Developing an artificial neural network model to predict the durability of the RC beam by machine learning approaches

The chloride diffusivity is a significant index to describe the degradation rate of concrete structures. At present, a lot of models have been proposed by previous studies to investigate the chloride diffusivity in concrete specimens. However, most of these models only consider the effect of exposure time, and erosion depth on the chloride diffusivity and ignore the uncertain factor in the environment. This may lead to a tremendous difference existed in the predicted-to-experimental ratios of all typical chloride diffusivity models. Machine learning approaches as potential alternative predictors are often used to cope with this bias. This paper adopted five contemporary ML approaches, namely, the backpropagation (BP) neural network, the decision tree (DT), the random forest (RF), the linear regression (LR), and ridge regression (RR) to explore the chloride diffusivity. Input features considered in this study are water-cement ratios, the thickness of concrete specimens, coarse aggregate fraction volume, ratios of environmental temperature to the maintenance standard temperature, and ratios of environmental humidity to the relative humidity. It was found that the backpropagation (BP) neural network is the most accurate prediction (85%) with the lowest root mean squared error. With this superiority, A study on the importance of the input parameters reveals that the volume fraction of coarse aggregate, water-cement ratios, the thickness of the protective layer, and ratios of environment humidity to the relative humidity are the most influential parameters on chloride diffusivity. In addition, a data-driven model was established to investigate the chloride diffusivity considering the effect of random factors both in material and environment. Compared with the test results, the results indicated that the prediction results are closer to the test results, which is helpful for in-depth exploring the durability of the concrete structures.