mr-prognosticsairbuspromo.jpg.crop_display.jpg Airbus

GE Ramping Up Results-Driven Big Data Analytics

Next up: capturing and analyzing live data streams.

GE Aviation spent the last 18 months quietly shifting its Big Data efforts from whiteboards to results—delivering on moving its fleet diagnostics to a new platform and establishing itself as the poster child in General Electric’s shift to a “digital industrial” corporation.

Late last year, GE migrated the last of its engine families from a three- decade-old system to Predix, the company-built technology platform that helps its fleet-monitoring team capture more data and analyze it more quickly and accurately. Each of GE’s engines comes with a data diagnostics package, and about 35,000 of them send back data “snapshots” taken at key flight phases—such as during takeoff and in cruise, explains GE Aviation Fleet Support Executive Director Vijay Singh. A CFM56 snapshot has about 280 parameters, while the newest models have about 480.

Before Predix, GE relied on engineering brainpower to connect the dots and find hard-to-spot trends. While the capability to identify specific problems was there, the ability to quickly predict what might happen by using data from different engines was not.

“That’s why we are the first [GE team] on Predix,” says Singh. “It gives us a committed team of resources that combines physics and data science.”

Migration to Predix began in late 2014, with GE90 and GEnx monitoring. One of the immediate benefits was multivariate analysis, Singh says. For example, an engine operated in a hot, sandy desert environment will exhibit different baseline parameters than one operated in less harsh conditions. Predix factors this in, building specific profiles for each individual engine and adjusting its alert-generation parameters as a result.

“We use the additional variables to filter out non-relevant information,” says Greg Coons, director of Customer Portals and Contract Fulfillment, GE Aviation. “We’re able to normalize the data using the additional variables, making the analysis more accurate.”

GE’s fleet generates about 100 million records per year. With the final migration of engine families to Predix at the end of 2015, the platform is now first in line to receive them. Predix analyzes the data and where its sees an anomaly issues a Level 1 alert to technicians in one of two fleet-support centers, in Cincinnati and Shanghai. Level 1 relies on analyzing historical data, Singh says. Unresolved issues go to Level 2, manned by fleet- monitoring team product-line experts. In the rarest of instances—0.5% of all cases—a Level 3 alert is issued, which loops in the product engineers who design and build the engines.

The outcome is either nothing—meaning there is no issue—or a customer notification record (CNR). In 2015, the system generated about 350,000 alerts resulting in 9,000 CNRs. The CNRs were “86% accurate,” Singh says, meaning that nearly nine out of 10 identified issues needed attention.

While 86% may sound low for an industry used to 99%-plus reliability rates, Singh notes that data is different. “This is not hardware. We’re trying to catch things we may not know about before they happen,” he explains. “There are chances that we will always take to detect those unknown unknowns.”

Under Predix, the figures are improving. CNRs are up, but “false” CNRs are down, meaning the platform is finding more issues while flagging fewer non-issues. It is also better at predictive analytics.

One example is GE90 oil consumption traced to a v-nut that worked itself loose on an oil feed-tube. Using Predix, GE’s data analysts built a process that compares pre- and post-flight oil levels, adjusting for longer flights that consume more oil. The goal: Improve the default alert that shows when oil consumption exceeds a value—often too late to avoid a service disruption—and give operators up to 10 days of lead time to fix the issue. “It’s detecting a slow increase that otherwise might be missed,” says Coons.

The process has led to more than 100 CNRs and many grateful Boeing 777 operators.

Predix focuses on data snapshots to address issues. The next step: analyzing live data streams and becoming more predictive. GE has about 30 customers providing “continuous engine operational data” (CEOD), or streams of raw information. Right now, bandwidth limitations mean that most of it is generated only when parameters are exceeded, and the data are offloaded post-flight. As connectivity expands, data will be available live, making preventive analytics routine.

“Predix will give us the capability to analyze CEOD online,” Singh says, noting that the platform’s next version is in development. “Predix 1.0 is the tip of the iceberg on the data-analysis side.”

Predix is core to what GE CEO Jeff Immelt calls the company’s “digital industrial” evolution. The image of a Leap engine on the cover of the company’s 2015 annual report emphasizes GE Aviation’s leading role in the shift. Fleet monitoring’s early results underscore why GE is heading this way in the first place.

“Everything is becoming software-defined,” says GE Digital CEO Bill Ruh. “People will look and say: Show me how I get better uptime; show me how I get better fuel efficiency; show me how I get better safety. They are not going to buy a technology capability. In the end, they want to buy an outcome.” 

Hide comments


  • Allowed HTML tags: <em> <strong> <blockquote> <br> <p>

Plain text

  • No HTML tags allowed.
  • Web page addresses and e-mail addresses turn into links automatically.
  • Lines and paragraphs break automatically.