The Mapping of Clay Bearing Fractures from Well Logs with a Neural Network and Their Implications for Rock Mechanics

Knowledge of petrophysical and mineralogical parameters in a geothermal reservoir is essential for the estimation of rock mechanical behaviour during hydraulic stimulation. The strength of a rock is determined by manifold petrophysical parameters. Most parameters can only be indirectly measured via different logging techniques. Logging data represent the petrophysical parameters in a multidimensional way. Neural networks are well-suited to deal with datasets of such large dimensions. We describe a neural network (NN) based method to map clay bearing fracture zones indirectly from spectral gamma logs. Thus, a semi-quantitative synthetic log is created showing the clay content along the wells. Laboratory measurements complement the study with respect to the implication of clay appearance for the mechanical behaviour of the rock. It is shown that the NN method is suitable to create synthetic clay logs. Combined with laboratory mechanical measurements this tool helps estimating the response of the reservoir rock to changes in stress field or pore pressure.