Structural Damage Detection Based On Autoregressive Model And Principal Component Analysis

During the life period of civil engineering structures, invasions by naturalenvironmental causes and man-made factors are unavoidable to these structures,which lead to degradation and damage in the structures. Failure to detect andrepair these functional deficits may lead to catastrophe events. Therefore, thestructural damage identification is developing into a research area which are ofgreat significance both in theoretical and application aspects. In this paper, timeseries autoregressive model and principal component analysis are employed forstructural damage detection, and concentration is made on the removal ofenvironmental effects on the structural damage identification results. T he maincontents are as follows:Studies on topics of environmental effects on the structural dynamicresponse and the damage identification theory based on structural vibration arereviewed. The basic theory of the autoregressive model is introduced. Modelorder determination criteria and parameter estimation methods are compared.The recursive least squares estimation method combined with the BIC criterioncan ensure the accuracy and fast calculation for parameter estimation, and it issuitable for online monitoring.Two damage identification methods based on the autoregressive model areintroduced, namely, the Euclidean distance discrimination method and theItakura distance discrimination method based on the model residuals. Throughexperimental studies on a truss model, the effectiveness of the two methods isverified, but they are susceptible to environmental factors.The damage detection method based autoregressive model is improved byintroducing the principal component analysis (PCA). Autoregressive model isconstructed using acceleration of structures. Principal component analysis isapplied to extract and eliminate the environmental components fromautoregressive model coefficients. Reconstructed coefficient vectors of theautoregressive model are used to define the damage index for damage detection.The effectiveness of this method is proved by experiments on a truss model, andit is immune to environmental effects.A damage detection method by combing the autoregressive model andkernel principal component analysis (KPCA) is proposed. KPCA is used forextracting the characteristics of autoregressive coefficient vector, which is takenas a reference sample. The sample vectors with the same characteristics as the reference autoregressive coefficient vector are searched in the feature space. Thedamage is identified by comparing the similarity of these two vectors.Experimental studies show that this method can eliminate the e nvironmentaleffects and identify damage accurately.Real monitoring data from the Foshan Pingsheng bridge are employed totest the two proposed method. Results show that the autoregressive modelcombined with KPCA is more suitable than that with PCA for practicalapplications.