested on how the cost per ton changes when variability increased. Push almost always had the highest cost per ton index. However, the graph showed that as volatility and complexity were added, cost per ton increased with Pull over time.

The results were then compared according to different parameters (service level, working capital, stock in hub and port, etc.), and the best/worst policies were chosen.


AnyLogic simulation modeling visually represented supply chain processes and proved Pull policy as optimal. This policy provided a higher level of service at the lowest cost per ton, with smaller working capital and capital investment requirements at the same time. It also showed how the additional storage capacity would help. Other Pull policy major benefits were as follows:

Supply Chain Sensitivity Simulation Analysis

Push policy, applied by the company before, provided poor service level, because it did not consider demand variability. The company used a multiproduct supply chain, and when customers started demanding a product, it might be missing due to lack of free storage space that was filled up with another product. Pull policy algorithm acted differently. It decided when to safely reduce the stock or increase it, depending on the demand, without paying the cost.

The model capabilities included:

The latter provided detailed results on various model parameters. For instance, the difference of Delta cost per ton/sold for Push and Pull policies was three dollars a ton. At 13 million tons per annum, it was 39 million dollars of net profit loss because the wrong policy was chosen. In Tons Sold parameter, there was 4.1 million tons difference between Push and Pull policy results, with the same capacities and volatility. Multiplied by $300 per ton, it was 1.2 billion dollars of revenue loss from choosing the wrong policy.

When the analysis was presented to the top executive, it was decided to choose pull strategy for business development.

Project presentation by Dr. Alan Barnard and Dr. Andrey Malykhanov

Similar case studies

More case studies