A Multi-Agent AI Framework for the Design of Self-Adaptive Production Systems

Modern production systems face increasing demands for flexibility, resilience, and responsiveness due to volatile market conditions, customized product requirements, and unexpected disruptions. Traditional manufacturing systems, designed for static or predictable environments, struggle to adapt in real time to dynamic changes. This paper presents a novel Multi-Agent AI Framework (MAAF) for the design of self-adaptive production systems. The framework integrates autonomous software agents—each responsible for monitoring, decision-making, learning, and actuation—with advanced artificial intelligence techniques including deep reinforcement learning, multi-agent coordination, and transfer learning. A layered architecture is proposed comprising: (۱) a perception layer for real-time system state monitoring; (۲) a coordination layer for agent interaction and negotiation; (۳) a learning layer for adaptive policy optimization; and (۴) an actuation layer for physical system reconfiguration. The framework is validated through a simulated flexible manufacturing cell and a real-world case study in automotive assembly. Experimental results demonstrate that the multi-agent system achieves a ۳۷% reduction in mean flow time, ۴۲% improvement in disruption recovery time, and ۲۸% increase in overall equipment effectiveness (OEE) compared to traditional hierarchical control systems. The framework's self-adaptive capabilities enable autonomous reconfiguration of production routes, dynamic resource allocation, and predictive maintenance scheduling without human intervention. Key contributions include a formal mathematical model for multi-agent coordination in production systems, a decentralized learning algorithm for adaptive policy generation, and design guidelines for implementing self-adaptive capabilities in legacy manufacturing environments. The proposed framework provides a pathway toward fully autonomous, resilient, and intelligent production systems for Industry ۴.۰ and beyond.

کلیدواژه ها:

Multi-Agent Systems (MAS) ، Artificial Intelligence (AI) ، Self-Adaptive Production Systems ، Reinforcement Learning ، Industry ۴.۰ ، Flexible Manufacturing ، Autonomous Control