Artificial Neural Network Model for Predicting the Tendon Stress in Unbonded Posttensioned Concrete Members at the Ultimate Limit State

Abstract

Existing design guidelines, codes, and literature provide different calculation models for the estimation of tendon stresses in unbonded posttensioned concrete members at the ultimate limit state. Most of these methods are based on theoretical (e.g., collapse mechanism and bond-reduction models) and statistically-based empirical models, with only a few or no surrogate models based on artificial neural networks (ANNs). This study presents an ANN-based model to predict stress in unbonded tendons at the ultimate limit state based on a database of 251 prestressed concrete members with unbonded tendons collected from the literature. The predictions from the ANN-based model show very good agreement with the experimental results given in the literature during training, testing, and validation. A sensitivity analysis has been performed to quantify the degree of influence of the input variables used in the developed ANN model. The analysis shows that the predictions of tendon stress using neural networks are more accurate than those results obtained using the models given in the design guidelines and the literature.