Every LNG operations team knows the gap between what a liquefaction train was designed to produce and what it delivers in a given year. That gap costs margin, and its causes are rarely confined to one section.
Across the industry, the shortfall is measurable. According to the IEA, global LNG supply grew just 2.5% in 2024, well below its 8% average from 2016 to 2020, with feed gas constraints and production issues at legacy plants among the factors holding output back.
Inside the train, output depends on how constraints interact across sections. The visible bottleneck often appears at the cold end, while the margin was used up upstream.
How the train's anatomy shapes output
An LNG train is a sequence of interdependent sections. A constraint in one section propagates downstream, often without an obvious signal at first.
Feed gas enters through a slug catcher for bulk liquid removal. Acid gas removal then strips CO₂ and H₂S to specification. Molecular sieve dehydration dries the gas to below 1 ppmv water. Moisture breakthrough threatens the main cryogenic heat exchanger (MCHE) directly, because ice reduces heat transfer and can block passages.
Regeneration gas from the sieves may be recycled to the molecular sieve inlet or to the acid gas removal unit (AGRU) inlet, depending on the configuration. That recycle couples the two sections and affects amine load. AGRU performance in turn sets the volume and composition of the regeneration gas, which affects sieve bed cycle times and available dehydration capacity. When the amine system runs near its design limit, a modest change in regeneration gas composition can shift the loading profile enough to reduce dehydration capacity.
How upstream conditions reach the cold section
Heavy hydrocarbon removal, usually in a scrub column, keeps heavier components from freezing in the cold section. Scrub column separation changes when feed composition moves away from the design basis, and feed that is leaner than design can still increase heavy hydrocarbon slip to the MCHE. These changes are not always visible in the control room until the cold section feels them.
The precooling circuit then lowers gas temperature before the MCHE completes liquefaction and subcooling. By that point, the cold section is carrying the consequences of upstream conditions. An operator watching the MCHE temperature approach widen may not connect it to a small shift in AGRU loading three sections upstream, but the link is real and is often the root cause of the lost margin.
What governs MCHE output
MCHE performance is the most critical variable in an LNG train. The exchanger responds at once to ambient conditions, upstream gas quality and the state of the refrigerant system, so its behavior drives the economics of the whole train.
Operating data from Sakhalin LNG, published in Gas Processing & LNG, gives a useful diagnostic framework. LNG production downstream of the MCHE moves with four variables:
- gas temperature at the MCHE inlet, from the precooling circuit
- LNG temperature at the outlet, reflected in the rundown setpoint
- available mixed refrigerant (MR) compressor driver power
- the MR cycle coefficient of performance (COP)
When output falls below target, one or more of these usually explains it. MR cycle COP deserves close attention because refrigerant composition directly affects how efficiently the cycle turns driver power into cooling duty. The mixture of nitrogen, methane, ethane, propane and butanes is chosen to match the natural gas heat composite curve inside the MCHE closely. When composition drifts, the cycle works harder for the same output.
Warm-end and cold-end temperature approaches inside the MCHE are useful indicators. A wider approach at either end points to reduced heat transfer effectiveness from fouling, off-optimum MR composition or flow distribution problems.
Nameplate capacity reflects a design basis: a reference ambient temperature, feed gas composition and equipment condition. It is not a production target that holds all year. The same train can run close to design in one period and lose margin when hotter weather, fouling or feed changes tighten several constraints at once.
The ambient, fouling and feed cascade
The most persistent throughput constraint is the interaction of ambient temperature, equipment condition and feed variability. According to LNG Industry, industrial gas turbine output falls by roughly 0.7% per °C above design ambient. At the same time, higher ambient temperature raises refrigerant condensing pressure and compressor power demand. Available power falls while required power rises.
Air cooler fouling adds to this. As fins foul, condensing temperature rises above what ambient temperature alone would cause, and compressor demand climbs in the same direction as on a hotter day.
Feed gas variability works through a different mechanism. In some operations, actual feed acid gas content has exceeded the design basis, which limited AGRU capacity and constrained throughput even when the liquefaction section had room. Over longer periods, reservoir depletion can produce leaner feed that shifts scrub column behavior and propane constraints.
These conditions often arrive together. The hardest operating periods are the ones in which ambient derating, fouling and feed changes have all tightened the operating room at once.
Where single-loop control loses the full train picture
Conventional regulatory control manages individual loops, with one PID controller holding one temperature, pressure or flow. In an LNG train, scrub column control affects both fractionation and liquefaction, so separate loops can miss how the whole train is moving.
MR composition optimization shows the limit clearly. Finding the best composition requires an invariant performance indicator, such as the Carnot factor, rather than isolated temperatures or compressor load. Direct composition control is impractical because analyzer latency and the interdependence of components slow the feedback path. At Sakhalin LNG, the operations team used indirect control through calculated inventories instead.
How loop interactions consume margin
Even when each controller performs well, the interactions between them can still consume margin. A temperature controller on the precooling circuit and a pressure controller on the MR compressor may each hold setpoint while their combined effect on MCHE performance no longer suits current conditions. When ambient temperature or feed quality changes, setpoints that were right an hour earlier can become suboptimal without either loop signaling a deviation. That coordination gap is where throughput improvement opportunities tend to concentrate.
Coordinated changes to compressor load, MR composition and temperature profile can recover capacity that single-loop operation leaves unused. Sustaining those gains requires tracking how multiple constraints shift with ambient conditions, feed changes and equipment degradation. That tracking works best when the model learns from the plant's own operating history, because feed, ambient conditions and equipment condition all move away from the design case.
Aligning planning, operations, maintenance and engineering
The coordination problem extends beyond the control room. Maintenance timing affects refrigerant efficiency, and production targets sometimes rest on planning assumptions that no longer match equipment health. When planning, operations, maintenance and engineering each work from a different view of the train, those conflicts stay hidden until throughput drops.
LNG plants typically run with lean crews. Even experienced operators cannot continuously reoptimize MR composition, compressor loading and temperature profiles while also handling alarms and routine work. A learned model can track more constraint interactions at once, though it cannot replicate every judgment experienced operators make about equipment condition and safety margins.
Building trust in steps
Advisory mode is a practical start. The model recommends setpoint changes and operators decide whether to apply them. Newer operators get a clearer view of how coupled constraints are moving, and experienced operators can compare recommendations with their own reading of the unit. The same model supports more consistent decisions across shifts, shows what-if trade-offs before changes are committed, and tracks how fouling or feed changes are tightening the train over time.
When all four functions work from one model built on the plant's own data, they can evaluate throughput targets, maintenance timing and realistic operating margin from a common understanding of where the train stands.
From advisory, plants move to operator-approved execution within operator-defined limits, and then to closed loop, with operators able to take over at any time and keeping authority over the operating envelope. Each step is earned through operating results. For many facilities, the margin available from better coordination across these functions can be larger than what any single control loop improvement delivers.
How the Imubit Platform applies
The Imubit Platform applies Map, Model and Control across the LNG train as it does across refining and gas processing. Map traces the gap between nameplate and actual output to the recurring decisions behind it, from AGRU and dehydration targets to MR composition and compressor loading. Model uses Deep Learning Process Control® (DLPC®) to learn the train's nonlinear behavior from its own history, guided by first principles. Control takes those decisions to closed loop, writing setpoints to the existing APC and DCS minute by minute, inside limits set by plant engineers.
