In biogeomechanics, which describes the mechanical responses to microbial-rock interactions and its succeeding alterations, there is complexity in the estimation and predictability of biological processes and biologically-altered properties of rocks at a greater scale which inhibits the upscaling of biogeomechanical properties and processes from laboratory-scale. However, the successful application of this emerging field of rock mechanics (biogeomechanics) relies on proper upscaling of treatment process of geomaterials with biological agents from a laboratory scale (core scale) to a larger scale (field scale) which could be achieved by adopting a machine learning technique. This work proposes a state-of-the-art machine learning (ML) approach to predict temporal biogeomechanical properties at a field scale. Four ML techniques of K-Nearest Neighbor (KNN), Artificial Neural Network (ANN), Decision Tree (DT), and Random Forest (RF), were adopted to develop our new ML approach for the prediction of biogeomechanical properties. Firstly, experimental tests were conducted to obtain time-lapse biogeomechanical properties [microbially altered Uniaxial Compressive Strength (UCS) and Poisson's ratio (ν)] of shale and carbonate formations at core- and bulk-scales, and subsequently, these core-scale experimental data (428 datasets for shale and carbonate) were utilized to predict the field-scale biogeomechanical properties. Further, we compared and analyzed the ML-predicted biogeomechanical properties. There is a high degree of correlation between the bulk-scale biogeomechanical properties obtained from uniaxial compression tests and the ML-predicted field-scale biogeomechanical properties. The most accurate results for carbonate formation are produced by the RF model (UCS: R2 = 0.9613; MAE = 6.15 MPa; MPE = 2.62%; VAF = 96.16%; a20-index = 0.9091), whereas for shale formation is the KNN model (UCS: R2 = 0.8576; MAE = 5.41 MPa; MPE = 0.65%; VAF = 85.82%; a20-index = 0.9841). This study provides a novel potential for predicting the changes in rock mechanical properties due to biologically-induced processes at multi-scales (micro-, meso-, and mega-scale). Further, this study provides the first insight and a robust predictive tool for evaluating biogeomechanical properties at field scales where there is limited or non-existent data to constrain geomechanical models and the design of target formation.
All Science Journal Classification (ASJC) codes
- Civil and Structural Engineering
- Geotechnical Engineering and Engineering Geology
- Geomechanical modeling
- Machine learning
- Machine learning methods