A gas processing plant burns fuel gas continuously to run three energy-hungry units at once — the amine reboiler stripping acid gas, the molecular sieve beds regenerating on a fixed schedule, and the fractionation train's reboilers holding product purity through every swing in feed composition. Each unit is usually tuned once, conservatively, and left alone, because nobody wants to be the operator who dialed back reboiler duty and shipped an off-spec cargo. That conservatism has a cost: most plants are running meaningfully more steam and fuel gas than the actual gas composition on a given day requires. This case study walks through a representative 200 MMscfd gas processing facility — spanning amine treating, dehydration, and NGL fractionation — where continuous AI optimization across all three units delivered a 12% reduction in total plant energy consumption and roughly $4.1 million in annual savings, without loosening product specification. The breakdown below shows exactly where that number came from, unit by unit, and what it took to get there — reach out through support to see how the same approach maps onto your own facility.
12% Plant-Wide Energy Reduction, $4.1M in Annual Savings
AI optimization across amine treating, dehydration, and NGL fractionation at a 200 MMscfd gas processing facility cut fuel gas and steam consumption without touching product specification.
The Facility
A 200 MMscfd inlet gas processing facility running MDEA amine treating for acid gas removal, molecular sieve dehydration ahead of a cryogenic turboexpander, and a full NGL fractionation train separating ethane through natural gasoline. The plant had been in stable operation for years, with each unit tuned to a conservative fixed setpoint that had not been revisited since original commissioning.
Where the Energy Was Actually Going
Before optimization, a baseline energy audit traced fuel gas and steam consumption to three specific, addressable patterns — none of them equipment failures, all of them setpoints that had never been revisited against real-time gas composition.
Amine Treating — Fixed Circulation Rate
Lean amine circulation ran at a single conservative rate around the clock, sized for the plant's highest historical acid gas loading rather than the loading actually present on a given day. The reboiler was frequently working harder than the treated gas specification required.
Dehydration — Regeneration on a Timer, Not a Trigger
Molecular sieve beds regenerated on a fixed cycle schedule regardless of actual moisture breakthrough, which meant regeneration gas was frequently burned to dry beds that were nowhere near their moisture capacity.
Fractionation — Reflux Ratios Sized for Worst-Case Feed
Depropanizer and debutanizer reflux ratios were fixed to hold product purity through the widest historical swing in feed composition, which meant most days ran with far more reboiler steam than the actual feed required to hit spec.
See This Kind of Audit Run Against Your Own Plant
Every gas processing facility carries some version of these fixed-setpoint patterns. A baseline energy audit against your historian data shows exactly where they show up in your operation, before any optimization is deployed.
What Changed, Unit by Unit
The AI model replaced each fixed setpoint with a continuously recalculated one, driven by real-time gas composition, moisture loading, and feed analysis rather than the worst case the plant was originally designed around.
Results: Where the 12% Came From
The total plant-wide reduction is the sum of three separate, unit-level improvements — no single change accounts for the full number, which is exactly why optimizing only one unit in isolation tends to leave most of the available savings on the table.
Contributed roughly 5 percentage points of the total plant-wide energy reduction and an estimated $1.7M of the annual savings.
Contributed roughly 3 percentage points of the total plant-wide energy reduction and an estimated $1.0M of the annual savings.
Contributed roughly 4 percentage points of the total plant-wide energy reduction and an estimated $1.4M of the annual savings.
Product specifications were held throughout — treated gas moisture content, H2S concentration, and NGL product purity all stayed within their original targets. The savings came from removing excess margin, not from relaxing the targets themselves.
How the Rollout Was Sequenced
The optimization was not switched on plant-wide on day one. It moved through a phased rollout designed to validate the model's recommendations before it was trusted to adjust live setpoints.
Baseline Audit & Data Connection
Historian data from all three units connected and analyzed to establish the true baseline energy consumption pattern against actual feed variability.
Model Training in Shadow Mode
The model ran alongside existing operations, generating setpoint recommendations without adjusting anything, so its logic could be validated against real outcomes.
Single-Unit Pilot
Live control handed to the model on the amine unit first, the unit with the clearest baseline inefficiency, while dehydration and fractionation stayed in shadow mode.
Full Rollout
Dehydration and fractionation moved to live optimization once the amine unit pilot confirmed stable performance against specification.
Continuous Optimization
Setpoints keep adjusting as feed composition, ambient conditions, and equipment condition change, rather than settling into a new fixed baseline.
What Made This Result Possible
None of these were exotic requirements. They were the specific decisions that separated a successful rollout from a stalled one.
Optimizing all three units together, since energy savings in one unit can shift load onto another if they are tuned in isolation.
Running the model in shadow mode first, so operations trusted its recommendations before it touched a live setpoint.
Holding product specification as a hard constraint throughout, not a target to be traded off against energy savings.
Piloting on the unit with the clearest baseline inefficiency first, to build confidence before expanding scope.
Treating optimization as continuous, not a one-time retuning that would drift back toward conservative setpoints over time.
Using existing historian and instrumentation data rather than waiting on new field hardware to start the baseline audit.
Frequently Asked Questions
Is a 12% energy reduction typical, or specific to this facility's baseline inefficiency?
The magnitude depends heavily on how conservative a plant's existing setpoints already are — a facility that has been retuned recently will see less headroom than one running on commissioning-era settings, which was the case here. Published engineering research on amine unit tuning alone documents reboiler duty reductions of up to 20% from solvent and circulation rate optimization, which lines up with the amine unit's contribution in this case. The most reliable way to know what a similar audit would find at your facility is to book a demo and run a baseline review against your own historian data.
How does the model avoid pushing a unit out of specification while chasing energy savings?
Product and process specifications are built into the model as hard constraints, not soft targets — treated gas moisture and H2S content, and NGL product purity, all have defined limits the model is not permitted to cross regardless of the energy savings on offer. The optimization search only considers setpoints inside that constrained space, which is also why shadow mode validation matters: it confirms the model respects those constraints against real operating data before it ever adjusts a live setpoint.
Does this require replacing our existing control system or DCS?
No. The model typically layers on top of the existing control system, reading historian and live instrumentation data and writing setpoint recommendations or adjustments through the same interfaces operations already uses, rather than replacing the DCS itself. This is part of why the rollout in this case study could move from baseline audit to live optimization in roughly twenty weeks instead of a multi-year control system replacement project.
What happens if feed gas composition shifts significantly, such as a new well pad coming online?
This is specifically what continuous optimization is built to handle — a fixed setpoint tuned for one feed composition profile is exactly the pattern that creates the inefficiency this case study addresses. As feed composition shifts, the model's recommendations shift with it, which is a meaningful advantage over a manual retuning that only happens when someone notices performance has drifted. Reach out through support to discuss how a feed composition change specific to your field would be handled.
How long until a similar optimization pays for itself?
Payback timing depends on the facility's baseline inefficiency and the scope of the rollout, but the phased approach used here — baseline audit, shadow mode validation, single-unit pilot, then full rollout — is designed so that savings begin accruing from the pilot unit well before the full plant-wide optimization is complete. That structure also means the investment case does not depend on a single, unvalidated leap to full deployment.
The Pattern Behind the Number
What made this result possible was not a single breakthrough on one unit — it was recognizing that amine treating, dehydration, and fractionation had all been tuned the same way: conservatively, once, for a worst case that rarely occurs. Continuous optimization did not change what any of these units are capable of. It changed how consistently each one operated close to its actual requirement instead of its historical worst-case margin, and it did that every day instead of at the next scheduled retuning.
Find Out What Your Plant's Number Looks Like
Book a 30-minute demo with an iFactory process engineer. Bring historian data from your amine, dehydration, or fractionation units and leave with a baseline estimate of where your energy savings potential sits.
