生产(经济)
计算机科学
生成语法
工业生产
帕累托原理
人工智能
帕累托最优
工业工程
工程类
生产模式
工业设计
生成设计
基于模型的设计
多目标优化
生成模型
机器学习
生产线
作者
Tao Wu,Jie Li,Qiang Liu,Jinsong Bao
标识
DOI:10.1080/09544828.2026.2633490
摘要
Under the high-mix, low-volume flexible manufacturing paradigm, industrial production design and operations encounter dual challenges. Although current intelligent scheduling methods enhance operational efficiency, their effectiveness remains constrained by inherent design deficiencies. To address these limitations, this paper proposes a causality-driven generative design framework, named CausalGID, which provides a new unified approach to physical layout optimisation and operational strategy generation. Initially, a multi-scale causal knowledge graph is developed, employing large language models to extract expert-rule triplets from domain literature and to perform cross-calibration with data-driven causal relationships, thus establishing a unified knowledge base for hierarchical design. Subsequently, a causality-guided Pareto-front layout generation method is proposed, which formulates production line design as a multi-objective optimisation task and uses structural causal knowledge to guide evolutionary algorithms in automatically generating non-dominated layout solutions, thereby enhancing global flexibility and production capacity limits. Finally, a knowledge-enhanced multi-agent reinforcement learning mechanism is designed, incorporating operational causal constraints and multi-objective reward functions to guide heterogeneous device agents towards efficient, robust, and reliable dynamic coordination. Compared with traditional baseline models such as D-GA, S-MOEA, CausalGID demonstrates an 18.5% increase in average throughput, achieves 89.1% takt matching, and reduces the idle resource rate from 15.3% to 4.2%.
科研通智能强力驱动
Strongly Powered by AbleSci AI