Machine Learning-Based Empirical Formulations for Strength Properties of Steel Fiber Reinforced Concrete

Document Type : Regular Paper

Authors

1 Ph.D. Candidate, Department of Civil Engineering, Faculty of Engineering, University of Qom, Qom, Iran

2 Associate Professor, Department of Civil Engineering, Faculty of Engineering, University of Qom, Qom, Iran

3 Professor, Department of Railway Engineering, Iran University of Science and Technology, Tehran, Iran

10.22075/jrce.2024.32743.1963

Abstract

The accurate approximation is a benefit of the modern machine learning technique, which also disappeared the problems of traditional empirical methods, such as human and technical errors plus environmental pollution. Although there are many good samples on the state-of-the-art regarding the machine learning prediction of strength properties of steel fiber reinforced concrete, fewer articles are dedicated to proposing empirical formulations. This paper brings some novel empirical formulations to identify the strength properties of macro steel fiber-reinforced concrete. A 2650 multi-national data records are used to perform the regression, which is an exclusive dataset. This archive is the largest available dataset used in the state-of-the-art steel fiber-reinforced concrete prediction process, which is beneficial for supervised learning. Since the user must be careful regarding overtraining with such a vast resource, a successful strategy provided by the authors in previous research is utilized in which various machine learning techniques are compared to forecast the considered properties. So the Ridge, Lasso, and linear methods are used as regressors to predict the strength properties and the constants. Symbolic regression, a powerful tool for producing empirical formulations, is used for creating mathematical expressions regarding the strength properties. The performance is also evaluated based on well-known error analysis metrics. The formulations are presented for flat, waved, and hooked end fibers, the most common fibers used in construction engineering. The machine learning-driven formulations are exclusive due to the utilized strategy and the resources, and the precision of the relations are denoted, which presents the superiority to traditional methods.

Highlights

  • This paper brings some novel empirical formulations to identify the strength properties of macro steel fiber-reinforced concrete.
  • A 2650 multi-national data records are used to perform the regression, which is an exclusive dataset.
  • Since the user must be careful regarding overtraining with such a vast resource, a successful strategy provided by the authors in previous research is utilized in which various machine learning techniques are compared to forecast the considered properties.
  • The Ridge, Lasso, and linear methods are used as regressors to predict the strength properties and the constants. Symbolic regression, a powerful tool for producing empirical formulations, is used for creating mathematical expressions regarding the strength properties.
  • The performance is also evaluated based on well-known error analysis metrics.
  • The formulations are presented for flat, waved, and hooked end fibers, the most common fibers used in construction engineering.
  • The machine learning-driven formulations are exclusive due to the utilized strategy and the resources, and the precision of the relations are denoted, which presents the superiority to traditional methods.

Keywords

Main Subjects



Articles in Press, Accepted Manuscript
Available Online from 12 May 2024
  • Receive Date: 27 December 2023
  • Revise Date: 07 March 2024
  • Accept Date: 29 April 2024