基于堆叠去噪自编码器的桥梁损伤定位方法研究

Damage Location Identification of Bridge Structures Based on Stacked Denoising Auto-Encoder

现有的损伤识别方法面对大量的桥梁健康监测数据存在处理能力有限、提取的信息不能全面反应桥梁的健康状态的问题。利用深度学习在大数据方面的优势,提出一种基于堆叠去噪自编码器的桥梁损伤定位方法。以一简支梁桥有限元模型算例对该方法进行验证,提取该桥L/5,2L/5,3L/5,4L/5处的竖向加速度时程响应值,并针对每一个时刻建立上述节点的加速度矩阵(4×1),然后将经过预处理的矩阵送入堆叠去噪自编码器进行特征提取完成模型训练,最后将测试样本送入该模型进行分类,完成损伤定位任务。结果表明本文提出的方法相比于传统的机器学习方法具备定位准确率高和抗噪性能好的优势,在今后的桥梁结构损伤识别领域具有一定的参考价值。

The existing damage identification methods are faced with the problem that the processing capacity of a large number of bridge health monitoring data is limited and the extracted information cannot fully reflect the health status of the bridge.Based on the advantage of deep learning in big data,a method of bridge structure damage location identification based on stacked denoising auto-encoder is proposed.The method is verified by a finite element model of a simply supported beam bridge.The vertical acceleration time history response value at L/5,2 L/5,3 L/5,4 L/5 of the bridge are extracted.The acceleration matrix(4 x 1)at every moment is established,then the preprocessed matrix is sent to the stacked denoising auto-encoder for feature extraction to complete the model training.Finally,the test samples are sent to the model for classification to complete the task of damage location identification.The results show that the proposed method has the advantages of high positioning accuracy and good anti-noise performance compared with the traditional machine learning method,and has certain reference value in the field of bridge structure damage identification in the future.