Industrial Operations - Maintenance AI & Inspection OCR Implementation
A major chemical storage and logistics operator across Australasia partnered with ELab to address critical maintenance coordination challenges following a merger. Over 10 months, we deployed a two-stage AI inspection system that achieved 95.5% accuracy, identified 39 defect references including critical safety hazards that manual processes had missed, and reduced form processing time from 40-60 minutes to under 3 minutes.
Industry: Chemical Storage & Logistics | Duration: 10 months | Region: Australia / New Zealand
Following a merger, this organisation faced significant operational integration challenges. Maintenance coordinators were spending 71-97 hours monthly processing work orders, with only 23-29% of that time on value-added activity. Processing delays of up to 2 months at some sites were creating safety and compliance exposure. And 30-40% of follow-on work orders were duplicates, cluttering the maintenance management system.
The core problem: paper inspection forms had to pass through multiple system handoffs before reaching the maintenance database, with manual data entry consuming 40+ hours per month per coordinator.
What We Did
Process Analysis & Foundations
We began with stakeholder interviews and a Lean Six Sigma process analysis that quantified the waste across the operation - identifying that the majority of coordinator time was non-value-added. This gave us clear priorities for where AI could have the most impact.
Maintenance Knowledge Base
We built a comprehensive knowledge base from existing maintenance documentation and deployed an AI assistant integrated into Microsoft Teams, giving coordinators natural language access to procedures and standards.
Inspection OCR & AI Processing
The centrepiece of the engagement: a two-stage AI system deployed within Microsoft Teams that transforms paper inspection forms into structured, system-ready data.
- Stage 1 receives uploaded forms, performs OCR on handwritten and printed data, extracts metadata, and auto-organises documents by site and asset.
- Stage 2 reads the inspection comprehensively, extracts all defects with criticality triage (P1-P4), validates readings against normal ranges, detects contradictions, and generates structured output ready for the maintenance management system.
Total processing time: 90-150 seconds per form, down from 40-60 minutes manually.
Results
| Metric | Result |
|---|---|
| Processing accuracy | 95.5% context correctness |
| Safety defects identified | 39 references across 4 sites |
| Processing time per form | 90-150 seconds (was 40-60 minutes) |
| Coordinator time (projected) | 60-70% reduction |
| Duplicate work orders (projected) | Reduced from 30-40% to under 5% |
Critical Safety Findings
The AI identified safety issues that manual processes had missed, including a missing earth strap creating explosion risk in chemical storage, a cluster of radar failures suggesting a systemic capital equipment issue rather than isolated events, and multiple active leaks requiring immediate action.
This cross-site pattern detection - spotting that multiple radar failures across different assets indicated a capital equipment problem - is the kind of intelligence that only emerges with systematic data extraction and correlation.
ELab continues to support through managed AI services. The next phase targets direct integration with the maintenance management system to eliminate the remaining copy-paste step, followed by digital forms at source to replace paper entirely.
