Manufacturing Large Language Model Development
Decision-making is a crucial activity in manufacturing systems, encompassing production, operations management, and continuous improvement. Decision-makers often rely on their expertise and practical experience to develop multiple alternatives tailored to the specific needs of a given site, ensuring well-informed choices.
These decisions permeate the entire manufacturing lifecycle, including the design and scheduling of production lines, the selection of appropriate processing methods, and the operation and maintenance of equipment.
Efficient and accurate decision-making plays a critical role in optimizing the entire production process, helping manufacturing companies streamline decision-making process, reduce uncertainty, improve operational efficiency, and enhance overall performance. This is especially true when unforeseen events, such as quality defects or equipment failures, arise during production.
In such cases, rapid and precise decision-making can prevent issues like product quality deterioration, production delays, and even safety incidents. However, achieving fully continuous optimal decision-making is nearly impossible, as the manufacturing decision-making process is inevitably influenced by various internal and external factors, leading to decision-making bias.
Therefore, developing an intelligent manufacturing decision-making framework to effectively mitigate decision-making bias is essential for organizations to address complex production challenges, ensure consistent product quality, and maintain efficient production capacity.
The causes of decision-making bias in manufacturing can be categorized into objective and subjective factors. On the objective side, the increasing complexity of modern manufacturing systems presents new challenges.
Manufacturing systems involve intricate interactions among equipment, information flow, process flow, and human–computer interfaces, creating new demands for decision-making methods. With the rapid advancement of AI technologies, such as machine learning and deep learning, these techniques are being gradually integrated into manufacturing decision-making support systems.
Unlike traditional rule-based decision trees and expert systems, which rely on fixed rule sets or templates, AI-driven approaches automatically identify patterns and make predictions by training models on manufacturing data [8], as illustrated in Fig. 1.
AI applications in manufacturing decision-making are primarily divided into two paradigms: data-driven and knowledge-driven. Data-driven approaches focus on extracting patterns from large datasets [9] and are particularly suited for applications like condition monitoring and predictive maintenance.
While data-driven approaches assist decision-making, they typically do not directly generate actionable recommendations. In contrast, knowledge-driven approaches leverage knowledge graphs or pre-trained models to derive decision-making plans by parsing in-depth manufacturing knowledge, thus compensating for the limitations of data-driven methods.
Furthermore, the emergence of LLMs has significantly enhanced the understanding of manufacturing domain expertise. With their advanced semantic parsing capabilities, LLMs can effectively identify and extract implicit semantic information from the manufacturing process, thereby enhancing the intelligence of the manufacturing decision-making framework.
Ref : An integrated framework of Fine-Tuning large language models for intelligent Decision-Making in manufacturing