Thank you Tina(1) to have invited us at the NPF conference for Reservoir Characterization which was last Tuesday and Wednesday in Stavanger. I was very honored to have the opportunity to detail the iQC solution to a panel of reservoir specialists, but at the same time scared about the audience. NPF reservoir conferences are about real geosciences. NPF Reservoir Characterization is a meeting where the geoscientists come with their well cores (not core pictures or measurements but real rock!) and group together around them to explain funny event such as how an early dolomitization of their carbonate reservoir has created some porosity and how a second wave of dolomitization has destroyed it except in some places. They touch the rocks, spit on them for a better view, smell them (they say that dolomite smells as a seal, I cannot confirm since I have never smell a seal ...) and then move to a core of sandstone drilled few meters lower where they can observe how a fluvial sand deposit was re-mobilized by some tidal events 300 million year ago with drastic consequences on the grain size distribution, hence the reservoir capillary pressure, hence the today production!
Before to start my presentation, I was really anxious. How those hard-rock geologists will consider the details about the ways we build learning models for data management I have tried to explain in my 35 slides?
I should have noticed the NPF geoscientists are not only hard-rock geologists but also techno-enthusiasts. OK, they play with their core in a great Stavanger hotel conference room as kids with marbles in the school yard, but with hololens on the nose to integrate the piece of core they are handling with well images, logs and SCAL measurements. Another indicator which should have reduced my level of stress could have been the discussion they have to compare the advantages of drones versus laser to measure the outcrops they integrate as a derivative in their reservoir modelling. Well, NPF geoscientists are really in the XXIst century.
My presentation finally went well, I received plenty of questions afterwards about the ML algorithms we use in iQC and what is our experience in deep learning. After my presentation went some others also dedicated to the digital subsurface. All of them were excellent because based on very interesting case studies. One of the NPF geoscientist (David Wade from Statoil ) has submitted cuttings pictures to DIGITS, the Nvidia GPU training systems (developer.nvidia.com/digits) and got excellent results as well on clastics than carbonates: The CNN (Convolutional Neural Network) has exactly the same difficulties than a skilled well site geologist to distinguish between a packstone and a grainstone: the Dunham classification system seems as well ambiguous for both! I appreciated also the work done by Pr John Howell and the students of the University of Aberdeen who are populating the Safari db with outcrop measurements to be used in any type of reservoir modelling, ML models including. Interesting to notice that like for the image-net db frequently mentioned in the NPF conversations this week, social media can contribute to the DB, so if you have an outcrop in your backyard, you are expected here! At Halliburton, Ben Saunders and al have used a ML to source the continental origin of the clastics of some Barents Sea formations and their results make sense.
Is it because I have been involved in the detection of seismic facies in the past with Stratimagic? I have to admit that the numerous results presented Peter Bormann from ConocoPhilips about the use of CNN (Convolutional Neural Network) and DNN (Dense Neural Networks) to detect faults or seismic geo-bodies let me without any voice. The CNN he has used not search for seismic facies using the features of traces but really seach for 3D objects using 3D features within the full cube.
He opened also some new perspectives for seismic resolution improvment using AI the same way than new AI tools can make your low resolution photos great again (a paper about super resolution by nvidia here another here ).
The bird above comes from the Max Planck Institute here
All these presentations open new area for AI implementation in geosciences, but the more important may be the fact that geosciences have started their digital revolution: Not only the geoscientists continue to make very high level interpretations, but now they have started to train the machines to do as well. The Artificial Intelligence has really sub-surfaced!
(1) Tina Todnem is managing The Digital Subsurface Project at Statoil, she was also acting as a chair at the NPF reservoir characterization conference this week.