A NEW HYBRID FRAMEWORK OF MACHINE LEARNING TECHNIQUE IS USED TOMODEL THE COMPRESSIVE STRENGTH OF ULTRA-HIGH-PERFORMANCE CONCRETE

The Ultra-High Performance Concrete (UHPC) as an efficient material in constructional projects needs to be investigated in terms of ingredients and their magnitudes to compute the compressive strength of concrete. Empirical determination of relationships between constituents can demand more energy a...

Full description

Saved in:
Bibliographic Details
Published inStavební obzor Vol. 32; no. 3; p. 329
Main Authors Zuo, Xin, Liu, Die, Gao, Yunrui, Yang, FengJing, Wong, Guohui
Format Journal Article
LanguageEnglish
Published Prague Czech Technical University in Prague Faculty of Civil Engineering 01.10.2023
Subjects
Accuracy
Artificial intelligence
Cement
Civil engineering
Compressive strength
Concrete
Constituents
Datasets
Gas solubility
Gases
Humanities
Ingredients
Machine learning
Modelling
Optimization algorithms
Particle swarm optimization
Support vector machines
Ultra high performance concrete
Online AccessGet full text

Cover

More Information
Summary:The Ultra-High Performance Concrete (UHPC) as an efficient material in constructional projects needs to be investigated in terms of ingredients and their magnitudes to compute the compressive strength of concrete. Empirical determination of relationships between constituents can demand more energy and cost. At the same time, the intelligent systems have enabled us to appraise the compressive strength based on ingredients' composition. Also, choosing eco-friendly materials in concrete as one of the widely-used items worldwide should be encouraged. This study has attempted to model the compressive strength of UHPC. Support Vector Regression (SVR), as a machine learning technique aligned with the Particle Swarm Optimization (PSO) and Henry's Gas Solubility Optimization (HGSO), have been used to simulate the compressive strength of concrete calculated based on different materials used in the present article. Eight constituents were tested to generate the compressive strength values. Various metrics were used to evaluate the modeling process. In this regard, the R2 of test phase modeling for SVR-HGSO was obtained 0.960 while for SVR-PSO, 0.925. In the training stage, the correlation rate was computed 0.902 for SVR-HGSO, which is 1.7 percent higher than SVR-PSO with R2 0.887.
ISSN:1805-2576
DOI:10.14311/CEJ.2023.03.0025