Evaluation of the relationship between the physical properties and capillary water absorption values of building stones by regression analysis and artificial neural networks

Abstract

The most important factor in the movement of groundwater or mineral precipitation in building stones is the capillary water absorption properties of the rock. Besides, capillary water absorption is one of the most important parameters in the degradation process of building stones. The determination of the capillary water absorption values of rocks is a very time-consuming and sensitive process. In this study, the capillary water absorption values of 100 different rock samples were predicted by simple regression (SR), multiple linear regression (MLR), and artificial neural network (ANN) method using physical properties (dry density, P-wave velocity, porosity, water absorption of weight). In the evaluation performed by the SR, although the correlation coefficients in the relationships between the physical and capillary water absorption properties of rocks varied between 0.676 and 0.911, it was observed that the values predicted from these relationships for the samples with high capillary water absorption (C > 200 g/m(2)/s(0.5)) were deviated from the experimental values. In the MLR analysis, the highest correlation coefficient was found to be (R-2 : 0.708). Among the physical properties used as input parameters in the ANN method, the dry density property indicated the best correlation coefficient in the training (R-2 : 0.9587) and testing (R-2 : 0.9603) results. Furthermore, it was determined that the approach developed with the ANN was more reliable in predicting capillary water absorption values.