Estimation of dynamic properties of sandstones based on index properties using artificial neural network and multivariate linear regression methods

The dynamic properties of the rock are very important for the design of geotechnical structures and the modeling of deep drilling. In the present study, the velocity of compressional and shear waves (Vp and Vs) and the dynamic elastic modulus (Ed) of sandstones were estimated based on index tests using artificial neural network (ANN) and multivariate linear regression analysis (MVLRA) methods. For this purpose, petrographic, physical, mechanical and dynamic tests were performed on 54 specimens. Petrographic results showed that the samples were classified as feldspathic litharenite. The results showed that the Vp/Vs ratio was equal to 1.78. Also, the effect of mineralogy on mechanical properties was more than dynamic properties and the effect of quartz on dynamic properties was more than other minerals. The presented relationships were evaluated using R-squared (R-2), root-mean-square error (RMSE), mean absolute relative prediction error (MARPE), variance account for (VAF) and performance index (PI). The results of the ANN to estimate the Ed, Vp and Vs showed that it is possible to estimate these parameters based on inputs with high accuracy. The accuracy of the ANN was higher than the MVLRA. Estimation of Vs, Vp and Ed by ANN showed correlation coefficients of 0.97, 0.86 and 0.92 and RMSE of 0.10, 0.31, and 3.98, respectively. The ANN was also conservative in predicting these variables, while MVLRA was conservative only in estimating the Vs and Ed of the studied sandstones.

Automated summary

The study focused on estimating the dynamic properties of sandstones, finding that mineralogy had a greater impact on mechanical properties than on dynamic properties, with quartz having the most significant effect on dynamic properties. Utilizing artificial neural network (ANN) and multivariate linear regression analysis (MVLRA) methods, the dynamic elastic modulus, compressional and shear wave velocities were accurately estimated, with ANN outperforming MVLRA in accuracy.