Experimental Testing and PSO-Enhanced Neural Networks for Impact Failure Analysis of H-Section Steel Members

H-section steel members, as a commonly used load-bearing receiving member in building structures, may be subjected to the impact of accidental loads during their service life, and therefore, the impact loads need to be considered when carrying out the design. In this paper, based on experimental testing, the particle swarm optimization algorithm (PSO) is used to optimize the hyperparameters of the multilayer perceptron (MLP), and a combined prediction model PSO-MLP for H-section steel members subjected to lateral impact loads is proposed to predict the damage of the H-section steel members after impact. The results show that the prediction model based on PSO-MLP can predict the damage of the H-beam columns more accurately, and compared to the random forest model (RF) and the support vector machine (SVM), the PSO-MLP model has better prediction accuracy and robustness. In addition, the effects of different features on the impact performance of the members were analyzed, in which the weakest impact location is 0.57 L away from the fixed end and the effects of axial compression ratio, flange, and web thickness were similar to the results of previous studies; the impact angle showed a strong nonlinear relationship with the critical impact velocity, which the weakest impact angle is around 50° from the strong axle; and the height and width of the cross-section showed a linear enhancement of the impact performance.