“A Skeptic’s Guide to the Value of Artificial Intelligence in Steel Mills”
13 minute video
On October 27, 2020 I joined a virtual panel discussion with Fastmarkets Steel Success Strategies. In this session, I address:
- How to be a profitable steel manufacturer
- The characteristics of a good artificial intelligence (AI) problem
- A view into our Product Quality AI application
- … and more!
We were pleased to participate in this virtual event.
Here’s a transcript of my comments on behalf of Noodle.ai:
Thanks, Richard. Hi everybody, I thought I’d take a little different approach in this presentation and hit head on, what I think a lot of people are thinking, which is, “is there really value out of artificial intelligence in steel mills?” and so, I’m calling this, “A Skeptics Guide to the Value of Artificial Intelligence.” A lot of us are trained as critical thinkers and are inherently skeptical. I thought it would be important to have a presentation to address that skepticism. Really, this is all around the challenge that a lot of metal manufacturers are facing, which is basically in this challenging economic time of how to really run a profitable mill, and to be more profitable and increasingly profitable. And I am absolutely convinced that the next frontier is really in the digital realm, in analytics, and in squeezing more profit per mill hour out of out of all of our operations.
So, first of all, how do you know when something is a good AI problem? First of all, it has to be a complex problem, there needs to be large amounts of data about the problem, has to be non-obvious, what’s causing the problem. So even if something’s complex, if it’s obvious, you really don’t need AI. It also needs to have a sufficient value – there’s a lot of times I’ve seen people make mistakes where they say, “Hey, let’s apply AI to this really small problem to see if it works.” The problem is, in a lot of those cases, that it’s more expensive to develop the AI and to implement the AI, then you get out of the problem. So, it needs to be a very valuable problem. And, if you come up with a recommendation out of your data, you have to have the ability to take actions; so that’s what a good AI problem is.
At Noodle.ai, we focus on two specific areas. One is around quality and the other is around asset health. For a typical steel manufacturer, about 10 million tons per year, on product quality, that’s about $130 million is lost due to bad quality, meaning defects or other things that go wrong in the manufacturing process. And AI can reduce about eight to 15% of this; from our experience, this is from actual customer results. From an asset health perspective, this is really unplanned downtime, that this is typically a 100 million dollar a year problem for most metals’ plants, and AI is very effective here, and can reduce about 25% of this. Now, I realize that this at this time, all of you are saying, “I’ve seen slides like this all the time and I think it’s a bunch of hooey. And simply we don’t believe it.”
So, I thought what I do is to kind of do a deep dive on one area and explain exactly how it works and where the value comes from. So that you get a really a flavor of how it creates value. And if you’d like to see other areas that we’re happy to do that, but in the interest of time wouldn’t really focus on this one area.
So, let’s focus on quality. As a lot of you know, a lot of things go wrong in the manufacturing process and there are defects that happen all the time. About 80% of these, from our analysis, that people will have some sense of that something went wrong, but really won’t know why initially, and will really try to figure out what went wrong. 20% of the time [they] can’t figure it out. Like we really don’t know what happened and [have] been unable to find out. And the process of doing this root cause analysis usually takes anywhere from four to 12 weeks in many cases. So, here’s basically the “as is,” like what’s happening right now, that someone in the mill identifies a quality issue. They send it to a metallurgist, the metallurgist gets details on the defect [and] to the best extent that they can, they gather data, which is a very time consuming process because a lot of these are not set up to be able to identify data quickly and to query data quickly, like it was an IBA files, it’s usually very difficult to query, then they consult the metallurgical literature, which is time consuming process, creating hypotheses testing means hypotheses. And trying to figure out of the 400 parameters in a typical line with 160,000 combinations of a manual process of what went wrong, and how can we prevent it; this is where AI really shines, to look at things of this complexity.
So this is how our, Noodle.ai Product Quality AI (PQAI) system works; it identifies of those hundred and 60,000 combinations of things, what combinations are leading to anomalous behavior. So, you will see, basically, the green areas are areas where there’s good coils being produced, and the red are the anomalous areas. It is a multivariate view of things, [PQAI] looks at what combinations of characteristics lead to quality issues. In this case, there is this looking at the combination of the speed of gauge and the force of an asset on the metal. It’s in this coil, you look at a 1/400ths of the coil like so, you’re looking at a very specific area of the of the coil, and it will basically analyze what is leading to defect, it’s the combination of this, and this is, and this, and this, leading to defects.
And then what it does is actually comes up with a recommendation of set points in the mill for given circumstances and allows a metallurgist to very quickly not only do root cause analysis, to but also to say how we can fix it and under what circumstances and makes a huge difference in preventing defects and understanding why things why things go wrong. So, basically, if you look at quality, this is the sort of the value tree in quality, that you have bad quality, which is defects, and then the good quality which is, “are you overpaying for things to ensure good quality?” So, we have an app for that also, but the sooner the better. The cost of bad quality are the root cause analysis scrap and the detection costs. So, you can see some of the really great results that you get from the app root cause analysis, we went through the process I just showed you can reduce that 60% of the time, and then likewise down to improve profitability.
So, for a typical 10-million-ton capacity, you get about 2% defects, 20,000 defective tons a year, about 40% of these are just scrapped, 40% are downgraded and go down to a secondary market, 20% have to go through rework. So, if you add up, the cost of this is about $50 million a year for just this type of quality issue. And with just this one module of product quality, typical savings would be about $7.5 million a year from AI. And I’ll just say that AI is nothing to think [of as a] daunting thing. AI is basically just better math and running on lots of data on supercomputers. It’s something that you simply couldn’t do 5-10 years ago. And I think that this is the real breakthrough in profitability in metals manufacturing, and we’re seeing amazing results.
So basically, just an overview of Noodle.ai and our products, we have two product suites, one is Vulcan, which is more focused on the manufacturing side. I showed you a little bit about Product Quality, we have two other applications, one for Asset Health, and one for Demand Signal analysis. And then we have our Athena suite, which is more focused on the supply chain side; which is around looking at the supply and demand matching it’s very, very effective on the supply chain side.
One of the things that I think makes these solutions really effective is the combination of the knowledge of AI and domain knowledge. So we have a really great partner, SMS, and SMS digital, who we work shoulder-to-shoulder with; we’ve been working with them for about four years now and are bringing the best of both worlds of absolutely world class AI, understanding and digital expertise with the over 100 years of metallurgical understanding, and obviously a deep understanding of how mills work. So, we’re really happy to have a great partner in in SMS.
And my final thought is that, I see that there are a lot of mills that are skeptical about AI and there are a lot of other mills that are trying it out and getting great results. It always reminds me of the [idea that] whenever there’s a breakthrough, there’s a lot of people that are skeptical. I’m sure Blockbuster was skeptical of streaming movies, and Kodak was skeptical of the digital camera. I think it’s really important that the people really understand this technology, because I think it is going to be the real breakthrough in in profitability, and we’re already seeing those results. So, I guess my final thought is to not get blockbustered and please get started in this area. I think it’s a really important and exciting area.
So, thank you very much, Steve Pratt if you want to reach me at: firstname.lastname@example.org.
- Push Up Open? Why Small Words Matter in UI Design
- Enterprise AI® Powered Supply Chains:
- “A Skeptic’s Guide to the Value of Artificial Intelligence in Steel Mills”
- CIOs Discuss “Life Since COVID-19: The Supply Chain Becomes a Strategic Asset”
- Noodle.ai’s Manufacturing Application Platform: An API-First Strategy