Machine learning integrated design and operation management for resilient circular manufacturing systems

The implementation of smart manufacturing devices in the manufacturing industry has provided an effective and intelligent way to monitor their facilities and thus control their integrated manufacturing processes. Despite the ongoing technological advancements, the same industry is often involved with complex problems that are associated with increasing costs related to the persistent deterioration of the manufacturing/remanufacturing systems. In literature, these systems are often called circular manufacturing systems . Furthermore, this deterioration affects the quality of the manufactured products. To address these problems, a novel integrated design and operation management-based framework is introduced in order to obtain joint policies for the authorization of production, recycling, maintenance and remanufacturing activities in the context of deteriorating circular manufacturing systems. This framework utilizes a reinforcement learning technique and ad-hoc production control policies, such as Base Stock and CONWIP, in an effort to improve the flexibility and the resilience of the examined systems to the ever-changing customer demand. Finally, a series of simulation experiments evaluate the behavior and the efficiency of the proposed mechanism within the context of single-stage and 2-stage deteriorating manufacturing/remanufacturing systems. Results suggest that the presented approach can effectively generate sufficient inventory of ready-to-be-sold products due to the enhanced awareness of the ongoing customer demand compared to the traditional reinforcement learning joint control.

Automated summary

The implementation of smart manufacturing devices has provided an effective way for monitoring facilities and controlling manufacturing processes. In response to the increasing costs associated with the deteriorating manufacturing/remanufacturing systems, a novel integrated framework utilizing reinforcement learning and specific production control policies has been introduced to improve system flexibility and resilience. Simulation experiments show that this approach can effectively generate sufficient inventory to meet customer demand compared to traditional reinforcement learning methods.