European Workshop on Structural Health Monitoring - EWSHM 2022 - Volume 3
European Workshop on Structural Health Monitoring - EWSHM 2022 - Volume 3
复制标题
欧洲结构健康监测研讨会 - EWSHM 2022 - 第 3 卷
DOI：
10.1007/978-3-031-07322-9_48
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Gibson S
中科院分区：
文献类型：
--
作者：
Gibson S
疲劳是结构失效的常见原因：作者先前的工作已经表明使用高斯过程回归来开发疲劳损伤累积的概率评估的有效性。通过传播未知载荷下结构响应（应变）的预测模型的不确定性，可以对结构的损伤状态进行更可靠的评估。虽然这些黑盒模型以前在准静态问题上显示出良好的结果，但用这种方法很难预测动态行为。这里探讨了一种有前途的、新颖的方法来解释这一点，即通过GP内核集成物理知识。这对疲劳损伤预测精度的影响是显著的，从概率角度来看，损伤方差大大减小。
Fatigue is a common cause of the failure of structures: previous work by the authors has shown the usefulness of using Gaussian process regression to develop a probabilistic assessment of fatigue damage accumulation. By propagating uncertainty from a predictive model for structural response (strain) under unknown loading, a more robust assessment of the damage state of a structure is enabled. Although these black-box models have previously shown good results for quasi-static problems, dynamic behaviour is difficult to predict in this way. Explored here is a promising and novel means of accounting for this, by integrating physical knowledge specifically through the GP kernel. The impact of this on accuracy of fatigue damage prediction is shown to be significant and the damage variance from a probabilistic perspective is reduced substantially.