A neural network method for analysing concrete durability

Abstract

This paper describes the use of an artificial neural network (ANN) method for the analysis of relationships between a number of input parameters and observed damage owing to reinforcement corrosion. Data on the effects of the environmental conditions, structure and properties of concrete on the degree of damage caused by steel corrosion have been gathered on 11 concrete bridge structures in a Croatian moderate continental climate. The main causes of deterioration were chloride ions, from de-icing salts, and accelerated carbonation owing to the higher carbon dioxide concentration on highways and in towns. The methodology of data gathering from surveys, diagnosis and remedial works to concrete structures is described. The damage was classified into six categories based on the type of remedial work necessary. As the parameters are time dependent and show high scatter, a probabilistic-like approach was adopted using an ANN for fuzzy feature categorisation as a tool for classification of the degree of damage. The ANN was successfully trained and validated for the range of data from the investigated bridges. The outputs of the work could be used for fuzzy prediction of the extent of damage in the structure service life and for planning the maintenance. The outputs can also be used to assist in the design and restoration of the reinforced concrete structures.