SafeSense, AI to go

Baker Hughes truck-mounted mobile edge analytics platform from Nvidia brings supercharged Street-View technology to the oil field.

Speaking at the virtual 2020 Nvidia global technology conference, Xiaoqing Ge unveiled Baker Hughes’ ‘SafeSense’, a truck-mounted mobile ‘edge analytics’ platform. The idea behind SafeSense is to collect and analyze video-recorded data from a Google Street View car-inspired device mounted on some of Baker Hughes’ 4,800 North American service vehicles. These already travel to remote production locations on a regular basis, making a great platform for in-the-field intelligence gathering. Ge suggested some use cases, HSE, road condition monitoring and scouting but there are undoubtedly many others.

The SafeSense device comprises a roof mounted housing for cameras that provide a 360° view of surroundings which are captured in real time and processed with mobile ‘edge’ analytics. The process starts with assisted machine learning with a human-in-the-loop approach. This trains the system to recognize oil country facilities and activities such as oil trucks, flares, drilling rigs and tank farms. For many applications the system bests satellite monitoring offering better resolution and more frequent coverage.

Training is performed on a high-end Nvidia DGX system using the SSD (single shot multi-box detector), MobileNet and TensorFlow/TensorRT inference engine. The resulting models are transferred to an Nvidia edge GPU computer (a Jetson TX2). A video analytics pipeline builds on the Nvidia DeepStream API, a component of Nvidia’s Metropolis platform for ‘transforming pixels and sensor data to actionable insights’. Frames captured from DeepStream are stored in a NoSQL database for further analytics and visualization.

Work on Safe Sense started in 2018. Generation 3 rolled out in 2019 with 6 2K resolution CSI cameras in the unit analyzing 20GB/hour of imagery in real time. A five-truck pilot in the Permian Basin covered over 17,000 miles and gathered over 3TB of data.

Ge thinks that the SafeSense infrastructure and pipeline for deep learning has broader application to other ‘visual domain’ problems. The neural net backbone can easily be swapped for other deep learning tools and the results are exportable as JSON files. Baker Hughes envisages various business models for SafeSense, selling the data, software or derived intelligence to customers in the energy industry, investors or regulators.

Comment – Safe Sense brings the kind of AI used in remote (satellite) sensing down to earth. To get an idea of the power of the technology, read our report from the Ersi EU PUG elsewhere in this issue where we quote WoodMac’s Stephen Bull who observed (in the context of satellite imagery), ‘If you can see it you can train a model’. Makes you think…

Click here to comment on this article

Click here to view this article in context on a desktop

© Oil IT Journal - all rights reserved.