Automatic operational modal analysis is necessary for long-term monitoring of structures when using modal information. Many algorithms have been proposed to accomplish this task; two examples are the fully automatic algorithm by Reynders et al. in 2012 and the semi-automatic algorithm by Kvåle and Øiseth in 2020; however, few in-depth direct comparisons exist. This work compares the two algorithms mentioned above with a new fully automatic algorithm recently developed by the authors, inspired by the best functioning elements from algorithms published since 2008 and following the standard three-step framework. The comparison is performed using 3600 experimental datasets from three real bridges – the Hardanger and Hålogaland suspension bridges and the Bergsøysund floating pontoon bridge – where the detections made by the algorithms are compared to the pre-defined reference modes of the bridges. The new algorithm (Dederichs 2023) performs equally to the Kvåle 2020 algorithm in correct detections and better than the Reynders 2012 algorithm. The two best algorithms detect the easily excited structural modes in over 90 % of the datasets. The Reynders 2012 algorithm is shown to have a more cautious approach to mode detections, making few correct detections and fewer errors. A sensitivity analysis of the criteria to compare detected modes to reference modes shows that this does not impact the comparison’s conclusions.

中文翻译：

---

悬索桥和浮桥三种自动运行模态分析算法的实验比较

使用模态信息时，自动操作模态分析对于结构的长期监测是必要的。已经提出了许多算法来完成这项任务； Reynders 等人的全自动算法就是两个例子。 2012 年提出了半自动算法，2020 年 Kvåle 和 Øiseth 提出了半自动算法；然而，很少有深入的直接比较。这项工作将上述两种算法与作者最近开发的一种新的全自动算法进行了比较，该算法受到 2008 年以来发布的算法中最佳功能元素的启发，并遵循标准的三步框架。使用来自三座真实桥梁（Hardanger 和 Hålogaland 悬索桥以及 Bergsøysund 浮桥）的 3600 个实验数据集进行比较，其中将算法进行的检测与桥梁的预定义参考模式进行比较。新算法 (Dederichs 2023) 在正确检测方面与 Kvåle 2020 算法相同，并且优于 Reynders 2012 算法。两种最好的算法可以检测超过 90% 的数据集中容易激发的结构模式。 Reynders 2012 算法被证明对模式检测有更谨慎的方法，正确检测很少，错误也更少。对检测模式与参考模式进行比较的标准的敏感性分析表明，这不会影响比较的结论。