AVEVA PI System is the historian behind many plants, refineries, water utilities and power stations. It records a constant stream of values from sensors and control systems, and the Asset Framework organises them into equipment hierarchies that engineers recognise. The data is all there. Getting a straight answer from it usually means building a PI Vision display, writing a spreadsheet query or finding the one engineer who knows the tag naming scheme.
Agentic CMMS connects your PI System to Claude, ChatGPT and other AI agents. Reliability engineers, planners and supervisors ask questions in plain words, and the agent finds the right elements, attributes and time ranges for them. The answer comes back with numbers and context, not a request for another report.
What your AI agent can do with AVEVA PI System
- Read a trend in plain words. "How has the bearing temperature on Fan F-204 changed over the last 30 days?" The agent finds the attribute in the Asset Framework and summarises the trend with its minimum, maximum and average.
- Compare equipment side by side. "Which of our six cooling water pumps ran hottest last week?" The agent reads the same attribute across sibling elements and ranks them.
- Review event frames. "List the downtime event frames on Kiln 2 this month and how long each one lasted."
- Check a value at a moment in time. "What was the suction pressure on P-101 when it tripped at 03:40 on Tuesday?" The agent reads the value at that time stamp and the minutes before it.
- Find the right tag. "Which PI points belong to the boiler feed pumps?" The agent walks the element hierarchy and lists the attributes with their units of measure.
- Spot drift early. "Show me any compressors whose vibration average this week is 20% above their three-month average."
- Summarise a shift. "Give me a short summary of the main process deviations on Line 4 during the night shift."
AVEVA PI System with your other systems
Process data tells you how an asset behaves. Your maintenance and business systems tell you what was done about it. Ask across them:
- When vibration on a fan passes its limit in PI, have the agent open a work order in Fiix or IBM Maximo with the trend summary in the description.
- Check whether the temperature rise on a gearbox started after the last repair recorded in SAP Plant Maintenance (PM).
- Compare running hours from PI with the operating limits in the equipment datasheet stored in Microsoft SharePoint, and flag anything outside them.
- Combine PI data from your older lines with tag history from Ignition by Inductive Automation on the newer ones, in one answer.
Process data with engineering discipline
Historian data is easy to misread. A flat line can be a stopped pump or a broken data link, and an average can hide a short spike that matters. The agent states the time range, units and summary method it used, so your engineers can check its working before they act on it.
Access stays deliberate. Each person approves their own agent, owners and admins decide which roles may query PI and which may not, and every query lands in the tamper-evident activity log.