| dc.description.abstract |
Uniaxial compressive strength (UCS) of rocks is the most commonly used parameter in geo-engineering application. However, this parameter is hard for measurement due to a time consuming and requires expensive equipment. Therefore, obtaining this value indirectly using non-destructive testing methods has been a frequently preferred method for a long time. In order to obtain multiple regression models, input parameters need many assumptions. Thus, the estimation of the mechanical properties of rocks using by machine learning methods has been investigated. In this study, UCS values of rocks were estimated by reformulating with artificial intelligence-based age-layered population structure genetic programming (ALPS-GP) which is one of machine learning methods. Artificial neural network (ANN) and ALPS-GP models were performed to predict UCS from porosity, Schmidt hammer hardness and ultrasonic wave velocity test methods. For this purpose, the mentioned three tests (porosity, Schmidt hammer hardness and P-wave velocity) were carried out on ten different stones from Turkey. ANN was performed to evaluate this new technique. Reliability of UCS values determined by models was checked with mean absolute error (MAE), coefficient of determination (R2), root mean square error (RMSE) and variance account for (VAF) values. These values were calculated as 1.64, 0.98, 2.11 and 99.81 for ANN, and 2.11, 0.98, 2.50 and 97.86 for ALPS-GP, respectively. It was observed that both methods used were quite successful in UCS estimation. The most important advantage of the ALPS-GP model is providing an equation for UCS estimation. In the light of the obtained findings, it has been revealed that this equation derived from ALPS-GP can be used in UCS estimation processes of similar rock types (limestone, dolomite and onyx). © 2021, King Fahd University of Petroleum & Minerals. |
|