Agentic AI on Operational Ontology Layer for Autonomous Manufacturing Intelligence

Semantic ontology orchestrates specialized AI agents to transform fragmented manufacturing operations into autonomous, proactive operational intelligence

Paul Kim

Merck Electronics

Specialty chemical and semiconductor manufacturing often suffers from siloed data across ERP, LIMS, and process historians. Investigating quality or process excursions require engineers to manually navigate disconnected systems, permissions, and cross-functional support - an inefficient workflow that delays root cause identification and allows anomalies to persist.  

To address this challenge, we developed an ontology-based semantic data layer that unifies these heterogeneous data sources into a single, AI-queryable knowledge structure. ERP batch and inventory data are semantically linked with LIMS quality measurements and historical trends. Process monitoring results are also connected, powered by batch-level and real-time multivariate SPC (M-SPC) models that evaluate equipment state from sensor data. These M-SPC models compress high-dimensional sensor signals into a small set of monitoring statistics, enabling near-real-time detection of abnormal equipment states and early process anomalies. This architecture enables AI systems to directly access and traverse complex relationships across the entire manufacturing data landscape without manual data retrieval.

On top of this ontology layer, we deployed a multi-agent AI architecture to overcome the context limitations of LLMs. Each specialized agent focuses on a specific domain such as quality, process, or operations, allowing it to maintain deep expertise within its manageable context scope. These agents collaborate with an orchestrating agent that synthesizes findings across domains and delivers proactive alerts and weekly operational summaries without human intervention. Deployed at manufacturing sites in South Korea and Taiwan, this system enables a single engineer to obtain comprehensive insights in minutes instead of hours, while providing 24/7 autonomous monitoring. By enabling faster detection and resolution of process deviations, this system ultimately ensures that consistently well-managed, high-quality products are delivered to customers.  

By bridging data silos with semantic ontology and orchestrated AI agents, this approach establishes a scalable pathway toward autonomous manufacturing intelligence - driving efficiency, adaptive process control, and resilience. For semiconductor industry where precision and speed are critical, it represents not just a technical advance but an enabler of future-ready, competitive operations.

Key Technologies Covered

• Ontology-based semantic data integration for manufacturing
• Agentic AI orchestration for autonomous data investigation
• Real-time and batch-level multivariate statistical process control (MSPC)
• Autonomous operational reporting and proactive alerting