Experimental investigation and comparative machine-learning prediction of strength behavior of optimized recycled rubber concrete

In the present paper, the design of optimized rubber concrete composite containing silica fume (SF) and zeolite (ZE) was undertaken using the literature, and the properties were assessed through destructive and non-destructive (NDT) methods. In order to optimize the rubberized cement composite, the optimum tradeoff between compressive strength as the main objective and rubber content, as well as the optimum fractions of the admixtures were taken into account. Main tests including workability, compressive strength, elastic modulus, and ultrasonic tests were carried out to fully assess the effects of rubber, ZE, SF, curing, and age on the rubberized composite behavior. Primary and secondary wave velocities, i.e. V-p and V-s were determined from ultrasonic test to characterize different mixtures. Static modulus results obtained from NDT were compared, and it was found that NDT results were in very good agreement with those of destructive test results. Moreover, the dynamic elastic modulus determined from compression and shear wave velocities (V-p, V-s) conforming to ASTM were compared with those estimated from six different relationships including BS, EN and ACI relationships along with other well-known equations available in the literature. In order to predict the compressive strength of the rubberized cement composite as a function of the influencing variables, a comprehensive comparative modeling was performed and different predictive models were developed using regressions and machine-learning (ML) techniques, i.e. nonlinear multi-variable regression (NMVR), Artificial neural network (ANN), genetic programming (GP), adaptive neuro-fuzzy inference system (ANFIS), and support-vector machine (SVM). Closed- form formulations were derived for NMVR, ANN, and GP models, and parametric study was conducted for ML models. Performance criteria such as root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R-2) were used to compare the models' performance. It was found that SVM outperformed the other models with the highest R-2 and the lowest RMSE equal to 0.989 and 1.393, respectively. (C) 2020 Elsevier Ltd. All rights reserved.