Characterizing fracture toughness using machine learning

Abstract

The existing models for fracture toughness characterization based on nanoindentations that account for the fracture length are limited to simple (ideal) geometries that are absent in shales. The present study proposes two conceptual models to estimate the fracture length created by nanoindentations in shales. It also presents a workflow to apply the conceptual models and uses machine learning, enabling a systematic and automated analysis. The conceptual models assume that the induced fracture is in the first mode to determine the fracture toughness. In this study, fracture toughness is also determined by the energy method that relates the load-displacement hysteresis to the fracture toughness without restricting the fracture mode. The present study sheds light on the complexities of characterizing fracture toughness using nanoindentations and has applications in the petroleum industry. The conceptual models are appealing for formation characterization using small pieces, such as drill cuttings, when large samples (~2.5 cm) required for conventional tests are unavailable. The conceptual models have applications in estimating fracture toughness when the induced fracture patterns become more complex.