Real-Time Predictive Maintenance with Edge-Orchestrated Multi-Agent Pipelines

Emerging technologies such as artificial intelligence, cloud-based infrastructures, and edge computing offer highly scalable solutions in modern manufacturing applications. This paper presents an innovative Machine Learning Function Orchestrator that seamlessly integrates with edge computing and multi-agent systems. Machine Learning Function Orchestrator features two key components: MAS pipeline manager and service management system. The MAS pipeline manager is responsible for the coordination and deployment of multi-agent systems. The  Service Management component gathers data from MAS agents and deploys performance supervisor service for each MAS pipeline. These components work together to dynamically deploy, configure, and manage Multi-Agent System (MAS) pipelines to enable efficient coordination and real-time control of distributed multi-agents. Machine Learning Function Orchestrator integrated with Open Horizon, an open-source decentralized edge computing framework is used to facilitate the deployment and reconfiguration of Multi-Agent System pipelines. The framework is demonstrated in a predictive maintenance use case using visual inspection with multiagent-based control. The edge agents keep track of equipment status, identify anomalies, and forecast likely failures, allowing agents to perform autonomously on intervention planning and optimization with minimal impact on the production. The proposed framework provides a robust foundation for flexible, high-performance manufacturing environments, enabling smart factories to achieve superior adaptability, resilience, and operational efficiency.