Prediction of the degree of steel corrosion damage in reinforced concrete using field-based data by multi-gene genetic programming approach

Unanticipated failure of reinforced concrete structures due to corrosion of steel rebar embedded in concrete causes to increase the demand for finding methods to forecast the service life of concrete structures. In this field, the success of machine learning-based methods leads to the use of multi-gene genetic programming (MGGP) method for classifying the degree of corrosion destruction of steel in reinforced concrete in this paper. Despite the common application of MGGP that is the symbolic regression, in this research, MGGP was adapted to use in classification tasks. Accordingly, a large field database has been collected from different regions in the Persian Gulf for modeling of MGGP and neural networks. Comparing the results attained from the MGGP procedure with neural networks revealed that both methods have a good ability to predict the degree of steel corrosion damage for the data range of examined reinforced concrete. But, MGGP gives a particular mathematic equation to estimate the outcome by using the input variables. Moreover, this method can also implement sensitivity analysis simultaneously. The selected input variables by MGGP via the evolution process were the most relevant to the class corrosion whereas there was not any redundancy between them. It is in good agreement with results obtained from sensitivity analysis.