Neural network analysis for corrosion of steel in concrete

Corrosion of the steel embedded in concrete plays a vital role in the determination of life and durability of the concrete structures. Several researchers have studied the corrosion behaviour of the embedded steel and the different types of protective measures that are available to control the corrosion. However, little work has been done to recognize, identify the performance and predict the behaviour of the steel over a long term. Hence, this work concentrates on recognizing the behavioural pattern of the embedded steel and predicting its potential characteristics using artificial neural network (since the potential of the embedded steel is used to determine whether the steel is corroding or not as per ASTM C 876-91). A systematic study to develop a suitable method that can accept, analyze and evaluate experimental data that are at random and/or influenced by external, unpredictable variables has been carried out, using the back propagation method. This method is fast and is able to produce an output that has minimum error for this experimental setup. This has resulted in the development of back propagation neural network, that can train and test the system, to calculate the specified parameter for different conditions and recognize the behavioural pattern. Using this methodology, the corrosion of the steel embedded in concrete is analyzed and it is observed that the error encountered is only about 5% for the predictions made from the analysis. (c) 2004 Elsevier Ltd. All rights reserved.