Prediction of Carbonate Aggregates Properties Through Physical Tests

Most engineering projects are involving carbonate rocks in many countries. These rocks are mainly used for various purposes as construction materials for road pavements, Portland cement concrete, building stone, etc. Two important parameters for these projects are the intact rock strength as uniaxial compressive strength (UCS) and crushed rock strength as aggregate crushing value (ACV). Sometimes it is impossible to obtain suitable samples for these tests. Therefore, predicting models have widely used as alternative methods. Since the rock physical properties affect its strength, these properties have been considered to predict UCS and ACV. The main purpose of this study is the application of simple regression, multiple regressions, i.e., linear and non-linear, and artificial neural networks (ANN) to predict the strength properties of carbonate aggregates. In the present paper, 28 samples of carbonate aggregates are studied. The simple physical experiment including porosity (η), density (r), and water absorption by weight (Wabs), and rock strength experiment including UCS and ACV are carried out. Consequently, the best relationships between carbonate aggregate strength and physical properties are determined. Different statistical techniques are used for evaluating and determining the accuracy of empirical equations. The results of the correlation coefficient and significant level indicate that physical properties have significant correlations with ACV and UCS. Subsequent linear and non-linear regression analyses revealed that Wabs and r are the most valid indirect tests to estimate ACV and UCS, respectively. Also, the results indicated that the ANN model showed higher accuracy for predicting UCS (R² = 0.92) and ACV (R² = 0.95) than regression models.