International
Association
for Mathematical Geosciences
ENSG - Nancy Université
Student Chapter
Research
Our research topics can be divided
into four main themes: model generation and interpretation, model
restoration, stochastic algorithms and visualization algorithms.Model generation and interpretation
Model generation is a complex matter, since data gathered on the field
only offer a limited glimpse of the actual subsurface, making the
generation problem underconstrained.Caroline Godefroy is involved in adaptative meshing, where the mesh can conform to reservoir heterogeneities and be refined around well data.
David Riffault works with implicit horizons representation. He generates constraints to interpolate the implicit horizons according to a given folding style. The interpolation is then performed using the Discrete Smooth Interpolation method (DSI).
Fabrice Levassor deals with model interpretation using a neural network algorithm. The neural network is able to generate semblance functions of an input model, given a set of typical geological template models.
Restoration
Restoration aims at unfolding and unfaulting the current subsurface
structure, in order to rewind to the initial depositional state. This
can be used to check for the local geological validity of the models.Marc-Olivier Titeux looks for restoration in the presence of salt domes. This task is especially challenging since salt domes and surrounding rocks have very different mechanical properties.
Pauline Durand-Riard works with restoration of implicit horizons. Due to the mathematical properties of the implicit approach, her work avoid typical meshing issues which occur in complex geological environments.
While rock compaction is often ignored in restoration, the restoration results can be inaccurate. Florian Basier integrates depaction in the restoration process. His method works on both explicit and implicit restorations.
Stochastic algorithms
The goal of stochastic algorithms is to provide several realistic
models from user-defined rules. Stochastic algorithms are based on
random number generators, i.e. two runs of the same algorithm will
produce different outputs.Florent Lallier tackles with the problem of well correlation. On most correlations, a single so called 'most likely' time line is proposed. The goal of Florent Lallier is to integrate uncertainty on the correlation process, by finding correlation probabilities. He uses the Dynamic Time Warping algorithm (DTW), which is often used in speach recognition and bioinformatics.
Vincent Henrion is concerned with the generation of discrete fracture networks (DFN) and karst systems. During the DFN generation, fracture development is constrained with existing fracture data in order to mimick mechanical fracture propagation rules. For karst systems, a new distance-based approach called Oddsim has been designed.
Nicolas Cherpeau works with the uncertainty on topology of fault networks. His method allows for stochastic changes of fault branching hierarchy, and could then be used to simulate fault networks with different topologies (e.g. fault connections).
Visualization
Visualization algorithms are challenging in geoscience, since the
volume of data to be explored grows exponentially whereas users are
more and more demanding in terms of interactivity and easiness of use.Gautier Laurent develops a volume rendering algorithm dedicated to curvilinear grids, based on the widely used slicing algorithm. His methodology extracts slice as isosurfaces using the Marching Cubes algorithm. The slice are eventually preintegrated before rendering.
Thomas Viard is working on uncertainty visualization. His work explores several methods to convey a sense of uncertainty on geological models, on either petrophysical or structural models. Displays are meant to be clearly and immediately understood, while carrying more information than traditional images.