The spatial distribution of the non-Gaussian characteristics of wind pressure and rapid and accurate determination of peak factors for long-span tri-circular cylindrical roofs with dominant openings is crucial for enhancing the efficiency of wind load assessment. This study, based on wind tunnel tests of a roof with openings under 36 wind directions, developed an optimized artificial neural network model to predict peak factors from skewness, kurtosis, and mean pressure coefficients. The model achieved R 2 > 0.95 across all measurement points, with a maximum absolute error of 0.35, demonstrating high accuracy. Statistical analysis of 32,256 samples revealed pronounced non-Gaussian characteristics near roof edges due to flow separation and vortex shedding, primarily driven by external pressure. The external pressure is the primary driver of the non-Gaussian characteristics of net pressure. Subsequently, the peak factors g peak are calculated using the translation process theory and compared to the peak factor g peak,G determined based on the Gaussian assumption. The results indicate that the peak factor derived from the Gaussian assumption and the standard-specified value of 2.5 are significantly smaller compared to that estimated through the translation process theory. Therefore, the non-Gaussian properties must be considered when estimating the extreme wind pressure in wind-sensitive regions of similar long-span roof structures. The study provides both insight into non-Gaussian wind effects and an efficient predictive tool for wind-resistant design and evaluation of similar structures.
Zhang et al. (Thu,) studied this question.