If the cryogenic pump fails, the cold chain breaks — and the shipping window goes down with it.

The problem

The bearing wears out slowly. The failure does not.

An LNG cryogenic pump — submerged or external — runs at −162 °C, lubricated by the fluid itself, with clearances in the order of hundredths of a millimetre. It is a brutally reliable component until it stops being one. And the path from "all OK" to "plant stoppage" is not abrupt: it is a vibration signature climbing over days or weeks, in bands the classic protection system does not look at until the overall threshold is crossed.

1. Bearing wear — a harmonic of the rolling element pass frequency appears. Catch it at this point and you change the bearing at the next scheduled shutdown.
2. Impeller clearance loss — the blade pass band rises. Catch it here and you plan the intervention without breaking the shipping window.
3. Two-phase cavitation — gas enters the column; random broadband energy. Catch it here and you adjust NPSH before the impeller suffers.
4. Terminal failure — protection trips, the pump stops, the regasifier loses flow, the shipping window collapses, and the cost is counted in hundreds of thousands.

The classic system — machine protection + maintenance rounds + a quarterly FFT analysis by an external expert — is designed for step 4. Steps 1, 2 and 3 happen silently between two rounds. The control room operator knows it; what they do not have is the permanent ear they need.

How it fits into the IRIS system

iLEAN Edge does not replace machine protection — it gives it the continuous ear between rounds.

The problem is not a missing transducer: in a serious LNG plant there are transducers on every pump. The problem is that the signal lives on an island — the protection rack, the FFT analysed every three months, the external expert's email that arrives late. iLEAN acts as the filler that closes that gap, without touching the SIL layer or the certified protection system.

Edge lives alongside machine protection and learns your pump's healthy signature. Connect crosses that signature with the demanded flow, the tank batch and the maintenance calendar. The agent prepares the file. The person decides and signs off.

The three iLEAN pieces applied to the LNG cryogenic pump:

• Edge — a terminal with a CNN trained to recognise the spectral signature of the four failure modes (bearing, clearance, two-phase flow, misalignment) over the signal from the proximity transducer or the external housing accelerometer. Local inference: if the plant loses the network, Edge keeps tracking how the anomalous pattern progresses and keeps raising alerts on the panel.
• Connect — captures the context: flow demanded by the regasifier, the tank batch being discharged, the pump's last maintenance noted on a maintenance manager's spreadsheet, the pump manufacturer's email with the technical note from the last incident in the fleet. All at second zero, without anyone forwarding anything.
• Agent — crosses the anomalous signature with the shipping window, the availability of redundant pumps and the estimated cost of diverting flow. It proposes the intervention that minimises stoppage risk: replacement at the next window, switch to pump B, NPSH adjustment. The person validates, signs off, and operations executes. And an evidence pack is left for the auditor.

See the full IRIS architecture →

Before and after

Round-based maintenance vs. anticipation through a continuous signature

Aspect	Classic cryogenic pump maintenance	With iLEAN Edge on the transducer
Vibration analysis	Quarterly FFT by external expert	Continuous signature recognition by local CNN
Bearing detection	When the overall threshold is crossed	At the first harmonic — days or weeks earlier
Response to an anomaly	Schedule a shutdown at the next window	Schedule the shutdown + divert flow with time to spare
Risk of unplanned stoppage	Accepted	Reduced — the signature warns before the trip
Operation without network	n/a	Edge keeps running as long as the cabinet has power
File for auditor / manufacturer	Rebuilt by hand	Per-event pack, automatic, with signature history

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data from your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Regasification or liquefaction plant with 3-6 cryogenic pumps at the head, 2-out-of-3 or 3-out-of-4 redundancy, shipping windows scheduled with the customer.
• Edge pilot on one critical pump (sensor + CNN trained on the healthy signature + alert to the control room). First value expected within a few weeks: baseline signature recorded, first useful alert of an anomalous pattern in a specific band.
• Expected reduction in unplanned stoppages per pump of ≥ 30% against the baseline of recent years.
• Indicative payback between 4 and 9 months, depending on the historical frequency of unplanned events and the average cost of a regasification stoppage over one window.
• The hard lever is a single unplanned stoppage avoided: shipping window saved, flow sustained, contractual clause not triggered. One alone pays for the pilot.

And the maintenance manager's reasonable doubt

"What if the CNN gets it wrong and tells us to stop a pump that was healthy?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI limits itself to recognising a signature over the sensor's real signal, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: Edge proposes, the control room technician signs off, SIL protection stays where it has to be. The three safety rings are there for exactly this.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.