Structural damage identification using Parzen-window approach and neural networks

Yuyin Qian, Akira Mita

研究成果: Article査読

13 被引用数 (Scopus)


The basis for proposed approach to damage identification is pattern changes in the structure's physical properties. This paper proposes a non-destructive testing technique based on modal analysis in order to develop a new, efficient and simple damage detection method for civil structures. The fundamental idea of the pattern classification approach is to use training data to determine the classifiers to evaluate the categories of the test data. However, a very large database is required to store training data for complicated damage cases if no technique to reduce the size is used. This paper presents an approach with the purpose of using smaller database. The structural damage location and damage extent are identified, respectively, by two different methods of pattern classification; firstly, the Parzen-window method for the structural damage location and secondly the feed-forward back-propagation neural network used to identify damage extent. The measured structural vibration responses data always contain noise. The output inevitably has some errors when the data with noise is inputted into the classifiers network. The proposed approach was thus enhanced to have stability against such noise by considering variations in signals. The results of numerical simulations show that our proposed approach can indeed identify the structural damage for shear building structures using small training data. Finally, a series of vibration experiments for the 5-storey shear frame structure were performed to verify the performance of our proposed approach. The results show that for shear buildings, damage degree and extent can be successfully determined through measuring the frequency change.

ジャーナルStructural Control and Health Monitoring
出版ステータスPublished - 2007 6 1

ASJC Scopus subject areas

  • 土木構造工学
  • 建築および建設
  • 材料力学


「Structural damage identification using Parzen-window approach and neural networks」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。