Modeling the Lateral Behavior of Concrete Rocking Walls Using Multi-Objective Neural Network

Concrete rocking walls (CRWs) are an appropriate alternative for common concrete shear walls due to lower repair and downtime costs. Knowing the exact behavior of structural elements is generally achieved by conducting comprehensive experiments, which are costly and time-consuming. According to previous studies and research trends in the field of rocking walls, it is necessary to provide a comprehensive model for design purposes. It is difficult to provide a theoretical model for predicting the behavior of concrete rocking walls, which includes a variety of parameters, and simultaneous calculation of the effect of those parameters requires a comprehensive model. Providing such a model through classical or numerical methods, due to the complexity of the problem or modeling difficulties, requires much time and complex calculations. However, solving this problem by soft computing methods makes it possible to simplify and speed up these calculations. Therefore, this paper aims to develop a multi-objective neural network (MNN) model to predict the lateral behavior of CRWs. To generate the required data for modeling, experimental results were employed, and all the parameters affecting the lateral bearing capacity of CRWs were extracted to use them as input parameters. Finally, the lateral force-displacement and neutral axis-displacement curves were predicted as the output parameters. Besides, the accuracy of the existing equations presented in international concrete codes and developed by researchers in predicting the flexural strength of CRWs was investigated using the results of experiments performed on these walls. This study shows that the MNN model predicts the lateral behavior of CRWs with acceptable accuracy. This model is able to correctly predict initial stiffness, secondary stiffness, maximum strength, and rotational capacity. According to the results, the terms of the concrete design codes of the United States (ACI 318-14), New Zealand (NZS 3101), and Australia (AS 3600) predict the flexural strength of CRWs in the range of 0.59 to 0.99 actual values. Also, Canada's concrete design code (CSA-A23.3-04) provides a highly unconservative prediction of the walls' flexural strength. Nevertheless, the MNN model showed more accurate predictions in comparison with the investigated codes and existing expressions.