Prediction of the Ultimate Load-Carrying Capacity of Aluminium Alloy Columns Based on Multi-Objective Particle Swarm Optimisation and Gaussian Process Regression

As a structural material characterised by low density, high strength, excellent corrosion resistance and recyclability, aluminium alloy tubes are finding increasingly widespread application in the construction sector. However, there is currently a lack of research on the prediction of the bearing capacity of aluminium alloy square tube columns. To investigate the failure behaviour of aluminium alloy square tube columns under axial and eccentric compression, this paper first designed 10 thin-walled aluminium alloy square tube column specimens with varying lengths, cross-sectional dimensions and wall thicknesses. Axial and eccentric compression tests were conducted, and the loading process and failure modes were analysed. Building on this, a hybrid load-bearing capacity prediction model combining Multi-Objective Particle Swarm Optimisation (MOPSO) with the Gaussian process regression (GPR) algorithm was proposed. This model is capable of automatically learning and capturing 236 sets of experimental data. Subsequently, using the established prediction model, the contributions of high-sensitivity parameters and cross-sectional influence parameters to the load-bearing capacity were determined. Based on the prediction results, a correction factor for the diameter-to-thickness ratio was introduced into the eccentric compression bearing capacity formula of the Chinese code to establish an improved calculation formula. Compared with the implicit formula provided by machine learning models, the explicit formula proposed in this paper is more suitable for practical engineering design. The results show that the prediction results agree well with the experimental results and can accurately predict the ultimate bearing capacity of aluminium alloy square columns. Compared with the bearing capacity calculation methods in existing codes, the proposed formula reduces the root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2) of the dataset by 70.91%, 70.85% and 64.27%, respectively, whilst increasing the coefficient of determination (R2) from 0.8107 to 0.9830 (a relative improvement of 21.25%).