Steps and pitfalls on the way to cognitive manufacturing
Cognitive manufacturing — the combination of data collection, massive data storage and artificial intelligence — holds the promise of transforming industry. We talked to IBM, one of the pioneers in the field, to get some pointers on what it takes to make cognitive manufacturing a success.
IBM initially developed cognitive manufacturing for quality early warning purposes in its own facilities, says Binny Samuel, Offering Leader — Watson IOT Connected Manufacturing. Finding it extremely useful he says, “we industrialized it and are selling it to the market.”
Steps in the process
Having blazed the trail, what advice can IBM give to companies contemplating a similar move? It’s best to proceed step by step, based on perceived value, says Samuel: “The first thing I would ask is, where do you see the highest value? Do you see a place where there’s a high potential to improve throughput? Do you see a place where there are higher wastages?”
The next step is to assess what data will be needed, and how much of that data is already available. “You need to have the data as a starting point because cognitive manufacturing is a data-driven optimization of manufacturing,” he explains. “Pick a value area and look at what kind of a data can be brought to the surface.”
It’s wise to note, he points out, that tremendous insight often comes from unstructured data that already exists. There tends to be a lot of it — from inspection cameras to logs to tech manuals that sit on a shelf unused; the problem is how to get insight from all of it. “I would say today 85 percent of the data used on the plant floor is there, it’s insightful, but we don’t use it or harvest it.”
But just grabbing all possible data is not the answer. “Many people think they need to have all kinds of instrumentation in their plant before starting a cognitive manufacturing journey,” says Samuel, “and it’s not true.” But it can be difficult to get a realistic understanding of what is the minimum data that needed to get some value and insight. So it’s essential to study and understand what data will really be needed to achieve tangible value.
Obstacles to implementation
Obstacles on the road to cognitive manufacturing tend to be management-related, rather than technical. First, the ability to prove value in a reasonable time: “In the past, it has taken months of engagement and high-level investment before being sure of the value,” says Samuel. Unfortunately, patience is often in short supply given today’s obsession with quarterly results. It’s vital to be able to be able to prove the value within eight to 12 weeks.
Another obstacle stems from organizational dynamics — specifically the divide between IT and OT. “IT on one side has certain control; OT on the other side has got certain control, and the solution in cognitive manufacturing lies between these two,” he explains. “Manufacturers not fully understanding how to deal with this IT-OT integration and who should do what has been a struggle to fix.”
Selling it to management
Unlike some other changes, cognitive manufacturing seems to have a resonance in the C-suites. But while senior management may be for it, lower levels may still slow things down, with questions about things like IT security issues, or how to finalize a cloud sourcing agreement. So while higher-level support helps, says Samuel, “you need to know how to play this within the working-level people, the middle-level management: how do you secure the security concerns, how do you secure IT-OT integration, how do you understand both sides’ concerns and come to a quick resolution process? Industry is still learning these things.”
Security: A balancing act
Industrial companies have traditionally protected their control systems with a combination of air gaps and robust firewalls, to keep any bad actors (human or otherwise) that might gain access to the plant network from interfering with the control system. Exceptions require considerable thought, and anxiety can mount when it becomes a question of relocating a database to the cloud. “But the main danger does not lie in the cloud itself, he goes on: “Every company like ours does its best to make sure that our cloud systems are secure, and people like me spend extra time to make sure our system is impenetrable.”
A lot of it comes down to the question of what data is put into the cloud and what to keep within a company’s four walls, Samuel continues. This involves consideration of not just security, but privacy (especially IP) and of speed of response. It’s best, he says, to use a hybrid approach: there are certain things that are definitely needed in a common platform, to leverage the elastic compute power, leverage common infrastructure and also bring in some common set of capabilities. But there are also opportunities for some key data to reside and run within the user’s four walls. Carefully understanding this distinction makes it possible to solve the privacy issue and also bring in more real-time content. But in the end it’s a judgment call — the user must find his or her own comfort level.
A look at the future
What will happen going forward? Eventually, Samuel says, most companies will adopt cognitive manufacturing, but not all the same way and not all to the same extent. “Large corporations that have more experience in collecting and getting data will look at higher value in terms of predictive analytics and prescriptive analytics and even cognitive computing, whereas maybe smaller companies, which don’t have substantial data collection capabilities, probably would first focus on descriptive analytics before getting into the predictive aspect, and then slowly graduate. Let the solution or the system guide you through the steps.”