Semi‐supervised structural linear/nonlinear damage detection and characterization using sparse identification
Semi‐supervised structural linear/nonlinear damage detection and characterization using sparse identification
复制标题
DOI：
10.1002/stc.2306
复制
发表时间：
2018-12
影响因子：
5.4
通讯作者：
Zhilu Lai;Satish Nagarajaiah
中科院分区：
文献类型：
--
作者：
Zhilu Lai;Satish Nagarajaiah
基于振动的结构损伤检测技术作为结构健康监测的一种有效手段，在土木工程界得到了广泛的研究。结构损伤通常被假定为线性刚度折减，这不能代表真实的世界中的非线性损伤，如塑性变形、裂纹、接头松动等。基于新奇检测和基于监督分类的方法已被用于非线性损伤检测，但无法获得损伤的直接物理解释。为了解决这一限制，这项工作的主要范围是以半监督的方式检测和物理表征线性/非线性类型的结构损伤。所提出的方法首先使用以前提出的稀疏识别建立一个基线（未损坏）模型。然后，损伤被认为是在结构系统中的恢复力的变化。它可以进一步作为一个伪力沿着施加到结构系统的外部干扰力。这样，损伤系统就转化为一个等效线性系统和非线性恢复力。通过比较基线模型预测的相应输出和新测量的数据，识别质量归一化恢复（伪）力的变化，其可用于进一步的损伤表征。一个说明性的例子和三个实验测试介绍在这项工作中，以验证所提出的方法的有效性。
As a useful class of techniques for structural health monitoring, vibration‐based structural damage detection has been intensively studied in civil engineering community for decades. Structural damage is commonly assumed to be linear stiffness reduction, which cannot represent nonlinear damage from the real world, such as plastic deformation, cracks, and joint looseness. Novelty detection‐based and supervised classification‐based methods have been used for nonlinear damage detection, but direct physical interpretation of the damage cannot be accessed. To address this limitation, the major scope of this work is to detect and physically characterize linear/nonlinear‐type structural damage in a semi‐supervised way. The proposed method first uses previously proposed sparse identification to establish a baseline (undamaged) model. Then, damage is considered as a variation of restoring forces in the structural system. It can be further treated as a pseudoforce along with the external disturbance force applied to the structural system. Thus, the damaged system is transformed into an equivalent linear system and nonlinear restoring force. By comparing the corresponding outputs from the prediction of baseline model and newly measured data, a variation of mass‐normalized restoring (pseudo) force is identified, which can be used for further damage characterization. An illustrative example and three experimental tests are introduced in this work to verify the effectiveness of the proposed method.