Soft computing approaches for comparative prediction of the mechanical properties of jute fiber reinforced concrete. (November 2020)

• Record Type: Journal Article
• Authors: Sultana, N.; Zakir Hossain, S.M.; Alam, Md Shah; Islam, M.S.; Abtah, Mahmoud Ahmed Al
• Is Part Of: Advances in engineering software. Volume 149(2020)
• Publication Date: 2020-11
• Subjects: Jute fiber -- Modeling -- Response surface methodology -- Artificial neural networks -- Support vector regression -- Reinforced concrete; Computer-aided engineering -- Periodicals; Engineering -- Computer programs -- Periodicals; Engineering -- Software -- Periodicals; Periodicals; 620.0028553
• DOI: 10.1016/j.advengsoft.2020.102887
• Languages: English
• ISSNs: 0965-9978

Abstract

Highlights: Jute Fiber Reinforced Concrete Composites (JFRCC). Response Surface Methodology (RSM), Artificial Neural Networks (ANN), Support Vector Regression (SVR). Prediction of optimum mechanical properties of concrete. SVR model predictions are closer to the experimental results. Abstract: The fibers in concrete is necessary to enhance its engineering properties. Different types of fibers are used as a reinforcing material. Due to low cost, availability, and environmentally friendly, in recent years, the use of natural fibers in concrete has increased attention widely. Among all the natural fibers, jute fibers are very cheap and available in tropical countries. This study assesses different soft computing approaches: RSM (Response Surface Methodology), ANN (Artificial Neural Networks) and SVR (Support Vector Regression) for development of nonlinear empirical models that predict the mechanical properties (compressive and tensile strengths) of Jute Fiber Reinforced Concrete Composites (JFRCC). These properties are mainly dependent on water-cement (W/C) ratio, length and volume of jute fiber. The codes for ANN and SVR were written in MATLAB (R-2019a), while Minitab® 18 statistical software was used for generating experimental design matrix via Box-Behnken design (an experimental design for RSM). The data for the properties of JFRCC were obtained based on this design matrix and these data were utilized to develop, compare and evaluate the suggested models. The results indicate that SVR model performs much better than ANN and RSM models with respect to various performance measuring parameters (e.g., correlation coefficient, residual, relative error, mean absolute error, root mean squared error, and fractional bias) for predicting both compressive and tensile strengths.