Multi-factor fuzzy prediction model of concrete surface chloride concentration with trained samples expanded by random forest algorithm

Abstract

Although many factors affect the surface chloride concentration (Cs) of concrete structures in ocean exposure environments, establishing a quantitative prediction model that considers multiple factors using traditional regression methods is still a challenge. Herein, a multi-factor hybrid model combining the advantages of the random forest algorithm and fuzzy logic system is proposed. This hybrid fuzzy logic system can achieve quantitative, continuous, and visual expressions of the influence law of various environmental and material factors on Cs. It has high precision in predicting time-dependent Cs with multiple factors considered. The calculation results indicate that the stable value of Cs increases with increasing water-to-binder ratio (w/b). Nonetheless, when w/b is greater than a critical value, the increment in w/b no longer affects Cs. Nonetheless, w/b does not affect the time required for Cs to reach a steady state. Additionally, the mineral admixtures affect both the final stable value and required time. The service times required to achieve a stable value of Cs for ordinary cement concrete in the submerged, splash, and tidal zones are 2.75, 3, and 3.5 years, respectively. For concrete with identical material conditions, the stable value of Cs in the tidal zone is the highest, followed by that in the submerged and splash zones, and the atmospheric zone is the least. Finally, a predictive model of Cs, which is expressed using a multivariate function with piecewise coefficients, is proposed for quantitative predicting durability design and service life.