Novel soft computing hybrid model for predicting shear strength and failure mode of SFRC beams with superior accuracy

The ability of steel fibers to enhance the shear strength and post-cracking behavior of plain concrete stimulated remarkable increase in using steel fiber-reinforced concrete (SFRC) in construction. However, steel fibers increase the complexity of assessing the shear behavior. Developing accurate models to estimate the shear capacity is crucial to satisfying requirements of design codes. While various empirical models have been developed for this purpose, they suffer from multiple shortcomings. Machine learning techniques have recently emerged as a strong contender for mitigating such drawbacks and providing better accuracy. In this study, a novel metaheuristic atom search optimization (ASO) algorithm based on molecular dynamics was coupled with artificial neural networks (ANN) to forecast the shear capacity of SFRC beams and overcome drawbacks of standalone models. Moreover, four classification models (naive Bayes, support vector machine (SVM), decision tree, and k-nearest neighbor) were used to forecast the failure mode of SFRC beams. Performance assessment of the models revealed that the ASO-ANN model achieved most reliable predictive accuracy for shear strength, while the k-nearest neighbors model was the most accurate for failure mode classification. The ability to predict simultaneously the shear strength and failure mode with superior accuracy opens immense opportunities for the shear design of new SFRC beams and for selecting innovative retrofitting strategies for existing shear deficient structures.