If you've ever passed an oil well or a petroleum refinery, you may have noticed a steel cage climbing high into the air, with an ever-burning flame at the top, and perhaps wondered about its purpose. The tower, known as a flare stack, is used to safely vent and burn off flammable gases, preventing dangerous pressure buildups. These stacks need regular inspections that can require mandatory shutdowns that last for days, a high-risk vertical climb, and workers enduring extreme temperatures.
The widespread advent of unmanned aerial vehicles has addressed some of these issues; now, camera-laden drones can be used to take close-up high-resolution photographs of a stack. But the images they capture still must be carefully inspected for any signs of damage or anomalies in the structure that could be missed by the naked eye. Avitas Systems, a GE (GE 3.74%) Venture company, has partnered with NVIDIA Corp. (NVDA 4.14%) to use graphics processing units (GPUs) and artificial intelligence (AI) to improve that process.
The NVIDIA DGX-1 combines lightning-fast GPUs with AI software. Image source: NVIDIA.
Addressing a hazardous and costly dilemma
The costs associated with inspections in this industrial setting and others can be high, but such checks are vitally necessary for personnel safety and regulatory compliance. Even the slightest corrosion, hot or cold spot, or microfracture could be a sign of a potentially dangerous development.
These industrial inspections are labor-intensive and slow operations, requiring workers to manually collect data; the associated shutdowns result in production delays. The cost of industrial inspections exceeds $40 billion annually, and some companies spend more than $100 million on them each year. Additionally, these examinations are typically carried out on a scheduled basis, increasing the likelihood that a hazardous development will go unnoticed until the next review.
AI to the rescue
Avitas will use NVIDIA's DGX-1 supercomputer to build an artificial neural network -- an AI system rooted in deep learning. A sophisticated computer model powered by complex algorithms mimics the human brain's capacity to learn, and is taught to recognize patterns and distinguish differences. These AI systems can detect even the most minute changes or inconsistencies, outperforming their human counterparts at the same task. This technology has been applied to a variety of groundbreaking uses, including the ability to recognize cancer in medical images.
NVIDIA's DGX Station can process the resulting data in remote locations, without requiring a reliable internet connection. Industrial inspections can now be conducted by a wide range of drones and other unmanned robots armed with computer-vision technology, and a variety of cameras and sensors to collect the necessary images and data. The system can also replicate observation conditions by placing the drone or robot in precisely the same position every time, which enhances the ability to detect changes that occur over a period of time.
NVIDIA's DGX Station allows for AI computing at the edge. Image source: NVIDIA.
AI is driving this train
NVIDIA has been on a tear lately, as artificial intelligence has taken hold and is finding new uses every day. In its most recent quarter, the company reported revenue that grew 56% year over year, but even more impressive was the 175% growth in its data center segment, which houses revenue from AI. As more and more companies and industries awake to the potential applications of AI, NVIDIA stands to gain even more.
