Prediction of occurrence of thermal cracking of RC abutments using artificial neural networks

Abstract

This paper presents the applications of artificial neural networks (ANNs) in predicting the occurrence of thermal cracking of RC abutments. The target cracks in this research are the externally restrained cracks in vertical walls and parapets. Dataset for training, testing and validation of the artificial neural networks was taken from the database of Yamaguchi prefecture, records of actual construction of concrete structures. In this study, a reliable dataset was selected by removing apparently wrong entries and cross verification was performed for all entries. ANNs for vertical walls and parapets were developed separately. K-fold cross validation was also performed. A high level of accuracy was achieved for both vertical and parapet walls. The results from this study exhibit the influential factors on cracking with their importance, which will be utilized in future cracking mitigation.