Generic multi-agent reinforcement learning approach for flexible job-shop scheduling
The dissertation of Dr.-Ing. Schirin Bär deals with the development of a reactive job store scheduling solution for flexible manufacturing systems. Reinforcement learning strategies are shown to train cooperating agents that are generalized to handle unknown jobs and situations.
We asked Schirin about his dissertation:
In what context was your dissertation written? Which projects or other factors particularly influenced your dissertation?
In my research unit at Siemens AG, we are constantly working on the production of the future. The question arose as to what a generic concept with low engineering for flexible order planning could look like.
That's why I investigated deep reinforcement learning strategies and their possible industrial applications in my research.
The IMA Institute at RWTH Aachen University and the TMDT at the University of Wuppertal were also researching deep reinforcement learning strategies and their use, which led to this wonderful collaboration.
What contribution does your work make to the field of research?
The research work shows a generic concept of how cooperating agents can be trained to navigate partial products through a production process. This makes it possible to develop a flexible production control system for a large number of product variants that reacts to events and changes by making situational decisions.
The contribution to the research shows which deep reinforcement learning strategies are suitable for the problem and which concepts for state, action and reward designs are advantageous for training cooperating agents that act according to a local and global optimization goal and are generalized to deal with unknown situations and orders.
What's next for you and the topic?
The cooperation between the TMDT and Siemens AG is being continued by other doctoral students who are working on how deep reinforcement learning can be used for industrial issues such as transportation processes in logistics.
In the last year of my doctorate, I started a leadership program at Amazon and worked with my teams to ensure that prioritized orders left the warehouses on time.
After two years at three different Amazon locations, I took parental leave and at the same time we moved to Suzhou in China, where our son was born. After a wonderful family time, I was delegated to China by Siemens and today I am introducing our new product portfolio of HMI panels to the market.
In addition to technology and research, I was able to gain operational store floor and leadership experience and am now learning to understand the Chinese market in order to strategically place products.
