Major industrial companies are moving into robotic inspection services, combining their knowledge of infrastructure maintenance with drones, robots and artificial intelligence (AI) for automating the collection and analysis of data.
Honeywell has launched a commercial inspection service using Intel’s Falcon 8+ industrial drone and targeted at the utility, energy, infrastructure, and oil and gas industries. The Honeywell InView package includes the drone, pilot app and a web portal to help customers create standardized routines and crisis-response inspections, as well as providing data analytics.
A General Electric startup is taking AI into the field to automate and optimize inspection of industrial assets by drones and robots. Avitas Systems, launched by GE in June, has partnered with computing specialist Nvidia to develop AI for robotic inspection and data analytics.
Replacing time-based manual inspections of assets such as transmission towers and flare stacks with automated checks based on assessing the risk of defects developing is expected to save customers time and money as well as being safer, says Alex Tepper, cofounder of Avitas Systems.
Boeing subsidiary Insitu launched Inexa Solutions in May to offer commercial aerial remote-sensing services for markets including linear infrastructure inspection—surveying pipelines, power lines and railway tracks. Lockheed Martin is also targeting the linear infrastructure market, while Airbus Aerial was formed in May to bring together commercial satellite and airborne remote-sensing capabilities.
A startup formed by GE Ventures—which creates, incubates and launches new businesses within GE—Avitas Systems is offering inspection services to the oil and gas, energy and transportation industries. It uses drones, crawler robots and autonomous undersea vehicles to automate inspections.
Avitas Systems is using Nvidia’s DGX computing systems to run the AI algorithms it is developing for use in planning the inspection paths, processing the images collected, and for the data analytics involved in automatically detecting defects such as corrosion, hot or cold spots or microfractures.
Nvidia’s DGX-1 supercomputing workstation is being used centrally for coding and training deep learning algorithms, such as convolutional neural networks for image classification and general adversarial neural networks for labeling captured images.
Additionally, Avitas Systems plans to deploy Nvdia’s compact DGX Station supercomputing system locally with the robots to help recognize defects automatically at inspection sites. “We are passionate about doing AI not just in the data center, but pushing it to the edge,” says Tepper.
“Our long-term vision is for the robots to incorporate AI, so they change their behavior on the basis of what they are seeing,” he says. “We are not there yet, but we are pushing AI from the data center to the field.”
Unveiled in May, DGX Station was designed as a deskside AI supercomputer, but Avitas saw the potential to deploy the system in vans, says Jim McHugh, Nvidia vice president and general manager: “We shared our prototype with Avitas and they will soon get the full production unit.” Based on Nvidia’s second-generation Volta architecture, this has three times the performance, he says.
Avitas Systems is using AI to plan flightpaths for drones that optimize the collection of data at points of interest on assets such as pipelines and refineries. AI is then used to layer the images collected on a 3D model of the asset and to perform automatic defect recognition.
According to Tepper, anomalies that can be detected automatically using AI algorithms trained by asset-inspection experts within GE range from sensing fugitive methane emissions from leaks in pipelines to using change detection to locate trees encroaching onto pipeline rights of way.
Information from inspections then goes into a central database, where it is fused with data from past inspections, maintenance histories and other operational information. A different set of AI algorithms assesses the riskiness of an asset and plans future inspections based on that risk.
It costs $4 million a year to conduct routine time-based manual inspections of a medium-size refinery, says Tepper. “We can do it for $3 million by automating the inspections and optimizing where they use the inspection resources based on the data collected,” he says.
“We are pro-autonomy not just for safety, but because of the repeatability of the data collected,” Tepper says. “We can create an autonomous flightpath, then fly the exact same flightpath in six months time and compare the data side-by-side.”
Avitas Systems is already working with customers in the oil and gas industry on inspections from the upstream oil wells to the midstream pipelines and downstream refineries. Industry reception has been good, he says, boosted by GE’s July acquisition of oil field service-provider Baker Hughes.