Machine learning techniques for estimation of Los Angeles abrasion value of rock aggregates

Rock aggregates are extensively used in the production of materials such as asphalt concrete and Portland cement concrete. Los Angeles Abrasion (LAA) value is one the basic characteristics of crushed aggregates that reflects their resistance against mechanical abrasive factors such as repeated impact loading. There have been several efforts to estimate the LAA value from surrogate physical and/or mechanical properties of the material. Previous works have mainly focussed on a limited number of data samples and thus may not be generalised to make predictions for different lithologies. Another drawback of the current approaches is that they are often in the form of one-to-one correlations between the LAA and a measure of mechanical behaviour such as the uniaxial strength. This paper investigates the capability of Machine Learning (ML) models for prediction of LAA value. Different material properties have been tested as the input parameters to achieve the best prediction results. It was observed that the ML models perform considerably better for predicting LAA compared to the existing correlations reported in the literature.