Edge computing is nothing new. General Electric has been collecting and processing data from jet engines and wind turbines for decades. But the dynamics at play in terms of matching the huge volumes of data generated in the field with the processing of data are changing.
The two key aspects driving edge computing at the moment are relatively simple, at least at a conceptual level, according to Colin Parris, who’s the CTO of GE Digital and VP of software and analytics research at GE: Demand for data processing on the edge is going up, and the costs are going down.
But that doesn’t mean that GE Digital can take an entirely standards-based approach to building edge solutions. Today, custom hardware still plays a big role.
For wind turbines, for example, GE Digital uses custom-designed application-specific integrated circuits (ASICs) to be able to run sensor data from windmills though production-optimization algorithms. The algorithms are extremely complex and must take various factors into account, including the mechanical limits of each individual windmill and the wake effect of the entire farm.
“The first wind turbine could disturb the wind such that less wind comes to the second and third turbines,” Parris explained to Datanami in a recent interview. “So in some cases, I’m going to take the first turbine, and I’m going to slow it down, so more wind hits the second or third [turbine]. As you optimize it like this, what you have is the entire fleet makes more energy.”
Calculating the optimal operational method is dependent not only the direction and velocity of the wind and the mechanical limits of individual windmills, but by the position of the windfarm as a whole. As conditions change, the algorithms constantly re-calculate the optimal position for the fleet, which requires a significant amount of computational heft.
GE Digital uses its own custom ASICs to run these algorithms on the wind turbines because conventional X86 CPUs are not up to the task, Parris said. The performance advantage of running algorithms closer to hardware, as is done with these ASICs, is needed not only due to the volume of sensor data on the turbines, but also because GE Digital may not know what protocols will be used in advanced, he said.
“As that sensor data comes in, it comes in as a time-series,” he said. “You want to process that very fast, whereas with a regular processor, you can’t process that fast enough, so the ASIC does that for you.”
This data-driven approach enables GE Digital to squeeze 2% to 3% more electricity out of windmill fleets on behalf of its customers, relative to no data-driven optimization. That translates into tens of millions of dollars per year per customer, Parris said.
The math behind sensor and processor selection gets even more complex for GE Digital when other factors, like remoteness and risk, are factored into the equation. Windmills operate in inherently dangerous and remote conditions, and the operator must balance the risks and costs of performing unexpected maintenance versus the added cost of additional sensors and faster processors (not to mention the higher odds of breakdowns and failures when using more sophisticated sensors and processors).
“In some of our earlier models that went out there, because these wind turbines last anywhere from 10 to 20 years, you want to ship stuff that doesn’t break often, so you may not ship the cutting-edge stuff,” he said. “You want to have a few sensors that are very reliable in key areas that help your algorithm. A lot of sensors doesn’t help you.”
In addition to measuring wind speed and direction, sensors on a windmill measure stress and heat in the turbines’ gearbox. If sensors indicate a problem, GE Digital can activate additional sensors to get more data. It uses drones to take video of the outsides of blades, which are analyzed for pitting and signs of stress cracks using computer vision algorithms. There is even a robot that can crawl down the length of a 100-foot blade to look for signs of wear that appear only on the inside.
As more windmills are built around the world–including the massive off-shore wind farms that are currently planned for the East Coast of the United States–the costs of unexpected maintenance go up, and therefore the number of sensors goes up too. More sensors and more data means more sophisticated algorithms, and therefore more processing power, too.
“When something breaks onshore and I go to it, if it’s a cat[egory] four or five problem, I may have to bring a crane and the cost of that is huge,” Parris said. “Think about when it’s offshore. The is crane is on boat. The cost is astronomical. So I’m to have a lot more data, I’m going to do a lot more computing out there to make sure nothing breaks on this platform. The data requirements there are massive.”
Edge computing will play a big role in driving energy efficiency projects all over the world, from big factories all the way down to individual homes. For example, GE Digital is using its edge computing solutions to help manufacturers manage the amount of carbon they produce.
“Because we’re dealing with this energy transition throughout the planet right now, what has to happen is you can’t wait for a centralized place to manage everything,” Parris said. “A factory may want to manage the amount of energy it consumes and the amount of carbon it produces. How much energy do I use in the smelter plant inside? How much energy do I use in a manufacturing process? How do I manage that such that I have the lowest possible energy consumption and lowest possible carbon production?”
On the home front, Parris foresees algorithms powered by edge computers helping to automate the energy usage patterns for individual homeowners. The transition to cleaner energy sources, such as solar and heat pumps, is already underway, particularly in places like California, where cities have banned natural gas connections on new homes. Balancing the local generation and consumption of energy with the state of the overall grid will be instrumental in optimizing the use of energy over space and time, thereby minimizing the carbon footprinted that’s needed to flatten the peaks and valleys.
“If you look at the UK, if you charge your electric vehicle between 12 and 3 a.m., you pay a lot less. Why? You have a ton of energy coming in from the wind at night, but the factories and all the homes are turned off, so that’s the cheapest price for wind energy,” Parris said. “So bit by bit, all of these things are adding up.”
Can Digital Twins Help Modernize Electric Grids?
Hitching a Ride to the Edge with Akamai
2021 Predictions from the Edge and IoT