Autonomous robots traversing an industrial site, exploring places where humans can’t tread – it sounds like science fiction, but it’s possible thanks to AI-powered services.
In many industries, such as oil and gas, transportation and energy, ensuring asset and facility uptime, as well as safety and regulatory compliance, is non-negotiable. Companies can spend upward of $100 million every year on industrial inspections, which in turn can drive five times that amount in maintenance costs.
Many companies employ traditional time-based inspection practices with the expectation that choosing the right inspection interval will identify a problem before the probability of failure is unacceptably high. This approach often incurs exorbitant, but necessary cost, while bearing the risk of missing critical defects that escape the inspection window.
Using Deep Learning to Power Risk-Based Inspection
In many sectors, the shift away from time-based inspection has begun. Risk-based inspection (RBI) involves extensive data and probability calculations, combined with detailed assessments of the consequences of component failure.
Information fuels RBI, and a new generation of drones and other unmanned robotic technologies is helping collect the sensor and video data that can drive the calculations leading to more intelligent maintenance scheduling. GPU-powered deep learning combined with advanced analytics and robotic inspection is helping to deliver a service that more accurately predicts when maintenance is required.
At the heart of this process is NVIDIA DGX-1, the AI supercomputing platform helping learn and train from immense volumes of captured data. Using computer vision techniques, the AI system learns to detect faults and create heat maps of target areas and components to repair or replace, prioritizing based on calculated risk. This trained model is then optimized for field deployment, where an array of drones and other unmanned robots equipped with sensors and cameras collect data and video footage of sites being inspected.
Bringing the Data Center to the Field
The story doesn’t end on a DGX-1 in the data center. The industrial sites being inspected are sometimes located near the “edge of civilization” — places far removed from 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.
To address this challenge, data centers are needed in the field. This is now possible in the form of the NVIDIA DGX Station. It packs the power of hundreds of CPUs (literally racks of x86 data center servers) into a compact form factor that sips electricity by comparison. The DGX Station can be used for inferencing on the data closest to where that data is being created, while enabling model refinement as the data comes in.
Robots That Get Smarter, Saving Companies Money, Saving the Environment
With every second of captured data comes the opportunity to leverage deep learning to improve the model, retrain it on the latest information, and increase the speed and efficacy of robotic inspection across all field sites. In this way, a complete lifecycle of deep learning value can be created that continuously evolves and improves service to customers, powered by the constantly growing footprint of sensor and video data.
Estimates are that service like this can reduce industrial inspection costs by as much as 25 percent, while also ensuring that industrial sites are properly maintained to avoid escalating emissions and damage to the environment.
- Artificial Intelligence