Experimental and Physics-Informed Synthetic Dataset for Mechanical Properties and Failure Modes of FRP Coupons

This dataset comprises both experimental and physics-informed synthetic data related to the mechanical properties and failure modes of FRP coupons. The experimental data were compiled from peer-reviewed publications spanning 1998-2025 and include key input parameters such as coupon geometry, material properties, specimen configuration, and test methods, along with corresponding outputs (mechanical properties and failure modes). To address data scarcity and enhance analytical capability, high-fidelity synthetic datasets were generated using a Physics-Informed Tabular Variational Autoencoder (PI-TVAE). The quality and reliability of the synthetic data were rigorously evaluated against the experimental dataset using multiple statistical measures, including range comparison (minimum and maximum), standard deviation, Kolmogorov–Smirnov test, entropy difference, and Cramér’s V. The results confirm that the synthetic data closely replicate the statistical characteristics and dependency structures of the experimental data. This dataset is intended to support research in: Development of probabilistic machine learning models for predicting mechanical properties and failure modes of FRP coupons Uncertainty quantification in material performance and structural behavior Data-driven and automated design in structural and civil engineering Benchmarking and validation of generative modeling approaches in engineering applications All datasets are provided in .xlsx format to ensure accessibility, reproducibility, and ease of reuse by the research community.本数据集包含与纤维增强复合材料（Fiber Reinforced Polymer，FRP）试样的力学性能及失效模式相关的试验数据与物理信息驱动合成数据。试验数据源自1998年至2025年间的同行评议文献，涵盖核心输入参数（如试样几何尺寸、材料属性、试件配置与试验方法）及对应的输出结果（力学性能与失效模式）。 为解决数据稀缺问题并提升分析能力，研究人员采用物理信息驱动表格变分自编码器（Physics-Informed Tabular Variational Autoencoder，PI-TVAE）生成了高保真合成数据集。研究团队通过多种统计指标对合成数据的质量与可靠性进行了严格评估，对比基准为试验数据集，指标包括值域对比（最小值与最大值）、标准差、柯尔莫哥洛夫-斯米尔诺夫检验、熵差与克莱姆V统计量。评估结果证实，合成数据能够精准复现试验数据集的统计特征与变量依赖结构。 本数据集旨在为以下研究方向提供支撑： - 面向FRP试样力学性能与失效模式预测的概率机器学习模型开发 - 材料性能与结构行为的不确定性量化研究 - 结构与土木工程领域的数据驱动与自动化设计 - 工程应用中生成式建模方法的基准测试与验证 所有数据集均以.xlsx格式提供，以保障研究群体可便捷获取、复现并复用相关数据。