An advanced method for predicting the axial compressive bearing capacity of fiber-reinforced polymer-strengthened hollow section metal columns

Abstract

Successful experimental and theoretical studies have promoted fiber-reinforced polymer (FRP) as an innovative strengthening material for metal columns in structural engineering, highlighting the necessity for an effective and reliable calculation method for such strengthened members. Here, based on existing experimental data, an advanced method using an artificial neural network (ANN) is developed to predict the axial compression bearing capacity (ACBC) of FRP-strengthened hollow section metal (FSHM) columns, where FRP strengthening includes wrapping/bonding FRP sheets/laminates; this method considers realistic conditions, incorporates a variety of parameters, and produces accurate and fast predictions. First, a multilayer perceptron (MLP) neural network was applied. By studying the optimization algorithm, learning rate, hidden vector and activation function, an optimal ANN model was established, which yielded better predictive accuracies and safety than existing calculation methods. Using the validated model, an prediction program was developed that promptly generated predictions of ACBC with a high reliability. Finally, a sensitivity analysis was performed to further assess the effects of key parameters on the behavior of FSHM columns to provide practical recommendations for their design.