Hitachi AI Agent for Manufacturing Equipment Failure Diagnosis (RAR + STAMP/CAST)

Hitachi, Ltd. developed an AI agent for manufacturing equipment failure diagnosis that uses a novel Retrieval Augmented Reasoning (RAR) approach combined with STAMP/CAST system safety analysis methodology. Core innovation: Rather than simple RAG (searching past records), RAR combines equipment structural knowledge graphs with expert diagnostic logic to handle unknown/new failures. Key technical components: Converts P&ID and electrical drawings into knowledge graphs that AI can understand Uses Bill of Materials (BOM) format to represent equipment structure Trained on STAMP/CAST (System-Theoretic Accident Model & Process) methodology for root cause analysis RAR paradigm: retrieves appropriate drawings/SOPs while combining equipment structure with expert analytical skills Joint demonstration with Daikin Industries: Trial operation started April 2025 Accuracy improved from 67% (general-purpose AI) to over 90% Response time: ~10 seconds per inquiry Field evaluation: experienced maintenance staff rated it "equivalent to or better than a typical maintenance technician" Also started joint verification with Mitsubishi Chemical Corporation end of 2025 The system runs on Hitachi's cloud platform and is part of digital service HMAX™ Industry (Lumada 3.0).