Ai-Butler PM (Preventive Maintenance)

A closed-loop preventive maintenance AI platform - detect, recommend, act - built on SCADA/AVEVA data

Preventive maintenance, advanced through automated prediction, recommendation and action.

Built on the equipment data collected by PI System (AVEVA),
bringing together anomaly detection, feature extraction, decision support and AI planning

Anomaly
Detection

Feature
Store

Decision
Support

AI
Planning

3D Ops
Dashboard

The practical limits of conventional PM

• Too few maintenance staff and hours for the number of assets (tens to hundreds of thousands) → PM completion and quality suffer

• A gap between the data and the floor leaves inspection results unreliable, raising the risk of breakdowns, defects and safety incidents

• Plenty of alarms but little explanation of cause → false and missed alerts, delayed response

• Reacting after the fact, so losses are hard to recover

How Ai-Butler PM works (AVEVA PI + AI PM closed loop)

How Ai-Butler PM works: the AVEVA PI and AI PM closed loop

Value for small and mid-sized manufacturers (End-Customer Benefits)

Safety and uptime

Fewer unplanned stops, higher MTBF, and early detection of the signs of failure to reduce the risk of incidents and losses

Cost and operating efficiency

A shift from time-based to condition-based maintenance (CBM), with work orders and reports automated (linked to SAP/CMMS)

Acceptance and visibility

The wall-size 3D operations dashboard is a highlight of any VIP tour - a genuine wow factor

Quick Matrix: what sets it apart (against conventional maintenance and alarms)

Item	Conventional practice	Ai-Butler PM
Detection	Mostly thresholds and rules	Rules + ML + DL ensemble Anomaly score with evidence
Analysis	Little explanation of cause	LLM + knowledge base Decision support that explains
Action	Carried out by hand	AI planning Sequences the follow-up work
Operation	Alarm fatigue and misses	Better quality (fewer false alerts) KPIs and governance

Item

Conventional practice

Ai-Butler PM

Detection

Mostly thresholds and rules

Rules + ML + DL ensemble
Anomaly score with evidence

Analysis

Little explanation of cause

LLM + knowledge base
Decision support that explains

Action

Carried out by hand

AI planning
Sequences the follow-up work

Operation

Alarm fatigue and misses

Better quality (fewer false alerts)
KPIs and governance