Investigation of strength behavior of thermally deteriorated sedimentary rocks subjected to dynamic cyclic loading

Abstract

This study focuses on the investigation of the strength behavior of thermally damaged rocks subjected to dynamic cyclic loading. For this purpose, NX size rock samples were prepared and heated in a cyclic manner to 200 °C with an increment of 50 °C. The thermally treated samples were tested at 5–15 kHz loading frequencies as per the ASTM-C215. At resonance, the overall increment in the values of the measured parameters was noted as 16–30 for longitudinal quality-factor, 9–20 for torsional quality-factor, 1.5–2.2 kHz for longitudinal resonance-frequency, 1.0–1.8 kHz for torsional resonance-frequency, 34–162 GPa for Young's modulus, 17–69 GPa for shear modulus, and 0–39% for strength improvement factor (SIF). Apart from multivariate statistics, this research addresses the utilization of advanced data modeling techniques: cascade-forward neural-network (CFNN) and adaptive neuro-fuzzy inference system (ANFIS) for the development of the predictive relationship between the SIF and dynamic properties. Results showed that the ANFIS- model (R = 0.97) performed better than the CFNN-model (R = 0.93). For SIF based classification of thermally damaged rocks, a hybrid meta-classifier (MC) was developed by stacking six other machine-learning classifiers including naïve Bayes (NB), logistic regression (LR), K-nearest neighbor (K-NN), multilayer perceptron (ANN-MLP), support vector machine (SVM), and random forest (RF). The outcomes of the analysis showed that for both training and validation of data, the MC had a comparatively higher value of performance indicators than the rest of the classification algorithms. The supervised classifiers were arranged in the following descending order based on their classification performance: MC > RF > NB > LR > ANN-MLP > K-NN > SVM. This study proposes MC as the best classifier for the SIF-based classification of rocks.