In the second of our four-part series on innovation, Steel Thoughts speaks to Carlos Alba, head of AI and digital R&D, to learn more about the company’s surprisingly long relationship with artificial intelligence and the impact it is having on every aspect of the business.

Let’s start at the beginning – when did ArcelorMittal first start experimenting with AI?

AI is all over the front pages these days, but ArcelorMittal was looking at the potential of AI technology more than two decades ago – well before ChatGPT captured the popular imagination. The company first set up an official AI research team in 2004. Back then, the focus was on letting mathematical optimisation, machine learning techniques and innovative algorithms get to work on corporate problems like production scheduling and supply chain management.

Since then, the range of applications and uses has mushroomed. Now, the AI division employs around 100 people providing support and services to ArcelorMittal units around the world.

In which areas is AI making the biggest difference?

We’re looking at all areas of our business and prioritising where we can add the biggest remaining value – and that’s the key point, our work must add value and deliver a competitive advantage to the group. Examples of areas where we are active include predictive maintenance and quality control, product development, patent analysis and scrap optimization. But it goes further than just creating categories. We’re developing deep specialisations in each of those areas to make ArcelorMittal stand out from the competition.

There’s a lot of off-the-shelf AI that can be applied to a range of problems – but we’re not interested in that. We’re only after the AI that is really going to make a difference, hence why we are focussed on developing our own solutions, albeit we do ensure we partner with some of the best AI minds on the planet to integrate the latest and greatest AI solutions into our work. We have a strict set of criteria to measure the potential of any breakthrough: it must have the capacity to differentiate ArcelorMittal from the competition; it must be applicable worldwide, in a secured IT environment, and it must have a significant impact by reducing costs

Tell us a bit more about the categories, for instance predictive maintenance.

As mentioned, we started our AI journey by looking at production scheduling, and this had a major impact on throughput and cost. We took inspiration from nature and developed bio-mimicking algorithms based on ants looking for food and bringing it back to their nests. They take the optimal path – a straight line – and this can be mathematically modelled for the production process. By that I mean converging the sequence of items in a process to reach the optimal point. We first implemented it in our finishing lines…and it worked. Throughput and yield increased and we delivered some impressive cost savings – almost $1 million a year on a hot dip galvanising line, for example. We then expanded this to our steel shops and the savings multiplied by three of four times.

The natural next step was then to look at quality and predictive maintenance. We’ve developed internal AI models for quality checks, especially for some of our higher-added value steel, such as that we supply to our auto customers. This is good, it ensures defect material doesn’t go to customers. But more interesting than detecting is predicting. Put simply, predictive maintenance is harnessing huge amounts of data about an industrial process like a steel mill to be able to detect a fault or a defect in a piece of machinery before it happens. That increase in reliability is a massive efficiency boost.

Take motors or hydraulic actuators, for example. A steel mill uses many motors and hydraulic motors. So, we first developed an offline AI platform that we called Sentinel to predict motor or hydraulic actuator failures, with a 100% success rate in the pilot cases. We ran these pilots in our plants in Canada and northern France and are now testing it in Brazil. The platform has been moved online and since the roll out we have had zero issues with the equipment at any of these mills. All the potential failures have been predicted, meaning the maintenance teams can come in and fix them before they go wrong. You’re shifting from a situation of relatively frequent failures to one where reliability is getting close to 100%. Our challenge now is to extend the platform to other equipment types and roll it out more broadly across our global asset base.

And the other areas?

It’s best to think about the second and third categories – product development and patent analysis – together because the impact of AI on them is primarily through speed increase.

For instance, with some grades of automotive steel, AI has managed to slash the product-development time from between three and five years to less than one year. How do we do this? Our teams around the world share models, data sets and experiments on a common platform. Once the data is available and the models for that family of products ready, we see a reduction of the design phase down to as little as three months, and then six to nine months for testing. Imagine that – a five-fold reduction compared to how we used to work!

But that’s not all. In patent analysis the savings are even more dramatic, up to 10 times as fast. Intellectual property is an increasingly important commercial battleground.

That is becoming a very fruitful area of research, largely because our AI tools are so efficient at analysing patents, the different chemical compositions etc.. Before, it used to take two to three hours per patent. Now it’s just a matter of minutes.

And the final category, scrap. How does AI help with recycling scrap metal?

Scrap is becoming a critical raw material for our steelmaking because recycling is one of the best ways to reduce our CO₂ emissions. But the problem with scrap is that it’s – to put it bluntly - scrap. It’s what somebody else has thrown away, making it very hard for us to know exactly what’s in it, what impurities and in what degrees of concentration. That’s very different from using iron ore and a blast furnace, where maintaining consistency is relatively easy.

And these impurities – for example metals like copper, tin, chromium, nickel – can all affect the qualities of the end-product. The ultimate challenge is to be able to make advanced steels from scrap, and AI is making a huge difference in helping us get there. We use it at every stage in the scrap process, from purchasing and procurement, to analysing for impurities, to looking at the effect on quality of different combinations of impurities at different levels.

And because we’re using millions of tonnes of scrap around the world to make millions of tonnes of steel, we have vast datasets that can be mined to improve our understanding. The sheer scale of our data gathering is one of the things that sets ArcelorMittal apart.

And what about the humans behind this artificial intelligence? Are they all 20-something maths geniuses?

Not at all! I was a 20-something when I first joined ArcelorMittal in the 2000s. But now I’m 49, having been with the company for 18 years. Of course, we have people in their early twenties, with PhDs in advanced modelling, statistics, computer science or machine learning. But they are junior engineers, junior researchers.

The important thing is that we offer them challenges and opportunities to grow and develop. And they get to collaborate with people who have the domain knowledge, who know about our products and who have worked for the company for a long time. It’s a great knowledge combination.

And our AI focus doesn’t just reside in R&D. We now have AI teams across the group in each of our segments. It really is a joint effort between R&D, our CTO teams, IT, and those segment teams. We’re really focussed on taking advantage of our scale. The segments typically take R&D developed models for core components in our mills, while we in R&D focus on new breakthroughs and watching advancements in the AI market so we can continue to develop new or improved AI solutions for the group. It’s a fascinating and fast-moving space, and one that I’m sure will offer a huge range of differentiation for our company in the decades ahead.