Industries / Mining and materials / Chemicals and petrochemicals / Specialty chemicals
Scaling from lab to plant
Scaling up takes a recipe proven in a one-liter flask to a vessel thousands of times larger, where heat is much harder to remove and a mistake involves far more material.
6
Milestones achieved, from 1976 to 2021.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1976
Vents sized for runaways
Twenty-nine companies formed the Design Institute for Emergency Relief Systems (DIERS) to develop methods for designing the emergency vents that let a runaway reaction release pressure. It spent $1.6 million studying how a mixture of vapor and liquid flows out through a vent, and how to size vents for it. DIERS became a users' group in 1985, and over 200 companies eventually asked to join.
Credit
DIERS, under the American Institute of Chemical Engineers (AIChE)
1985
Bench calorimeters go on sale
After a runaway in one of its plants, Ciba's chemical engineers began in 1965 to measure the heat a reaction gives off in a temperature-controlled lab vessel, and used the data to plan safe temperature programs for plant reactors. A project with Mettler from 1981 produced the RC1, a bench-scale reaction calorimeter (an instrument that measures heat released as a reaction runs), which went on sale in 1985.
LessonWriting in 1997, Regenass said only a few hundred such instruments were in use because they were costly, large or hard to use, so a safety tool spreads only as fast as it is easy to use.
Credit
Willy Regenass, H. Martin and colleagues at Ciba; Mettler (now Mettler-Toledo)
1988
Planning for lost cooling
R. Gygax published a way to judge a reaction's risk at plant scale from lab data: measure the heat of the wanted reaction and of any decomposition, compare it with the plant's ability to remove heat, and work out what happens if cooling fails. F. Stoessel later sorted processes into five "criticality classes" using four temperatures, among them the highest temperature the batch reaches if cooling stops. The German guideline TRAS 410 uses these classes.
Credit
R. Gygax, F. Stoessel and J. Steinbach, the authors Regenass names for the approach
2004
A toll batch runs away
On 12 April 2004 a runaway reaction over-pressured a reactor at MFG Chemical in Dalton, Georgia, during its first full batch of triallyl cyanurate, made under a tolling contract (making a product to another company's order). Toxic allyl alcohol vapor escaped and 154 people went to hospital. MFG had tested three batches in a 30-gallon reactor but used a different recipe in the 4,000-gallon one, and its bench tests, designed for yield and cost, missed a strongly heat-releasing decomposition.
LessonLab tests must look for hazards as well as yield, and the plant must run the recipe that was tested.
Credit
US Chemical Safety Board
2007
The flask gave no warning
On 19 December 2007 a 2,450-gallon reactor making MCMT, a gasoline additive that raises octane, burst at T2 Laboratories in Jacksonville, Florida, killing four people. The process had been developed in a one-liter glass reactor that needed no cooling and was never run above 193 °C; the plant reactor did need cooling, and when cooling failed the reaction ran away. In response to the CSB's findings, AIChE proposed and ABET, which accredits US engineering degrees, approved in October 2011 a requirement for proficiency in chemical process hazards across a broad range of engineering disciplines.
LessonThe cooling a small flask needs says little about the cooling a plant vessel needs.
Credit
US Chemical Safety Board; AIChE and ABET for the curriculum change
2021
Software picks the next experiment
Researchers at Princeton University and Bristol Myers Squibb released open-source software that uses Bayesian optimization, a statistical method that chooses the most informative next experiment, to tune reaction conditions. In a test run as an online game tied to real lab results, it beat expert chemists and engineers on the average number of experiments needed and on consistency. Europe PMC lists 439 papers citing it.
Credit
Benjamin Shields, Ryan Adams and Abigail Doyle (Princeton) with Bristol Myers Squibb's chemical process development group
Watch for
Chemify's robot farm opens
Chemify, a University of Glasgow spin-out, opened its first "Chemifarm", a 21,500-square-foot site in Glasgow that combines AI molecule design with robotic synthesis. It calls the site the most advanced chemical synthesis facility in the world and says it will design and scale up molecules on demand, with hundreds of robots.
Achieved whenan outside customer or a published paper reports molecules made there, with amounts, yields and success rates.
Claimed by
Chemify and Lee Cronin, June 2025
Robot process chemist
Researchers at the University of Liverpool report a modular platform in which a mobile robot runs late-stage process chemistry, the lab work done before a route goes to a pilot plant, joining synthesis, work-up (separating and cleaning up the product) and analysis. They say it matched human results and ran unattended for long periods.
Achieved whena manufacturer or another lab reports using it on real process work, with measured time saved and repeatability.
Claimed by
Emma Brass, Andrew Cooper and colleagues, University of Liverpool, March 2026
Open
01
Open table of reaction heats
A public table of measured heats for the main reaction in published process-development studies: heat of reaction, heat capacity of the mixture, adiabatic temperature rise (how hot the batch would get with no cooling at all), the highest temperature reached if cooling fails, the onset of decomposition, the instrument used, and a link to the page. Then a test of how closely free estimating methods predict the measured heats.
The briefCredit · unclaimed02
Open plant cooling check
Free, open-source software that takes lab calorimeter results (heat of reaction, heat release over time, heat capacity, decomposition onset) and a plant vessel's size and jacket, and reports whether the vessel can remove the heat, the highest temperature reached if cooling fails, and the Stoessel criticality class from 1 to 5, with a table of common vessel sizes and every step of the calculation shown.
The briefCredit · unclaimed
Open 01 · the brief
Open table of reaction heats
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of measured heats for the main reaction in published process-development studies: heat of reaction, heat capacity of the mixture, adiabatic temperature rise (how hot the batch would get with no cooling at all), the highest temperature reached if cooling fails, the onset of decomposition, the instrument used, and a link to the page. Then a test of how closely free estimating methods predict the measured heats.
Why it is still open
The Open Reaction Database, the main open format for recording reactions, stores exothermicity only as yes or no. Japan's National Institute of Occupational Safety and Health published DSC data for 358 reactive chemicals, which measures the heat of a substance breaking down, not the heat of the reaction a plant runs. Searches for this file found large tables of computed reaction heats but no open table of measured heats of desired reactions.
Built with what exists
Open-access process papers, such as the free CHIMIA archive; a model that reads long documents and tables, such as Muse Ember (see the Note "The jobs agents repeat"); the Open Reaction Database's open schema, extended with heat fields; thermochemistry estimates from the open-source Reaction Mechanism Generator (RMG), maintained at MIT and Northeastern University.
First step
Copy every heat-of-reaction value from 20 open-access process papers into a spreadsheet with these fields, starting with Lonza's 2006 CHIMIA paper, which reports 110 kJ/mol measured against 120 kJ/mol estimated for a quaternary ammonium biocide.
Done looks like
300 reactions from at least 100 papers, each row linked to its page. Two people check 50 rows separately and agree on at least 95% of values. The table is published under an open license, with the gap between estimated and measured heats reported as a mean error in kJ/mol.
Open 02 · the brief
Open plant cooling check
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
Free, open-source software that takes lab calorimeter results (heat of reaction, heat release over time, heat capacity, decomposition onset) and a plant vessel's size and jacket, and reports whether the vessel can remove the heat, the highest temperature reached if cooling fails, and the Stoessel criticality class from 1 to 5, with a table of common vessel sizes and every step of the calculation shown.
Why it is still open
The method is set out in the German guideline TRAS 410 and in textbooks, and reviews still apply it by hand: a Purdue University student review worked through the classification step by step on paper. The free calculators found on GitHub are single-author teaching tools with no published validation, such as a tube-reactor scale-up calculator posted in October 2026.
Built with what exists
Python with SciPy for the heat balances; jacket areas and volumes from vessel makers' published data sheets; worked examples from papers and textbooks; a coding agent that writes and runs its own tests, such as one built on Inkling 2 Small (see the Note "Checked without being asked").
First step
Collect 10 worked examples with full inputs from open papers, starting with the estimated 60 °C adiabatic rise in Lonza's 2006 CHIMIA paper, and code the adiabatic rise and worst-case temperature for them.
Done looks like
For at least 20 worked examples from at least 5 independent sources, the software matches the published highest temperature within 2 °C and gives the same criticality class in every case. Code, inputs and the comparison table are public, and two chemical engineers have reviewed the method notes.