MEASUREMENT AND PREDICTION OF P-WAVE VELOCITY UNDER ANISOTROPIC STRESS CONDITIONS FOR SYNTHETIC AND ACTUAL ROCKS

P-wave velocity is critical to reveal grain bonding and fluid type in rocks under subsurface conditions. This helps in building the geomechanical earth model of the subsurface formations, which is utilized to optimize well drilling and completion programs. Previous studies investigated the impact of isotropic stresses, temperature, pore fluids on the P-wave velocity. However, the impact of stress anisotropy has not been evaluated when the overburden stress is greater than the horizontal in-situ stresses for normal faulting regime environments. Therefore, the objective of our research is to investigate the impact of stress anisotropy on the P-wave velocity for both synthetic and actual rock samples.

The experiments have been conducted on cylindrical core samples, gypsum cement synthetic cores (i.e., homogeneous medium) and actual sedimentary cores including sandstone, limestone, and shale. The sample is placed in a temperature/pressure-controlled cell, and it is mounted by vertical and horizontal acoustic transducers for velocity measurements. Set of experiments has been conducted under anisotropic stress conditions, the confining stress equals to 0.7, 0.8, and 0.9 of the axial stress, at room temperature and 85 C. The results show that the temperature varies inversely with the P-wave velocity due to the thermal opening of microcracks. In addition, the P-wave velocity increases as the ratio between the confining stress to the axial stress increases due to the gradual closure of microcracks.

On the other hand, 2 supervised machine learning models have been utilized, multi-variate linear regression (LR) and artificial neural network (ANN), to predict P-wave transit time (i.e., inverse of velocity) from conventional well logs (i.e., bulk density, neutron porosity, gamma ray, and depth) under 2 scenarios. The first scenario is when both the bulk density and depth are included as inputs to incorporate the confining stress effect, while the second scenario excludes these 2 parameters from inputs. The results show that the prediction accuracy of the transit times is higher when the confining stress parameters are included as inputs to the models. In addition, the ANN model shows better extrapolation capabilities for test cases with inputs that have out of range values with respect to the training data.