IBM bets big on cognitive; ropes in Watson for manufacturing

Organizations today face rising resource costs in traditionally low-cost production markets. Cognitive manufacturing helps swing the balance back in favor of manufacturers, believes Bruce Anderson, IBM.

The electronics manufacturing industry faces tumultuous times with increasing customization, shorter lead times, frequently changing environments, and shrinking order sizes – all while managing a sophisticated supply network.

Thinner margins and increased competition threaten consistent quality, risk greater downtime and reduce desired flexibility. Investments in new equipment and automation systems are increasing the amount of data available from the shop floor, but most is not used to its full potential.

Now this is where cognitive manufacturing can make the cut – by helping organizations fully harness the data generated by the machines on the shop floor.

A tête-à-tête with Bruce Anderson, global MD, Electronics Industry, IBM highlights the current challenges faced by the manufacturing space, and how cognitive steps in to make the difference.

Why cognitive manufacturing matters in electronics?

What we pay for electronics has shrunken over the last few years. There's a reverse pressure on the manufacturing cycle.

What matters most in manufacturing is that you have to run the plant at a cost that's lesser than what the finance guys proposed. Cognitive can help you cut down the response time to fix manufacturing anomalies. 

Some companies actually exited the business because they figured they couldn't make money out of it. In the semiconductor manufacturing space, as designs got more and more complex, you had to get better and better at the manufacturing side.

Anderson believes that to strike a balance between manufacturing costs and production overheads, automation and data helps equally. The companies that survived did so solely because of the way they handled data.

“But what they're doing now, is that they're taking dark data - data that's stuck in the machines. So, the idea is to get that data out, put it into an analytical environment, derive insights, and then act on it fast enough to make a change in the manufacturing economics,” adds Anderson.

What's changed is the amount of data which is now available, and the speed with which you can make sense out of it. 

Sensors installed in machines collect data, and the data is then stored in either the datacenter or the cloud, and no one ever gets any insight out of it. It's just a complete waste, opines Anderson.

What can cognitive manufacturing achieve that conventional methods of manufacturing cannot?

Let's take conventional approach - you take data set, take that back to your office and work on it. And then you come up with ideas on what can be changed in the manufacturing process. Let's be generous, and give this process a cycle time of 24 hours. So, for 24 hours, that problem still existed in the manufacturing cycle. During this phase, you've lost throughput, yield, and quality.

Also, the data that they were looking at, was by our estimates, only half the available data. It actually all boils down to the dark data - what did you not know, that could help you come up with a better answer.

When you start to unlock a lot of the data using IoT-like technologies, you wind up with a richer data set in order to analyze the problem. 

What you really need is the ability to derive meaningful insights with all the data, and have the experience go in to the repository for the next guy to get answers to the problems.

When you apply cognitive to a problem, you can get a range of answers immediately, about how you can solve that problem. What we're doing here is augmented intelligence, where a human being makes the decisions on which strategy to use.

So, this eliminates the need for people to manually collect the data. In addition to this, we were able to shrink the time taken down by a factor of 24 or more.

Companies in India that have invested in data collection got a head start - it's a data-hungry process.

All swell, but why haven’t manufacturers embraced cognitive yet?

The survey indicates that 38 percent of companies haven't yet implemented cognitive, but have plans of doing so; 26 percent responded saying that they have roped in cognitive to a limited degree; whereas 24 percent of companies said that they already have a pilot program in place.

One of the biggest reasons is insufficiently skilled human resource. So, the people in manufacturing have to have the kind of background to implement cognitive. 

You will as a company, still go through that evolution of starting with basic analytics, then move towards predictive, as you learn your problem better. You'll start to get to a prescriptive approach from a preventive one. And then finally, you'll start using cognitive as your data set is actually in shape to do that.

Companies in India that have invested in data collection got a head start - it's a data-hungry process. What you need to remember is that when you just get the raw data in, you get about 30 percent of the questions right. What you really need to do is train the data. So, you have human experts who work with the data till you come up with the right answers. 

After we trained the data, the accuracy went up to the high 90s - we were able to rank-order all the right answers.

So how does one go about training the data?

Basically, the algorithms can sift through all of the text, and it's able to understand nouns and verbs based on natural language processing. It can ask you questions, and your answers remain in its memory.

In each domain of manufacturing, there's a certain taxonomy. You have to spend enough time with it so that Watson can build a taxonomical understanding of the information.

So, it's going to build the taxonomy and apply it to the data you've already given it. And the next time, it's going to ask you fewer questions, till it comes to a point where it asks you almost no questions. It will still want to know from its rank-order answers if it's doing a good job, though.

Impact of opensource on electronics manufacturing

One of the best examples of IBM's role in opensource is the Hyperledger--an open source, modular, multi-channel transaction network. I think blockchain for supply chain is going to be huge. A multi-company system like blockchain can reduce the friction in manufacturing to a large extent.

What matters most in manufacturing is that you have to run the plant at a cost that's lesser than what the finance guys proposed. Cognitive can help you cut down the response time to fix manufacturing anomalies. 

Watson can grab all the information, analyze, and tell you: Here're the things you need to check in this order, because essentially, it has the brains of every service person, every manufacturing designer incorporated into it.