Hi, I'm Amogh — CFD Engineer!

At the National Laboratory of the Rockies I build simulations of an experimental furnace that makes iron out of ore using hydrogen. The most detailed of them is three-dimensional and follows the pellets of ore as individual objects, with the gas flowing around them, built in Ansys Rocky and Ansys Fluent and run on the laboratory's supercomputer. One thing stood in the way: Rocky can move particles about but it has no way of making them react, so the chemistry had to be written from scratch as a custom module. I developed that module with Ansys's own engineers.

Writing it is skilled, slow work. So I trained an AI agent on the software's developer documentation, along with example code and tutorials. It now writes and compiles a working module on its own.

The second job was duller and cost more time. A simulation could only run about an hour before it had to be stopped and saved, adjusted by hand in the interface, written out as a restart file and started again — then the same thing an hour later, for every case, each covering a full day of reactor time. Days of my week went on that. An agent now runs the whole loop in Python: it submits the jobs, handles the restarts and keeps going without me.

The third piece is AMReX, an open-source simulation framework the laboratory's SAMS group has already built codes on. I trained an agent on those codes and set it to work on iron ore reduction. It can reproduce a model of one ore pellet turning into iron, which I had built in COMSOL during my PhD.

• 95 %+ of the ore turned into metal, guided by the model
• $15M pilot plant investment informed by the cost analysis
• 1st reactor in the USA to make iron with hydrogen, modelled and checked

Iron is made in a shaft furnace: pellets of ore go in at the top, hot reducing gas comes up from below, and reduced iron leaves at the bottom. That gas has normally been made from reformed natural gas.

In 2021 the United States Department of Energy started the Grid-Interactive Steelmaking with Hydrogen project, or GISH, to lower the risk of investing in the equipment needed to make iron with hydrogen instead and turn it into steel in an electric arc furnace. A pilot reactor was built as part of that project. It was small, but built to the same design as a full plant so its results would carry over. One of the things we wanted to know was how it would behave when the feed was changed from pure hydrogen to a mixture of hydrogen and reformed natural gas.

I built a FEA multiphysics model of that reactor. What it predicted for the quality of metal produced was close to what the plant actually made, and so were the temperatures it gave for the inside of the furnace. That agreement is what made it usable on the plant rather than only on paper. When the pilot clogged during commissioning, the model was used to work out how to bring it back into operation, and to say in advance how a startup would go. The work is described in this paper. The pictures below show the simulation results at a steady state, showing how the soild species and gaseous species vary inside the pilot reactor geometry.

The same model was then scaled up to industrial size, to a shaft of the proportions a commercial plant would actually use. No plant of that kind existed to compare it against, so the scaled version rests on the agreement already established at pilot scale. One result stands out on its own: the reactor needs appreciably more hydrogen than the chemistry alone would require, and the surplus leaves unused. The pictures below are the same set of profiles for that larger reactor, with the temperature of the pellets added.

• 22 citations on the resulting paper
• 3 temperatures, from 800 to 900 °C
• 3 gas flow rates measured at each one

For as long as anyone has made steel, the job has been the same one: get the oxygen out of the iron ore. What changes is the gas you use to do it. That gas has usually been made from coal, natural gas, and commercial plants have decades of experience running on it. Then hydrogen arrived as the alternative — and when someone asked how a furnace would actually behave once it switched over, there was no confident answer.

You cannot find the answer by watching. An industrial furnace is millions of pellets stacked several storeys deep, sinking slowly through a reducing gas mixture, and nothing about the inside of that pile is visible from outside it. So the question had to be made smaller. We took a single pellet, hung it on a wire, and lowered it into a furnace on its own — one pellet at a time reduced using different gas mixtures, at different temperatures, followed from start to finish. How long did it take to turn iron ore into iron, and what held it back? These experiments helped us obtain the chemical kinetics of the reduction process.

Do that enough times and a pattern appears. The data obtained from the pattern goes into a model, and once the model can reproduce what the pellet did on the laboratory scale, you can start asking it about pellets nobody has tested — a different gas, a different mixture, a furnace that has not been built yet. That is what the whole exercise is for. A question that used to be settled by trial and error on a running plant gets answered on a computer first. The measurements and the model are set out in the paper.

• 84 citations on the resulting review paper
• 99+ different gases can be tested
• 50 % less time to test a membrane

A membrane is a thin sheet of polymer that pulls one gas out of a mixture without ever having to turn it into a liquid. All it needs is higher pressure on one side than the other, and one gas slips through faster than the rest. The catch is a stubborn one, and it is about the material rather than the machinery: the sheets that are fussiest about which gas they let through are also the slowest, and the quick ones are not fussy enough. You can have a clean separation or a fast one, but not both — which is what has kept membranes out of the harder jobs. Mixing tiny crystals into the polymer is one way around it. The most promising ones are called metal–organic frameworks: they are riddled with holes of a very exact size, so small molecules pass through and larger ones cannot, their structure can be tuned to the job, and they sit comfortably in the polymer rather than fighting it. My co-authors and I set that case out in a review of the field for the Journal of Polymer Science.

At Arizona State University I built a setup that makes electrospun fibers with those crystals inside fibers. It draws a single thread out of two liquids at once, one wrapped around the other, so the crystals and other tiny particles end up sealed inside the thread rather than sitting on its surface. Fibers made this way separate a gas by a different trick from the sheets: instead of letting one gas through, they catch it and hold on to it in the adsorption process at high pressure. The two approaches were running side by side. I also built the test rig that measures how the sheets themselves perform, in both of the shapes they are made in — flat sheets, and bundles of fine hollow tubes.

Earlier, at IIT Bombay and the Tata Centre, I built a test rig of my own. What drove that work was biogas — the gas that comes off rotting organic waste such as cattle dung. It burns, but it comes out watered down with carbon dioxide and other unwanted gases. Take those away and what is left is concentrated methane, the same gas piped into kitchens. I learned to make the hollow tubes there on the Membrane Lab's existing equipment, under Prof. Jayesh Bellare.