A drive-by bridge inspection framework using non-parametric clusters over projected data manifolds
A drive-by bridge inspection framework using non-parametric clusters over projected data manifolds
复制标题
一种基于投影数据流形上非参数聚类的桥梁巡检车框架
DOI：
10.1016/j.ymssp.2022.109401
复制
发表时间：
2022
影响因子：
8.4
通讯作者：
P. Cheema;M. M. Alamdari-M.;K. Chang;C.W. Kim;Masashi Sugiyama
中科院分区：
文献类型：
--
作者：
P. Cheema;M. M. Alamdari-M.;K. Chang;C.W. Kim;Masashi Sugiyama
发展基于车载传感的桥梁健康监测(BHM)框架，即所谓的直接结构健康监测(SHM)或路桥检测，目前是一个迅速发展的研究领域。其目标是使用安装在桥上的车辆上的传感器测量的振动响应来监测桥梁结构完整性的任何变化。为此，我们提出了一种新的数据驱动方法，将基于统一流形逼近和投影的非线性降维技术(UMAP)与基于带噪声应用的基于层次密度的空间聚类(HDBSCAN)的非参数聚类技术相结合。最后，通过UMAP和HDBSCAN的组合使用，证明了这种方法与分析路桥检测数据时存在的问题和限制很好地吻合。也就是说，在驾车通过桥检查环境中，数据集可以具有低基数，并且各个簇可以具有不同的大小和密度，并且不需要具有球形。此外，随着新的损害案例的演变，新的集群可能会随着时间的推移而出现。UMAP和HDBSCAN的配对将被证明以原则性的方式协助解决这些问题。为了验证，我们首先利用车桥相互作用系统的理论公式，对该系统进行了有限元仿真，得到了五种不同的桥梁状态下移动车辆的响应。在此基础上，以实验室简支桥梁模型为例进行了验证。通过增加多块加劲板，在跨中逐渐增加集中质量，逐步改变桥梁模型的结构。然后，将所提出的方法应用于识别实验室桥梁模型中发生的变化。损伤表征的实验结果表明，可以成功地分离出各种桥态。本文的工作主要有三个方面：(1)首先，提出并验证了一种基于UMAP和HDBSCAN相结合的特征提取和分类框架。(2)在大量的数值和试验研究的基础上，提出了一种早期成功的桥梁检测方法，用于监测桥梁结构的渐进变化。(3)本文的研究成果为桥梁网络状态监测提供了新的思路。
Developing robust bridge health monitoring (BHM) frameworks based on the vehicle-mounted sensing, or so-calledindirect structural health monitoring (SHM)orDrive-by Bridge Inspectionis currently a rapidly growing research area. The goal is to monitor any change in the structural integrity of a bridge using the vibration responses measured from sensors installed on a vehicle passing over the bridge. To this aim, we present a novel data-driven approach by integrating a nonlinear dimensionality-reduction technique using Uniform Manifold Approximation and Projection (UMAP) together with a non-parametric clustering technique using Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN). Ultimately, through the combined use of UMAP and HDBSCAN, it is demonstrated that such methods align perfectly well with the problems and restrictions present when analyzing drive-by bridge inspection data. That is, in the drive-by bridge inspection context data sets may have low cardinality, and the individual clusters may have different sizes and densities and do not need to have a spherical shape. Furthermore, new clusters may emerge over time as new damage cases evolve. The pairing of UMAP and HDBSCAN will be shown to assist in these issues in a principled manner. For the validation, we first make use of the theoretical formulation of the vehicle-bridge interaction system and conduct a finite element simulation of this system to obtain the moving vehicle response once traveling over five different bridge states. Further, a simply supported bridge model in the laboratory is considered for the experimental validation. The structure of the bridge model is gradually changed by adding multiple stiffener plates, and by gradually increasing a concentrated mass at mid-span. The proposed method is then applied to identify changes that occur in the laboratory bridge model. The experimental results for damage characterization demonstrate that various bridge states can successfully be separated from one another. The contributions of the work are three-fold; (1) First, a novel feature extraction and classification framework based on the combined use of UMAP and HDBSCAN is developed and validated. (2) Second, this paper demonstrates one of the early successful attempts of drive-by bridge inspection for monitoring the progressive change in the structure of a bridge based on the extensive numerical and experimental investigations. (3) Finally, the research presented in this work can open up new opportunities for condition monitoring of bridge network.