Two deals for small startups, completed recently and announced on Tuesday, add interesting technologies that should help the company develop new capabilities and help enterprises meet challenges in industrial IoT. Both were announced at GE Digital’s Minds + Machines conference in San Francisco and didn’t come with publicly announced price tags.
In mid-October, the company bought Wise.io, a startup that GE says can refine machine-learning algorithms through a process like natural selection.
In the industrial world where GE Digital operates, recommendations based on machine learning carry higher stakes than they do on e-commerce sites. Instead of spitting out advice that says you should spend a few dollars on a book, industrial IoT software may recommend removing an engine from an airliner at a cost of hundreds of thousands of dollars.
“We have to be very low on the false positives and very low on the false negatives,” said Harel Kodesh, CTO of GE Digital and vice president of Predix, the company’s industrial IoT platform.
The way it’s done is to run thousands of possible algorithms to figure out which is best for the job. Wise.io specializes in techniques for carrying out that winnowing process quickly and at a relatively low cost, Kodesh said.
It’s also developed ways to do this with relatively sparse data sets. That’s useful because industrial systems are highly reliable and don’t generate many examples of how failures happen, he said. That compares with millions of customer histories that an e-commerce site may have to work with to make a product recommendation.
Last week, GE Digital also bought Bit Stew Systems, a startup that helps to determine what the bits streaming out of an edge device mean.
When devices like sensors start sending data to IoT systems that are supposed to process it, those systems may not know what the bits represent, Kodesh said. In one packet, one set of bits may represent temperature readings, another pressure readings, and so on.
GE grapples with this problem when it’s developing new products, and it’s often down to trial and error to parse out the data. This can take months.
Bit Stew’s software can do the same thing in half an hour or less, Kodesh said. It does so by using machine learning to generate hypotheses about what different bits represent and then test them out by looking at other bit streams from the same device. GE also plans to use Bit Stew’s machine-learning technology in other ways and take advantage of a lightweight application generation system from the company, he said.
Separately, on Tuesday at its Minds + Machines conference, GE Digital introduced a system for enterprises to spread out IoT tasks to be more effective.
The new set of capabilities, called Predix Edge System, turns the company’s cloud-based Predix IoT platform into a distributed operating system. With it, organizations can place each of their IoT applications wherever they will run most effectively, including in sensors, controllers, gateways or the cloud.
“You have to take advantage of all those compute nodes so you can choose the right tradeoff of latency and cost,” said Harel Kodesh, CTO of GE Digital and vice president of Predix, in an interview. For example, the code that shuts down a pipeline in case of a leak should run as close as possible to the leak sensor rather than up in the cloud, where there will be a longer delay before the shutdown command can reach the site.
Predix has resided overwhelmingly in the cloud so far, though it has included some software that runs on edge devices to stream data into the cloud. With the addition of Predix Edge System, enterprises that buy Predix will also get a variety of new components that can run elsewhere. For example, some applications, such as software that controls a turbine, will be virtualized so they can run in different places.