Application of neural network in predicting damage of concrete structures caused by chlorides

In recent years artificial neural networks (ANN) have shown exceptional performance as a regression tool, especially when used for pattern recognition and function estimation. Artificial neural networks mimic the structure and operation of biological neurons and have the unique ability of self-learning, mapping, and functional approximation. They are highly nonlinear, massively parallel, and can capture complex interactions among input/output variables in the system without any prior knowledge about the nature of these interactions. The ANN for feature categorization was used as a tool for classification of damage and prediction of expected future degree of damage. Data on the effects of the environmental conditions, structure, and properties of concrete onto the degree of damage caused by steel corrosion have been gathered on three concrete structures in Adriatic marine environment. The data were gathered at seven different ages of concrete structures. The damages were classified into five categories based on type of remedial works required to repair the damage. The model that was developed can be useful for planning the monitoring, design of remedial works as well as in improvement of their protection. The influence of variability (sensitivities) of the principal influencing parameters, the ranges of values for principal influential parameters associated to certain categories, and interactions among influential parameters were investigated.