Soft computing approaches for evaluating the slake durability index of rocks

In this research, some predictive models were constructed for estimating the slake durability index of sedimentary rocks. Sandstone, limestone, travertine and conglomerate were collected as studied rocks, and comprehensive laboratory investigations such as mineralogical study and geotechnical properties including dry unit weight, porosity, Schmidt rebound hardness, P-wave velocity, uniaxial compressive strength and slake durability index were determined based on standard procedures. Results of mineralogy studies and XRD analysis showed studied rock samples are dominantly composed of quartz and calcite with different textures. The durability test, up to three cycles, was performed in fluids with different pH conditions. Based on the results, the slake durability index is affected by the pH of the test fluids, and in initial cycles, the decreasing rate of slake durability index is higher than the end cycles. Also, in most of the samples, a constant pattern exists between the slake durability index and pH of the testing solutions so that the weight loss increases with decreasing pH from 7 to 4. Soft computing techniques including multiple regression analysis (MRA), artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) were developed for estimating SDI for pH = 7 and 4 conditions. Experimental equations were obtained by MRA with correlation coefficients from 0.74 to 0.85 that show porosity, Schmidt hardness, P-wave velocity and UCS are the good parameters for estimating the SDI2 of rocks. In order to evaluate the performance of predictive models, some statistical coefficients, including R, RMSE, VAF, MAPE and PI were also calculated. ANFIS models have the best coefficients, and the results demonstrate that the ANN models are efficient when compared to MRA. Therefore, all methods obtained acceptable results, but the ANN and ANFIS are more reliable methods for estimating the SDI of rocks.