Mechanical and Physical Based Artificial Neural Network Models for the Prediction of the Unconfined Compressive Strength of Rock

Abstract

Basalt rocks as building stones were used in many historical buildings in Jordan, and maintenance of these buildings is usually required with time. As part of the effort to collect the necessary information that is needed for future works in repair/strengthen the basaltic structures against any possible future damage. The dry density, Ultrasonic Pulse Velocity, Schmidt Hammer Rebound test, Brazilian Tensile Strength Test, Slake durability, and Point Load test were recorded for specimens tested in the lab to develop indirect methods of estimating the rocks Unconfined Compressive Strength (UCS). Simple regression (SR) analyses were performed to establish correlations between UCS and the results of each above-mentioned rock indices. The SR results showed that a regression model with multiple inputs is needed. In this study, the Back Propagation-Artificial Neural Network (BP-ANN) approach was utilized to predict the USC of Basalt Rock. Two ANN models were developed; one using the physical properties of rocks and the other one using the mechanical properties of rocks. Part of the data collected was used to train the ANN, and a set of independent data was used to validate the developed model. The performance of the ANN model in predicting UCS was compared to that of Multivariate Regression (MVR). The obtained results showed that the ANN model gave higher prediction performance compared to other models. A sensitivity analysis for the developed ANN model was performed to verify the importance of each input. The prediction of UCS can be used to design the proper conservation and repair/strengthen strategies that will allow dealing with the current conditions and the future natural hazards to which these structures are exposed.