Fouling cost and cost optimal cleaning of the preheat train.

Cleaning on a calendar leaves fuel, carbon and throughput on the table.

In a crude distillation preheat train the incoming crude is heated across tens of shell and tube exchangers before the furnace. As fouling builds on each shell the recovered heat falls and the coil inlet temperature entering the furnace declines, so the furnace fires harder to hold the coil outlet temperature. That raises fuel and carbon, and once the furnace reaches its duty limit the only remaining response is to cut crude throughput. Cleaning restores recovery but takes a shell out of service and carries a cost, and because the shells are coupled through a shared flow network the effect of cleaning any one unit propagates across the train, so cleaning on a fixed calendar leaves recoverable fuel, carbon and throughput on the table.

A twin that prices fouling and a scheduler that chooses the cure.

EntroMetrix builds a physics informed twin of the preheat train that prices fouling on every shell. Each unit is a mechanistic simulation of deposition and heat transfer, wrapped in a network coupled reduced order model that tracks fouling resistance, duty and pressure drop and rolls them up to the coil inlet temperature that sets furnace fuel and carbon cost. On top of the twin a deterministic cost optimal scheduler, posed as a mixed integer problem, chooses which shell to clean, when to clean it and how to split the crude flow, to minimise the total of furnace fuel, carbon, cleaning and downtime over the run subject to the furnace duty limit. It is solved on a receding horizon, so the schedule is updated as the run advances and cleans are not bunched before an artificial end date.

Validated across 37 shells in eight networks, to 0.31 K.

The twin is built from data the plant already holds. Historian records of temperatures, flows and pressure drops across the train, with the log of past cleans and the crude slate, are enough to fit each shell and to reproduce how heat recovery declines between cleans. Around a hundred and eighty days of history gives reliable forward prediction, and the fouling law and operating limits are set with the plant's own engineers before any schedule is proposed.

On a comparable preheat train the reduced order twin reproduced the coil inlet temperature of a high fidelity reference to within about a third of a degree across a full year, and matched its annual operating cost to within a few percent, across thirty seven shells in eight networks. That fidelity is what lets the twin price fouling shell by shell and rank where a clean is worth most.

Cleaning and flow solved together, on a receding horizon.

Optimising cleaning timing and flow distribution together delivered materially more than retiming cleans alone, while adjusting flow on its own recovered little, so the two decisions are solved jointly. Solved on a receding horizon, the schedule stays inside the plant's proven operating window and does not cluster cleans before an artificial end date. The study is retrospective and offline, validated against the plant's own cleaning history before any schedule reaches the floor.