Cylindrical shells in a wind flow: regression analysis of the wind pressure distribution coefficient

The article considers the behavior of thin-walled vertical cylindrical steel shell structures (capacities range from 300 to 5000 m³) in a gas flow. Under wind load, the main danger for such structures is global or local buckling. In technical applications, in order to simplify buckling analysis, we usually perform a transition from a complex distribution of a wind pressure q2w to an equivalent uniform pressure q2we=q2w·kw. According to the design standard for storage tanks, it is assumed that the wind pressure distribution coefficient kw equals to 0.5 for all tanks, regardless of geometric parameters of the shells. However, this generalization is not accurate since the value of the coefficient depends on the parameters of cylindrical shells subjected to a gas flow with specific characteristics. The value adopted in normative documents does not provide the required level of reliability. However, under certain conditions, kw can be calculated as the ratio of the critical value of the external uniform pressure pcr2 to the critical value of the frontal wind pressure qcr2w. In this article, the critical frontal quasi-static wind load qcr2w (which causes shell buckling) is determined through numerical simulation of a wind flow in OpenFOAM using the k-ω Shear Stress Transport model. The critical uniform pressure pcr2 and, thus, buckling of the shell are determined by verifying the singularity of the tangent stiffness matrix of the corresponding shell structure. The kw coefficient is calculated using obtained values of the critical pressures, as described above. Using machine learning methods in the R statistical package, we have built a regression model, which allows us to calculate the kw coefficient using the geometric parameters of the shell. The adjusted Akaike Information Criterion (AICc) and the leave-one-out cross-validation method have been used to verify the adequacy of the model, along with the estimation of the mean squared error (MSE) and mean absolute error (MAE). The final regression model for the kw coefficient quite accurately corresponds to the data obtained through numerical simulation.