Application of multi-agent reinforcement learning for effective production scheduling in industry 4.0
In recent times, rapid progress can be seen in the field of artificial intelligence. These techniques have served many interesting applications from image and speech recognition to playing, and even beating humans in a game of Go. At the same time, there is a major shift in the manufacturing paradig...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2020
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/141677 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | In recent times, rapid progress can be seen in the field of artificial intelligence. These techniques have served many interesting applications from image and speech recognition to playing, and even beating humans in a game of Go. At the same time, there is a major shift in the manufacturing paradigm following the introduction of Industry 4.0 and Smart Factories. This has led to the development of tools based on modern artificial intelligence techniques to exploit big data produced by these factories to improve operational performance. Many different approaches have been studied to apply machine learning to production scheduling, using methods such as genetic algorithms[1], traditional reinforcement learning[2], and multi-agent systems[3-6]. However, high computational costs[7], slow reactions[1, 8], and low scalability[6] are some of the challenges that these methods face. This report explores a machine learning technique that attempts to improve on current efforts and overcome their disadvantages. The aim of this study is to evaluate the effectiveness and feasibility of applying an automated production control system that requires only minimal human interaction. Multi-agent reinforcement learning (MARL) is the main machine learning technique used for this report. Through the brief exploration conducted in the experiment, MARL has shown promising results in fulfilling the objective of the study. The system combines neural networks with MARL which allows the system to handle the large amounts of data from machines. Finally, this is implemented through a decentralized MARL system to control machines on a shop floor. This decentralized approach shows good performance in controlling machines in both stable and dynamic environments |
|---|