Research on application of the artificial neural network to prediction behaviour of concrete subjected to uniaxial compression

Abstract

ABSTRACT
Analysis, evaluation and prediction of compression behavior of concrete by analytial methods, numberial simulation is one of the essential and important things in minimizing experimental compression on concrete, reducing testing costs and the amount of concrete discharged into the environment. From the researches of many authors over the years have suggested behavioral models such as Hognestad, CEB-FIP, Wee & Mansur, Almusallam…However, above for the behavior curve of stress – strain relationship has not really followed the experimental behavior line. The study proposes to use artificial neural network (ANN) to predict compressive strength of concrete from different aggregate components through data set of 55 experimental compression samples. The compressives value recorded after the approximation process will be the input parameter for the proposed LIT target function. The genetic evolution algorithm (GA) is applied to find the optimal coefficients to optimize the LIT behavior function, to provide the final proposed behavioral model. In order to ensure that the prediction curve is reliable after the optimal process, the study compares the prediction results with 3 experimental sample groups to stress and strain relationships. From the recorded results, the postoptimal behavioral curve closely followed the test behavior curve with low error. From there, the proposed behavior function is highly reliable.

Keywords: Uniaxial compression, behaviour model, genetic algorithms, artificial neural network, behaviour model optimization.