Reynaldo Gomez, who manages Nvidia’s energy partner ecosystem, presented some poster child deployments of ‘GPU-accelerated computing in oil and gas’. This leverages the Nvidia GPU Cloud, a ‘free’ online code repository with containers and pre-trained models (BERT, ResNet 50, etc...) for deep learning and other HPC applications. Gomez cited Beyond Limits work on well placement optimization with reinforcement learning, Bluware’s InteractivAI (GPU powered deep learning-based seismic interpretation), Helin’s Red Zone computer vision-based safety systems for offshore rigs and Abyss Solutions’ semantic segmentation models for corrosion monitoring. Gomez also presented work performed by InstaDeep for Total on deep learning-based fossil identification in 3D micro CT scans which leveraged a ‘distributed 3D Mask R CNN’ developed by DKFZ, the German Cancer Research Center.
FuelTrust’s Darren Shelton opined that ‘almost all carbon/GHG reporting is using rough estimates, inaccurate formulas, and has significant duplication across the supply and distribution chains’. GHG traceability is ‘complex’. Enter FuelTrust’s ‘from source to smoke’ solution for end-to-end visibility, compliance and anti-fraud needs in the global fuel lifecycle. An AI-based ‘digital chemist’ tracks, predicts, and validates outcomes at every step in the emissions lifecycle. Results are ‘securely recorded on the blockchain’. ‘DNA tracing’ of marine bunker fuels is said to address regulatory compliance and mitigate frauds in quantity and quality.
Ananth Natarajan (Cybereum)
showed how ML can be used to improve oil and gas megaproject planning.
There is indeed room for improvement. ‘For decades, capital project
leaders have relied on practices that attempt to optimize individual
investments, such as a nuclear power plant, an oil refinery, or a
pipeline. Cost overruns approach $1.2 billion on the average project
(some 79% over budget) and delays run from six months to two years.’
Oil and gas megaprojects are not the worst, but still on average show a
23% cost overrun. The problem is down to three causes: unpredictable
project complexity, bias and over-optimistic estimates of cost,
schedules and ‘principal agent issues’ including overstatement of
benefits and ‘data hiding’. Megaprojects may suffer from the
‘impossibility of deterministic planning’. A variety of cognitive
biases may come into play from optimism, Parkinson’s Law, whereby ‘work
expands to fill the allocated capacity’, and ‘Students’ Syndrome’ where
work is ‘procrastinated to the last moment’. All of which and more are
elegantly summarized in the DesignHacks cognitive bias codex.
Natarajan traced the evolution of project management from the GANTT
approach of the last century through Pert, Primavera and MS Project.
The last few years have seen the emergence of the cloud, AI, and ‘reference class forecasting’
a Cybereum specialty. Another favored approach is AI-based network
analysis, said to be a ‘step forward from traditional PERT and critical
path methods’. These are illustrated on the dynamic graph-based web page from the Cybereum-backed Blockchain for Projects
site. BfP is described as a ‘Web3’ development that encodes projects
into a ‘responsive, committed, symbiotic organism comprised of all the
stakeholders, by the combination of incentive engineering and
cryptographic methods’!
Amardeep Sibia from Drishya.AI
presented his company’s ‘Artisan’ engineering digitalization
application. Artisan can be used to create a digital twin of a
brownfield asset. The toolset reads scanned engineering drawings
(notably P&ID diagrams), understanding engineering logic to create
the digital model. Artisan detects pipe naming conventions, symbols and
lines and prepares symbol-to-line associations for manual QC. The
system recreates CAD plant diagrams and builds 3D models of legacy
plant along with a master tag list and asset inventory. Poster child is
Shell Canada’s Peace River SAGD brownfield. Drishya recently joined
Scovan’s PadX partnership that sets out to ‘accelerate innovation of SAGD well pad design and execution through Western Canada’.
More from Energy Conference Networks.
© Oil IT Journal - all rights reserved.