Evaluating the structural performance of axially loaded stainless steel circular tubular columns with ultra‐high‐performance concrete using an optimized attention pyramid convolutional neural network

Abstract The high rigidity and load‐bearing capacity of concrete‐filled stainless steel‐tube (CFSST) structures make them a common choice for high‐rise building construction. CFSST columns exhibit intricate interactions between concrete and stainless steel, leading to challenges in accurately modeling their behavior under axial loading conditions. Stainless steel tubular circular columns with axially loaded concrete and ultra‐high strength concrete are predicted to perform using a hybrid technique. The proposed hybrid strategy is the collaborative implementation of Attention Pyramid Convolutional Neural Network (APCNN) and Giant Trevally Optimizer (GTO). It is hence called the APCNN‐GTO technique. The primary goal of the proposed approach is to increase the CFSST columns' ultimate axial load‐carrying capability in comparison to conventional reinforced concrete (RC). The data is initially extracted from the dataset of steel tubular columns filled with concrete. They then feed the data into a pre‐processing approach based on federated neural collaborative filtering (FedNCF). To extract the best features, the dual‐tree biquaternion wavelet transform (DTBWT) is used for the pre‐processing result. For circular CFSST columns subjected to axial loading, the ultimate axial load is predicted using the APCNN technique. Next, APCNN's weight parameters are tuned with GTO. The proposed technique's implementation in the MATLAB platform is described as follows: the proposed solution outperforms existing approaches such as CNN, FNN, and RNN by 98% and has a 0.1% lower error rate, paving the path for advanced applications in high‐rise and heavy‐load construction.