Maintenance Intelligence for
Hydrocarbon Processing Plants.

KeplrAI predicts asset failures, guides maintenance teams, and coordinates parts readiness before downtime happens.

The thesis

The plants that process hydrocarbons are drowning in sensor data.

Every critical asset streams pressure, temperature, vibration, and flow data into historians nobody can fully leverage. Plants still run maintenance on calendars and tribal knowledge. The result: over $1M per day in unplanned downtime, parts that arrive weeks after failures are predicted, and institutional knowledge that walks out the door when a senior reliability engineer retires.

We believe every asset should have its own brain. Not a plant-wide black-box model. Asset-level reasoning: the compressor learns what a healthy compressor looks like. The exchanger learns its own fouling signature. Every wrench turn makes the model sharper, and that intelligence compounds across the fleet.

The closed loop

From raw signal to sharper model, start to finish.

One continuous loop per asset. Field data flows back into the model. Every repair makes the next prediction sharper.

01 / 06
INGEST
Live
DCSOPC UA
SCADAModbus
OSIsoft PIPI Web API
Aspen IP.21SQLplus
Honeywell PHDODBC
KeplrAI
Asset Intelligence
Read-only
TLS 1.3
On-prem
Air-gap
5 historians · 1,247 tags ingested● synced
Step 01

Connect

Plug into the plant's existing DCS, SCADA, PLCs, and historians (PI, Aspen IP.21, Honeywell PHD). Read-only integration. No rip-and-replace.

02 / 06
TRAIN
Live
Asset · K-101
Centrifugal Compressor
Fitted
Vibration
mm/s
Pressure
barg
Temperature
°C
Physics-informed · per-asset baseline● learning
Step 02

Model per asset

Every critical asset gets a physics-informed model trained on its specific operating conditions and failure modes. Not generic anomaly detection. Bearing wear, fouling signatures, valve degradation — each detected with engineering-grade reasoning.

03 / 06
FORECAST
Live
Asset · K-101 · Health Index
Time to failure
Estimate
25d 6h
Failure mode
Bearing wear
Confidence
92%
Severity
Critical
Step 03

Predict

Assets flag degradation early with time-to-failure estimates, confidence intervals, and supporting evidence. Your reliability team gets a prioritized view, not a stream of alerts.

04 / 06
ORDER
Live
Predicted failure · K-101
Parts coordination
T-18d
Part identified
Mechanical seal · SKU-44918
Bearing N-3 · K-101
Inventory checked
0 on hand
Not in stock
Lead time · 22 days · Vendor: John Crane
Procurement order
Auto-approved
PO-2486 issued
Delivery · Apr 14 · Cost center 03-MAINT
Triggered 14h after prediction● order placed
Step 04

Coordinate parts

Predictions auto-check inventory and trigger procurement workflows in your existing procurement system. Long-lead items surfaced weeks before failure.

05 / 06
EXECUTE
Live
Maintenance window
Apr · Optimal slot
WO-1138
M
T
W
T
F
1
2
3
4
5
6
7
8
WO
9
10
11
12
13
14
15
Work order
Apr 16 · 06:00
K-101 · Mech seal replacement
  1. 1.Isolate K-101 · LOTO
  2. 2.Drain lube oil reservoir
  3. 3.Replace mech seal SKU-44918
  4. 4.Recommission · verify vib < 2.8 mm/s
Dispatched · J. Ortega · Field App● synced to CMMS
Step 05

Schedule and execute

Work orders flow to your existing CMMS at the optimal window. Technicians get asset history, predicted failure mode, and repair procedure on their mobile device.

06 / 06
FEEDBACK
Live
Asset library
Predicted vs Actual
Assets learning
1,246
+1 reconciled
MAE ↓ 0.4d · sharpened● continuous
Step 06

Learn

Technicians log what they actually found. Field data flows back into the asset's model. Every repair makes the next prediction sharper. The asset library compounds across every plant we deploy.

Measured outcomes
Downtime
0%
Reduction in unplanned downtime
Utilization
0%
Improvement in asset utilization
Incidents
0%
Faster incident resolution

Tell us what you operate.