Summary

Engineering Group, a global engineering firm and technology consultancy with a strong practice in simulation, worked with Pathmind to apply reinforcement learning to intelligently route heavy industrial parts over a complex assembly line in a factory. Reinforcement learning was able to solve the manufacturing optimization problem, minimize factory flow bottlenecks, and discover paths that reduced unnecessary moves on the line by 11%, while increasing the number of parts whose movements are coordinated by 66%.

Challenge

Heavy parts that require multiple complex manufacturing processes, such as engines and turbines, can make it difficult for factories to allocate machine resources as each part moves down the line, and time their movements in coordination. Production line managers often consider each individual production cycle manually.

The average number of moves required to complete processing of all components in the factory was decreased by 11% — from 79 moves to 70 moves — when compared to the best previous heuristic. Additionally, the intelligent policy coordinates the movement of 66% more objects in the factory — 10 components vs. 6 components — than the best previous heuristic.

Why Pathmind Was Needed

Many factories attempt to formalize the knowledge of experts in custom heuristics that involve thousands of lines of code. But those expert systems are cumbersome and hard to adapt when data changes. Often the domain expertise of decades is implemented in brittle systems that are equally hard to adapt or replace. But reinforcement learning has shown that it is able to learn complex sequential decision-making protocols when trained on a valid simulation.

The factory owner needed a more adaptable system that was capable of responding to changes in data, while also moving heavy parts through its manufacturing processes more quickly. They needed to increase throughput within a more robust decision-making system.

Outcomes

Engineering Group’s team developed and implemented a reinforcement learning policy with Pathmind to control the movement of heavy objects in a factory. The policy coordinates movement of heavy objects on the assembly line more efficiently than any previous heuristic.

The average number of moves required to complete processing of all components in the factory was decreased by 11% — from 79 moves to 70 moves — when compared to the best previous heuristic. Additionally, the intelligent policy coordinates the movement of 66% more objects in the factory — 10 components vs. 6 components — than the best previous heuristic.

The results are an increase in manufacturing throughput and a decrease in cost associated with moving objects. This work was an extension of Engineering Group’s previous blog post on modeling factory flow.

About Engineering Group

With over 12,000 professionals in 65 global locations, Engineering Ingegneria Informatica S.p.A designs, develops and manages innovative solutions for the business areas where digitalization is having the biggest impact.