Inside Advanced Scale Challenges|Saturday, November 18, 2017
  • Subscribe to EnterpriseTech Weekly Updates: Subscribe by email

Drones and GPUs Provide AI Boost for Inspection Firm 

Shutterstock_Christian-Lagerek

A company backed by GE’s venture capital arm is using NVidia GPUs, deep learning frameworks, and drones to automate the inspection of physical assets in the oil and gas, electrical distribution, and transportation industries.

Avitas Systems, owned by GE Ventures, today unveiled a business plan centered on saving industrial firms millions of dollars annually by improving how they inspect for flaws and other issues impacting their field assets. The plan also calls for lowering the risk to human workers by utilizing a fleet of drones, underwater submarines, and robots to gather data.

“We’re automating industrial inspection using robotics and advance data analytics,” said Alex Tepper, founder and head of corporate and business development at Avitas. “We’re doing this specifically for oil and gas, power, and transportation industries. In each of these industries, customers spend hundreds of millions of dollars on inspections and maintenance regimens for their infrastructure and equipment.”

Inspection is done manually in most of these industries, which raises the cost and also the risk. “They literally were having people walk around with sensors or ride in trucks looking up at power lines for example, or looking down from helicopters, or in a lot of cases, attaching themselves to a harness and ratcheting themselves up to the top of an asset to inspect it,” Tepper said during a briefing for reporters yesterday.

Avitas Systems plans to install NVidia’s DGX Station in a van to crunch drone data at remote industrial sites

Avitas is using the NVidia’s newest GPU platforms, including the DGX-1 and the DGX Station, for various data processing tasks, including: developing flight or operation plans for drones or robotic; crunching the data gathered by sensors on the drones and robotics; rendering three-dimensional models of the data; merging the field data with other data gathered; looking for anomalies in the data; and building and refining operational models of the assets that it’s monitoring.

“We’re dealing with massive amounts of data,” Tepper said. “So storing and processing that very large amount of data is a very difficult task, and creating models and algorithms off of that requires a huge amount of computing power, which is one of the reasons we partner with NVidia to partially solve that problem.”

The NVidia platforms are basically self-contained supercomputers that have everything needed to run AI workloads, including operating systems, middleware, GPU hardware, and the latest deep learning frameworks, such as TensorFlow, Theano, pytorch, Caffe2, and MXnet.

The plan calls for Avitas to use the DGX-1, which contains eight Tesla V100 GPUs and a 300GBs NVLink connection, in its data center for the purpose of refining and training the neural networks behind its inspection routines. The company aims to install a DGX Station, which features four V100s and sports 480 TFlops of processing power, into a van that it drives into the field to provide processing power at remote locations, such as for upstream oil wells or midstream pipeline inspection.

Avitas is also planning on making its drone- and GPU-powered inspection fleet available to oil refineries. “For a typical midsize refinery, they spend about $4 million per year on inspection,” Tepper said. “We can do a similar inspection for about $3 million, and the reason we can do that is because we have the ability to automate a variety of those inspections, and then optimize….the inspection and maintenance resources based on the data that we’ve gathered.”

A midsize refinery will spend $4 million annually on manual inspections, according to Avitas Systems (Christian Lagerek/Shutterstock)

While drones and robots will automate the physical inspection of assets and lower the risk to humans, Avitas’ AI framework will eventually allow it to predict which of its customers’ assets is the most likely to require maintenance. That information will allow the company to optimize its inspection routine, which is where the money savings comes in.

“What we’re allowing them to do is to think about the way they view inspection and maintenance in a fundamentally different manner,” Tepper said. “We [inspect] an individual asset…and then we string together the information from that asset with all the other assets, so that we can then help them allocate their resources in a much better way.”

Avitas, which started working on this system in 2016, is looking for erosion, hot spots, cold spots, and micro-fractures in oil and gas pipelines. It’s also building systems to detect fugitive methane emissions, to detect corrosion under insulation, and to detect right-of-way encroachments.

Jim McHugh, a vice president and general manager at NVidia, said he’s really excited about what Avitas is doing. “What Avitas has shown here is a great example of how [DGX products] can work, not just in the data center but out on the edge.”

Avitas is aware that it’s pushing the state of the art in edge computing. “In a lot of cases, the technology is not quite there yet,” Tepper said. “But we’re working hard, as are many others, to continue pushing the AI from the data center, where you have massive amounts of centralized computing power, to the field where it’s local and remote, but you have slightly less, to the edge where you have … space constraints.”

About the author: Alex Woodie

Alex Woodie has written about IT as a technology journalist for more than a decade. He brings extensive experience from the IBM midrange marketplace, including topics such as servers, ERP applications, programming, databases, security, high availability, storage, business intelligence, cloud, and mobile enablement. He resides in the San Diego area.

Add a Comment

Share This