Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Energy efficient control of production lines with the use of Reinforcement Learning

Manganas Grigorios

Full record

---

Summary

Optimization of the production processes is a major topic in industrial engineering. New technologies such as Reinforcement Learning (RL) have emerged, allowing for novel approaches to the issue of optimal production control in all its aspects, including production planning, maintenance and safety administration, logistics etc. Research on this field focuses on training of agents in simulation environments mimicking the behavior of their real-world counterparts. In this thesis, a simple 2-machine environment with intermediate buffers and a client queue is developed and fitted with an RL agent as its master controller, using the Proximal Policy Optimization (PPO) method. This agent is then trained on a number of different scenarios, simulating environment variable fluctuations often met in real-world production systems, such as increased demand or production costs. Though in a smaller scale than a real industrial facility, training showed effective control of the production line and, in some of the trained agents, resilience in more demanding scenarios. These findings may assist in the future development of RL agents for use in the field of industrial control and pave the way for more challenging problems to be tackled.

Available Files

Services