R&D Projects
Current research projects focus on innovative methodologies for Digital Rock Analysis using Deep Learning and Joint Seismic Inversion with various subsurface realizations, enabling connectivity and reservoir facies probability analyses.
1 – Digital Rocks Analysis with Deep Learning
(LTrace and Petrobras)
The main goal of this partnership between LTrace and Petrobras, is identify pretophysical properties and lithofacies from CT and micro-CT with deep learning.
Goals
- Predictions of elastic properties and lithofacies
- Understand the relationship between micro and macro rock scales
- Workflow’s unification and uniformity
- Correlation between facies and petrophysical/elastic properties
- Generalized rock physics models to better describing reservoir heterogeneity
Tools
- Python 3
- TensorFlow + Keras + PyTorch
- 3D Slicer
Partnership
Petrobras/CENPES (Leopoldo Américo Miguez de Mello Research Center)
Publications
- EAGE 2019 – “Deep Learning for Grain Size Distribution Estimation in Micro CT”
- SBGf 2019 – “Deep Learning for Grain Size and Porosity Distributions Estimation on micro-CT Images”
- SEG 2019 – “Deep 3D convolutional neural network applied to CT segmented image for rock properties prediction”
Authors: Fernando Bordignon, Leandro Passos de Figueiredo, Rodrigo Exterkoetter, Bruno Barbosa Rodrigues (Petrobras) and Maury Duarte (Petrobras)
2 – Connectivity Analysis
What separates us from other seismic inversion methodologies is the ability to generate joint realizations of petrophysical properties together with facies in a single inversion technique. Our realizations are perfect for connectivity analysis and facies probability analysis, as well as a probabilistic reservoir volume calculation.
In addition to the seismic inversion tool, we are bringing state of the art geostatistical tools to a software package with several techniques to allow the posterior distribution sampling of the properties. With multiple subsurface scenarios, it is possible to perform an uncertainty analysis of the reservoir, volume and connectivity probability. Which are important information for risk analysis and decision making in the reservoir modeling and characterization.
3 – Joint Petrophysical Inversion
The methodology discussed above is a sequential approach for uncertainty quantification. However, recently works defend that the joint facies-properties inversion yield in more accurate results. In fact, our R&D team has proposed the Joint Bayesian Petrophysical Inversion with rock-physics-based prior model, and they conclude that the joint approach yields better and sound estimation of the subsurface properties.
- Joint Bayesian inversion of seismic data with rock-physics-based prior model for the joint estimation of lithofacies, elastic and petrophysical properties
- Joint prior distribution of elastic and petrophysical properties is defined by a Gaussian mixture model, where each component of the mixture model represents a specific lithofacies and it is associated to a specific rock-physics model
- The inverse solution is a joint posterior distribution of lithofacies, elastic and petrophysical properties, which is also a Gaussian mixture. The posterior distribution is computed by Monte Carlo sampling combined with geostatistical methods to incorporate spatial correlation of subsurface properties.
