Comparison between the ANN and MLR performance in prediction the UCS and E of sedimentary rocks

The accurate measurement of geoengineering properties of intact rock such as uniaxial compressive strength (UCS) and modulus of elasticity (E) using laboratory methods requires a lot of time and cost. Hence, development of predictive relationships and models to estimate the UCS and E of rocks seems to be essential in rock engineering. This study deals with the prediction of uniaxial compressive strength and modulus of elasticity of sedimentary rocks using artificial neural networks (ANNs) and multivariable regression analysis (MLR). For this purpose, ۱۹۶ samples from ۴ rock types (sandstone, conglomerate, limestone and marl) were cored and comprehensive laboratory tests have been done on them. To develop the predictive models, physical properties of studied rocks such as P-wave velocity, dry density, porosity and water absorption were considered as model inputs, while UCS and E were the output parameter. The performance of the MLR and ANN models were assessed based on coefficient correlations (R), root mean square error (RMSE), mean absolute error (MAE) indices. The Comparison of results between the ANN and MLR models indicates that the ANN predictive model was more acceptable for predicting the UCS and E than the other.

Keywords:

UCS , Modulus of elasticity , ANN , MLR , Sedimentary rocks