Title:
A Multi-Gene Genetic Programming Model for Predicting Shear Strength of Steel Fiber Concrete Beams
Author(s):
Mohamed K. Ismail, Ahmed Yosri, and Wael El-Dakhakhni
Publication:
Structural Journal
Volume:
119
Issue:
2
Appears on pages(s):
317-328
Keywords:
data-driven empirical models; genetic programming; sensitivity analysis; shear strength prediction; steel fiber-reinforced concrete (SFRC) beams
DOI:
10.14359/51734345
Date:
3/1/2022
Abstract:
The superior performance of steel fibers (SFs) in enhancing the shear capacity of reinforced concrete (RC) beams supports their use in regions that would otherwise require minimum shear reinforcement (stirrups). Subsequently, several researchers proposed different predictive models to capture the contribution of SFs to the shear capacity of RC beams without stirrups. However, most of these models were empirically developed using respective limited data sets, restricting their generalizability. Artificial intelligence-based models have shown high efficacy in imitating the behavior of complex systems based on large data sets; however, such
models are usually criticized for being impractical for design and too complex to use. To address this issue, a new methodology was adopted in this paper using multi-gene genetic programming (MGGP)—a class of artificial intelligence techniques—to develop an elegant shear strength prediction model for steel fiber-reinforced concrete (SFRC) beams. Guided by mechanics and previous research findings, the governing parameters and their key combinations were first identified and selected for the model development. Subsequently, the model was trained, validated, and tested using observations from 752 experimental tests, and different measures were applied to assess its performance and generalizability. The MGGP-based model developed in this study outperformed 15 shear strength prediction models developed in previous studies and is also presented in a standards-ready format to facilitate adoption. The outcomes from this study demonstrate the promising capability of a mechanics-guided artificial intelligence approach to develop interpretable models that can efficiently predict complex structural behaviors of other components and systems.