Content area
Abstract
This work is composed of 2 parts: a theoretical part, where we explore the links between geostatistics and finite elements, and a practical part, in which we apply a new method of approximation, the logarithmic finite element method, to Darcy filtration problem.
Theoretical part (appendix). Generally, two types of information are available on a physical phenomenon: (1) field law (partial differential equations); (2) series of measurement. The purpose of this work is to develop a method for the modelling of physical phenomena that will integrate in a single formulation those two types of information. To this end, we shall study the connection between two families of methods: (a) the finite element method which is used mostly for the numerical approximation of boundary value problems, hence to process deterministically informations of type 1; (b) geostatistical techniques which permit to analyze series of measurements in order to perform estimations on the phenomenon (random approach).
The theoretical framework which is presented in appendix has enabled to establish a link between the intrinsic random functions of order k (I.R.F.-k) of MATHERON (1,2) and finite elements. Then the notion of random vector modulo a vector subspace S is introduced. It permits to present in an unified manner (cf. DUCHON (1)) the theory of best linear unbiased estimator (kriging). When S is the set of constant functions, we obtain ordinary kriging. When S is the vector space of polynomials of degree $\leq$ k, we obtain universal kriging. At the limit, when S is a finite element subspace, the best linear unbiased estimation is given by the finite element expansion.
All these considerations have led to propose a new approximation and estimation method: the logarithmic finite element method (LFEM), which will be used for the numerical solution of Darcy filtration problem.
Practical part (application to Darcy problem). A new random finite element model is presented in chapter 2 to solve the direct Darcy problem. Permeability is represented by a stationary random function, the expectation of which being constant in each flow zone (random direct problem). The use of this model is valid only for porous media with slowly varying permeability.
In chapter 3, the logarithmic finite element method is applied to solve Darcy direct problem when permeability is constant in each flow zone (deterministic direct problem). This method is mainly interesting for boundary value problems with singular data, such as point sources or wells in the case of Darcy problem. We show also how to incorporate in this model waterhead measurements in a number of wells.
Finally, in chapter 4, we explore the solution of Darcy inverse problem when permeability is represented by a logarithmic finite element expansion. We are interested in a heuristic solution technique which permits to incorporate in the numerical model a maximum of physical informations on the porous medium: physical constraints, geological informations, permeability and waterhead measurements.
In this thesis, we intend to apply mathematics to practical engineering problems. Citing MATHERON (14) (introduction, page 12), we are fully conscious "that symmetric critics can be addressed from both praticians and mathematicians: for the first ones this work being unnecessarily abstract, for the others insufficiently rigourous." This problem is difficult to avoid. Therefore, in order to make the reading easier, the theoretical considerations have been presented in the appendix, so that the application to Darcy problem can be read independently.