Rio Tinto Documents 30-Year-Old Manufacturing System Using AI

Why It Matters

The assistant transforms fragmented, undocumented legacy knowledge into actionable insight, accelerating innovation and lowering change‑risk for Rio Tinto’s critical aluminium production line. It demonstrates how heavy‑industry firms can modernize without costly system replacements.

Key Takeaways

• AI domain assistant maps 30‑year‑old Metpro MES dependencies
• Amazon Bedrock Knowledge Base and AgentCore power Rio Tinto’s solution
• Llama 3.1 8B fine‑tuned to capture operational logic
• Engineers resolve change impacts in minutes, not days
• Enables low‑risk modernization without replacing legacy system

Pulse Analysis

Legacy manufacturing execution systems (MES) like Rio Tinto’s Metpro are the backbone of heavy‑industry operations, yet they often suffer from decades of undocumented tweaks and scattered manuals. Over time, this knowledge erosion hampers onboarding, inflates maintenance costs, and creates hidden risks whenever a seemingly minor change is made. By recognizing that rewriting or replacing such entrenched software is impractical, Rio Tinto turned to generative AI to capture the tacit expertise embedded in Metpro’s code and process documentation.

The technical solution leverages Amazon Bedrock Knowledge Bases to ingest the Metpro code repository, design specifications, and operational procedures, creating a unified, searchable corpus. Using SageMaker AI JumpStart, Rio Tinto fine‑tuned a Llama 3.1 8B foundation model, granting it deep domain awareness of the system’s logic and decision pathways. An AgentCore layer continuously pulls fresh information from the knowledge base, ensuring the assistant stays current without constant re‑training. This hybrid of static fine‑tuning and dynamic agent retrieval bridges the gap between static model knowledge and the ever‑evolving reality of plant operations.

From a business perspective, the AI assistant reduces the time to assess change impact from days to minutes, freeing engineers to focus on process improvements rather than reverse‑engineering legacy code. The approach also mitigates risk, enabling incremental, low‑disruption modernization of critical infrastructure. For the broader mining and manufacturing sectors, Rio Tinto’s deployment showcases a scalable pathway to preserve and operationalize decades‑old system knowledge, turning legacy liabilities into strategic assets while avoiding the massive expense of full system replacement.