Buckling failure prediction of storage tanks in fire scenarios: a machine learning approach

Shenbin Xiao, Lijun Wei, Tao Zeng, Haishun Wang, Guoliang Yang, Chao Chen,
Buckling failure prediction of storage tanks in fire scenarios: a machine learning approach,
Engineering Failure Analysis,
Volume 182, Part C,
2025,
110157,
ISSN 1350-6307,
https://doi.org/10.1016/j.engfailanal.2025.110157.
(https://www.sciencedirect.com/science/article/pii/S1350630725008982)
Abstract: Storage tanks are common pressure vessels in chemical industrial parks. Their risks in fire scenarios are characterized by dynamics and coupling. Therefore, the failure analysis and prediction of storage tanks are of great significance for the safety of chemical industrial parks. Previous studies focus on the static characteristic analysis of storage tanks, while research on dynamic thermo-mechanical coupling response mechanisms and failure prediction remains limited. Additionally, conventional numerical methods typically incur substantial computational costs in multi-parameter coupling analysis. The methods also lack universality across different application scenarios and exhibit insufficient timeliness in delivering analysis results. This study establishes a dual-driven “numerical simulation-machine learning” framework and employs finite element analysis to develop a solid double-layer flame model combined with the artificial damping method (ADM) to simulate dynamic thermomechanical responses and buckling failure processes. This study incorporates three machine learning models—Backpropagation Neural Network, Convolutional Neural Network, and Random Forest—to develop a predictive methodology for tank buckling failure under multi-parameter coupling conditions. The methodology is based on numerical analysis results of the height-to-diameter ratio, liquid height, and pool flame height. The models are trained using these geometric and thermal parameters as inputs, with the buckling failure time of the storage tank as the output. This method systematically compares the mean relative error to identify the optimal predictive model, effectively capturing the complex nonlinear characteristics of storage tank failure mechanisms. The results reveal the liquid level and height-to-diameter ratio delay failure through specific heat capacity and structural effects, while pool flame height prolongs failure via thermal radiation. The Backpropagation neural network demonstrates optimal performance. This prediction model enables real-time prediction of tank failures in fire scenarios and provides an innovative solution for the intelligent upgrade of safety protection in tank farms.
Keywords: Storage tank; Buckling; Machine learning; Multi-parameter coupling; Failure prediction