Analysis and prediction of the axial compression properties of desert sand concrete with steel tube restraint based on an improved BP neural network model

Accurate analysis and prediction of axial compression are important for ensuring the construction quality and safety of desert sand recycled aggregate concrete confined by steel tubes. In this study, the axial compressive strength and elastic modulus of recycled aggregate concrete with different sand contents, water-cement ratios, and steel constraints were tested to evaluate the effects of these factors on the axial compressive performance of the recycled aggregate concrete. It was determined that a steel tube restraint could effectively improve the ductility of desert sand recycled aggregate concrete. However, with increases in the sand content and water-cement ratio, the peak stress slightly decreased. The axial compressive strength and elastic modulus of the recycled sand aggregate concrete confined by steel tubes exhibited little change in the elastic stage under a functional load. During the initial stage of loading, the lateral strain exhibited strong discrete characteristics. In the peak stress stage, the transverse coefficient gradually increased. Overall, our analysis revealed that axial compressive performance exhibits evident engineering uncertainty under the comprehensive influence of factors such as steel constraint, desert sand content, and water-cement ratio. Therefore, an improved backpropagation (BP) neural network model of the axial compressive properties of recycled aggregate concrete with steel-tube-confined sand was established with the presence of steel constraints, desert sand content, and water-cement ratio serving as inputs, and axial compression strength and elastic modulus as outputs. Engineering verification calculations indicated that the BP neural network model can predict concrete performance under actual working conditions with a small error rate. Compared with traditional models, the neural network model has comprehensive advantages in terms of fitting accuracy, reduced overfitting, and enhanced stability.