Structural damage diagnosis based on image processing of response data and deep residual shrinkage network(PDF)
《建筑科学与工程学报》[ISSN:1673-2049/CN:61-1442/TU]
- Issue:
- 2026年01期
- Page:
- 28-40
- Research Field:
- 智能检测与建造技术专栏
- Publishing date:
Info
- Title:
- Structural damage diagnosis based on image processing of response data and deep residual shrinkage network
- Author(s):
- LI Shujin; ZHANG Jieling; ZHAO Yuan
- 1. School of Civil Engineering and Architecture, Wuhan University of Technology, Wuhan 430070, Hubei, China; 2. Sanya Science and Education Innovation Park of Wuhan University of Technology, Sanya 572025, Hainan, China
- Keywords:
- damage diagnosis; deep residual shrinkage network; convolutional neural network; Gramian angular field; multilabel classification; data enhancement
- PACS:
- -
- DOI:
- -
- Abstract:
- Utilizing the advantages of convolutional neural network(CNN) in processing two-dimensional images and the good performance of deep residual shrinkage network(DRSN) in noise resistance, stability and robustness, a new method of structural damage diagnosis using DRSN after two-dimensional processing on dynamic response signal was proposed. Taking the node damage diagnosis problem of planar and spatial frame structures under complex damage conditions as the research object, starting from two aspects of model sample input and feature learning, the one-dimensional structural dynamic response signals were constructed into image enhancement samples by means of Gramian angular field(GAF) transformation and data enhancement processing. At the same time, a DRSN multi-label classification model suitable for the diagnosis of damage location and damage degree of frame structure nodes was constructed, and the damage diagnosis performance was studied from the aspects of training convergence speed, diagnosis accuracy, training sample category and network structure. In addition, in order to further verify the effectiveness and practicability of the proposed method, the anti-noise performance under strong noise interference and the generalization performance when processing new samples were studied. The results show that the DRSN model trained with image enhanced samples performs better than the common CNN model in terms of diagnostic accuracy, iterative convergence speed and stability, and shows good robustness and adaptability on different diagnostic objects. The adaptive threshold reduction mechanism of DRSN has excellent anti-noise performance and generalization performance, which makes it more advantageous in the case of strong noise and small sample.
Last Update: 2026-01-20