Journal of Structural and Construction Engineering (Jun 2019)

Providing a method for predicting the concrete slump based on Adaptive Neuro-Fuzzy Inference System

  • meysam effati,
  • pooneh shahmalekpour

DOI
https://doi.org/10.22065/jsce.2018.91259.1252
Journal volume & issue
Vol. 6, no. شماره ویژه 1
pp. 127 – 140

Abstract

Read online

Concrete performance is of very high importance in civil engineering projects. One of the most common ways to measure the performance of concrete, is the slump test. To save time, money and materials, it is better to use intelligent methods in predicting the slump. Therefore, in this study a method based on soft computing is used, so without the need to perform arduous physical experiments, one can obtain an estimate of the slump.In this study, an adaptive neuro-fuzzy model which has the benefits of both neural network and fuzzy inference system, is used to predict the concrete slump. In order to train the algorithm for future use, comprehensive experimental data is essential .So by collecting data related to 44 concrete slump experimental tests, variables such as water-cement ratio, sand, gravel, silica fume and super plasticizer which are the principal components of concrete, are considered as input variables and the amount of slump is considered as the output variable in the proposed model.In order to evaluate the performance of the proposed model and accuracy of the results, the results of the adaptive neuro-fuzzy model is compared to that of artificial neural network model, which is obtained in a parallel research done by author, by statistical parameters such as correlation coefficient and root mean square error. By averaging the results of ten different classifications of experimental input data, the correlation coefficient is approximately equal between adaptive neuro-fuzzy and neural network slump. While the root mean square error obtained by using adaptive neuro-fuzzy model is 0/4477 which is less than 0/6964 by neural network model. The difference in the output error of the two models are due to different learning algorithms used in two models and unknown number of hidden layers and neurons in the desirable artificial neural network model.

Keywords