Here's why we need to build a smarter IoT
High-voltage capacitors for power plants undergo a complicated manufacturing process. Metal coils are carefully wound around a core, baked to remove all moisture, and then injected with different types and amounts of industrial-grade oils. Product failure can occur at any stage of the process, potentially turning thousands of dollars' worth of materials into little more than scrap.
Now a network of thousands of tiny sensors placed throughout the production line is adding greater certainty to this manufacturing process. Measurements from these sensors can be correlated and analyzed in real time to identify potential failure conditions, and operators can be alerted to resolve problems before they become irreversible.
These sensors are part of the Internet of Things (IoT), a network of connected devices interacting over the Internet that is projected to grow to billions of devices over the next decade. Yet while these devices are eradicating product failures, the cost of transmitting data collected every second from hundreds of sensors to an analytics engine in the cloud or a remote data center can amount to millions of dollars annually in bandwidth and storage charges. In addition, millisecond responses can make the difference between success and failure of the manufacturing process, and the latency caused by round-trip responses from the cloud could spell the difference between a working product and costly failure.
“A year ago, the focus of IoT edge processing was simply getting data from endpoint agents into an edge gateway, normalizing the data, and shipping all the data to the cloud as quickly as possible,” says David King, CEO of Foghorn Systems, a developer of software for industrial and commercial IoT applications. But while the cloud excels as a computing environment for creating fleetwide predictive models by analyzing data from many devices, it’s insufficient for the real-time needs of many manufacturing processes and other industrial applications.
Connecting IT and OT to realize market opportunity
The need to react to changing conditions in real time is leading companies to process and analyze data in closer proximity to the devices out on the “edge” of their networks. "You don’t want to actually run those models in the cloud," King says. "The right approach is to provide real-time edge analytics and machine learning for monitoring, maintaining, and optimizing the performance of individual industrial assets while utilizing the cloud for aggregating the metadata from all the machines into a global picture.”
This is bringing together two fairly distinct disciplines—information technology (IT) and operational technology (OT)—to form what some have labeled the "intelligent edge." OT is the realm of programmable logic controllers (PLC), supervisory control and data acquisition (SCADA) systems, and other hardware and software that control the physical equipment and processes used in manufacturing, pharmaceuticals, utilities, drilling rigs, and pipelines. IT has long been a separate realm of business systems and data centers. The two disciplines don’t necessarily make a natural match, but edge intelligence is rapidly bringing them together to allow consolidation of business, engineering, and scientific insights.
A recent McKinsey report predicts that the IoT's economic impact could total as much as $11 trillion per year by 2025 in nine settings, ranging from homes to factories and commercial locations. “Even though consumer applications such as fitness devices garner the most attention, B2B applications offer far more potential economic impact than purely consumer applications,” according to McKinsey. “And even more value can be created when consumer devices are connected to B2B systems.”
In active pursuit of this emerging market are scores—perhaps hundreds—of companies, ranging from venture capital-backed startups to tech and industrial giants. IoT startups raised $4.3 billion globally in publicly disclosed equity funding last year, according to Marcelo Ballvé, research director with CB Insights. It’s not hard to justify that figure when you consider that General Electric’s industrial software and services revenue amounted to $5 billion in 2015 and was projected to grow to more than $7 billion in 2016, with its Predix cloud-based IoT platform at the center of a strategy aimed at generating a $15 billion business segment by 2020.
Ultimately, many industrial, commercial, and consumer applications may become interwoven as applications such as connected cars and smart cities forge a machine-human digital interface that delivers new capabilities and services ranging from predictive maintenance to guiding a driver to an open parking spot in closest proximity to the next appointment on their electronic calendar.
Zettabytes of data demand smart IoT
Gartner reckons 6.4 billion connected devices were in use worldwide in 2016, up 30 percent from 2015, and expects that number to reach 20.8 billion by 2020. Estimates of the volume of data generated by IoT devices vary wildly, starting at 1.6 zettabytes by 2020 and going much higher, even outrunning projected total data center traffic. But the burden of broadband, storage, and processing costs may be offset by the 5.6 billion enterprise- and government-owned IoT devices that Business Intelligence projects will be using edge computing for data collection and processing in 2020.
At present, most IoT data goes unharvested. For example, McKinsey notes that “only 1 percent of data from an oil rig with 30,000 sensors is examined.” Unleashing greater value is dependent on sophisticated analytics that can use historical data to create models to determine how to optimize equipment and processes, and how to spot anomalies before they turn into problems that produce temporary shutdowns or catastrophic failures.
But the volume of data involved can be staggering. In the transportation segment and in many remote locations where IoT serves a need, real-time operational data must be transmitted via satellite or cellular service back to cloud services or data centers. Even in environments that have access to fixed broadband, these data volumes are likely to require either additional infrastructure outlays or greater service provider fees.
The railroad industry in North America, with an estimated 26,000 locomotives in service in North American during 2015, demonstrates the potential and challenges. GE’s latest-generation Evolution Series Tier 4 road locomotive utilizes more than 200 onboard sensors that process more than 1 billion instructions per second and generate about 10 gigabytes of data annually. In the air, Pratt & Whitney’s next-generation PurePower PW1000G jet engines collect 5,000 parameters of data continuously throughout a flight, and P&W expects to be streaming 11 petabytes of data across the fleet by the time that engine reaches maturity in 2030, according to a spokesperson for the company.
While centralized cloud services or data centers have greater resources to apply analytics to operational data, create models, and determine best-case scenarios, the costs and latency involved in sending data on a round-trip cycle—up to a central resource to apply analytics and determine an action and subsequently back down to the device to initiate an action—may be impractical.
When dealing with locomotives, airplanes in flight, autonomous vehicles, and wind turbine farms, a millisecond may be enough time to avert a failure, whereas the second or more required to round-trip data between a device and the cloud may not. “There has to be kind of a filtering taking place at the network edge, so that what is really relevant, what really matters, is what gets transferred,” says Aapo Markkanen, a principal analyst at Machina Research.
GE’s locomotives employ an onboard GoLINC hardware and software platform that comprises a “mobile data center” incorporating processing, wireless communication, networking, video, and up to 8 terabytes of data storage. “Today’s locomotives are generating more data than ever and far more data than can be economically off-boarded—hence decisions have to be made concerning the storage and extraction of what matters most,” according to GE. A Predix-based rules processing engine enables configurable, logical processing of data on-board, off-board, or at both locations.
These innovations are at the core of what GE envisions as a future of “self-aware trains in a smart ecosystem” that can utilize onboard sensors, track sensors, and cameras to improve performance and avoid failures. Small efficiencies can add up to large savings for the U.S. rail industry, says GE, with a 1 percent reduction in wait times at stops worth $2.2 billion and a 1 mph increase in speeds valued at $2.5 billion.
Achieving semi-autonomous IoT
The key to getting the most out of IoT devices is giving them some level of capability to compute, store, and analyze the data they generate, notes Craig Partridge, director of data center platforms consulting at Hewlett Packard Enterprise (HPE). “The intelligent edge has to have some kind of semi-autonomous capability or capacity,” he adds. "That’s especially true with solutions that are designed to handle high-bandwidth content."
Partridge's organization worked with a major city's transportation network to deploy real-time analytics for facial recognition to understand patterns of behavior and spot anomalous behavior. Drawing on experience from the program, Partridge explains a fundamental weakness of relying only on a cloud-based analytics system: That real-time feed of video data from every point of the city's transportation network "can’t simply be transported all the way back to the core in order for the core to apply analytics and return a result back to the edge in order for some action to be taken."
Typically, video monitoring systems archive data for post-processing and analysis after an event has been detected. IP-based cameras feeding into HPE’s IDOL Media Server running directly on hardened HPE Edgeline IoT Systems are able to perform rich media analytics directly at the edge. These edge capabilities enable security operators to focus on activities that require immediate attention from a security professional, such as potential threats or illegal action. That type of video recognition system can also be used in retail locations, utilizing an edge device to recognize the faces of repeat visitors, such as high spenders or known shoplifters, and notify shop floor staff of the need to act accordingly.
Many IoT devices will be generating streams of data that rarely deviate from the norm and add little value to a cloud-based analytics algorithm—storing information on every revolution of every wind turbine in a wind farm, for example, would produce huge volumes of essentially useless data. The deviations help fine-tune models and provide indications that a piece of equipment or a process is headed for a failure or suboptimal result. Edge analytics enable an automated system to respond quickly.
Software giant SAP is integrating IoT with its HANA cloud platform to help pharma customers achieve greater consistency in the drug manufacturing process. Pharmaceutical companies want every pill to be exactly the same, and quite often they will analyze how their machines are working to determine whether quality standards have been met, says David Jonker, SAP's senior director of predictive analytics. “We now do analytics in real time," he says. "As medicine is being made, we can predict how we’ll compare to a ‘golden batch.’”
Models created with SAP BusinessObjects Predictive Analytics and HANA Cloud Platform can be pushed down to the device or site level for real-time control. At the device level, embedded SAP SQL Anywhere and SAP Stream Processing Engine software can monitor, analyze, and store data locally as well as synchronize some or all of that data to upstream systems. Localized versions of enterprise data center or cloud applications can run transactions locally when connections are unavailable and subsequently synchronize with centralized resources when a connection is reestablished.
Keep on trucking
In 2013, the U.S. transportation system moved a daily average of about 55 million tons of freight valued at more than $49.3 billion, according to the U.S. Department of Transportation. Trucks accounted for more than 67 percent of the weight and 65 percent of the value of all shipments annually. Freight movement automation savings could add up quickly with the elimination of driver downtime and greater efficiency in determining driving routes and speeds via real-time processing of sensor, location, weather, and traffic data.
Autonomous cars are all the rage these days, but trucks are far more interesting, says Mike Krell, lead IoT analyst at Moor Insights & Strategy, a high-tech analysis and consulting firm. “The efficiencies you can get from trucking and changes you can make to that industry will probably save people a lot more money and be a lot more effective and efficient than simply automating your car or my car,” he says.
Otto, a startup acquired by Uber Technologies in 2016, is in the business of building kits to enable self-driving vehicles. Late last year, Otto demonstrated its technology with a 120-mile beer delivery trip. Camera, radar, and laser sensors mounted atop a truck communicate with an onboard computer system and software “that make real-time driving decisions based on those sensors” and maintain the truck’s safety systems, the company says.
Maintaining safety at the edge
When talk turns to autonomous vehicles, connected devices, and surveillance video, security and privacy issues inevitably come up. “On a macro level, what worries me about IoT is that there is a bit of a headlong rush into doing it, building connectivity into so many devices,” says Josh King, who spent a decade working in the wireless industry and is now chief legal officer at Avvo, an online marketplace for legal services.
“If manufacturers—who often do not have a competency in software—overreach, it is inevitable there will be security breaches or abuse of privacy rights,” says King. “And if they take a cavalier approach, something bad is going to happen and then there is going to be a regulatory overreaction.” Nonetheless, he says his instinct is to “let people test this out, and hope they learn and iterate quickly.”
Markkanen of Machina Research argues for not letting fears get in the way of IoT investment.
“The technology to secure IoT deployments is already there, especially in the embedded computing space [and] industrial automation," he says. "It’s pretty advanced, but having top-notch security doesn’t help a lot if you still have basic human factor mistakes such as bad password management.”
Still, it’s probably safe to assume that companies will increasingly invest in the area of intelligent edge once they get their arms around the concepts. “When you talk about the edge, I definitely see this where cloud was 10 years ago," says Srdan Mutabdzija, global solution offer manager with Schneider Electric, a global supplier of energy management and automation products that provides self-contained micro data centers and other products for edge applications.
"Everybody is talking about it, but everybody is still trying to figure out what it is,” he adds.
The transformational force of IoT will force a realignment of traditional data management roles. CIOs must expand their scope beyond business processes to encompass and integrate the realm of OT and the management of physical things. Configuring, managing, and securing data at the intelligent edge represents new IT architectural challenges, along with massive business opportunity.
Intelligent edge: Lessons for leaders
- Database and analytics capabilities must be evaluated to see how they can support the manufacture and deployment of physical products.
- Getting a handle on current and future uses of sensors is essential for evaluating storage and bandwidth needs.
- Using analytics at the edge requires a new way of architecting centralized applications, whether those are in the cloud or in enterprise data centers.
- If not already in process, a crash course to align IT and OT is essential for determining roles and responsibilities, and agreeing on goals.