Motivation
Supply chain, product, and controller are closely associated with each other, since the design of one of them affects the design of the rest.
Therefore, it is desirable to have a holistic design paradigm, such that a manufacturing company can adapt to rapidly changing market trends and social responsibility.
Challenges
However, design iterations of supply chain, product, and controller are separate rather than integrated, where each design iteration involves a complex optimization problem.
This limits the flexibility of the entire design process, which is critical when an important change in one of the stages should be readily reflected (e.g., re-design due to a pandemic or environmental regulations).
Approach
This research aims to utilize reinforcement learning, which enabled prompt and flexible decision-making in StarCraft II by incorporating resource harvesting, unit production, and real-time control.
Since the aforementioned tasks are analogous to supply chain, product, and controller design respectively, I claim that reinforcement learning has the potential to realize a holistic design paradigm in a real-world scenario.
Value
This work will validate the feasibility of reinforcement learning to develop a holistic approach across supply chain, product, and controller design, allowing the manufacturing industry to catch up with evolving trends and social responsibilities.
Also, this approach can be generalizable to different application domains, where holistic decision making is desired to incorporate multiple interacting components (e.g., service design, material design).