Unplanned shutdowns down 45%, inspection record completeness at 99%

Every unplanned shutdown of a continuous process unit cost several million yuan. Inspection records were on paper and often incomplete, equipment problems were usually noticed only when an alarm fired, and the experience from maintenance work orders was never captured, so the same failures kept recurring.

The company wanted to combine inspection data with DCS data to catch anomalies early, and to keep its maintenance experience.

Digitize inspections first, then connect prediction and maintenance knowledge, rolled out in three phases.

• Inspection Agent: inspectors record by voice or photo, entries are structured automatically and linked to the equipment register, and missed checks trigger reminders
• Anomaly warning: DCS trends combined with inspection records give an anomaly probability and recommended action for critical rotating equipment and heat exchangers
• Maintenance assistant: when a work order is raised, past solutions for the same failure type and the parts list are pushed automatically
• Every shutdown review is archived automatically, building a unit-level failure knowledge base

Industrial data is synced one way through an isolation gateway to the analytics environment. The time-series anomaly detection model and the LLM have separate jobs: the former produces the warning signal, the latter explains it and recommends action. Field terminals meet explosion-proof requirements.