Intellegens Blog – Stephen Warde, September 2022
Discussing applied machine learning for chemicals, materials and manufacturing – see all blog posts.
Metals were first used over 10,000 years ago – a copper pendant discovered in what is now northern Iraq has been dated to 8,700 BCE. Metals have been so critical to our civilisation that entire ages have been named after innovations in their production and use. The bronze age axes pictured were the literal cutting edge of technology in their day.
It seems almost impossible that there can be anything else to discover in the world of metallurgy, particularly when compared with the relatively new kids on the materials block, like plastics (which we discussed here recently) or composite materials.
And yet metals remain the best, sometimes the only, solution to many problems. And the boundaries of metal R&D continue to be pushed. NASA recently announced a new oxide dispersion strengthened (ODS) alloy that can endure temperatures over 2,000 degrees Fahrenheit, is more malleable, and survives more than 1,000 times longer than existing state-of-the-art alloys. The driver for innovation here was to find ever-more performant materials for extreme conditions. This is a common factor in applications such as aerospace engines, nuclear power, or chemical processing.
Other motivations for development include lightweighting (for example, in automotive alloys) and enhancing resistance to corrosion (a problem that costs the global economy $2.5tn per year). Then there is climate change – can we reduce the energy consumption of steel production and processing, which alone is responsible for 7-11% of global carbon emissions?
Of course, when you are talking about perhaps the longest-lived technology area in the world, each new advance has to be eked out by identifying and tuning the tiniest details that impact performance. Examples might be tweaks to the proportions of alloying elements, the introduction of different trace elements to the alloy recipe, small changes to the regime of heat treatments and other processing steps applied to the alloy, or the use of novel surface treatments. The marriage of metal innovation with new manufacturing technologies like additive manufacturing (as in the NASA case) also offers exciting new possibilities.
Metallurgy is a complex, multi-dimensional problem, where progress comes from understanding and exploiting the often subtle, hidden relationships between ingredients, structure, processing, and properties. This is rich territory for machine learning, a technology designed to establish and use exactly these types of relationships. Indeed, the Intellegens Alchemite™ machine learning tool was first applied to aerospace alloy development at Rolls-Royce, where it was able to suggest alloy designs that were then validated, enabling substantial savings in experimental costs and time. Welding Alloys Group have applied machine learning to design a cost-effective and more environmentally-friendly hardfacing material to combat problems with wear. And, recently, OCAS NV were able to use machine learning to extract valuable insights from microstructural images, supporting development of advanced steels.
Who knows where the next hundred or thousand years of metal innovation will take us. Researchers will continue to test the limits of strength, durability, and ductility, traded-off against both environmental and economic cost. They’ll need every tool at their disposal to do that, and we’re sure machine learning will be part of that story.
Studying metal properties with Alchemite™ machine learning
Image credit: Birmingham Museums Trust (source file). Image sourced from Wikimedia (source page). Licensed under Creative Commons Attribution 2.0 license.
.jpg){kind=link}