Compressive strength and Young's modulus (YM) are relevant geomechanical properties for assessing the brittleness of particular formations, the potential for shear fracturing, and the complexity of fractures created during stimulation. In a previous study [1], unconfined compressive strength (UCS) and Young's modulus (YM) were determined on 25 micro-pillars (indicate dimensions) machined from a Woodford shale sample from uniaxial. In this study, we have trained a machine learning algorithm - Least Square Support Vector Machine (LSSVM) - to predict these properties as a function of the elemental compositions of rock. Elemental analysis of the 25 micro-pillars was obtained using energy dispersive X-ray spectroscopy (EDS). The weight fractions of five key elements, oxygen, silicon, aluminum, iron, and potassium, are used as input to the model. Data are divided into training (80%) and test (20%) data sets. Repeated K-fold cross validation is used during the training. Average values of the hyperparameters (regularization parameter γ, and Kernel parameter,σ) are accepted from the trained model. The results show that the machine learning algorithm is capable of predicting geomechanical properties with reasonable accuracy despite the limited amount of experimental data.

1. EXPERIMENTAL

In this study, a core-plug rock sample was collected from a core-slab (stored at the U.S. Geological Survey (USGS) Core Research Center in Denver, Colorado) recovered from 1,982-1,983 m depth interval of the 1 Olie Smithson well drilled in the Woodford Shale (Anadarko Basin in Northern Oklahoma).

A set of twenty-five micrometer-sized samples – micro-pillars – was focused-ion-beam (FIB) nanofabricated on the top of the core-plug rock sample and was used for micro-compression testing, as shown in Figure 1. This microscopic uniaxial compression test (UCT) measures the unconfined compressive strength (UCS) of materials in addition to their pseudo-static loading Young's moduli (E). Each micro-pillar was 10 μm in diameter and 20 μm tall, maintaining a 1:2 ratio between diameter and length for the UCT experiment. Once prepared, the micro-pillars were crushed using a Hysitron TI-950 TriboIndenter nanomechanical test instrument, also known as nanoindenter. To break the samples, a 20-μm diameter flat diamond tip of the nanoindenter was brought in contact with each of the micro-pillars and the force was applied until the samples failed. The UCT experiment was performed under a constant load rate control of 10 μN/s and a maximum load of 10,000 μN – load high enough to crush the rock. Load and displacement data were collected, and UCS) and E of each micro-pillar was calculated. A more detailed description of the experimental procedure, along with the micro-compression results are provided in the recent paper by the authors [2].

This content is only available via PDF.
You can access this article if you purchase or spend a download.