|
چکیده
|
Determining mechanical properties such as static modulus of elasticity (SME) and uniaxial compressive strength (UCS) is necessary in civil engineering and rock mechanics projects. Despite the high accuracy of laboratory methods, direct measurement of these parameters is time-consuming, costly, and difficult. Hence, current research aims to determine UCS and SME by indirect statistical and intelligent methods. For this purpose, first, the physical properties, Schmidt hardness, P-wave velocity (PW), UCS, and SME of the sandstone, limestone, and dolomite were measured in the laboratory. Then, different models were used to estimate these parameters using multivariate linear regression, random forest algorithm, Gaussian process regression, backpropagation multilayer artificial neural network, and support vector regression. Error level, a10 and a20 indices, spider diagram, and performance index were used to appraise the models. Dolomite samples showed the highest resistance values, and porous marl limestone samples showed the lowest. By comparing sandstone samples from different sites, it was found that sandstones with carbonate fragments and gypsum and clay cement have the lowest resistance as compared to the chert fragments. The support vector regression precision (R2=100% for UCS and R2=97% for SME) was higher than other methods. Mean difference values of actual and estimated UCS and SME were equal to+0.03% and −0.75%, which shows the best effectiveness of models in predicting properties.
|