NMR relaxation techniques are widely used for petrophysical characterisation [1]. Relaxation in rocks is defined by a list of physical properties such as bulk relaxation time, susceptibility contrast between solids and fluids, solid phase morphology, effective diffusion coefficients and hydrogen indices of micro porous regions. Although certain properties can be determined through measurements, others require sophisticated or destructive techniques (e.g. MICP). Recent advances in digital rock analysis enabled NMR relaxation simulation on rock digitized representation [2]. For a given saturated porous system measured and simulated magnetization decay and T2 distribution have to be identical. We propose a non-invasive approach which can extract a comprehensive set of physical properties by matching simulated and experimental T2 distribution curves using Particle Swarm Optimization (PSO) algorithm [3] -- a population-based stochastic method of which particles search for better positions in the search domain by changing its orientation and speed following the behavior of bird flocking. The particle swarm is iteratively and concurrently updated within the boundary of parameter values specified. The parameters are finally extracted when the optimizer finds the global optimum (minimum of the fitting residuals) for this nonlinear fitting problem. We also improved the speed of convergence and exploration ability of this algorithm by introducing a dynamic topology into our optimizer.
We apply this approach to a number of core samples: outcrop sandstones and carbonates and successfully deduce a set of petrophysical properties from the good fitting we achieved. Moreover, this algorithm is tested to be highly efficient in searching high dimensional, continuous solution domains. It is implemented for parallel computing which further improves its efficiency for as particles can be split up among a range of processors but sharing local and global best solutions.