Concrete chloride diffusion modelling using marine creatures-based metaheuristic artificial intelligence

Chloride-induced steel reinforcement corrosion endangers the durability of concrete structures and may cause substantial economic loss and environmental impact. An accurate estimation of chloride diffusion of steel-reinforced concrete assists in the reliable prediction of its service life. Artificial neural network (ANN) trained by classical optimization algorithms suffer from the overfitting issue. However, metaheuristic algorithms keep the balance between exploration and exploitation to overcome this issue. Four marine creatures-based metaheuristic optimization algorithms, i.e., Jellyfish search optimizer (JSO), marine predators algorithm (MPA), salp swarm algorithm (SSA), and whale optimization algorithm (WOA) are synthesized by ANN in this study. The proposed algorithms are served to model the apparent chloride diffusion (D_ap) of concrete exposed to atmosphere, tidal, splash, and submerged environments. A total of 216 data from relevant field experiments was extracted from the literature. The results indicate that the synthesized algorithms with simpler architectures perform better than the traditional algorithm. The Wilcoxon rank-sum test shows that the ANN-JSO algorithm performs better than other ANN algorithms, and there is no preference among the average performances of the ANN-MPA, ANN-SSA, and ANN-WOA in the testing phase. The global sensitivity analysis denotes that the exposure time, curing conditions, and exposure are the most critical parameters in predicting the D_ap of concrete.