Sri Elaprolu’s Post

🧵 Real Stories of Generative AI in Action (Feature 2 of a multi-part series, you can access full series at #AWSGenAIinAction) 🏗️ Most people don't give much thought day-to day on how the materials that support our daily lives are delivered. Thankfully Rio Tinto does! Rio Tinto is a leading multinational corporation in the metals and mining sector that plays a vital role in supplying materials critical for modern infrastructure and technology. The AWS Generative AI Innovation Center (#GenAIIC) recently worked with Rio Tinto to develop a solution supporting the ongoing maintenance and operations of their assets -- basically making sure that Rio Tinto can keep us driving, riding, cycling, and living. In "industry" terms, the solution addresses asset fault classification and defect elimination. ❓ What does that even mean? This is the process by which the health data of physical assets (e.g. equipment, infrastructure) is assessed to determine a problematic asset, and subsequently put it in line for corrective action. Sounds simple enough, but actually it's quite complex (and ripe for innovation). Faults have to be classified at the most granular level (e.g. which individual part in the asset, what is the specific nature of the defect) by experienced reliability engineers who manually review and reconcile each fault, referencing disparate systems and documentation. What's more, there are limited reliability engineers with the specialist knowledge required to do the work! 💡 What's the solution? Together with AWS, Rio Tinto developed #ReconAI. Built on Amazon Bedrock with Anthropic Claude models and leveraging foundational services like Amazon EC2, Amazon S3, Amazon Athena, and Amazon Glue, ReconAI uses a RAG-based approach to describe the failed equipment, retrieving relevant information from knowledge bases to identify the failed equipment and failure mode. Multi-agent collaboration for reflection and re-ranking algorithms provide further context-relevant improvements to enhance the existing algorithm for fault classification. 💪 The results: Initially tested on Rio Tinto's locomotive fleet, the solution achieved 80% fault classification accuracy (a 10% improvement), but has since been improved to an impressive 96%. Even better, manual effort to update classifications as new maintenance becomes available is reduced by 80%, and engineers can focus their time on delivering engineering projects to eliminate high-value fleet-wide defects, instead of manually classifying failure events. This is a great example of how generative AI can transform a "traditional" industry and enhance operational resilience through innovative technology. Learn more from Rio Tinto's recent presentation at AWS Summit Sydney: https://shorturl.at/28E7o #AWS #GenerativeAI #Innovation #IndustrialTech #Mining #Infrastructure #DigitalTransformation #Bedrock #AWSGenAIinAction