Kaze Du , Bo Yang , Keqiang Xie , Nan Dong , Zhengping Zhang , Shilong Wang , Fan Mo
{"title":"LLM-MANUF: An integrated framework of Fine-Tuning large language models for intelligent Decision-Making in manufacturing","authors":"Kaze Du , Bo Yang , Keqiang Xie , Nan Dong , Zhengping Zhang , Shilong Wang , Fan Mo","doi":"10.1016/j.aei.2025.103263","DOIUrl":null,"url":null,"abstract":"<div><div>Intelligent decision-making is pivotal for unlocking the value of industrial knowledge and enhancing the manufacturing sector across diverse scenarios. However, traditional decision-making methods in manufacturing fail to fully capture the complex interrelationships among various components, often resulting in biased decisions. As a novel productivity tools, large language models (LLMs) have strong contextual semantic parsing capabilities. Therefore, this paper proposes a fine-tuned LLMs integration framework for intelligent decision-making in manufacturing. The framework enables the extraction of decision-making information from diverse feature subspaces through multiple parallel fine-tuned LLMs, which generate several preliminary decision-making plans. Subsequently, the framework models the probabilities of these plans to derive a ranked list of candidates. It then employs RoBERTa and a Dynamic Weighted Mixture of Experts Ranking Method (DWMOE) to perform multi-dimensional feature extraction and candidate ranking, guided by a multi-metric head. Finally, the best fine-tuned LLM is used to fuse the top-ranked candidates, minimizing bias in the final decision-making process. To evaluate the efficacy of LLM-MANUF, we construct a dataset of manufacturing product equipment operation and maintenance texts based on a specific automotive enterprise. The results indicate that the LLM-MANUF not only outperforms individual fine-tuned LLMs but also matches the performance of LLMs with 30B parameters, achieving a BLEU-4 score of 83.37 points, which demonstrates exceptional reliability and effectiveness. LLM-MANUF provides a powerful intelligent decision-making support tool for manufacturing decision-making models.</div></div>","PeriodicalId":50941,"journal":{"name":"Advanced Engineering Informatics","volume":"65 ","pages":"Article 103263"},"PeriodicalIF":8.0000,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Advanced Engineering Informatics","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1474034625001569","RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Intelligent decision-making is pivotal for unlocking the value of industrial knowledge and enhancing the manufacturing sector across diverse scenarios. However, traditional decision-making methods in manufacturing fail to fully capture the complex interrelationships among various components, often resulting in biased decisions. As a novel productivity tools, large language models (LLMs) have strong contextual semantic parsing capabilities. Therefore, this paper proposes a fine-tuned LLMs integration framework for intelligent decision-making in manufacturing. The framework enables the extraction of decision-making information from diverse feature subspaces through multiple parallel fine-tuned LLMs, which generate several preliminary decision-making plans. Subsequently, the framework models the probabilities of these plans to derive a ranked list of candidates. It then employs RoBERTa and a Dynamic Weighted Mixture of Experts Ranking Method (DWMOE) to perform multi-dimensional feature extraction and candidate ranking, guided by a multi-metric head. Finally, the best fine-tuned LLM is used to fuse the top-ranked candidates, minimizing bias in the final decision-making process. To evaluate the efficacy of LLM-MANUF, we construct a dataset of manufacturing product equipment operation and maintenance texts based on a specific automotive enterprise. The results indicate that the LLM-MANUF not only outperforms individual fine-tuned LLMs but also matches the performance of LLMs with 30B parameters, achieving a BLEU-4 score of 83.37 points, which demonstrates exceptional reliability and effectiveness. LLM-MANUF provides a powerful intelligent decision-making support tool for manufacturing decision-making models.
期刊介绍:
Advanced Engineering Informatics is an international Journal that solicits research papers with an emphasis on 'knowledge' and 'engineering applications'. The Journal seeks original papers that report progress in applying methods of engineering informatics. These papers should have engineering relevance and help provide a scientific base for more reliable, spontaneous, and creative engineering decision-making. Additionally, papers should demonstrate the science of supporting knowledge-intensive engineering tasks and validate the generality, power, and scalability of new methods through rigorous evaluation, preferably both qualitatively and quantitatively. Abstracting and indexing for Advanced Engineering Informatics include Science Citation Index Expanded, Scopus and INSPEC.