Base oil control with AI — an off-spec batch is several m³ that get downgraded.

The problem

A base oil is not judged by eye — and a downgraded batch is thousands of liters lost.

Base oil control lives in three realities that almost never come together in time:

1. The chemistry of the cut — kinematic viscosity at 40 ºC and 100 ºC, viscosity index (VI), saturates and sulfur content. It determines whether the base oil is Group I, II or III under API 1509. The premium gap between groups is real and worth money per liter.
2. The blend — the recipe that combines cuts to hit the customer's target viscosity (SAE 0W-20, ISO VG 32, and so on). It lives in the DCS and in the ERP, and often in a spreadsheet kept by the blending manager.
3. The customer specification — the finished-lubricant formulator, the OEM, the export buyer. Each one has its minimum CoA, its tolerance and its rejection criteria. It arrives by email, sometimes as a WhatsApp message from the sales rep.

The blending manager knows it — and knows that a batch of several m³ that drifts off the target VI is a downgrade the plant pays for. Group III dropped to Group II is several euros per liter multiplied by thousands of liters. The buyer does not negotiate the CoA. The classic system works — and when it fails, it fails by quarters of margin.

How it fits the IRIS system

iLEAN does not replace the viscometer or the DCS — it seals the cracks between what the plant measures and what the customer asks for.

The problem with base oil control is not a lack of measurements: it is measurements that reach the point of decision too late. iLEAN acts as the putty that seals those cracks, without touching the DCS or the viscometer, respecting OT isolation through the three rings.

Edge sees the viscosity before dispatch. Connect reads the recipe, the DCS and the LIMS wherever they live. The Agents cross-reference it with the customer's spec. The person signs — never the other way round.

The three iLEAN pieces applied to base oil control:

• Edge — a terminal that captures the in-line viscometer (Stabinger, vibrating element) through a dry contact or a panel reading. It reads viscosity and the calculated VI in real time, triggers a light stack and holds the batch in the intermediate tank if it falls outside the target SAE/ISO window. It works with no network.
• Connect — captures the blending curve from the DCS through a passive mailbox (without touching the PLC). Captures saturates and sulfur results from the LIMS. And it captures what comes from outside: a customer specification that arrived by email, an OEM tolerance change, an API/ATIEL certificate. At second zero.
• Agents — they live in ring 3. They cross-reference measured viscosity with the planned recipe, the history of the upstream cut and the customer specification. They anticipate drift within hours. They keep the per-batch CoA current, ready for the buyer and for the API audit.

See the full IRIS architecture →

Before and after

Manual blending + slow LIMS vs. cross-referenced blending with iLEAN

Aspect	Blending with DCS + LIMS + spreadsheets	With iLEAN Edge + Connect + Agents
Reading the in-line viscometer	The instrument's local display	Integrated into the cross-check with recipes and customer spec
Detecting a batch out of VI	In the LIMS, hours after the blend is closed	At the viscometer, before it is closed
Updated customer spec	An email in the sales rep's inbox	In the system at second zero
Per-batch CoA for the buyer	Assembled by hand before dispatch	Maintained automatically, signed by the person
API/ATIEL audit	Rebuilt by hand	Dossier available when the batch closes
Batch downgraded from III to II	Discovered after the buyer rejects it	Anticipated, with the recipe adjusted before closing

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 of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Base oil plant with hydrocracking + dewaxing, Group II and Group III production, blending for several OEM formulators.
• Edge pilot over the in-line viscometer + Connect + a blending Agent covering one group (Group III, where the premium is higher). First value within a few weeks.
• Indicative payback between 4 and 9 months, depending on how often batches have been downgraded in recent years and on the price gap between Group III and Group II.
• Expected reduction in downgraded batches of ≥ 30% (a conservative estimate, to be validated).
• The hard lever: a single multi-m³ batch rescued from a downgrade pays for the pilot.

And the blending manager's reasonable doubt

“What if the agent changes my recipe without me seeing it?” — it cannot. The Agents propose recipe adjustments; they do not execute them in the DCS. The signature on the adjustment still belongs to the blending manager. And even for that proposal, reliability on anchored tasks (cross-referencing recipe, viscosity and spec) is below 1.5%^[1]. The AI here does not invent a new recipe — it proposes an adjustment based on the history of the cut and on the customer's active spec. The person sees the reasoning and signs.

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

Related topics you may find useful: biofuel control · asphalt and bitumen control · refinery cracking control.