Estimating the friction angle of black shale core specimens with hybrid-ANN approaches

Abstract

Shear strength parameters of rock play a significant role in the design stage of various geotechnical structures such as earth dams, embankments, foundations and tunnels. The direct determination of these parameters in laboratory is time consuming and expensive. Additionally, preparing core specimens with a good-quality is sometimes difficult, especially in weathered and highly fractured rocks. This paper presents an indirect determination of internal friction angle of shale rock specimens through two hybrid neural net based models that combine artificial neural net with genetic algorithm (GA-ANN) and imperialist competitive algorithm (ICA-ANN). In fact, GA and ICA were utilized to improve the efficiency of ANN predictive model via the weights and biases adjustment. To achieve this aim, an extensive experimental program was designed, according to which a series of black shale specimens were characterized using various laboratory tests, including p-wave velocity, Schmidt hammer, point load and triaxial compression. After establishing a proper database for the analysis, simple and multiple regression as well as hybrid intelligent models were developed to predict the internal friction angle of the shale specimens. To compare the obtained results from the models, several performance statistical indices were computed. The results indicated that simple and multiple regression models are not good enough in predicting the internal friction angles. Concluding remark is that the proposed intelligent models are superior in comparison with simple and multiple regression models. Using the coefficient of determination as performance measure, the quality of the developed GA-ANN model was evaluated as 0.917 and 0.909 for training and testing datasets, respectively whereas these values were achieved as 0.960 and 0.956 for the ICA-ANN model. This means that the ICA-ANN model can provide higher performance capacity in estimating the internal friction angles as compared to the GA-ANN. In addition, the results of other performance indices, i.e. variance account for and root mean square error confirmed that the hybrid ICA-ANN predictive model can be introduced as a new technique for predicting the internal friction angle of shale rock specimens in practice.