Over the past four decades, complex, large-scale manufacturing processes have been defined, measured, analyzed, refined, improved upon, Six-Sigma-ed, into efficient producers of products that are safe, reliable and profitable.

But, there are signs that manufacturers are reaching the limits of quality improvement using only the old methods that reshaped American manufacturing in the 1980s. 

That's where Fero Labs, a New York-based startup that makes actionable machine learning software for improving processes and increasing quality of manufacturing facilities comes in. In a telephone interview, CEO Berk Birand explained:

The reality is that despite the tremendous improvements there are still a lot of things that can go wrong, a lot of uncertainty, in manufacturing. You have a lot of variability in terms of your raw materials.  You have equipment that's degrading continuously. You have prices that are changing globally, which means that you need to adapt continuously. Typically, in many plants that are armies of engineers whose job is to exactly do that to make sure that the plant not only keeps running, but keeps making high quality products in a better way. 

The problem, Birand notes, is that the industrial sector has so successfully addressed the challenges of reducing uncertainty and reducing variations that much of remaining uncertainty is just inherently there and there's not much to be done about it using the old tools. Said Birand: 

If you're if you're a steel plant for example and you're taking scrap steel and you're melting scrap steel, you will never know ahead of time what it is that you're melting. You will never know if it has high chromium, low copper or high copper, you're just melting some dishwashers and car bodies and you never know what's going to be inside and there's no way you can you forecast-until Fero came along--what's going to be inside, so there's always going to be uncertainty.

So, the obvious question is, how does Fero Labs help solve those problems? Said Birand:

What we do is build explainable machine learning software that doesn't try to reduce uncertainty but actually tries to measure it using ML and IoT sensors and then recommend ways to adapt and make decisions based on this information. So, before we melt some scrap, for example, we try to plan and maximize what's going to be inside of this particular batch. 

What Fero allows workers to do is melt a sample and measure what's inside. And then within seconds, they can change the entire production plan so that it maximizes the quality of the products and minimizes the emissions that are taking place.  So basically the software allows, let's say, a chemical plant or a steel plant to measure these added sensors and compute a recipe to advance specific outcomes.

Brazil-based Gerdau, the largest producer of long steel in Latin America, which has more than 320 plants in North America, installed the Fero platform to first help identify data that was bad or missing so its quality departments could identify data they had to drill down into and clean or create. Said Birand: 

They've connected our software Fero to their plant and piped all of the different sensors in their plant to our software. And then they communicated to the software the goals that they want to achieve. So things like, I want to minimize my raw material costs. Plus, I want to make sure that the steel beams that I make are within spec and have a certain specific strength.

After cleaning up its data, Gerdau began getting more and more accurate predictions about the mechanical properties of the steel it made based on the data coming in. The company then focused on creating live-prediction dashboards that could show operators real-time results as they worked the machines. 

Fero calls its software "explainable ML" because it tells the operators, who don't have to be engineers, why it is making its specific recommendations.   

The company then experimented with those predictions, creating models that could detail how they could optimize every batch and move leftover materials to different products to further cut down on waste.  Birend explained:

They communicate all of these different goals-and they have a very complex goals-- about how to maximize quality and other factors.  The software takes all of the information and raw data then computes the best way to make steel so that they can minimize the raw material costs, maximize quality, and even minimize emissions. 

As a result, in the first year of use, they reduced their raw material costs by 9%. So they're using fewer raw materials now, which, of course, saves money but also has a great environmental side effect.   

 © 2021 LeackStat.com