ControlIQ
The tuning was never the problem.
Control algorithms are settled engineering — the backbone every plant already runs on, and they do their job. What decides whether a loop holds is the model underneath them: how well anyone actually knew the process when the controller was set up. ControlIQ builds that model from your process's own behaviour, tells you whether the data supported it, and commissions the controller only when it did.
Where loops actually fail
A test that is too short does not look wrong.
On real hardware, a test run for 900 seconds against a process that takes 3881 seconds to respond produced a model that fit its own data almost perfectly — and described a process four times faster than the real one. Goodness of fit cannot catch that: it measures whether the curve passes through the points, not whether the data ever pinned the answer down. The controller that comes out is confidently wrong, and nothing in the loop's behaviour announces it.
What we do
Measure the loop. Decide whether to trust it. Then commission.
The same sequence on any single loop, whatever the plant makes. The step that is ours is the third one — refusing a model the data did not determine, rather than returning it anyway.
Learn how the loop behaves
From a supervised test where the loop will take one, from setpoint moves under the controller already running where it will not, or from history the plant has been recording all along.
Build the model
The process physics as the baseline, a correction fitted to your own data, and a stated confidence — the same decomposition the rest of the platform uses, applied to how a process moves in time.
Accept or refuse it
Checks that ask whether the data determined the model, not merely whether the curve fits. If it did not, we say so and stop — a refusal is a result, not a failure.
Commission into your system
The controller the loop actually needs — classical or model predictive — commissioned into the control system already running it. Nothing of ours executes inside the loop.
How long the test ran
0.2 τ
Fit
R² 0.9922
Time constant
0.20× true
Goodness of fit never drops below 0.99. The time constant is what collapses — and only the acceptance check can see it.
Capabilities
Built for loops you cannot afford to guess at.
Models from your own plant data
The route depends on what your loop will allow — an open-loop test is the cleanest, and it is not the only one. The loops worth the most are rarely the ones you may bump freely.
A verdict on the data, not just the fit
The distinctive part, and the one nothing in a standard workflow provides: an explicit statement of whether the data was ever enough to support a model, before anything is built on it.
One model, either controller
Classical and model predictive control come out of the same identification, so the choice is made on evidence rather than on what a vendor happens to sell.
Into the system you already run
Whatever is already running your loops — a DCS, a PLC or PAC, a panel controller, a supervisory link — takes the settings on the block already in service. No new hardware in the loop, no safety re-certification, and nothing of ours to fail.
What you get out of it
Answers before you commit plant time.
Commissioning window
~3τ + ~5τ
Supervised testing then a hold-out check, scaled to your loop's own dynamics — so we can quote a schedule for a process we have never seen.
If it fails
Nothing changes
Your controller keeps running. There is nothing of ours inside the loop to fail.
When simple is enough
We say so
Not every loop needs model predictive control. Where a well-tuned classical loop wins on the evidence, that is what we recommend.
Who it is for
Plants where a mistuned loop is expensive.
Existing loops, existing hardware
Teams who want better control out of the equipment already installed, without adding anything to the loop that can fail.
Lines that have not started yet
Where the model comes from the process design instead of the plant, and the controller is specified before commissioning rather than tuned afterwards.
Anyone evaluating advanced control
Engineers who have been quoted a control project and want to know what it will cost in plant time, what it will disturb, and what happens if it does not work.
Frequently asked questions
- What is ControlIQ?
- ControlIQ commissions process controllers from a plant's own data. It builds a model of the loop, states explicitly whether the data was enough to support that model, and then commissions the controller the loop needs — classical or model predictive — into the control system already running it.
- How long does ControlIQ need on my plant?
- It scales with your loop's own dynamics rather than a fixed calendar: roughly three time constants of supervised testing, and about five more checking the model against data it was never fitted to. That is the only window that touches production — the modelling and commissioning are offline, and the shadow period runs alongside your existing controller without changing anything.
- What if the loop cannot be taken off its controller for a test?
- Then the route changes but the verdict does not. The model can come from setpoint moves under the controller already running, or from history the plant has been recording all along, or from the process design before a plant exists at all. What never changes is the last step: if the data was not enough to determine a model, we say so and do not tune from it.
- Does anything of yours run inside my control loop?
- No. ControlIQ commissions the controller block already in service, so your controller keeps running whatever happens to us. There is no new hardware in the loop and nothing to re-certify.
- What is the difference between ControlIQ and Chemical Engineering Agents?
- ControlIQ is a controller: it learns how your process behaves and then holds it there, running inside the control system you already have, unattended, every cycle. Chemical Engineering Agents are advisory — they read the same chemical record to answer questions, review QA and flag deviations, and a person approves anything that acts. Agents never move a valve; ControlIQ never gives an opinion.