Air separation and cryogenics: wide margins, slow ramps, boil-off

Gas mixtures drift with temperature and pressure

What happens. Mixtures are filled by pressure or flow, and the components behave differently as cylinder temperature and pressure change during filling. The mixture leaves the filling station close to target, then the analysis shows it slightly off. Some batches are topped up or refilled.

What it costs. Mixtures outside their tolerance, especially for calibration and speciality gases where the tolerance is tight. Rework, extra analyses and filling time, and occasionally a complaint from a customer whose instruments were calibrated with it.

Why the usual fix fails. Fixed correction factors cover the average day. Filling more slowly reduces the temperature effect but costs capacity, and the next component or cylinder size needs new factors.

How a fixed-price remote diagnosis checks it. From filling logs (pressures, temperatures, flows, weights) and the analysis results I check how much of the deviation follows from gas behaviour that can be computed in advance. The fixed-price remote diagnosis shows whether model-based mixing control could hit the target on the first fill more often, and what it would need.

Plants run close to relief valves and trips

What happens. Pressure and cryogenic plants run near their relief pressures and trip limits, because that is where they are most efficient. When a disturbance comes, a pressure spike lifts a valve or trips a compressor. Operators then set the targets further away from the limit.

What it costs. Every trip means downtime, a restart and often lost product. Every step back from the limit costs efficiency for all the hours without disturbance.

Why the usual fix fails. Standard controllers react only when the pressure is already rising, and tuning them faster makes them nervous. More margin moves the problem from trips to lost efficiency.

How a fixed-price remote diagnosis checks it. From your trip and alarm records and the history around them (pressures, flows, valve positions, compressor data) I check which disturbances lead to the trips and how early they show. The fixed-price remote diagnosis answers whether predictive control that respects the limit could keep the plant closer to it with fewer trips, and what it would take.

Background: Predictive control of process plants

Liquid hydrogen boils off along the chain

What happens. Liquid hydrogen evaporates in storage tanks, trailers and during every transfer. The gas that boils off is vented, flared or recompressed. How much is lost where along the chain is often known only roughly.

What it costs. Product produced with a lot of energy and then lost. Liquid hydrogen is expensive to make, so boil-off directly cuts the margin of every delivery.

Why the usual fix fails. Better insulation is a hardware question with long lead times. Operating rules for transfers, cool-down and tank pressure are set conservatively and seldom revisited.

How a fixed-price remote diagnosis checks it. From tank levels, pressures and temperatures, transfer logs and venting records I compute where and when the boil-off arises and how much of it is due to the way tanks and transfers are run. The fixed-price remote diagnosis shows how much loss process control of tanks and transfers could avoid, and what it would need.

Air separation units change load slowly

What happens. The air separation unit is run at a steady load, and load changes are made in slow, careful steps. The columns need time to settle, and operators know that a fast change can upset purity for hours.

What it costs. The plant cannot follow cheap or expensive power hours, cannot reduce load quickly when the grid asks for it, and loses the chance to earn from flexibility. Its large compressors run at full power when power is dearest.

Why the usual fix fails. Ramping faster with the existing controls risks off-spec product. Standard controllers on the columns do not see how a load change will move purity long before it happens.

How a fixed-price remote diagnosis checks it. From the history of past load changes (air flow, product flows, column temperatures and pressures, purities) I check how the columns respond and how far faster ramps would stay within purity. The fixed-price remote diagnosis shows whether predictive control could make faster load changes possible, and what it needs.

Background: Predictive control of process plants

Air separation ignores swings in the power price

What happens. Production follows customer demand and the habit of running steady. Power prices change from hour to hour, but the plant does not use its liquid storage to shift production away from expensive hours.

What it costs. Power is by far the main cost of air separation. Making product in expensive hours that could have been made in cheap ones raises the bill every day.

Why the usual fix fails. Planning by hand around price forecasts, tank levels, contracts and load-change limits is too much for a daily routine. So the plan stays flat.

How a fixed-price remote diagnosis checks it. From production and storage data, power prices and your load-change limits I compute how production could have been shifted over past weeks and what it would have saved. The fixed-price remote diagnosis shows whether planning to prices with liquid storage pays off and what it would need in your plant.

Background: Economic predictive control

Product purity is held with a wide margin

What happens. The columns are run so that oxygen, nitrogen or argon purity stays well above the specification. Analysers are slow or sit at the end of the line, so operators keep a cushion.

What it costs. Every bit of purity above the specification costs compressor power or recovery. Over a year that cushion is a large part of the power bill that buys nothing.

Why the usual fix fails. Lowering the margin by hand is risky: one off-spec delivery causes more trouble than months of extra power. Faster analysers help, but a reading still comes after the column has moved.

How a fixed-price remote diagnosis checks it. From column temperatures, pressures, flows and the analyser values I check whether purity can be computed earlier from the column data and how much margin would then be needed. The fixed-price remote diagnosis tells you whether computed purity and tighter control could run closer to the specification, and what it would take.

Background: Computed measurements

Liquefiers take long to recover after a disturbance

What happens. After a compressor trip, a load change or a disturbance in the cold, the liquefier takes a long time to return to its operating point. Operators bring it back by hand, carefully, to avoid a second trip.

What it costs. Lost production and liquid during recovery, and sometimes a warm-up that costs days. Customers waiting for liquid feel it.

Why the usual fix fails. The plant’s standard controllers are tuned for steady operation and are too slow or too nervous for large disturbances. Recovery procedures depend on who is on shift.

How a fixed-price remote diagnosis checks it. From the history around past disturbances (temperatures, pressures, flows, valve and turbine data) I check how the plant responds and how a faster, safe recovery could look. At CERN, the European particle-physics laboratory, I worked on predictive control and state estimation for the superfluid-helium cooling of the Large Hadron Collider. The fixed-price remote diagnosis answers whether predictive control could shorten recovery in your plant, and what it would need.

Background: Predictive control of process plants

Helium losses go unnoticed

What happens. Helium is recovered, purified and reused, but some is lost through leaks, purges and venting. The balance between delivered, recovered and stored helium rarely closes exactly, and small losses disappear in meter tolerances.

What it costs. Helium is expensive and supply is unreliable. A small loss that runs for months adds up to a large purchase.

Why the usual fix fails. Leak searches are done when the loss is already obvious. Meters drift, and nobody reconciles all the flows and inventories regularly.

How a fixed-price remote diagnosis checks it. From flow meters, tank and storage pressures and temperatures, and purchase and delivery records I reconcile the helium balance over time and look for where and when losses appear. The fixed-price remote diagnosis shows whether loss estimation by balance can locate the losses early, and what it would take to run it continuously.

Nobody knows the temperature profile inside the cold box

What happens. The cold box has only a few temperature measurements, mostly at the inlets and outlets. What happens inside the heat exchangers and columns is inferred from experience. When efficiency drops, nobody can say exactly where.

What it costs. Heat exchangers that are fouled or poorly balanced cost compressor power. Problems such as freezing or maldistribution are found late, sometimes only when the plant has to be warmed up.

Why the usual fix fails. Adding sensors inside a cold box means opening it, which is rarely possible. Design calculations describe the new plant, not today’s.

How a fixed-price remote diagnosis checks it. From the existing temperatures, pressures and flows and the design data of the exchangers and columns I check how well the internal temperature profile can be computed. The fixed-price remote diagnosis shows whether computed internal states would reveal the efficiency losses, and how they could guide operation.

Background: Computing unmeasured states for control

Turbo-expanders run with wide margins

What happens. Turbo-expanders are kept well away from their speed, temperature and liquid-formation limits. The setpoints were chosen at commissioning to be safe and have hardly changed since.

What it costs. An expander that could produce more cold runs below its potential, so the plant needs more compressor power for the same output.

Why the usual fix fails. Moving closer to the limits by hand risks a trip or damage to an expensive machine. The standard controllers cannot see a limit coming, so the margin is the only protection.

How a fixed-price remote diagnosis checks it. From expander data (speed, inlet and outlet conditions, bearing data, trips) I check how close the machine has really come to its limits and how much margin it needs under which conditions. The fixed-price remote diagnosis answers whether model-based control near the limits could gain efficiency safely, and what it would need.

Compressors in a train are staged by habit

What happens. Several compressors share the load, and which runs, at what load and with how much recycle is decided by habit or a fixed order. Anti-surge recycle is often open more than necessary.

What it costs. Compressors are the largest power consumers of the plant. Poor load sharing and unnecessary recycle cost power every hour.

Why the usual fix fails. Each compressor has its own controller, but nobody optimises the whole train. Changing the order by hand is tried once and dropped when something trips.

How a fixed-price remote diagnosis checks it. From compressor data (flows, pressures, power, recycle valve positions, operating hours) I compute how efficiently the train has run and how the load could have been shared better. The fixed-price remote diagnosis shows whether load-sharing optimisation pays off and what it would take.

Background: Economic predictive control

Tank pressure is managed by venting

What happens. In storage tanks for liquid nitrogen, oxygen, argon or carbon dioxide, the pressure rises as heat leaks in. When it reaches the setpoint, gas is vented. Fill and withdrawal patterns are not used to keep the pressure down.

What it costs. Product lost through the vent, and in some cases emissions. At customer sites the losses are hard to see and often blamed on the tank.

Why the usual fix fails. A simple pressure switch on the vent is reliable and cheap, so nobody touches it. Better insulation helps only when a tank is replaced.

How a fixed-price remote diagnosis checks it. From tank pressures, levels, temperatures, fill and withdrawal data and venting records I compute how much product is lost and how much could be kept by controlling pressure with a model of the tank. The fixed-price remote diagnosis tells you whether model-based pressure control would pay off across your tanks.

Large hydrogen liquefiers are new territory

What happens. Hydrogen liquefiers are being planned at sizes well beyond the plants running today. Their process, machines and control have little operating history behind them.

What it costs. Long, uncertain start-up, a plant that misses its design output for months, and the risk that the control concept does not fit the real plant.

Why the usual fix fails. Control concepts are scaled up from smaller plants and filled in during commissioning. Problems in the interplay of compressors, expanders and heat exchangers then appear on the live plant.

How a fixed-price remote diagnosis checks it. From the process design, the equipment data and any data from pilot or smaller plants I check where the control concept is weakest and which situations (start-up, load changes, trips) need a model. The fixed-price remote diagnosis answers whether a process model and control concept built before commissioning are worthwhile, and what they would take.

Tanker routes and fill levels are planned by hand

What happens. Dispatchers plan tanker routes to customer tanks from experience, telemetry where it exists, and phone calls. Some tanks are filled too early, others run low before the tanker arrives.

What it costs. Fuel and driver hours for half-full deliveries, emergency trips for tanks that run low, and boil-off in the tankers. Customers notice stock-outs.

Why the usual fix fails. Route planning software often plans routes but not when each tank should be filled. Without a forecast of each customer’s consumption the plan stays reactive.

How a fixed-price remote diagnosis checks it. From tank telemetry, delivery records and routes I forecast consumption per customer and compare past deliveries with an optimised fill and route plan. The fixed-price remote diagnosis shows how much route and fill optimisation could save and what data it needs.

Background: Optimised schedules and setpoint plans

Two-phase flow makes long transfer lines unstable

What happens. In long transfer lines liquid partly evaporates, and gas and liquid travel together. Pressure and flow oscillate, and the delivery end sees surges or gas instead of liquid.

What it costs. Unstable delivery, extra boil-off, valves and pumps working hard, and sometimes trips at the receiving end. Cool-down of the line takes longer than it should.

Why the usual fix fails. Controllers at either end see only their local pressure or flow and fight each other through the line. Running more slowly calms the line but costs capacity.

How a fixed-price remote diagnosis checks it. From pressures, temperatures and flows along the line and at both ends I check how the oscillations arise and whether a model distributed along the line can predict them. The fixed-price remote diagnosis answers whether such a model with matching control could stabilise the transfer, and what it would need.

Sensors in the cold fail or were never fitted

What happens. In cryogenic zones sensors are hard to place and to replace. Some fail and stay failed until the next shutdown; others were never fitted because there was no room or it was too costly.

What it costs. Operators run parts of the plant blind, control loops run on substitute values, and faults are found late.

Why the usual fix fails. Replacing a sensor in the cold means warming up part of the plant. Simple substitute values work in normal operation and fail exactly when conditions change.

How a fixed-price remote diagnosis checks it. From the warm-side measurements and the remaining cold sensors I check how well the missing values can be computed, and how early a failing sensor would show. The fixed-price remote diagnosis shows whether states estimated from warm sensors could replace the missing ones in operation and control.

Background: Computing unmeasured states for control

Commissioning a new plant takes weeks of tuning

What happens. A new plant or a new section is mechanically complete, and then the controllers are tuned on the live plant, loop by loop. Every cool-down and every trip during commissioning costs days.

What it costs. Late start of production, commissioning teams on site for weeks longer than planned, and controllers that end up tuned for one operating point only.

Why the usual fix fails. Tuning on the plant is the habit because a model is seen as extra work. But the same questions come back at every start-up.

How a fixed-price remote diagnosis checks it. From the process design, equipment data and data from similar plants I check which loops are critical and whether they can be tuned in advance on a model. The fixed-price remote diagnosis tells you whether pre-tuning on a model could shorten commissioning, and what it would take.

Unmanned plants respond slowly to faults

What happens. On-site generators and smaller plants run unmanned and are watched from a remote centre. When an alarm comes, someone has to work out from a few trends what is going on, often at night.

What it costs. Slow fault response, unnecessary call-outs, and outages at customer sites that a timely intervention could have prevented. Experienced people are needed for remote diagnosis and are scarce.

Why the usual fix fails. More alarms create more noise, not more insight. Each plant is a little different, so fixed alarm rules miss the unusual cases.

How a fixed-price remote diagnosis checks it. From the alarm and event history and the plant data around faults I check which faults announce themselves early and which signals tell them apart. The fixed-price remote diagnosis shows whether automatic fault diagnosis could support the remote centre, and what it would need.

Carbon dioxide liquefaction for new capture chains

What happens. Carbon dioxide captured at cement plants, waste incinerators and other sources has to be cleaned, compressed and liquefied for transport. The flow and composition of the captured gas follow the source plant, not the liquefier.

What it costs. A liquefier that cannot follow the source runs inefficiently, vents, or forces the capture plant to throttle. Product off-spec for transport causes rejected deliveries.

Why the usual fix fails. Control concepts are taken from merchant carbon dioxide plants that run steadily on clean sources. A capture chain needs control across capture, compression, liquefaction and storage together.

How a fixed-price remote diagnosis checks it. From the design of the chain and the expected behaviour of the source I check where the chain is most sensitive and what its control has to coordinate. The fixed-price remote diagnosis answers what a control concept for the chain needs and where a model would reduce start-up risk.

Plant controls are decades old and nobody dares to change them

What happens. Many plants still run on controls installed decades ago. The logic works, but documentation is thin and the people who wrote it have left. Changes are avoided because nobody knows what else they would affect.

What it costs. Improvements that would save power or product are never made. A failure of the old system is a growing risk, and replacement projects stall because nobody can specify what the old system does.

Why the usual fix fails. A like-for-like migration copies the old logic, including its workarounds, into new hardware. The plant gains new hardware but not better control.

How a fixed-price remote diagnosis checks it. From the existing control logic, drawings and plant data I document what the controls actually do and where the operation could be improved. The fixed-price remote diagnosis gives you a basis for a new, documented control concept, and an estimate of what it would take.