基于传感器集群BLSTM模型的结构损伤定位  

Structural damage localization based on BLSTM model with sensor clustering

作　　者：韩庆华[1,2,3] 马乾 党大智 徐杰[1,2,3] HAN Qinghua;MA Qian;DANG Dazhi;XU Jie(Key Laboratory of Earthquake Engineering Simulation and Seismic Resilience of China Earthquake Administration,Tianjin University,Tianjin 300350,China;Key Laboratory of Coast Civil Structure Safety of Ministry of Education,Tianjin University,Tianjin 300350,China;School of Civil Engineering,Tianjin University,Tianjin 300350,China)

机构地区：[1]天津大学中国地震局地震工程综合模拟与城乡抗震韧性重点实验室,天津300350 [2]天津大学滨海土木工程结构与安全教育部重点实验室,天津300350 [3]天津大学建筑工程学院,天津300350

出　　处：《振动与冲击》2022年第22期33-41,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金联合基金项目(U1939208);高等学校学科创新引智计划资助(111计划-B20039)。

摘　　要：为进一步提高无监督模式下的结构损伤定位精度,提出了一种基于传感器集群BLSTM模型的结构损伤定位方法。通过传感器集群BLSTM模型的建立,对不同位置处传感器的虚拟脉冲响应函数进行了预测,进而通过基准工况与未知工况模型预测残差的信息熵,构造了一种新的损伤识别指标,并依此进行损伤定位。通过8自由度质量弹簧系统试验、工字钢梁模型试验,以及桁架结构数值模拟,对方法的有效性和适用性进行了验证。结果表明,该研究提出的基于传感器集群BLSTM模型的结构损伤定位方法,可以对结构的单一位置损伤和多位置损伤进行准确定位,且可通过同一位置处损伤指标的变化对损伤程度进行判别,即使在外界环境变化和较高的噪声干扰下,仍能取得较好的损伤定位效果,具有良好的环境适应性和抗噪声干扰能力。To further improve the accuracy of structural damage localization in unsupervised mode, a structural damage localization method based on the BLSTM model with sensor clustering was proposed. Firstly, the virtual impulse response functions of sensors at different locations were predicted through the establishment of BLSTM models for different sensor clusters, and then a new damage identification index was constructed by the information entropy of the predicted residuals between the baseline condition and the unknown conditions, and damage localization was carried out accordingly. The validity and applicability of the method were verified by an experiment on an 8 DOF mass-spring system, an experiment on a steel I-beam and a simulation of truss structure. The results show that the proposed method not only can accurately localize both single-site and multi-site damage to a structure, but also identify the degree of damage by comparison of damage identification indexes at the same location. Even under the change of external environment and high noise interference, the proposed method can still have good damage localization results, which has good environmental adaptability and noise interference resistance.

关 键 词：损伤定位 双向长短时记忆神经网络 信息熵 统计分析 传感器集群 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3