Shear strength prediction for RC beams without shear reinforcement by neural network incorporated with mechanical interpretations

Abstract

In this study, the neural network model for predicting shear capacity of reinforced concrete (RC) beams without shear reinforcements is developed. Several explanatory variables concerning shear resisting mechanisms are derived from mechanical interpretation in order to optimize the number of variables employed in neural network model and the prediction performance. The experimental results gained from several research institutes are gathered and utilized as a teaching data and a validation data. The former is used for training the artificial intelligence (AI) model whilst the latter is applied for evaluating the prediction capability. Then, the predicted results gained from neural network model are evaluated against the shear capacity computed using national design standards. Regarding the result, the developed AI model yields better shear capacity prediction than that predicted using conventional AI model by thoroughly considering the variable relating to features of tensile reinforcements. Also, the AI model can predict shear capacity in a unified manner for RC beams having extensive range of shear span to effective depth ratio. Furthermore, the hidden shear capacity which cannot be achieved through current design equations are attained using the neural network approach.