.webp){kind=logo}
A steady-state model is like a photograph: it assumes the well has settled and calculates what conditions look like if nothing changes from here forward. A transient model is like a film: it carries the history of the well forward in time, so it knows not just where the operation is but how it got there and where it's heading in the next few minutes. Almost everything expensive in drilling happens in the moments a photograph can't represent: starting and stopping pumps, accelerating pipe, making a connection, cuttings still settling from the last hour of circulation. Stuck pipe, losses, kicks, and poor hole cleaning aren't caused by the average condition of the well. They're caused by the transitions. A model that only describes averages is blind during the exact minutes that produce non-productive time.
Our transient models, running on Intelie Live, update continuously against live data every few seconds, watching the same operation the driller is watching at the same moment, not as a batch study someone ran in the office days before the bit went on bottom.
What Steady-State Misses, Mid-Operation
The standard steady-state study is run back at headquarters during planning, days or weeks before the well is drilled, on assumed parameters. By the time the bit is on bottom, the well has already diverged from those assumptions, and the driller ends up comparing live sensor readings against a table printed from a model of a well that no longer exists.
Equivalent circulating density is not a constant. Every pump start, every acceleration of the pipe, every surge and swab event moves downhole pressure through a window that in a tight-margin well can be less than half a pound per gallon wide. A steady-state answer gives the settled value; the formation experiences the peaks. You can fracture a formation, or swab in a kick, entirely inside the space between two steady-state answers.
Hole cleaning is the same problem on a longer clock. Cuttings beds build gradually, as a function of the whole history of flow rate, rotation, and time. A steady-state check can say the well is fine at the current flow rate while a transient model, tracking concentration along the annulus hour by hour, shows a bed silently growing on the low side of the lateral. That bed is tomorrow morning's stuck pipe, and it's invisible to a model with no memory.
Without a live model, the only reference for whether torque or standpipe pressure looks normal is the driller's own memory of the last few hours. With a continuous transient model, every measurement has a physics-based expected value sitting next to it, so the operation notices the moment reality starts drifting from what the physics says should be happening. That drift is usually the first symptom of trouble.
Five Models, One Well
Hydraulics, thermal, hole cleaning, torque and drag, surge and swab: run them as five independent tools and every one is working off stale or assumed values for the quantities the others compute. Temperature changes mud density and rheology, so thermal feeds hydraulics. Hydraulics determines how well cuttings get transported, so hydraulics feeds hole cleaning. The cuttings bed changes the friction the drillstring sees, so hole cleaning feeds torque and drag. Pipe movement generates pressure surges that ride on top of the circulating pressure, so surge and swab and hydraulics are two views of the same pressure field.
That's exactly how the desktop world works today: an engineer runs a torque and drag study with an assumed friction factor, a hydraulics study with an assumed temperature profile, a hole-cleaning check with an assumed rheology, each in a separate application, each frozen at planning time. None of the models are wrong individually. They're wrong jointly, because the couplings between them are where the real behavior lives.
On Intelie Live, the five run as one integrated digital twin, exchanging outputs continuously and recalibrating against live measurements from the rig, so the twin tracks the actual well instead of the planned one. Even vendors with real transient physics tend to sell it as separate desktop applications, one per discipline. The coupling, running continuously, calibrated by live data, on the same platform that handles the data quality underneath, is the part that doesn't exist elsewhere as a product.
Why This Has to Run at the Edge
Three things break if the physics moves to the cloud. The first is the decision loop: a round trip from a rig to a cloud region and back through a satellite link adds delay and jitter to a calculation that wants to run in seconds, right when a driller is about to move pipe. The second is connectivity itself. Rigs sit behind satellite links that saturate, degrade in weather, and drop, and if the physics runs in the cloud, every connectivity event blinds the rig exactly when something is most likely going wrong.
The third is specific to transient models and is the real architecture requirement, not a preference. A transient model is stateful. Its current answer is built on the entire history it has integrated: the temperature field, the cuttings distribution, the calibrated friction. Feed it a gap and it doesn't just miss those minutes, it loses its state, then has to re-initialize and re-converge while the well has already moved on. A steady-state calculation tolerates dropouts because every answer is self-contained. Transient physics punishes them, because the model's memory is the product. So the model has to live where the data is born, at full resolution, independent of the weather between the rig and the cloud. One customer made that call explicitly, choosing to run well integrity and production optimization at the edge for those same three reasons.
Why the List Stays Short
The field sorts into three groups: steady-state planning tools, transient physics sold as an offline desktop simulator for design and post-analysis, and a short list that runs transient physics continuously against live data. On our analysis, that third group has exactly one other name in it besides ours.
The hard part was never the equations. Transient wellbore physics is published science, and every serious vendor in this space employs people who can code it. The hard part is keeping a stateful numerical model alive, stable, and correct for weeks at a time against real field data instead of the clean inputs an engineer prepares for a desktop run. Sensor spikes, dropouts, unit errors, clock drift: a single bad value absorbed into a model's state corrupts everything downstream of it, and that's not something you can solve inside the model. It takes an industrial-grade real-time data quality layer underneath it, which most physics vendors don't have and can't cheaply build.
Then come the pieces that never make a demo slide: continuous automatic calibration, so the twin doesn't drift from the well it's supposed to represent; numerical schemes stable over unbounded run times, not just a simulation horizon; edge deployment across hundreds of rigs unattended, not one engineer's laptop. Each takes years and produces nothing to show for it individually, which is exactly what makes the gap durable. It isn't one insight a competitor can license. It's a stack of operational engineering that has to be lived through, and for planning-software vendors selling licenses into engineering departments, crossing into a real-time operations product with uptime obligations isn't an upgrade. It's a different company.
Proof in the Field
A super-major's 10,000-foot-in-24-hour drilling record wasn't produced by heroics. It came from standardizing one consistent way of drilling a well, monitored in real time, and the record fell out of executing that model well. At high rates of penetration, the constraint stops being the bit and becomes the wellbore: can you clean the hole as fast as you're making it, and can you keep downhole pressure inside the window while pumping hard and moving pipe fast. Those are exactly the transient hole-cleaning and real-time ECD problems, and the models gave the team a live answer to both, continuously, so they could hold parameters near the physical limit with confidence instead of backing off to a margin sized for ignorance. Without them, the operation runs on planning-phase steady-state limits plus experienced conservatism, slowing down on a fixed schedule whether the well needs it or not, because nobody can see whether it does.
Operators doesn'tdon’t co-file patents nor co-author with vendors. TheyIt co-files and co-authors them with companies that solved a problem it couldn't buy off a shelf. Stuck pipe is one of the most expensive recurring failures in drilling, and our joint work with Aramco, presented openly at SPE conferences (see the OnePetro library SPE-214604-MS, SPE-215528-MS, IPTC-24078-MS), came out of pooling their incident history and domain knowledge with our real-time platform and modeling. It's a pattern, not an exception: our early drilling-assistant research with Shell was done jointly and published (OTC-30488-MS), and our work with Petrobras on real-time data collection, and quality control and well planning followed the same shape (OTC-31816-MS, OTC-30574-MS, SPE-201612-MS). The largest operators in the world are, in effect, our development partners.
What the Twenty-Year Skeptic Asks First
The reaction from a drilling engineer who's run steady-state models for twenty years is rarely resistance to the physics. They were taught exactly what steady-state neglects, and they ran it for two decades because it was what a desktop and a planning workflow could deliver, not because they believed a wellbore reaches equilibrium. The first reaction is usually recognition: this is the model they always knew was right, running in the way they never thought was practical.
The first question is almost always what happens when the data is bad, and it's the right one to ask. It's why the physics sits on top of a data platform rather than beside one: quality control, normalization, and sensor arbitration run at the edge before anything reaches the models, and the models are continuously recalibrated against trusted measurements. The second question is whether it will match their numbers, and in stable phases of the operation it does, which builds trust. Then the interesting conversation starts about the moments in between, where their old tools went silent. Most engineers can name a specific well, a pack-off or a stuck string nobody saw coming, and at some point in a demo they realize the model would have shown the drift for hours.
What closes the skeptic is never the claim. It's a trial on their own wells, run side by side against the steady-state tools they already trust, watching the twin track reality trip after trip. That's the right note to end on: this isn't an argument we need anyone to accept on faith. It's one the physics wins in the field.
Intelie Live's transient engineering models run continuously against live data at the edge, hydraulics, thermal, hole cleaning, torque and drag, surge and swab, coupled as one real-time digital twin of the well. To learn more, visit intelie.com.
Featured Blog
Handpicked insights from the Intelie team.
