Abstract Hurricanes are responsible for approximately $28bn of damage every year in the United States alone, which may reach $151bn by 2075 due to the intensification of climate change according to certain prediction models. Approximately 35% of this damage is estimated to come from anchorage failures of non-structural components (NSCs). Severe exposure of NSCs to the adverse environments (such as elevated temperatures and long-term concrete cracking) and wind-induced bending effects during hurricanes promote anchorage failures. Three-dimensional (3D) nonlinear finite element (NLFE) analysis methods are currently required for simulating the anchor behavior due to the 3D phenomena involved; however, these models are rather complex and computationally prohibitive for analyzing large systems commonly encountered in practice. This study proposes a 2D analysis methodology that combines the strengths of 3D numerical modeling with the artificial neural network techniques to rapidly simulate the anchorage behavior while accounting for the effects of the adverse environmental exposure, concrete cone failure, and wind-induced bending effects. The methodology, which is validated with experimental data and 3D NLFE analyses, employs three distinct techniques as follows: (i) a novel modeling approach, ‘the Equivalent Cone Method,’ to accurately simulate the concrete cone breakout failure, (ii) analytical equations developed to account for wind-induced beam bending and elevated temperatures, and (iii) a multilayered feed-forward artificial neural network, trained and tested with the experimental data from a worldwide database, to rapidly account for long-term concrete cracking experienced by rooftop slabs. By employing these techniques, the proposed methodology permits the use of 2D NLFE models for anchor analysis with accuracies comparable to advanced 3D NLFE models but at a fraction of the computational cost.

中文翻译：

---

一种使用有限元和神经网络的混合方法，用于分析暴露于飓风和不利环境的粘合剂锚

摘要 飓风每年仅在美国就造成约 280 亿美元的损失，根据某些预测模型，由于气候变化加剧，到 2075 年可能达到 1510 亿美元。据估计，这种损坏中约有 35% 来自非结构部件 (NSC) 的锚固失效。NSC 严重暴露于不利环境（如高温和长期混凝土开裂）和飓风期间风引起的弯曲效应会导致锚固失效。由于所涉及的 3D 现象，目前需要 3D (3D) 非线性有限元 (NLFE) 分析方法来模拟锚的行为；然而，这些模型对于分析实践中常见的大型系统来说相当复杂且计算量大。本研究提出了一种 2D 分析方法，该方法将 3D 数值建模的优势与人工神经网络技术相结合，以快速模拟锚固行为，同时考虑不利环境暴露、混凝土锥体破坏和风致弯曲效应的影响。该方法经实验数据和 3D NLFE 分析验证，采用以下三种不同的技术：(i) 一种新颖的建模方法，“等效锥法”，以准确模拟混凝土锥体破裂故障，(ii) 分析方程开发用于解释风引起的梁弯曲和升高的温度，以及 (iii) 多层前馈人工神经网络，使用来自全球数据库的实验数据进行训练和测试，快速解决屋顶板经历的长期混凝土开裂。通过采用这些技术，所提出的方法允许使用 2D NLFE 模型进行锚点分析，其精度可与高级 3D NLFE 模型相媲美，但计算成本仅为一小部分。