Coupled experimental assessment and machine learning prediction of mechanical integrity of MICP and cement paste as underground plugging materials

Abstract

Compromised integrity of cementitious materials can lead to potential geo-hazards such as detrimental fluid flow to the wellbore (borehole), potential leakage of underground stored fluids, contamination of water aquifers, and other issues that could impact environmental sustainability during underground construction operations. The mechanical integrity of wellbore cementitious materials is critical to prevent wellbore failure and leakages, and thus, it is imperative to understand and predict the integrity of oilwell cement (OWC) and microbial-induced calcite precipitation (MICP) to maintain wellbore integrity and ensure zonal isolation at depth. Here, we investigated the mechanical integrity of two cementitious materials (MICP and OWC), and assessed their potential for plugging leakages around the wellbore. Further, we applied Machine Learning (ML) models to upscale and predict near-wellbore mechanical integrity at macro-scale by adopting two ML algorithms, Artificial Neural Network (ANN) and Random Forest (RF), using 100 datasets (containing 100 observations). Fractured portions of rock specimens were treated with MICP and OWC, respectively, and their resultant mechanical integrity (unconfined compressive strength, UCS; fracture toughness, K_s) were evaluated using experimental mechanical tests and ML models. The experimental results showed that although OWC (average UCS = 97 MPa, K_s = 4.3 MPa·√m) has higher mechanical integrity over MICP (average UCS = 86 MPa, K_s = 3.6 MPa·√m), the MICP showed an edge over OWC in sealing microfractures and micro-leakage pathways. Also, the OWC can provide a greater near-wellbore seal than MICP for casing-cement or cement-formation delamination with relatively greater mechanical integrity. The results show that the degree of correlation between the mechanical integrity obtained from lab tests and the ML predictions is high. The best ML algorithm to predict the macro-scale mechanical integrity of a MICP-cemented specimen is the RF model (R² for UCS = 0.9738 and K_s = 0.9988; MAE for UCS = 1.04 MPa and K_s = 0.02 MPa·√m). Similarly, for OWC-cemented specimen, the best ML algorithm to predict their macro-scale mechanical integrity is the RF model (R² for UCS = 0.9984 and K_s = 0.9996; MAE for UCS = 0.5 MPa and K_s = 0.01 MPa·√m). This study provides insights into the potential of MICP and OWC as near-wellbore cementitious materials and the applicability of ML model for evaluating and predicting the mechanical integrity of cementitious materials used in near-wellbore to achieve efficient geo-hazard mitigation and environmental protection in engineering and underground operations.