An Artificial Neural Network Model for Predicting Microbial-Induced Alteration of Rock Strength

In bio-mediated geotechnical techniques, the estimation of microbially altered geomechanical properties has mostly been conducted at core-scale due to the complexity of in situ field-scale measurement of biogeomechanical properties. However, the successful in situ field application of this emerging field of geomechanics relies on proper upscaling of this process from core-scale to field-scale (reservoir-scale). Here, we developed and applied a machine learning (ML) algorithm (artificial neural network, ANN) to model and predict the reservoir-scale biogeomechanical-altered properties of shale and carbonate rocks. We first obtained experimental data of the core- and bulk-scale mechanical properties (uniaxial compression strength, UCS) of the core samples impacted by a microbial strain. These core-scale data were then subsequently used as input variables to predict the field-scale biogeomechanical altered properties. The results show a high degree of correlation between the ML-predicted field-scale biogeomechanical properties and the laboratory-obtained bulk-scale biogeomechanical properties in shales (11.2% mean absolute percentage error) and carbonates (13.5% mean absolute percentage error). In addition, the result shows that the degree of correlation in rock mechanical properties and new mineral precipitations may be higher with increasing pore spaces in the tested rock types. This study provides a first leap from the laboratory and core-scale investigations toward field-scale geotechnical and geo-environmental applications of biocementation and biomineralization by predicting in situ biogeomechanical alterations.