Industries / Industrial equipment and services / Industrial services / Heat exchanger services
Performance monitoring
Performance monitoring works out, from the temperatures and flows a plant already measures, how fouled each exchanger is and when cleaning it will pay.
4
Milestones achieved, from 1959 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1959
Fouling that levels off
D. Q. Kern and R. E. Seaton, in "A theoretical analysis of thermal surface fouling" (British Chemical Engineering, 1959), treated fouling as deposits arriving at a steady rate while the flowing fluid tears them away faster as the layer grows, so that the layer approaches a fixed maximum. The model is still used to predict fouling and the savings from cleaning, for example by the 2002 Korean study of sponge-ball cleaning cited in the 2015 Sydney chiller study above.
Credit
D. Q. Kern and R. E. Seaton
1992
Cleanliness tracked by computer
A condenser's cleanliness factor is its measured rate of heat transfer divided by the design rate, read from a few temperatures and flows. Brad Buecker, working at City Water, Light & Power in Springfield, Illinois, wrote the calculation into a program using Heat Exchange Institute data, ran it three times a week and published it in Power Engineering in 1992. His records caught scale building up in a dry summer (from about 80% to 45%) and slime after a biocide pump failed (80% to 40%).
Lessona drop in a day or two, such as from the mid-70s to 45%, is usually air leaking into the condenser, not fouling; his was a cracked shell, so the cause has to be found before anyone cleans.
Credit
Brad Buecker; General Physics Corp., whose training course supplied the method
1995
A threshold for crude fouling
William Ebert of Chevron Research and C. B. Panchal of Argonne National Laboratory reanalyzed crude oil coking data from an Exxon and US Department of Energy project and built a simple model of when crude oil starts to foul hot exchanger tubes. It estimates the combinations of surface temperature and shear at the wall below which fouling does not build up. They presented it in 1995 at San Luis Obispo, California, at the first of the Heat Exchanger Fouling and Cleaning conferences. A 2015 review by D. I. Wilson, E. M. Ishiyama and G. T. Polley says such threshold models have since been used to predict fouling in exchanger design, in networks of exchangers and in planning cleaning.
Credit
William Ebert; C. B. Panchal
2019
A refiner tracks every shell
TOTAL put HTRI's SmartPM software into its refineries to monitor crude oil preheat trains, the chains of exchangers that heat crude before distillation. The software checks the plant's readings against each other, works out the fouling resistance of each shell in a series, predicts how fouling will grow and builds a cleaning schedule. TOTAL and HTRI staff published case studies from the Normandy and Grandpuits refineries.
Credit
HTRI, a research consortium at work since 1962 with over 1,000 members in more than 60 countries; TOTAL Research and Technology
Watch for
Virtual sensors spot fouling sooner
Researchers at the University of Zagreb trained two machine learning models, a long short-term memory network and gradient-boosted trees, on readings from a crude unit preheat exchanger taken every five minutes over about seven months, using the period when it was clean, then watched the gap between predicted and actual outlet temperatures grow as it fouled. Both picked up fouling more sensitively than the standard effectiveness-NTU calculation, and they estimate that one fouled exchanger added about 175 t of CO2 and EUR 12,000 in costs over two months. The data are available only on request.
Achieved whensomeone else reproduces the result on released data, or a refinery runs the models live and checks their warnings against what is found when the exchangers are opened.
Claimed by
Ž. Ujević Andrijić and N. Rimac, University of Zagreb, in Sensors, August 2025
Open
01
An open fouling benchmark
A public, anonymized set of real exchanger operating records (hourly flows and inlet and outlet temperatures, plus the date of every cleaning), with fixed scoring tasks: predict each exchanger's fouling resistance 30 days ahead, and flag the start of fouling. Baseline scores from the effectiveness-NTU method and two learned models come with it.
The briefCredit · unclaimed02
Test TEMA's numbers against data
An open table of measured fouling resistances from published work, each with fluid, flow speed, surface temperature, run length and whether the fouling leveled off, set beside the TEMA design value for the same service, to show where the design numbers are high, low or about right.
The briefCredit · unclaimed
Open 01 · the brief
An open fouling benchmark
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public, anonymized set of real exchanger operating records (hourly flows and inlet and outlet temperatures, plus the date of every cleaning), with fixed scoring tasks: predict each exchanger's fouling resistance 30 days ahead, and flag the start of fouling. Baseline scores from the effectiveness-NTU method and two learned models come with it.
Why it is still open
Each published method is tested on its own plant's data, which stays private. The 2025 Zagreb study's data are "available on request". A University of Tokyo team predicted crude oil fouling from refinery data in 2022 without releasing it. A 2020 University of Illinois deep learning model was trained on a simulated exchanger. Searches of Zenodo and Kaggle found no plant data with cleaning dates. On GitHub, most fouling models are trained on made-up data; one August 2026 repository posts 18 months of three-hourly temperatures and flows for a single crude unit exchanger with its design data, but with no license, no cleaning dates and no fixed scoring task, and a July 2026 Universiti Teknologi PETRONAS project kept its three-year refinery record private under a data-sharing agreement.
Built with what exists
Historian exports from a cooperating refinery, chemical plant or campus cooling plant; pandas; XGBoost and PyTorch for the baselines; Zenodo or Hugging Face Datasets to host the data with a DOI and a held-out test set.
First step
Ask the Zagreb authors, and the refinery whose experts they thank, whether their preheat-train data can be released under an open license, and ask the owner of the August 2026 GitHub data for a license and the exchanger's cleaning dates.
Done looks like
At least 20 exchangers with two years of data each and at least 30 cleanings; results on the held-out exchangers published for the three baselines; at least one team outside the project submits a scored entry.
Open 02 · the brief
Test TEMA's numbers against data
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table of measured fouling resistances from published work, each with fluid, flow speed, surface temperature, run length and whether the fouling leveled off, set beside the TEMA design value for the same service, to show where the design numbers are high, low or about right.
Why it is still open
HTRI's 1999 literature survey of refinery fouling named improving TEMA's recommendations as a main need and said "there are insufficient reliable data to accomplish these goals". A 1999 Argonne review compared TEMA values with plate heat exchanger values for a handful of fluids only; for cooling tower water it gives 0.044 m²K/kW against TEMA's 0.18 to 0.35. Since then the Heat Exchanger Fouling and Cleaning conferences have put three decades of measurements online free, and no search found a systematic comparison.
Built with what exists
The free conference papers at heatexchanger-fouling.com and Argonne fouling reports on OSTI; a document-reading AI model to pull values into rows, with a person checking each one (Notes: "Seven more flagged"); WebPlotDigitizer for values shown only in graphs; pandas; a licensed copy of the TEMA Standards for the comparison, publishing TEMA's own values only where open sources already print them and ratios elsewhere.
First step
List every paper in the 2011 to 2024 conference proceedings that reports a measured fouling resistance together with its flow speed and temperature.
Done looks like
At least 300 measured values from at least 50 sources, each linked to its source; a second person re-extracts 50 rows and agrees on at least 45; charts for cooling water and crude oil showing measured values against TEMA's.