Novel Full-Stack Treatments for Getting in Touch with Your Sensors
Zackary Flansberry, Quantum Tensor Gradiometry Lead at SBQuantum
An endeavour as difficult as geophysical exploration requires an extensive base of knowledge and know-how in order to deconstruct its results. Not only do cutting edge sensors leverage increasingly abstract physical principles, but the data acquisition tools have to be able to record more and more sensor inputs with increasing synchronization, processing and transmission requirements. And this is not even getting into the convoluted mathematical treatments that will translate the raw data into actual information of interest!
This makes it very natural to posit the following: in order for the machinery to reliably result in successful surveying, these three broad components — sensors, acquisition tools and interpretation tools — need to be developed as close as possible, with an ideal of an entirely self-contained pipeline.
Even before a survey, the acquisition tools should pre-emptively be designed to ensure that the layout will garner the necessary data for the interpretation tools to extract as much information as possible. As the collection happens, the output of the sensors should be visualized, analyzed and recorded into reports so that we already have a partial sense of the results. The data should be routed automatically to those intense, drawn-out post-processing algorithms that find a ruby in a mountain of rocks. Ultimately, the rule of “no surprises” is extended to the hardware: the quality of the knowledge derived from the analyses will be tied to the quality of the raw data to such a degree that the hardware development should consider the finer details of the analysis methods.
It then stands to reason that sensor companies should move towards developing data processing and interpretation tools of their own. At the minimum, this helps guiding the development of their products by chasing the right specs; for more ambitious outfits, it leads to the creation of a monolithic data pipeline that most efficiently brings you from the physical equipment to detection and classification. In this age of big data and of readily available processing power, we can even anticipate breaking the usual sequential paradigm as one could immediately feed the outputs of the analysis right back to the sensors in order to improve their raw output itself!
Keeping in mind the “full-stack” picture has recently sparked innovation in induced polarization (Seequent, 2015) gravimetry (Carbone et al., 2020), ground-penetrating radar (Sensoft, n.d.), hyperspectral imaging (Freitas, 2018) and muon tomography (Los Alamos, 2017) to only name a few measurement modalities. Even an age-old, established method like magnetometry is starting to integrate more of this philosophy to push its capabilities (USGS, 2014). Given how ubiquitous magnetic fields are, it can be argued that the conventional approaches to their recording leave a lot of their object identification potential on the table. Part of it stems in the inscrutability of its outputs: one of the advantages of the magnetic method is that you see everything, regardless of line of sight, but on the other hand, you see everything at the same time, unlike a camera that can focus on specific elements of the picture. The whole world is blended together.
By working on a self-contained Magnetic Intelligence Platform, SBQuantum is taking the bet that our magnetic world can nevertheless be distilled into the information you actually care about, as quickly as possible. Is this vague cloud of magnetic data indicative of the presence of a value-rich ore? How deep would it be? Where is this vehicle that we are tracking going? What model is it? The solution makes the raw data even more complicated than conventional methods to extract more knowledge, as we move away from scalar magnetometry (which acquires a single map of the overall strength of the magnetic field) towards vector magnetometry (which records the direction as well as the strength), and even tensor magnetometry (which records the direction, the strength, and the spatial rate of change of the magnetic field). Of course, this added complexity to an already dense dataset is only admissible if it is hidden from plain view and ultimately makes the outputs of the pipeline clearer, simpler and more insightful — all of which intrinsically require a “whole picture” approach to the hardware development.
And there you have it: the power of acquiring eight magnetic field maps instead of a single one is abstracted away in one accessible package that would make you think it’s really no big deal at all. Sprinkle in an eventual integration of the solution in your favorite cloud services — a topic you may soon hear about — and it becomes clear that such novel full-stack data treatments are the way to bring you as close as possible to the information of interest.
References
Carbone et al., D. (2020). The NEWTON-g Gravity Imager: Toward New Paradigms for Terrain Gravimetry. Récupéré sur https://www.frontiersin.org/articles/10.3389/feart.2020.573396/full
Freitas, S. (2018). Hyperspectral Imaging for Real-Time Unmanned Aerial Vehicle Maritime Target Detection. Récupéré sur https://link.springer.com/article/10.1007/s10846-017-0689-0
Los Alamos. (2017). Muon detector developed for subsurface borehole imaging. Récupéré sur https://www.lanl.gov/discover/science-briefs/2017/September/muon-detector.php
Seequent. (2015). Insightful Geophysics Real-Time Approaches to Guide Drill Programs. Récupéré sur https://www.seequent.com/insightful-geophysics-real-time-approaches-to-guide-drill-programs/
USGS. (2014). Investigations into Near-real-time Surveying for Geophysical Data Collection using an Autonomous Ground Vehicle . Récupéré sur https://pubs.usgs.gov/of/2014/1013/pdf/ofr2014-1013.pdf
