TDWI | Training & Research | Business Intelligence, Analytics, Big Data, Data Warehousing

Putting IoT Analytics to Work

Microsoft's latest customer reference for its Azure Machine Learning service is a showcase example of IoT analytics at work.

• By Steve Swoyer
• May 18, 2016

It's a shame that Microsoft Corp. didn't play up the Internet of Things (IoT) angle in trumpeting its new customer win with manufacturer Jabil Circuit Inc. Make no mistake about it -- Jabil's use of Microsoft's Azure platform is a great example of IoT analytics at work.

To recap: Jabil is a global contract manufacturer that generated $18 billion in revenue last year. It's using Microsoft's Azure Machine Learning (AML) service to analyze the sensor data generated by the machines on its factory floors. The idea is that AML's predictive technology will help Jabil improve efficiency and reduce waste in the manufacturing process.

The manufacturing floor is home to assembly-line robots, spot-welding machines, milling machines, lathes, shrink-wrap machines, and scads of other equipment, andall of this gear is packed with sensors.

It used to be that these sensors were standalone things. They'd keep track of hours-of-use, max/min values, and other metrics. Increasingly, these sensors are connected -- the machines in which they're installed have physical network addresses and are capable of transmitting sensor readings, events, and other telemetry information.

This has created a kind of positive feedback loop in which machine manufacturers pack more and more connected sensors into their gear, and so the factory floor of today is home to tens of thousands of sensors.

All of this sensor information is grist for predictive analytics. One obvious use case is to predict impending part or equipment failures. Another is to optimize production yields and quality control. This is precisely how Jabil is using Microsoft's Azure Machine Learning, a relatively new service that's been available for slightly over a year now.

Jabil says it's using AML and other Azure services to capture, process, and analyze "millions" of data points from machines at all levels of its manufacturing process. The company claims it's able to predict machine-process slowdowns with 80 percent or greater accuracy.

Jabil says IoT analytics has also helped it reduce the frequency of "scrap and rework" manufacturing jobs. Prediction hasn't just been a boon to Jabil's bottom line; its improved manufacturing process helps it keep contract-manufacturing costs low. This gives Jabil a competitive advantage when marketing to prospective customers: it can offer lower prices and better production yields.

The Analytics of Things?

Such use cases are the reason industry luminary Tom Davenport, a senior advisor to Deloitte Analytics and a fellow of MIT's Center for Digital Business, calls IoT analytics the "analytics of things."

This isn't (necessarily) just another buzzword. Instead, it gets at the twofold promise of IoT analytics.

First, IoT is a source for telemetry data from connected devices. Analysis of this data can permit manufacturers to optimize equipment performance, optimize power usage -- both for individual machines and for the factory floor -- predict impending part or machine failure, improve ongoing (and future) product development efforts, and even target interventions to address machine misuse.

Here's a quick example: Willi and Lotte use the same lathe. Data from the lathe indicates Willi is harder on the lathe when he's using it than Lotte is. Willi will need to be retrained -- or reassigned.

This is the first-level use of IoT analytics. It doesn't reflect what's new and even transformative about IoT, however. That would be the second-level use of IoT analytics: the ability to use IoT data to enrich other analyses, from BI technologies (such as reports, dashboards, and scorecards) to advanced analytics practices.

Enriching BI analytics with IoT data primarily serves an informational purpose, e.g., "We record a sale as 'pending' until the shipment leaves the loading dock. We know that N sales are still pending because of equipment failures with our vehicle fleet."

An even better example: "Our analysis of IoT data suggests that the Turbo Encabulator on a proportion of our short-haul trucks has a higher-than-normal failure rate. We have confirmed this with its manufacturer, and we're working with them to have the parts replaced. Nevertheless, our fleet will be operating at 16 percent diminished capacity for the next three months."

In the past, a manager or executive might have had to pick up a phone (or dash off an email) to find out what was going on. Now she'll probably know before the event or anomaly even shows up on her dashboard.

IoT data will be used to enrich new product development efforts; it will likely play a role in projects to identify and develop new markets, too.

Imagine a collaboration between, say, Subaru and Starbucks -- or, better still, the connected marketing service with which Subaru has signed an exclusivity deal -- such that a person who's been driving on the interstate for four hours is offered a limited-time coupon good for a drink at the Starbucks coming up at exit 341.

If the driver (or one of her passengers) decides to redeem said coupon, they might be offered another limited-time coupon, this time for a gas station at the same exit. That evening, after eight to ten hours of driving, the same driver might receive limited-time offers for a meal, lodgings, and other amenities.

This isn't just a new marketing opportunity, it's an entirely new market -- one in which traditional automotive manufacturers and automotive parts suppliers will partner with digital marketers, among others.

If IoT analytics means just the analysis of data from connected devices, it's valuable, but it doesn't redefine how enterprises use analytics. If, by contrast, IoT analytics also means using IoT data to enrich advanced analytics use cases, it will be legitimately transformative.