Double Deep Q-network Adaption Based Dynamic Task Scheduling in Manufacturing Workshops

Machining workshop computer aided process planning (CAPP) is quite common in general factories crossing various industries, such as automobile, food processing, and textile factories. Current scheduling methods are limited by varying process environments and operations. This paper presents a flexible and demand-based CAPP using Double Deep Q-network. During robustness testing, the system reveals a workshop containing three kinds of workpieces and five kinds of machines which constitutes three kinds of work operations. Eventually, this research represents an alternative to JSP and the traditional scheduling strategy. It also provides a more flexible system and has the potential to improve workshop scheduling efficiency, and it can be adapted to more application scenarios. To achieve our results, to begin with, we set a background for the CAPP research. We then adapt the Double Deep Q-network (DDQN) to the dispatching rules we produce to obtain a suitable dispatching rule when confronting various process and demands. We eventually conclude a CAPP that can be widely used, and we also present the method regarding adapting the rule to different circumstances. The advantage of our CAPP has been emphasized by comparing the performance of our DDQN and the traditional dispatching rule.