Using an Artificial Neural Network to Validate and Predict the Physical Properties of Self-Compacting Concrete

K Thirumalai Raja,R Krishnaraj,V S Kaushik,N Nagaprasad,N Jayanthi,Jule Leta Tesfaye,Fuat Kara

doi:10.1155/2022/1206512

Abstract

SCC (self-compacting concrete) is a high-flowing concrete that blasts into structures. Many academics have been interested in using an artificial neural network (ANN) to forecast concrete strength in recent years. As a result, the goal of this study is to confirm the various possibilities of using an artificial neural network (ANN) to detect the features of SCC when Portland Pozzolana Cement (PPC) is partially substituted with biowaste such as Bagasse Ash (BA) and Rice Husk Ash (RHA) (RHA). Specialist systems based on the fully connected cascade (FCC) architecture in artificial neural networks (ANN) are used to estimate the compressive toughness of SCC. The research results are confirmed with the forecasting results of ANN utilizing 73 trial datasets of differentiation focus proposals of cement, BA, and RHA containing parameters such as initial setting time (IST), final setting time (FST), and standard consistency. Experiments to determine compressive strength for a wider range of mixed prepositions will result in higher project expenses and delays. So, an expert system ANN is used to find the standard consistency, setting time, and compressive strength for the intermediate mix propositions according to IS 10262:2009. The experimental results of compressive strength for 28 days are considered, in which 70% was used to train the ANN and 30% was utilized for testing the accuracy of the predicted compressive strength for the intermediate mix proposition. Using all of the datasets, the number of hidden layers used for compressive strength prediction for intermediate mix proposal is determined in the first step. The compressive strength for the intermediate mix preposition was identified in the second phase of the research, using the number of hidden layers determined in the first phase. The results were validated using the correlation coefficient (R) and root mean square error (RMSE) obtained from ANN, resulting in an acceptance range of 97 percent to 99 percent.

Highlights

SCC has high compactness without requiring external vibration, which through imperfections bleeding, permeability, and segregation is eradicated [1]
Some of the input parameters values used in artificial neural network (ANN) are as given in Table 4. e number of neurons used in the simulation is 2/3 of the input layer size plus the output layer
E predicted compressive strength of concrete is recorded as the output of ANN network. e simulation results were recorded by varying hidden layers from 2 to 10 to identify the best architecture using the correlation coefficient (R) and root mean square error (RMSE) as stopping criteria for the epochs value of 500, the learning rate of 0.3, and momentum of 0.2. e R and RMSE values are calculated using equations (1) and (2). e 73 experimental datasets are used to train the ANN network to identify the best architecture [32,33,34,35,36,37]

Summary

Introduction

SCC has high compactness without requiring external vibration, which through imperfections bleeding, permeability, and segregation is eradicated [1]. E vital task in SCC is determining the mix proposition of ingredients. To address this challenge, many researchers proposed various methods like controlling the maximum coarse aggregate particle size to the total volume, reducing the volume ratio of aggregate to cementitious material, using various viscosity-enhancing admixtures. ANN has been used in various civil engineering research processes such as damage detection in structures performance, concrete analysis, materials behavior modeling, monitoring the groundwater, and optimization of structures. Since ANN is a nonlinear model, researchers are applying this approach to predict the viable mix propositions to be taken as inputs to ANN for analyzing the mechanical properties of hardened concrete. Some of the research works addressed by researchers in this direction are elaborated on below

Objectives

Results

Conclusion