To achieve a high-reliability design of high-temperature structures with a feasible balance between accuracy and efficiency, the physics-based probabilistic assessment for creep-fatigue failure is proposed under the probabilistic Linear Matching Method (pLMM) framework. At the physical level, the structural failure mechanism is reflected in the prepared training database, which is generated by the direct method procedures. And to efficiently express the relationship between design parameters and structural responses implicitly, the direct method-driven artificial neural network is built with the superior fitting quality of damage and lifetime. With the benchmarks provided, the applicability of the proposed probabilistic analysis approach for risk management of critical infrastructures is demonstrated, where the reliability-based creep-fatigue evaluation diagram is established according to different requirements. Furthermore, a novel data classification scheme is proposed to deal with the randomness in creep damage-dominated probabilistic assessment.

为了实现高温结构的高可靠性设计，并在精度和效率之间取得可行的平衡，在概率线性匹配方法 (pLMM) 框架下提出了基于物理学的蠕变疲劳失效概率评估。在物理层面，结构失效机制反映在准备好的训练数据库中，该数据库由直接方法程序生成。为了有效地隐含地表达设计参数与结构响应之间的关系，构建了具有优异的损伤和寿命拟合质量的直接方法驱动的人工神经网络。通过提供的基准，证明了所提出的概率分析方法对关键基础设施风险管理的适用性，其中根据不同的要求建立了基于可靠性的蠕变疲劳评价图。此外，提出了一种新的数据分类方案来处理蠕变损伤主导的概率评估中的随机性。