An airplane’s “black box” flight data recorder helps investigators understand how a crash happened. But with ubiquitous sensors and the ability to analyze petabytes of data online, technology could some day predict how future crashes will happen, and how to prevent them. “I don’t want to promise that we will build it…but it is definitely possible,” says Harel Kodesh, manager of GE’s Predix cloud-computing service.
Predix Cloud is designed to serve the industrial Internet–machines talking to each other to monitor and improve the efficiency of planes, trains, oil wells, wind farms, or medical equipment. GE has been running Predix in-house with its own machinery for about a year. This week, GE announced plans to offer a cloud-based version of Predix to anyone–even makers of competing industrial equipment such as Siemens–starting in 2016.
Industrial machines started carrying sensors and spitting out data years ago, says Kodesh, but the processing of that data was very low-tech. “The way to deal with this was with a pickup truck,” he says. “To drive around and write down LCD numbers and go back and put them in an Excel spreadsheet.”
Only in the past few years have technologies emerged that can manage and analyze this data en masse. One of them is Apache Hadoop–open-source software that can link and coordinate thousands of computers to store and analyze data with what Kodesh calls “pretty much unlimited computing power.” Another is the rapid advance in machine learning–algorithms that allow a computer to discern patterns from a swarm of data and eventually make predictions for what will happen next.
Ganesh Bell, who heads up software and analytics for GE’s Power & Water division, likens the advances in digital modeling to the progress of racing video games. In the 1980s, the cars were lumbering blocks of pixels. Today, advanced games exactly model how specific cars act in the real world.
GE has essentially built a video game for wind farms. Its PowerUp system, running on Predix, consists of “digital twins” for each real wind turbine in the field. Data from the actual turbines is fed into the model, allowing Predix to figure out how they might run better. Sometimes the solutions are unintuitive, such as running the upstream turbines less efficiently so that the downstream turbines catch more wind and the farm can generate more power overall. Tricks like that have allowed GE to get up to 5% more electricity from the same facility.
Predix also extends to carbon-burning technology, says Bell, like GE’s gas turbines. “We’ve been collecting data on our turbines for years. Now we have the technology to process it,” he says. “We estimate that just in GE’s fleet, we reduced fuel consumption on gas-fired plants by $5 billion per year.”
Predix adds another entry to the list of tech acronyms: PaaS, for platform as a service. GE knows plenty about how the kinds of equipment it builds, such as turbines, works; but in opening up Predix to customers, it’s not figuring out how other gear, such as mail-sorting machines, function. It’s just selling access to the technology platform–a system that can gather gobs of data, sort it, and allow advanced database queries to manipulate it at high speed. It’s up to clients who run the machinery to figure out what data to collect and how to analyze it.
In the case of an airline, for example, each flight generates hundreds of gigabytes of data–way more than a black box collects–such as cabin temperature, cabin pressure, and vibrations. “The data is there, but I don’t know how people are going to utilize the data,” says Kodesh. That’s why he can’t promise there will be a system to predict crashes. It’s not up to him. The airline might process data for any number of reasons, such as to improve baggage handling, he says.
The first company to try out Predix after GE was Pitney Bowes, maker of massive mail and parcel sorting machines. Kodesh calls Pitney Bowes an alpha customer and says it helped figure out how other companies could use the platform. (Oil giant BP, infamous for its mechanical glitches, has also signed on to use Predix.)
Pitney Bowes is now able to estimate when a machine is likely to break down and schedule a preemptive maintenance call. “You want to know what kind of signals you are going to have when the performance is going to deteriorate,” says Kodesh. “You ask the devices, the last time you failed, two weeks before that, what kinds of signs did you have?”
One thing GE does have to figure out for customers is how to measure the savings that Predix makes possible in order to show if the service is worth it. (GE might even set pricing for Predix based on these measures, says Kodesh.) In the case of an airline, that might be, “how many more flights, how many more passengers, how much less fuel,” says Kodesh. “We have to develop those metrics so we can report to you what is the business outcome.”