Two conversations occurred this week involving variations in feedstock that affect the process and the quality of the end product. One conversation involved coffee beans. The other steel.
The farmer fortunate enough to sell directly to the roaster (Direct Trade Coffee) can afford to pick ripe berries more often during the season than the farmer selling through a large broker. This reduction in the variability of the coffee bean feedstock allows the roaster to achieve tighter control in the process. Coffee beans purchased through a “middle man” have more variability requiring the roaster to over-roast (burn) the beans during the process (think large chain coffee). The coffee I’m drinking this morning has only the roaster between me and Diego Chavarria, who farmed those beans in Nicaragua. It’s delicious.
Ah, but steel, you wonder. Similar philosophy at least through the process. The other conversation involved Fero Labs’ CEO and co-founder, Berk Birand. Fero Labs has developed a software product to help operators and engineers reduce process volatility, gain more control over that process, and consistently produce higher quality product. The two main markets the company serves are steel and cement.
Birand explained the company’s focus by beginning with the main enemy of manufacturing—volatility. There simply exist too many variations for which manufacturing software can account. These include feedstock variability, operator variations from shift to shift, and many other environmental factors. This leads process engineers to over-design in order to allow for worst-case scenarios. In fact, he states, the average factory uses 9% more resources than necessary to account for these variations.
Birand and his co-founder are machine learning (should I say?) experts. “Machine Learning works better today because we have ten years of improved data infrastructure along with much statistical modeling thanks to such initiatives as Six Sigma. Our premise is to reduce overdosing by building a statistical model that can capture and process hundreds of parameters.”
The software builds real-time operations atop its modeling and machine learning engine. One example is steel processing. This type of manufacturing involves a feedstock of scrap steel. I had a customer once who was in this market. The pile of scrap I saw was amazing. Anyway, operations cannot really know the composition of the feedstock going into the furnace. Fero Labs’ software allows input of composition analysis in real time such that operators know what might be necessary to add to that particular batch to make the steel to spec.
Trust in “artificial intelligence” or machine learning models is hard to build because it is almost always a “black box.” Engineers cannot peer inside to understand what is going on. Birand describes Fero’s product as a “white box.” Engineers do have the ability to peer inside and see what’s happening. Building trust at this level makes it more likely that the customer will actually use the product.
I seldom have CEO interviews with the depth of detail of this conversation. It has been many years since I actually worked in manufacturing. These conversations bring back the excitement of finding a better way to make things.