Use Cases for Simulation & AI

Pathmind helps a wide range of industries respond better to unexpected shifts in data, surface new decision points, and find clear paths through complexity using AI and simulations.

Sign Up For Free

When Should I Consider Reinforcement Learning?

Randomness

Reinforcement learning excels at dealing with stochastic simulations compared to a heuristic or optimizer.

Emergent Behavior

Sometimes, you may want to test reinforcement learning to see if it can discover novel insights into your simulation.

Robustness

Reinforcement learning’s strategy is to train itself using as many unique scenarios as possible, allowing it to generalize. This enables you to apply the same policy to all scenarios.

Industrial Applications

Pathmind AI can help reach business goals across key applications, including manufacturing, mining, and supply chain optimization.

Supply Chain Optimization

Busy container stack yard as part of a supply chain

Business Problem
Embedded predictive models allow you to respond immediately to new situations as they arise, without the need to recalculate the optimal decisions.

Current Solution Limitations
An order-stocking solution was originally developed with a mathematical optimizer. The optimized solution worked well under ideal conditions, but when confronted with variance and change (unexpected drops in stock, delays or availability), it performed sub-optimally.

Pathmind Solution
Pathmind’s trained AI policy performed better than existing optimized heuristics in the face of variance and change for supply chain optimization. In situations where a novel situation was presented to an optimizer and to Pathmind’s AI policy, Pathmind’s policy was more robust in selecting decisions that minimized waiting times and cost without needing additional retraining, unlike the original optimizer-based solution.

Manufacturing: Assembly Line Order Sequencing

Business Problem
Complex production pipelines have varying assembly line requirements. The goal is to minimize mistakes by having all components available, while maximizing throughput.

Current Solution Limitations
Extremely difficult for human line managers to track orders and assembly-line components. Company was looking to automate routing to maximize throughput while minimizing errors.

Pathmind Solution
Successfully trained an AI policy for routing assembly orders through assembly lines based on component and line staff availability. The trained AI policy beat both existing human heuristics and a mathematically optimized simulation model.  

Industrial assembly line in a factory

Manufacturing: Field Service Optimization

Worker making vehicle repairs in the field

Business Problem
The equipment is geographically distributed within a certain area. Each piece of equipment generates revenue with a probability that is based on its age, and it requires maintenance and repair, which adds to costs. The goal is to maximize profit.

Current Solution Limitations
The previous benchmark is a FIFO heuristic. This approach does not take into account the depreciation rate probabilities and geographical distances.

Pathmind Solution
Pathmind’s trained AI policy is not only able to match the performance of the simple heuristic but do so while taking into account both depreciation rate probabilities and geographic distances giving the field service operator a more robust and flexible solution.

Note: This is an ongoing project where we expect to see further performance gains. Look for a case study blog post soon.

Mining: Open-Pit Trucks and Equipment

Business Problem
In an open-pit mine operation, trucks will shuttle between excavators and ore preparation plants. The goal is to manage the truck fleet to maximize the network’s throughput, despite uncertainties and equipment failures.

Current Solution Limitations
An optimization heuristic based on the shortest queue was the benchmark. This heuristic cannot account for the equipment failure probabilities or geometric distances covered by the trucks.

Pathmind Solution
A Pathmind-trained AI policy performed significantly better than the baseline. AI interacts with the environment and finds new and better decisions without needing to be explicitly programmed to do so. In this case, Pathmind’s AI maintains a longer queue of trucks at the closest facilities (either excavator or processing plant), rather than sending the trucks to more distant facilities with a shorter queue. The longer waiting time nonetheless coincides with a higher production rate, which is the true business goal.

Wheel loader and truck at a mining site

Not Sure Where to Start?

Get in touch for a consultation. We’re happy to showcase examples and discuss if reinforcement learning is applicable to your use case.