Engineering intelligence for process industries
Turn plant data into decisions that cut fuel, cost and CO2.
Twin Process AI builds an AI-powered digital twin of your process units. First-principles engineering, machine learning and constrained optimisation work together to reduce fuel consumption, raise thermal efficiency, catch faults early and hold product quality. Your existing control system stays exactly as it is.
- Digital twin
- Soft sensing
- Constrained optimisation
- Fault detection
- CO2 accounting
- Efficiency
- 87.1 %
- Stack O2
- 3.73 %
- Stack temp
- 233 C
- CO
- 20 ppm
- Feed
- 189,546 kg/h
- Outlet
- 415 C
- Fouling
- 0.75
- Detected fault
- fuel_upset
Trim excess air 19.8 % to 9.9 % (stack O2 3.73 % to 2.08 %)
$45.44/h~$381,710/yr at this state
- +1.26 ptsefficiency
- −1.52 %fuel
- −0.617 t/hCO2
Watch Move in steps and hold. Stop if CO passes 100 ppm before the O2 target, or if the stack falls below 98 C.
Why +−
Every percent of excess air above what the burners need is nitrogen heated from 20 C to the 233 C stack and vented. Cutting stack O2 by 1.65 points removes that parasitic mass flow, which is why the stack falls 16 C and dry flue-gas loss drops with it. The duty is unchanged, so the whole of the recovered loss shows up as less fuel through the burners. The move stops where it does because CO 167 ppm at the 200 ppm limit.
Engine output, unedited. The annual figure is this one operating state held for a year, not a year at that rate. Across the whole history the average is $96k per heater per year. First-principles simulation of a 53.4 MW crude preheat heater driven through 120 days of hourly operation. Not a plant installation.
What we optimise
Built for the units that burn the energy.
One method, applied unit by unit. Fired heaters first, because they burn the largest share of a refinery’s fuel, then the columns and the utility system around them.
Fired heater AI
AvailableFuel firing, heat duty, thermal efficiency, combustion quality and CO2, optimised hour by hour against the real operating limits. The twin names the constraint that stops the move going further, so nobody has to guess why the number is what it is.
See the unit
Soft sensing
AvailableFouling and true excess air inferred continuously from tags you already record. No new instruments to install, and any state that cannot be inferred from the available tags is switched off rather than guessed at.
See the sensors
Fault detection
AvailableDegraded burner tips and skewed air registers caught while the stack O2 analyser still reads normal, by comparing the unit against what a healthy one would do at the same duty and the same feed.
See the faults
Distillation AI
In developmentReflux ratio, reboiler duty, feed rate and product specification solved together, so energy per tonne falls without the top or the bottom spec leaving its window.
See the method
Plant optimisation
In developmentHeat and material balance closed on live data, utility consumption traced from OSBL supply through to ISBL demand, and the steam, power and fuel gas losses that never trip an alarm found and ranked.
See the platform
Energy and CO2 reduction
AvailableFuel, steam and power savings reported as tonnes of CO2 against target. Energy optimisation and decarbonisation are the same move, which is what makes a net zero commitment affordable rather than aspirational.
See the numbers
The number
Fuel saved and carbon avoided are the same move.
A heater trimmed to the right excess air burns less gas and emits less CO2 in exactly the same proportion. There is no trade-off to manage here, which is why energy optimisation is the cheapest decarbonisation a refinery has available, and why none of it needs new capital equipment.
805 t/yr
CO2 avoided, per heater, per year
A mid-size refinery runs dozens of fired heaters, and the same move on each one compounds into the site emissions figure.
$96k
Saved per heater, per year
Averaged across every hour of a 120-day history, then scaled by the fraction of the predicted saving that the physics actually delivered.
up to $380k
At the hours furthest off optimum
A single operating state, not a year held at that rate. Shown because those are the hours worth catching, and never shown on its own.
- 0.63 %
- Less fuel fired
- +0.65 points
- Mean thermal efficiency gain
- 2,675 of 2,783
- Hours where a move was available
Reported the way a plant already reports it: energy per tonne of feed against budget, steam and fuel gas against norm, tonnes of CO2 against target. Optimisation moves lines a general manager reads every month, which makes the case without translation and makes net zero progress something the operating budget can pay for.
Physics-corrected throughout. Assumptions: Fuel at 6 USD/GJ, carbon at 40 USD/t, 8,400 operating hours per year. First-principles simulation of a 53.4 MW crude preheat heater driven through 120 days of hourly operation. Not a plant installation.
Why this is hard
The data already exists. Almost none of it can be trusted as it stands.
Billions have gone into sensors, historians and control systems, so the measurement problem is largely solved. What is missing is the layer between the tag and the decision.
- 01
Instrument tags that no longer match the plant
Mis-mapped P&ID tags, dead transmitters and drifted instruments. A historian will show you a reading that has not moved in six weeks without ever telling you it is broken.
- 02
Sources that never reconcile
DCS, historian and laboratory results disagree with each other, nobody owns the difference, and every report built on top of them inherits it.
- 03
Balances that do not close
Mass and energy fail to add up across the unit, so the real performance of the plant stays hidden inside the gap between what went in and what came out.
- 04
Utility losses nobody can see
Steam, power and fuel gas leave through leaks, venting and start-up load that no monthly report separates out. None of it trips an alarm. All of it reaches the fuel bill.
- 05
Engineering that lives in spreadsheets
Manual calculations are slow, hard to audit and out of date the day after they are signed. A simulation case takes days to set up, so the operating window is reviewed once a year at best.
- 06
Dashboards instead of decisions
Plenty of trends and traffic lights, almost no automated recommendation that an operator can act on before the end of the shift.
These gaps compound, and they cost energy, throughput and reliability every day. Machine learning on its own makes them worse: applied to data that has never been reconciled, it learns the errors and reports them back with confidence.
Technology
Plant data in. Priced operating moves out.
A closed loop, and every stage of it is open to inspection. The physics can be checked line by line, the model can be scored against held-out operation, and the recommendation carries the reason it was made.
- 01
Data ingestion
Tags pulled from the historian, the DCS and the laboratory systems.
In progress - 02
Validation and balance
Every tag scored for confidence, then mass and energy reconciled until they close.
Planned - 03
Physics model
The unit from first principles: combustion, heat transfer, hydraulics, loss ledger.
Built - 04
Digital twin
Learns your unit, infers the state nobody measures, solves in microseconds.
Built - 05
Optimisation
The best safe setpoint against real limits, priced in fuel, CO2 and money.
Built - 06
Recommendation
A written action for this shift, and the limit that stops it going further.
Built
Nothing is written to your control system.
The platform reads the historian and returns a recommendation. Your DCS, your advanced control and your interlocks are untouched, which is what keeps the approval short and the risk on our side of the fence rather than yours.
The physics gets the last word.
Machine learning searches, the rigorous model checks. Every recommended setpoint is re-solved in the physics before it is shown to anyone, and the saving that survives that check is the saving we quote.
Why the numbers hold
Every recommendation is re-solved in the physics before it ships.
Machine learning is fast and optimistic. The rigorous model is slow and honest. We use the first to search a year of operation and the second to check what it found, then publish the result of that check rather than the prediction that went into it.
- 0.9934
Blocked-CV R2 on thermal efficiency. Nine of twelve targets clear 0.99
First-principles simulation of a 53.4 MW crude preheat heater driven through 120 days of hourly operation. Not a plant installation.
- 1,631x
Faster than a rigorous solve. 1.75 s becomes 1,073 microseconds
Measured on the same operating points, single core.
- 81.5 %
Of burner-fault hours caught that a stack O2 analyser cannot see, at zero false alarms
83 of 102 labelled hours across burner degradation and register imbalance. Zero flags across 2,499 healthy hours. Scored against the labels, built without them.
- 0.84
Of the predicted saving that physics actually delivered when we re-solved it
Ten recommended setpoints re-solved in the rigorous simulator. The surrogate predicted 1.52 GJ/h, physics delivered 1.27 GJ/h.
- 100 %
Of re-solved setpoints cleared every CO, tube metal, dew point and draught limit
The surrogate overstates the value of a move, not its safety. Every re-solved setpoint was valid in the rigorous model.
- ASME PTC 4
Full loss ledger. API 560 bands. NIST Shomate data
41 outputs, each with formula, assumptions, units, reference and validation.
The sensor that did not work is published too.
Three soft sensors were built for the heater and two of them ship. Fuel hydrogen cannot be inferred from the available tags, so it is switched off and nothing downstream reads it. A refinery that wants it online needs a Wobbe meter or a fast GC on the fuel header. More modelling will not recover a signal the tags do not carry.
Where we are, stated plainly.
First-principles simulation of a 53.4 MW crude preheat heater driven through 120 days of hourly operation. Not a plant installation. Calibration against a live heater is our next milestone. Every figure on this site traces back to a generated results file, and the engineering basis behind each output is written down with its formula, assumptions, units and reference.
Where this goes
One plant, six layers. Three of them exist today.
The end state is every unit on a site modelled, reconciled and optimised continuously. We are building it in the order that makes each stage useful on its own, and we mark honestly which stages are finished.
- 01In progress
Heat and material balance
Reconcile flows, temperatures and pressures until mass and energy close. Nothing downstream is trustworthy until this does, and in most plants it has never been done on live data.
- 02Planned
Tag validation
Find the instruments that are drifted, frozen, mis-mapped or simply wrong, and score the confidence of every tag before a model is allowed to read it.
- 03Built
First-principles simulation
Build the equipment properly: combustion, heat transfer, hydraulics and an efficiency ledger that a process engineer can audit line by line.
- 04Built
Constrained optimisation
Find the best safe operating point against real limits, price the move in fuel, CO2 and money, and name the constraint that stops it going further.
- 05Built
Machine learning layer
Soft sensors for the state nobody measures, a surrogate fast enough to search a year of operation, and anomaly detection that does not depend on fault labels.
- 06Planned
Operator copilot
Every unit on the site, answering questions in plain language and carrying the reasoning of engineers who have retired.
3 of 6 phases built and tested, for one unit type. Everything marked in progress or planned is a statement of direction, not a capability you can buy today.
Partner with us.
Send us 90 days of tags from one unit. We will tell you what it is leaving on the table, which of those hours we would have caught, and where our model disagrees with your plant. The last one is the useful part.
Refining, petrochemicals, fertiliser and power. If a process burns fuel to make heat, the method applies to it.