Performance evaluation of recycled aggregate concrete-filled steel tubes under different loading conditions: Database analysis and modelling

This paper presents a set of data mining analyses on the structural performance of recycled aggregate concrete-filled steel tubes (RACFSTs) conducted using grey relational evaluation and Back-Propagation (BP) neural networks. A comprehensive experimental database containing the results of 20 flexural, 105 compressive and 85 lateral cyclic loading tests of RACFSTs manufactured using conventional recycled aggregates (RAs) (i.e., sieve grading varying from 5 mm to 31.5 mm) and demolished concrete lumps (i.e., sieve grading varying from 50 mm to 300 mm) were compiled through a critical literature review. The influential experimental variables identified through the review of the literature, namely the geometric ratios (i.e., steel tube diameter-to-wall thickness ratio and length-to-diameter ratio), steel tube strength grade, effective water-to-cement ratio, RA content and axial load ratio were selected as the input parameters to evaluate their influence on the structural performance (i.e., load carrying capacity, stiffness, peak strain, ductility and energy dissipation) of RACFST beams and columns. The results of the grey sensitivity analysis indicate that the effective water-to-cement ratio, steel tube strength grade, and steel tube diameter-to-wall thickness ratio and length-to-diameter ratio are in general the most influential set of parameters on the structural performance of RACFSTs, respectively; whereas, the overall performance of RACFSTs is less sensitivity to the RA content when compared with other parameters. It can also be seen from the results that the RA content influence on the seismic performance of RACFSTs is larger when conventional RAs are used in concrete mixes instead of demolished concrete lumps. To extend the parametric range of the experimental database, BP neural networks were employed to estimate the load carrying capacity of RACFSTs, and the results demonstrate that this machine learning algorithm can simulate the effect of RA content on the load carrying capacity of RACFSTs. Based on the extended database, two simple expressions were proposed to model the RA content influence on the axial and lateral load carrying capacities of RACFST columns.