Fiber-Reinforced Polymer Confined Concrete: Data-Driven Predictions of Compressive Strength Utilizing Machine Learning Techniques

Abstract

Accurate estimation of the mechanical properties of concrete is important for the development of new materials to lead construction applications. Experimental research, aided by empirical and statistical models, has been commonly employed to establish a connection between concrete properties and the resulting compressive strength. However, these methods can be labor-intensive to develop and may not always produce accurate results when the relationships between concrete properties, mixture composition, and curing conditions are complex. In this paper, an experimental dataset based on uniaxial compression experiments conducted on concrete specimens, confined using fiber-reinforced polymer jackets, is incorporated to predict the compressive strength of confined specimens. Experimental measurements are bound to the mechanical and physical properties of the material and fed into a machine learning platform. Novel data science techniques are exploited at first to prepare the experimental dataset before entering the machine learning procedure. Twelve machine learning algorithms are employed to predict the compressive strength, with tree-based methods yielding the highest accuracy scores, achieving coefficients of determination close to unity. Eventually, it is shown that, by carefully manipulating experimental datasets and selecting the appropriate algorithm, a fast and accurate computational platform is created, which can be generalized to bypass expensive, time-consuming, and susceptible-to-errors experiments, and serve as a solution to practical problems in science and engineering.