“If you build enough infrastructure, build enough compute, build enough APIs … data scientists can do magic.”
Data is the currency of the Internet of Things. The ability to predict the outcomes of that data for contextualized outcomes will eventually transform the infrastructure of the world.
Modern machines are now generating self-aware data on a continual basis. Engineers, developers, scientists and researchers are just now beginning to build the processes that will allow people to harness that data and turn it into automated, contextual and powerful decisions.
The problem is most industrial-based enterprises have, as yet, failed to appreciate that data provides answers to potential problems and insights into how more efficient their assets could be. It then follows that if data makes an industrial asset more efficient, the logical step is to take that data, apply it to other machines doing the same job and limit the chances of something going wrong at another time in a different location.
Eliminating Points Of Failure With Predictive Analytics
Himagiri Mukkamala, the head of engineering for the Predix industrial analytics platform at GE Digital, believes that the ability to predict when something like a jet engine turbine will fail is the next step for the Internet of Things. According to a GE Digital document, investment in the Industrial Internet of Things will exceed $60 trillion by 2030 with 50 billion assets (machine parts, sensors, monitors, appliances etc.) attached to industrial machines connected to the Internet within four years.
“This is the vision of the Industrial Internet of Things,” said Mukkamala, in an interview with ARC at the GE Digital campus in San Ramon, California. “To merge physics-based algorithms with data learning and machine-learning algorithms … and do a better job of predicting when that blade might fail.”
In the past, the scale and the complexity of asset failure data would be overwhelming for anyone without a serious scientific or engineering background. Analyzing the information took time and often led to an asset being mothballed while the cumulative damage of potential failure was assessed.
Simply being aware that a blade in a jet turbine might have reached the end of its lifecycle was not enough. Companies could not take the risk that it might fail at an inopportune time … like mid-flight over the Atlantic ocean. Engineers and data analysts would apply the laws of physics and existing coefficients to a potential problem and then make a decision based on what they found.
The analysis and decision-making could sometimes take anywhere between six months to a year. In a modern and connected society, that is very slow. Technology and software surrounds us, underpins almost everything we do … and data is the glue that holds it all together.
One of the key aspects of predictive analytics is appreciating what the data actually means, said Mukkamala. An oil pipeline could have a crack that affects the flow, while a truck driver who displays erratic driving patterns while on the road may have fallen asleep behind the wheel. The Industrial Internet of Things gives data scientists and analysts a better idea of what is happening at any one time … which means that the scale of the data is different to consumer-facing devices.
The Potential Is Not In The Things, But In The Internet
The Internet of Things is flooded with devices that generate data. For the most part, the device itself is not that smart (though that is starting to change). Fitness trackers monitor health and fitness. A smart water filter can monitor how many gallons of water have been filtered. True value does not come in the raw data, but the contextual decisions and automated decisions made from that data.
Fitness trackers help people make healthier decisions. A smart water filter can order its own replacement.
The Industrial Internet of Things is all about assets. Planes, trains, automobiles, wind turbines, power plants, factories, oil rigs, health monitors … and all of the tiny modules and sensors that go into these things. All of the potentially connected devices can provide insights into how an asset has performed in the past and how it can be optimized in the future.
Mukkamala cited the example of a standard freight train.
In the United States, trains travel across the country at an average speed of 20-25 miles per hour, said Mukkamala. GE designs and builds these assets to run at 70 mph. Trains have over 200,000 parts and 250 sensors. GE Digital can now leverage the information that those sensors generate. If you improve the speed of the entire train system in the U.S. by just one mile per hour, freight companies can save over $2 billion per year. Add to that the potential to transfer material from door to door and an optimized—and connected—asset makes perfect sense.
If we highlight fuel savings specifically, a 1% industry-wide improvement in fuel efficiency across North America could equate to $140 million in annual savings.
“We call this the Internet of Real Things,” he said. “If you look at the things we are trying to control and analyze—locomotives, jet engines, wind turbines etc—then if you take away the tech part of it, it comes down to what outcomes are we trying to deliver. Controlling or getting the data from Fitbit is fine … but if you are trying to reduce the planned downtime for a jet engine … the scale of the data and the analytics are very different.”
Let’s just think for a moment about how much data a fitness tracker produces. According to Mukkamala, the majority of fitness trackers produce around one gigabyte of useful data every day. Compare that with a wind turbine that generates at least one terabyte of real-time data on a daily basis. The ability to extract actionable information that leads to an outcome is very different.
“How do we connect to these assets?” said Mukkamala. “If you connect to a Fitbit or something, the protocols are pretty standardized … newer devices, newer assets. [GE’s] industrial assets often go through a 40-year replacement cycle. What this means is that we have jet engines or locomotives that have been operating in different environments for 40 years so the connectivity and protocols are old school.”
The Platform Is The Service
The question is, how do you replace the current “break and then fix” model that has existed since the end of the Second World War and turn it into a “predict-and-prevent” model? The answer is to build a platform for developers that leverages the volume of data with the technology available, said Mukkamala.
When GE releases those cheeky “What’s The Matter With Owen?” commercials, this is what it is talking about.
In February of this year, GE launched Predix, an industrial cloud-based platform-as-a-service. Developers can build apps on the platform that turn asset operational data into actionable insights into how that asset performed in real-time. Assets operate in different environments and in disconnected locations—underwater pipelines, for example. Predix was developed by GE Digital as a standardized platform for industrial companies to build systems that not only reduce the chances of failure but scale up when necessary.
Predix has only been available for a couple of months but GE’s experience in the engineering sector has been valuable in cementing the concept that every company is a software company, Mukkamala said. Data scientists can now identify a problem with an asset, write an algorithm, deploy that algorithm and monitor the asset.
Developers visualize an outcome through a digital twin—imagine Tony Stark taking apart a holographic version of his inventions—and use algorithms to predict the results of that outcome. Once that has been analyzed in the cloud, an industrial fix or adjustment can be made in the real world.
The public cloud platforms (like Amazon Web Services or Microsoft Azure) do not currently support all the demands of the industrial sector. GE Digital created Predix to fill the gap that it believed was essential for the Industrial Internet of Things to reach its potential. In October 2015, General Electric’s CEO Jeff Immelt said in an interview with McKinsey that the company needed to be more like Microsoft and Oracle, adding that industrial companies were in the information business “whether they wanted to be or not.”
Industry Relies On Failure Prevention
Everything in the industrial world depends on an outcome, said Mukkamala.
As the volume of data increases, companies will have the ability to leverage that information for their own use cases and scenarios. Time equals money in the industrial sector. Reduced complexity for developers is essential to a platform-as-a-service product. Developers have full access to a framework of virtual models, configuration protocols, workflow tools and data management and analysis—many of which are open source and hosted in the Predix Cloud.
“The good thing is that if you understand that particular damage model and how it applies to, say, a wind turbine, then we can use that data and give that optimization to a different customer after de-identifying,” Mukkamala said. “By getting access to that data, we can make life better.”
A locomotive can topple over in extreme wind conditions. Imagine if a nearby wind turbine was connected to the same network as the train. The turbine could send wind data automatically to the train, allowing it to correct course to eliminate a potential derailment. In this case, the power of the mesh network created by the Internet of Things is not just conceptual but has very real and practical consequences.
The Internet of Real Things indeed.
A machine talking to other machines in the cloud without human interaction is often cited as a dystopian future. But the power of predictive analytics and automation creates a simpler system that has the ability to self correct itself. A connected wind turbine can relay information when it needs to adjust the pitch of its blade due to a change in wind direction to other turbines in the immediate vicinity, an efficient (and quick) means of maintaining power generation without human input.
“When you think about when a device is going to fail, it is a combination of the environmental context and the device or asset context,” said Mukkamala.
Predictive Asset Analytics Saves Time
Mukkamala declined to name individual industrial clients that have adopted Predix but said that increased access to demonstrated failure conditions would lead to more accurate predictions in the future. Once connected to the Predix Cloud, data is captured and stored in a multi-tenant community—all of whom have been vetted and screened before access is granted—which creates a library of use cases and industrial Internet information.
“That’s the beauty of the platforms that are available now,” he said. “If you build enough infrastructure, build enough compute, build enough APIs … data scientists can do magic and that is why the entire process of a year to determine an outcome has been brought down to a week or two. That is where our job gets interesting.”