Structures 28 (2020) 2203–2220 Available online 29 October 2020 2352-0124/© 2020 Institution of Structural Engineers. Published by Elsevier Ltd. All rights reserved. Application of ANN to the design of CFST columns Mohammadreza Zarringol a , Huu-Tai Thai b, * , Son Thai c, d , Vipulkumar Patel a a School of Engineering and Mathematical Sciences, La Trobe University, Bundoora, VIC 3086, Australia b Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC 3010, Australia c Faculty of Civil Engineering, Ho Chi Minh City University of Technology (HCMUT), 268 Ly Thuong Kiet Street, District 10, Ho Chi Minh City, Viet Nam d Vietnam National University Ho Chi Minh City (VNU-HCM), Linh Trung Ward, Thu Duc District, Ho Chi Minh City, Viet Nam A R T I C L E INFO Keywords: Artifcial neural network Concrete-flled steel tubular column Empirical equation Reliability analysis ABSTRACT In this paper, artifcial neural network (ANN) is used to predict the ultimate strength of rectangular and circular concrete-flled steel tubular (CFST) columns subjected to concentric and eccentric loading. Four comprehensive datasets are compiled and used for developing ANN-based predictive models. Empirical equations are also derived from the weights and biases of the ANNs to predict the ultimate strength of CFST columns. The proposed empirical equations can be used for both normal strength and high strength CFST columns with different section slenderness ratios (compact, non-compact and slender sections), and with different length-to-depth ratios (stub and slender columns). The test results are then compared with those predicted from the proposed empirical equations, American code, European code and Australian code. The comparison study shows that the ultimate strengths predicted from the proposed equations have the best agreement with the experimental results with the least mean square error (MSE). In addition, strength reduction factors for the proposed equations are derived using Monte Carlo simulation (MCS). The use of the proposed strength reduction factors will ensure that CFST columns designed by the developed ANN-based equations are safe because their reliability indices meet the target value of 3.0 required by American code or 3.8 required by European and Australian codes. 1. Introduction Artifcial neural networks (ANN) were developed to model the brain’s neural learning process for computer-based artifcial intelligence systems [1]. The ANN has the ability to recognize and learn the re- lationships between given information (inputs) and desired values (outputs) and to analyse complex datasets that are hard to process with the conventional mathematical methods [2,3]. An ANN consists of non- linear interconnected processing elements working in unison to produce an output (outputs). The learning process of ANNs can be divided into three types: (i) supervised learning which is based on the direct compare between the target and output values, (ii) unsupervised learning which is only based on the correlation of the inputs, and (iii) reinforcement learning which is a special case of supervised learning [4]. Since the predictions obtained using ANN are based on the experimental data, the predictions could be more realistic than other methods e.g. fnite element (FE) analysis [5]. The neural network approach can be used in structural design to replace the structural analysis process with a neural network [6]. ANNs have been successfully used in solving various structural problems such as structural dynamics [7,8], damage assessment [9,10], prediction of compressive concrete strength [11,12], modelling of confned concrete [13,14], stability problems [5,15], structural analysis and design [16,17], vibration of structures [18,19], modelling of fatigue crack growth [20,21], fre resistance [22,23] and reliability assessment of structures [24,25]. One of the applications of neural networks in structural design is predicting the ultimate strength of concrete-flled steel tubular (CFST) columns. CFST columns are composite members that have been widely used in tall buildings, bridge piers and long-span bridges due to the composite action between the concrete and steel tube, resulting in higher strength, higher energy ab- sorption capacity and better ductility compared to conventional steel and concrete columns [26,27]. In CFST columns, the steel tube provides confnement to the concrete core, and the concrete core delays the occurrence of the local buckling of the steel tube. Many design codes such as Australian code AS 5100 [28], American code AISC 360-16 [29], European code EC 4 [30], Chinese code DBJ 13-51 [31] and Japanese code AIJ [32] provide provisions to predict the strength of CFST col- umns. However, each design code shows a different level of accuracy and is only applicable to a certain range of material and geometry properties [33,34]. * Corresponding author. E-mail address: tai.thai@unimelb.edu.au (H.-T. Thai). Contents lists available at ScienceDirect Structures journal homepage: www.elsevier.com/locate/structures https://doi.org/10.1016/j.istruc.2020.10.048 Received 3 August 2020; Received in revised form 16 October 2020; Accepted 20 October 2020