GE Venture, Nvidia Bond over Inspection AI

　　The movers and shakers in a host of different industrial sectors are awakening to the looming impact of artificial intelligence. Few companies, however, have solid strategies to cope with the way their business will change, nor do most have a clear idea about when and how to implement AI, and with whom to partner to make it happen.

　　Avitas Systems, a GE Venture, and Nvidia are taking steps to bring some clarity to the use of AI in the industrial sector. The companies announced Thursday (Sept. 7) a partnership to work together enabling AI in inspection services for the oil, gas and transportation industries.

　　Nvidia posted in its latest blog, "How do you send a human being to inspect a petroleum refinery flare stack — one that operates at hundreds of degrees and requires negotiating a high-risk vertical climb? The answer is you don’t."

　　Where and how AI is used

　　Presumably, drones and crawling robots will do those jobs. But where and how exactly will AI fit into such an inspection process for the industrial market?

　　We asked Alex Tepper, founder and head of corporate and business development at Avitas Systems, to break it down. “There are many different spots where AI will be used,” he said.

　　First, AI can develop “optimal flight patterns” for drones to collect data in images and video.

　　Further, AI can create 3D models of an “asset” — a transmission tower, pipelines or oil refineries. It can then layer “points of interest” on top of such 3D models to enable drones and robots to spot anomalies such as cracks or corrosion, thus automating the defect detection process.

　　AI also has powers to “fuse” sets of different sensory data, said Tepper, developing algorithms that will help with risk analysis and predict when the next inspection is necessary, for example.

　　Of course, lots of companies are beginning to use drones and robotics to remotely inspect industrial infrastructure. Naturally, AI helps map out optimized “repeatable” paths for drones.

　　The key word is “repeatable.” Tepper explained that the repeatability of a drone’s path, for example, can create an opportunity for deep learning. Ultimately, the benefits of AI are in “risk-based data collection” and “development of master algorithms for risk analysis,” he added.

　　Division of labor

　　As for the collaboration between Nvidia and Avitas Systems, which company does what? How’s this going to work?

　　Avitas Systems is tasked with providing subject matter experts and AI data scientists. Nvidia offers DGX-1 and DGX Station systems for AI training — involved in automated defect recognition.

　　“We provide tools — AI development frameworks and CUDA platform — which helps unleash the skill set of Avitas’ data scientists,” explained Jim McHugh, General Manager of DGX Systems for Nvidia.

　　Tepper added, “As we collect a massive number of high definition pictures collected by drones, we need a place to store and process data, and run models and algorithms. For that, we need a huge amount of computing power,” which Nvidia provides.

　　AI’s benefits to the inspection services industry are clear. AI can bring more accurate data-based risk analysis, compared to traditional time-based inspection practices, which can be expensive and which could miss critical defects that escape the inspection window.

　　AI's missing link?

　　But what are the limitations of today’s AI? Still missing, according to Tepper, is an AI ability to compute on the very edge device itself.

　　The goal, he said, is to run AI algorithms on the edge, inside a drone, for example. The drone, then, could change its behavior in flight — literally on the fly, he explained.

　　Avitas Systems does AI training in its data center using Nvidia’s DGX-1. It’s also pushing AI into the field, by using a van-mounted Nvidia DGX Station. “But we want to push it further — really onto the edge,” stressed Tepper. Obviously, the industry is not there yet.

　　DGX-1, DGX Station

　　No doubt, Avitas Systems is pushing the boundaries of AI and inspection services. Nvidia’s McHugh confirmed that “Avitas Systems is breaking new ground by bringing NVIDIA DGX Station beyond the deskside and into the field for the first time.”

　　As McHugh explained, it was Avitas Systems’s Tepper who approached Nvidia, asking if a DGX Station — designed for use in an office — could be ruggedized and used “on location.”

　　By “on location,” Tepper meant in the field, in 110-degree heat or on the frozen Nordic Sea. Of course, the DGX Station wouldn’t sit naked in the desert. It would be housed in a van, but still subject to conditions far more severe than an office.

　　Using the DGX Station on location is particularly important, because “the industrial sites being inspected are sometimes located near the ‘edge of civilization’” explained Nvidia. Facilities that require inspection are often far removed from a robust networking infrastructure. “The voluminous flow of data returning from field drones and robots is often too large to be fed back to the data center for deep learning inferencing,” Nvidia added.

　　Avitas Systems stores deep-learning models in an AI Workbench, a solution that can process inspection data in real-time and retrain models to adapt to new uses. Tepper said the plan is to expand the company’s AI Workbench capabilities using the new NVIDIA DGX Stations with Volta.

　　Asked how industrial companies sector are embracing AI today, Tepper said, “There is general excitement,” partly because of cost pressure.

　　As companies fly unmanned drones, they make inspection both safer and cheaper. More important, repeatable analytics enable Avitas to know exactly when to inspect and when best to initiate maintenance measures.

　　According to Avitas Systems, industrial inspections are, yearly, a $40 billion business. “Some companies can spend $100 million annually on inspections, and five times as much on maintenance,” according to Tepper. He believes his company can lower inspection costs by as much as 25 percent.

　　Avitas Systems was one of the new business projects created in GE Venture’s incubator. Fully funded by GE, it is now a 100 percent GE-owned subsidiary.

　　The company, however, declined to disclose either the names or number of customers it’s working with.