Prediction of bond performance of tension lap splices using artificial neural networks

Recently, machine learning has been widely used in civil engineering, because better design can be achieved using the advanced computer intelligence and test results. In the present study, to improve design reliability and to extend design application range for the development and lap splice lengths, an artificial neural network model (ANN) was presented using 1008 existing experimental studies for splice tests. Although some of the test parameters are out of the limitations of design codes, all test results were used for the ANN to extend design application range considering present-day construction materials and practices. From a parametric study with the ANN, the effect of design variables was investigated, and predictions by the ANN were compared with existing design equations. Finally, based on the parametric study result, modifications were proposed for existing design equations to consider the effect of non-uniform bond stress distribution and the effect of cover concrete and transverse bars, as well as to extend design application range. Comparisons showed that the modifications improved the accuracy of the design methods. The high accuracy to the large number of existing test results confirms that the modifications based on the ANN can improve design reliability and also can extend design application range for the development and lap splice lengths.