See the slow faults coming, without inventing a date.
Some faults arrive in minutes. Fouling, scaling, corrosion and filter plugging take days. Prognose fits those channels over a three-day look-back and projects each one against the threshold that would declare the fault — with an uncertainty band, and without a number we cannot defend.
MVPEvery figure above is counted in the minimum viable product: the amine train modelled for gas sweetening and post-combustion carbon capture.
What it does
Lean/rich exchanger fouling, reboiler fouling and scaling, general corrosion, and solvent filter plugging. These are the faults where a projection is meaningful. Fast faults belong to Detect, and mixing the two produces confident nonsense.
Linear, exponential, power and logistic fits are all tried and compared by information criterion. The winner is reported with its R², so an engineer can see how much of the trace the model actually explains.
Each channel reports where it stands now, the threshold that declares the fault, the margin between them, the rate of change per day, and where the projection lands at the end of the horizon.
In-fault, approaching, deteriorating, stable, or insufficient data. An engineer should be able to read the outlook without reading the manual.
How it works
A three-day trailing window ending at the moment you select.
Unsteady rows are dropped and the remainder is reduced to hourly medians. A degradation trend fitted through load swings is a fit to the load, not to the wear.
Linear, exponential, power and logistic, selected against each other by information criterion rather than by whichever looks best.
Six hours ahead, carrying a ±2σ uncertainty band on the projection.
The projection is set against the fault threshold, and the verdict follows from the margin and the direction of travel.
Where it applies
The platform is focused on acid gas removal — gas sweetening and post-combustion carbon capture — because that is where the model is calibrated and where the fault library is real. The method underneath is not amine-specific: a physics reference model, residuals against expected-healthy, fuzzy diagnosis over graded indicators, and costed consequences. Extending it to another unit means new datasheets and a new calibration, not a new platform.
The platform runs on a real amine train with thirteen days of one-minute history behind it. Nothing on these pages is a mock-up.
Open the platform