While the steel industry may not be synonymous with AI adoption, steel manufacturers are not different from any other business in ensuring efficiency in their output and throughput. There are almost certainly opportunities for AI implementation across an industry worth hundreds of billions of dollars that currently lags behind in adopting AI technology.
Several potential ROI-adding use cases exist for AI and machine learning in steel manufacturing. A casual search in a reputable database (e.g., Researchgate or similar) will reveal various works published by SMEs within the industry.
Many of the reported use cases revolve around machine learning. Because steelmaking involves complex scientific processes (e.g., physics) and the combination of science and technology (e.g., metallurgy), many correlations and variables need to be made fully transparent and understood.
Machine learning can ingest large data stores and uncover said correlations and variables. Teams can then use this data for various purposes: identifying production line bottlenecks, detecting equipment malfunctions, tweaking the values (e.g., temperature) in the heating process, classifying defects, and more.
Business results of these cases can include: expediting production, increasing returns and shareholder satisfaction, producing better steel, generating higher yields on the market, replacing or supplementing manual labor in hazardous production processes, and more – if heavy manufacturing leaders can successfully adopt machine learning capabilities.
In a recent episode in the AI Success Factors series on Emerj’s AI in Business podcast, Emerj CEO Daniel Faggella spoke with Nikunj Mehta, CEO and Founder of Falkonry. Falkonry is an AI solutions firm based in the San Francisco Bay area focusing primarily on AI applications in heavy industry. This article will analyze their conversation, focused on two critical takeaways for applying AI capabilities in heavy industry environments:
- Identifying data issues using juxtaposition: Addressing multiple business problems at once, and comparing results, can reveal where systems aren’t extracting enough data for AI capabilities to produce genuine insights.
- Transparency in data-gathering methods for SMEs: Early adoption projects can turn SMEs into champions by having a self-evident, easily understood framework in place for identifying and explaining what behaviors are beneficial to the adoption process.
Nukunj discusses all of the above points in the context of a specific use case with a large North American steel production company.
Listen to the full episode below:
Guest: Nukunj Mehta, CEO and Founder, Falkonry
Expertise: Computer Science, Start-ups, SaaS, Cloud Computing
Brief Recognition: Falkonry has been listed on the Forbes AI50 as a “Top AI Firm to Watch.” The company’s prominent customers include the US Air Force and the US Navy, Siemens, and Ternium. Nukunj holds a Ph.D. in Computer Science from the University of Southern California.
Two-Factor Approach for Identifying Core Business Problems
In this use case, Nukunj discusses a production problem involving a large US steel manufacturer, North American Stainless. The company needed help to keep up with production throughput for reasons unknown.
Discovering the underlying issue was vital for this company as production and profitability were being directly affected in two ways: [1] Decreased steel throughput and [2] lower market earnings due to selling a lower-grade product.
In this use case, Nukunj discusses a production problem involving a large US steel manufacturer, North American Stainless, who turned to Falkonry to help in keeping up with production throughput. Accessing the operational environment for investigating the production issue was challenging and dangerous. In this case, the setting was hot rolling mills.
Hot rolling is the mechanical process whereby hot steel slabs are passed through a pair of steel rolls, exerting thousands of tons of pressure to produce the desired shape and dimensions. Nukunj states that the process involves a “fairly common but undesirable set of activities” for workers due to the operation of dangerous machines and the extremely high temperatures.
Nevertheless, the end-product of hot rolling is incredibly important, explains Nukunj, “This is the bread and butter of steel makers. They’re hopefully making these coils consistent enough from one slab to the next so that they can sell first-rate products to their customers.”
Overall, American Steel had two major concerns about the unknown production issue. First, because of the volatile and dangerous environment, the company worried that a serious malfunction could cause significant damage to other parts of the production plant. Second, a decrease in production quality would yield second-rate market prices, ultimately impacting profitability.
As for many production issues reflected in the present use case, the root cause for struggling throughput at North American still needed to be discovered. The presence of multiple influential variables or some other factor(s) obscures the ability to pinpoint the problem.
Nukunj elaborates on what are perhaps the two most crucial elements of preparing for AI in steel manufacturing and other heavy industries:
“Picking a good problem is usually the first hurdle. We solve it collaboratively. The second problem is often: How do you get data out of your existing data sources into an AI system? While it has become easier over time in the industrial world, a lot of the systems that people have been using are somewhat dated and do not necessarily play very nicely with the new data protocols.”
– Falkonry CEO & Founder, Nukunj Mehta
To uncover the fundamental production problem, North American ultimately chose to investigate – and, if necessary, address – two problematic production lines for potential data discrepancies. By comparing the data from seemingly production lines seemingly “independent from each other,” Falkonry’s team discovered that one production line only appeared to be suffering from larger issues because it wasn’t producing data relative to predictive maintenance.
Once both systems were producing relevant data, the causes at the root of North American’s throughput issues turned out to be the same.
Normally, early AI adoption projects want to focus on a single, important business problem – or at least as few as possible. However, because of Falkonry’s A-B testing style approach in comparing data collection in two distinct business problems, they were able to locate and address underlying production issues.
Transparent Frameworks for Empowering SMEs
A proactive, respected SME with a digital-first mindset to solving problems is an invaluable asset for sponsoring and supporting an AI project. In Falkonry’s case, it was a plant maintenance manager (who concurrently serves as the digitalization manager) who’d been at the company for over 30 years.
Nakunj says such a champion is invaluable in manufacturing, where SMEs often find it challenging to engage with AI engineers and data scientists. As a company, Falkonry has designed and implemented its own process for working with client SMEs and various AI champions.
Over the past couple of years, Nakunj states that Falkonry has focused almost exclusively on developing a protocol for partnering and working with SMEs. Nakunj elaborates on the protocol ethos of working with SME champions within a client company:
“The one thing we’ve learned is you have to be able to pinpoint where and when something happened that needs human attention. Try to give it a name, and provide them with backing evidence to say why it is that behavior is important to look at and provide them ways to mark their findings or their conclusions against those findings so that it can be incorporated into improving the learning.”
– Falkonry CEO & Founder, Nukunj Mehta
Nakunj continues by explaining how Falkonry puts its technology at the center of SME/client interactions and how it organizes feedback between SMEs and the client.
These Client/SME interactions are carried out within the system using the findings and explanations uncovered by the technology and confirmed by the SME. In American’s case, Falkonry’s technology would reveal a manufacturing/production issue. The SME would then meet with stakeholders within the same week to confirm what was found and provide a root cause analysis.
“[The SME] was critical to [American] developing confidence in what the technology was doing, that it was not misleading them, it was not missing much, and that it was leading to actions that they could take,” Nakunj explains.
While the champion is a critical success factor to the initiative, Nakunj tells Emerj that vendors like Falkonry tend to Falkonry must “quarterback” the SME’s actions since it is the vendor’s technology at the center of it all.
However, taking that sense of investment and initiative away from the champion can be detrimental to the entire process, says Nikunj: “The champion relied heavily on the technology doing its work, on the findings being understandable for which you have to provide them the interface, and to make the process of the subject matter expert’s work be streamlined.”
One critical element was how Falkonry worked with the SME champion to present results to stakeholders. As Nukunj explains below, this is not a function that should be left solely for the champion. You must explain what the results mean, help form findings into a cohesive story, and present it in a way that demonstrates value:
“How do these champions tell the story of success to the executive team? This is not something that champions can pull off on their own. So we had to quarterback them on taking the results — everything that they had verified — and forming it into a story that they could repeatedly tell, not just to their bosses, but to the executives, and potentially to executives in other geographies that are looking to achieve some similar results.”
– Falkonry CEO & Founder, Nukunj Mehta
