Application of group method of data handling technique in assessing deformation of rock mass

Abstract

Strength and deformation characteristics of a rock play a remarkable role in designing any geotechnical structure connected to rock mass. This study aims to propose a practical intelligence system, namely the group method of data handling (GMDH) for indirect rock deformation prediction. Direct measurement of rock deformation in laboratory is time consuming, difficult and costly. In the current study, several rock index tests were conducted, together with unconfined compressive strength tests, on collected granitic block samples. In this study, in accordance to the first set objective, four empirical equations were proposed based on predictors, including Schmidt hammer rebound number, p-wave velocity, porosity and point load strength index, aiming to predict rock deformation. The results of these analyses confirmed that there is a need to develop new multiple-input models in predicting rock deformation. To this end, a GMDH model was designed to forecast rock deformation. Aiming to obtain a fair comparison, a pre-developed artificial neural network (ANN), as a benchmark model of intelligence systems, was also developed to predict rock deformation. Then, through the use of some well-known performance indices, the GMDH and pre-developed ANN models were assessed and their results were compared to select the best predictive model amongst them. Results confirmed that the GMDH is a powerful and robust technique to the reliable prediction of rock deformation.