Big Data // Big Data Analytics
Commentary
10/17/2013
03:52 PM
Chris Murphy
Chris Murphy
Commentary
Connect Directly
Google+
LinkedIn
Twitter
RSS
E-Mail
50%
50%

Can Analytics Outperform The Machine Whisperer?

The Internet of things promises to spot industrial failures before a human expert could. Will you trust that data?

Thanks to analytics software and the Internet of Things, we're opening a new round in the never-ending bout of man vs. machine.

During a panel discussion this week at a General Electric Industrial Internet event (GE's name for the Internet of Things), the insight that got the most heads nodding came from Jay Neidermeyer, CIO of GE Aviation's supply chain and manufacturing operations.

Neidermeyer was describing GE Aviation's early efforts using GE Intelligent Platforms software for predictive analytics to anticipate when a CNC machine used to make airplane engine parts is about to break down, so that staff can do preventive maintenance and avoid delaying the delivery of a finished engine. On many factory floors, he noted, "there's the person who can predict the outage based on hearing the rattle." Neidermeyer said:

"It's hard to convince that person that this computer might actually have some insight that he or she doesn't. We're still kind of learning our way through that. I think our strategy is pretty simple -- let's find simple ways to demonstrate success. And by the way, maybe not even taking action on them yet. Just have the signals in place so you can say, 'I think something might be coming. Let's see if it's reality,' and let people observe an outcome together. The theory is that if some of these things work, it's going to start showing itself that the insight you gain from these methods is as good or frankly better than that person -- than the 'machine whisperer.'"

My sense is that GE Aviation is typical in being only in the earliest stages of using predictive analytics to understand when a machine or some other process might fail.

Global CIO
Global CIOs: A Site Just For You
Visit InformationWeek's Global CIO -- our online community and information resource for CIOs operating in the global economy.

Rather than applying true predictive analytics based on actual performance metrics, what I see companies doing most often is relying on averages. To use a car example, we're told to change our fuel pump at 65,000 miles because company research shows that the pump generally fails sometime after that point. That scenario is a lot different from changing the fuel pump because the real-time data tells us that pump is running poorly and will fail and leave you stranded on the roadside at 45,000 miles.

Averages won't be good enough to achieve GE CEO Jeff Immelt's Industrial Internet vision of guaranteeing customers no unplanned downtime on their critical GE equipment. It's the exceptions -- the pump that fails 20,000 miles too soon -- that cause unplanned downtime.

Predictive analysis based on actual operating data is one of the big promises of the Internet of Things, as sensors gather data from devices and analytics software allows constant monitoring for problems. Some technical barriers remain, including the ability to gather the right data on which to base decisions.

But Neidermeyer's example points to another barrier: Do you really trust the data? His example resonates because it's human -- we can picture that wise veteran who has seen it all before and doubts this newfangled data. But we're all machine whisperers, relying on our experiences to make decisions about the future, so any Internet of things initiative must prove its data is strong enough to overcome our past.

Comment  | 
Print  | 
More Insights
Comments
Oldest First  |  Newest First  |  Threaded View
D. Henschen
50%
50%
D. Henschen,
User Rank: Author
10/18/2013 | 2:19:48 PM
re: Can Analytics Outperform The Machine Whisperer?
Humans are very good at picking up on patterns. Think of the doctor who has sense of where the patient's condition is headed. Or the guy on the shop floor who knows the machine is going to go down when he hears a certain sound. But by the time the symptoms are pretty obvious, it might be too late to get preventative. One promise of sensoring is to pick up on the earliest warning signs that humans might not recognize quite so soon. The concept of learning from history and patterns of behavior is not new. The internet of things holds the promise of a computer-assisted, methodical way of putting the approach into practice where it can really pay off.
RobPreston
50%
50%
RobPreston,
User Rank: Author
10/18/2013 | 2:29:03 PM
re: Can Analytics Outperform The Machine Whisperer?
In many cases, predictive analytics won't replace human judgment and analysis, but augment it. I don't think I want computer software diagnosing my illness, but I would like to see my doctor drawing on all the data she can.
EmailZola
50%
50%
EmailZola,
User Rank: Apprentice
10/18/2013 | 7:21:00 PM
re: Can Analytics Outperform The Machine Whisperer?
A better if not 'the correct' answer here is "both" predictive analytics and the Internet of Things (IoT) in the grand vision of a cloud mediated info-sphere (CMI(?)) - as those of us blessed by or from more than one information revolution weathered can relate, since the steam drill challenges John Henry most by mimicking his behavior, the best evaluators are often cooperating practitioners, but once the respective learning curves align sufficiently for a repertoire transfer, the machines are stuck with the work tasks and the surviving human interest relates to managing the machines.

Chris' article mentions the fuel pump infant mortality scenario as favoring IoT over predictive analytics - it would certainly make a real-world driver happier - yet in actual practice, broader benefits inure as fault/exception data from the pump reaches the (above posited) CMI, where the faulted device's profile would be touched, perhaps updating manufacturing quality data sets used by predictive analysis.

It's also become clear to me recently that IoT oriented designs can prove valuable in project development long before there's data sufficient to drive predictive analytics, leading to 'preventive analytics'. Here an anecdote: what appeared to be a firmware bug in a www.cloudfridge.io alpha test turned out to be a failing 3rd party thermal sensor chip. I think the best "whisperer: would be compelled to do mechanical replacement to ID the problem, locally inconvenient, highly intractable in deployment.
Drew Conry-Murray
50%
50%
Drew Conry-Murray,
User Rank: Ninja
10/18/2013 | 9:11:42 PM
re: Can Analytics Outperform The Machine Whisperer?
I wonder about false positives (or false negatives, for that matter) whenever I hear about how great it will be to have sensors monitoring all these complex systems. More data doesn't always mean the right data. Imagine how quickly these sensors will get shut off if a few false positives halt production on a factory floor.
ChrisMurphy
50%
50%
ChrisMurphy,
User Rank: Author
10/21/2013 | 3:08:52 PM
re: Can Analytics Outperform The Machine Whisperer?
You make an excellent point. GE often cites the 1% example -- a 1% productivity increase in some factor (airplane fuel efficiency, for example) could save $XYZ million a year. You raise the flip side of bad data that costs that $XYZ million. There is a risk-reward analysis here for sure.
ChrisMurphy
50%
50%
ChrisMurphy,
User Rank: Author
10/21/2013 | 3:12:36 PM
re: Can Analytics Outperform The Machine Whisperer?
Interesting use case for IoT in product development and 'preventive analytics', one I haven't heard voiced. It's certainly true that catching a design flaw in development is the cheapest fix.
EmailZola
50%
50%
EmailZola,
User Rank: Apprentice
10/21/2013 | 5:38:28 PM
re: Can Analytics Outperform The Machine Whisperer?
I was going to relate to ACM that in the specific www.cloudfridge.io case, the failing sensor's presentation wasn't false (either positive or negative) per se, but an out of range value (a digital '11111111' string), apparently intended, but not that clearly documented, as a 'warning flag' - as an example of how IoT designs can among other benefits provide greater process transparency.

Similarly, I'd thought about mentioning that my own decade in computer manufacturing quality (when the US was still exporting) included stopping the line when a audit deviation was encountered, not as it were to "shut off" the sensor, but to replace it.

My enthusiasm along these lines was more than a bit chilled on the "all these complex systems" topic when the (English language) Fukushima report links found below reached me today:

http://www.osaka-gu.ac.jp/php/...

http://www.osaka-gu.ac.jp/php/...

Reading past all the blame-mongering is made clear that even the most advanced designs are vulnerable to human misunderstanding - perhaps the IoT redemption lies in lucid system semantics.
6 Tools to Protect Big Data
6 Tools to Protect Big Data
Most IT teams have their conventional databases covered in terms of security and business continuity. But as we enter the era of big data, Hadoop, and NoSQL, protection schemes need to evolve. In fact, big data could drive the next big security strategy shift.
Register for InformationWeek Newsletters
White Papers
Current Issue
InformationWeek Must Reads Oct. 21, 2014
InformationWeek's new Must Reads is a compendium of our best recent coverage of digital strategy. Learn why you should learn to embrace DevOps, how to avoid roadblocks for digital projects, what the five steps to API management are, and more.
Video
Slideshows
Twitter Feed
InformationWeek Radio
Archived InformationWeek Radio
A roundup of the top stories and community news at InformationWeek.com.
Sponsored Live Streaming Video
Everything You've Been Told About Mobility Is Wrong
Attend this video symposium with Sean Wisdom, Global Director of Mobility Solutions, and learn about how you can harness powerful new products to mobilize your business potential.