Predictive Maintenance Platform
Time-series + vision models predicting equipment failure 72 hours in advance across 14 plants.
- Client
- Vertex Manufacturing
- Year
- 2024
- Duration
- 16 weeks
- Team
- 5 engineers, 1 reliability SME
Vertex was losing an estimated $42M a year to unplanned downtime across 14 plants. The telemetry to predict most of those failures already existed in their historian — it had simply never been modeled, because labels were sparse and an earlier vendor pilot had produced so many false alarms that the floor stopped trusting alerts entirely.
The brief
Predict equipment failure with enough lead time to schedule maintenance, at a false-positive rate low enough that operators will actually act on the alert.
- Manufacturing
- Forecasting
- Vision
The decisions that mattered.
Every engagement turns on a handful of calls. These are the ones that decided whether this system reached production.
Precision was the product requirement
We tuned explicitly against false positives after learning the previous pilot ran near 40% and had been muted plant-wide. Landing at 9% is what made the system credible on the floor.
Labels reconstructed from maintenance records
Failure labels did not exist in usable form. We derived them from work orders and operator logs, then had Vertex's reliability engineers validate every label before training.
Multimodal beat any single signal
SCADA telemetry alone plateaued. Adding vibration spectra and floor-camera vision pushed the usable lead time out to 72 hours across the main asset classes.
Inference at the edge, inside the OT boundary
Vertex's OT network is air-gapped from IT. Models run on edge hardware inside the plant, with only aggregated predictions crossing the boundary.
What shipped, and what changed.
Outcomes
- Unplanned downtime reduced 34% across the monitored lines
- False-positive rate held at 9% through the first full year
- OEE improved by 6 percentage points
- 47 production lines instrumented across 14 plants
“The last vendor gave us a model. This team gave us something the floor supervisors actually check before they schedule a shutdown.”
What we built it with.
- PyTorch
- Python
- Kafka
- TimescaleDB
- Docker
- Azure
Other projects from the shelf.
Clinical Documentation Copilot
Ambient AI scribe that drafts structured notes, ICD-10 codes, and prior-auth letters in real time.
Read moreEquity Research Terminal
Multi-modal RAG over 10 years of filings, transcripts, and broker research with citation-grounded answers.
Read moreAI Concierge for Retail
End-to-end conversational agent resolving 62% of post-purchase requests without human escalation.
Read moreTell us about the problem.
A 30-minute call with the senior team. We will tell you what is realistic, what it would cost, and whether we are the right people for it.