Engineers have long ruled the realm of heavy industry, but as machines get smarter and more tightly connected to the cloud, mathematicians and data scientists are gaining greater control over the world’s industrial infrastructure.
The Internet of Things is more about bits than atoms, and this is especially true of the industrial Internet. The industrial Internet’s things—jet engines, gas turbines, oil rigs—are components of high-tech systems and also of business processes. GE is leveraging its status as a major builder of industrial technology, user of information technology, and seller of operational and engineering services to foster and guide the development of the industrial internet.
The company’s initiative to develop the smart infrastructure and data analytics that will enable real-time monitoring and predictive maintenance of industrial equipment is dubbed Software and Analytics at GE (SAGE). The goal is to increase efficiency and reduce downtime. The potential benefit, according to GE, is $276 billion in savings over 15 years across the aviation, electric, healthcare, oil and gas and rail industries.
GE is ramping up its software and analytics capabilities with a $1 billion investment in industrial Internet applications, plans for a Global Software Center in California, and a joint venture with consulting firm Accenture aimed at improving airline and air freight operations by collecting and analyzing aircraft performance data. GE has also launched a pair of industrial Internet “quests” on the data analytics open competition platform Kaggle. The quests are to improve aviation and healthcare productivity.
A Quest for Value in Machine Analytics
Aviation is a particularly good fit for performance monitoring and predictive maintenance because fuel efficiency is a big part of air carriers’ bottom lines. “Fuel burn for airlines is over 50% of their operating costs,” said Gary Mercer, vice president and chief engineer at GE Aviation. Mercer discussed SAGE at the 2012 MIT System and Design Management Conference on Systems Thinking for Contemporary Challenges in October.
“The industrial Internet will dwarf what we think of the Internet today by a factor of 100,” said Mercer. “It’s big data in the biggest sense.” Building the industrial Internet is a challenge, and it will require more than just slapping sensors on machines, he said. “How do we get these machines smarter, how do we connect them, and how do we use the analytics of that data to drive customer value?” he said.
GE is putting sensors in nearly all of its products. Products like jet engines and gas turbines have 10 or more sensors, said William “Bill” Ruh, vice president of GE’s Software and Analytics Center. In an interview, Ruh said each gas turbine sensor generates 588 gigabytes per day. “If you think about 20 sensors generating that, which is about the number on a gas turbine, you get a big number. And then if you multiply that by about 12,000, which is our gas turbine fleet, you get a really large amount of data per day,” he said.
Today GE is using the sensors for real-time monitoring. Longer-term, the company plans to use the data to do predictive analytics. The idea is to collect and analyze data when problems occur and be able to identify when a system’s behavior begins to match the historical data. “That’s going to be the power of sensors and analytics tied together,” said Ruh.
GE’s customers will be able to use the analytics to improve their operations. The company is also planning to feed the data to its engineering teams to help them improve existing products and develop new ones, said Ruh.
Analytics and Infrastructure Challenges
There are two aspects to building the industrial Internet: data infrastructure and analytics. A good data infrastructure will allow companies to collect and store data. And once it’s captured it needs to be analyzed. “You have to have the ability to consume all the data coming out of [the sensors], because it’s orders of magnitude greater than any of the big data problems that people talk about today,” said Ruh.
GE has identified four aspects of the industrial Internet that need to be built up: equipment, advanced analytics, system platforms and business processes.
- Equipment: sensors with local processing capabilities and fast data transmission, built into new machines and retrofitted into deployed systems.
- Advanced analytics: algorithms for transforming data into information, and standards for integrating data from different types of machines and machines built by different companies.
- System platforms: shared application-building frameworks used by suppliers, OEMs and users
- Business processes: practices that include machine information in decision-making, and processes for monitoring data quality.
Another important aspect of building the industrial Internet is data center capacity. Data processing demand is more than doubling every two years, which implies a 40-times increase in data processing demand by 2025.
Even with an adequate supply of data center capacity, the industrial Internet can’t rely on shipping raw data from sensors to data centers, said Ruh. “You’re not going to bring everything back into the cloud,” he said.
Handling sensor data in real time will require some local processing, which means sensors will need to be based on system-on-a-chip technology that combines data storage and processing as well as collection and transmission, said Ruh. In 5 to 10 years, cars, jet engines and gas turbines will have more processing power than some of the largest of today’s data centers, he said.
One consequence of all this local processing of sensor data is that the Internet will have to evolve to better support distributed processing, said Ruh. Today’s Internet, aimed at consumers, is designed to support client devices that access data on backend systems. The Internet does support some distributed processing, most notably the various projects that make use of idle desktop computers. But applications that span millions of far-flung devices operating in real time is another matter. “The architectures we have today don’t work for this kind of model,” he said.
IBM, Siemens Projects Also Target Industrial Internet
IBM takes a similar view. The industrial Internet is one aspect of the company’s five-year-old Smarter Planet initiative, said Scott Hebner, vice president of cloud and smarter infrastructure at IBM. “You can’t just take all that data and centralize it on some backend IT system and then analyze it,” Hebner said. Infuse machines with intelligence, and data transactions span the whole infrastructure as well as the backend data center, he said.
IBM’s Pulse 2013 conference in March in Las Vegas is scheduled to include a Smarter Infrastructure track that will cover many of the technologies and practices IBM uses in developing the industrial Internet.
GE’s Europe-based rival Siemens is also turning its attention to the industrial Internet. The company is leading a European Union research project dubbed Internet of Things at Work (IoT@Work). The project is focused on improving communications between industrial machines and Internet technologies. It will aim to make assembly lines more flexible by making component replacement plug-and-play and by allowing production systems to respond autonomously to problems.
Ultimately, the industrial Internet will need to be both autonomous and responsive to human control. Today’s focus is on developing predictive analytics. The next generation of analytics will focus on machine-learning algorithms, said Ruh. But even as industrial Internet applications gain the ability to determine that they’re encountering new types of problems, they’ll need to be designed for human decision-making, he said. You have to keep the person in the loop — “and be able to inject this into a workflow,” he added.
Eric Smalley is a freelance writer in Boston. He is a regular contributor to Wired.com. Follow him on Twitter at @ericsmalley.