The Los Angeles abrasion value (LAAV) of rocks is a critical mechanical aggregate property for designing road infrastructures and concrete quality. However, the determination of this critical aggregate property is labour-intensive and time-consuming and thus, in the literature, there are many predictive models to estimate the LAAV for different rock types. However, most of them are based on classical regression analyses, limiting their broader usage. In this study, several soft computing analyses are performed to develop robust predictive models for the evaluation of LAAV of rocks in the Ilıca region (Kütahya – Turkey). The main motivation for implementing soft computing analyses is that precise predictive models might be useful when exploring suitable rock types that are manufactured in crushing–screening plants. For this purpose, a comprehensive laboratory schedule was established to obtain some inputs for the evaluation of LAAV. As a result of the soft computing analyses, four robust predictive models are developed based on artificial neural networks (ANN), multiple adaptive regression spline (MARS), adaptive neuro-fuzzy inference system (ANFIS) and gene expression programming (GEP) methodologies. The performance of the proposed models is investigated by some statistical indicators such as R2 and RMSE values and scatter plots. As a result, the ANFIS-based predictive model turns out to be the best alternative to estimate the LAAV of the investigated rocks.
Read full abstract