Optimization of Fiber Radiation Processes Using Multi-Objective Reinforcement Learning

Abstract

With the advancement of technology, a new paradigm that utilizes artificial intelligence (AI) has emerged in the smart manufacturing industry. The adaptability and flexibility of AI are gaining significant attention as they offer solutions suitable for dynamic environments and support complex decision-making processes. This intelligent trend is creating new opportunities in the global manufacturing industry and enabling more flexible and personalized production processes. This study explores a novel approach that employs multi-objective reinforcement learning to optimize two objectives, namely, production quality and yield (productivity), in non-digitalized manufacturing processes. Through this methodology, we investigate how AI and data can be leveraged to digitalize and optimize production processes in non-digital industries. Moreover, this approach can effectively derive optimal parameters for manufacturing processes through multi-objective reinforcement learning. This research has potential to address complex problems in the manufacturing industry and emphasizes the ability to find the optimal balance between production quality and yield. These findings contribute to the continuous development of intelligent manufacturing systems and are expected to enable efficient and adaptable production processes within the industry, thereby playing a crucial role in guiding the direction towards active utilization of data and AI in non-digital industries. This research achieved an 85.24% accuracy in predicting fiber strength and a 87.02% accuracy in predicting fiber elongation, resulting in a 7.25% improvement in productivity.