A Study On Structural Damage Probability Method Based On Time Series

As an important step of structural health monitoring, structural damage identification is crucially important for the subsequent structural load capacity assessment, and is also currently a research area of much interest and importance in field of structural health monitoring system. Most of the traditional structural damage identification methods are based on the model, and need accurate finite model. So they are difficult to adapt to the requirements of large scale civil engineering structural continuous online monitoring. Because of the independence of the structural model, the model-free damage identification methods have certain advantages in the application of damage indentification. As one of the model-free damage identification methods, the damage identification method based on time series analysis needs not to build complex model, and it can use fewer model parameters to express the system information which is contained in a mass of structural response data. Meanwhile the obtained structural time history response data can be directly used to carry out damage identification. Therefore it does not have to canduct the response data’s transformation in time and frequency domains, which is conductive to the structural continuous online monitering.Because of the environmental factor, measurement noise and the error of time series modeling, the damage identification results usually have some randomness. So the damage identification method based on time series analysis usually combines with some statistical pattern recognition methods. But these methods can only give a simple judgement whether the damage occurred, which can not provide the probability for damage. To address this issue, this dissertation develops the study on the time series-based structural damage probability computing method on the basis of statistical pattern recognition. The main work of this dissertation is as followed:(1) The necessity of the study on damage identification together with engineering applications and system architecture of structural health monitoring system are first introduced, and then the principle as well as the classification of the structural damage identification methods is systematically elaborated. Furthermore, the dynamical characteristic-based structural damage identification methods are intensively introduced. Finally, the research status at home and abroad of the time series analysis-based damage identification methods is comprehensive reviewed. In addition, the shortcomings of the existing damage identification methods are pointed out.(2) The time series basic theory is systematically introduced, including basic conceptions and common models of time series. Then the the time-domain characteristic and the modeling procedure of time series model are intensively elaborated. Finally the theory of principal component analysis is introduced.(3) The procedure of the time series-based structural damage probability computing method is established, and the theory and operational approachs of the method’s four basic steps are elaborated, including data sample construction, data normalization, damage index computing, damage threshold determination and damage probability computing. And three damage indexs including the index based on residual, the index based on first three model parameters and the index based on mahalanobis distance, are detailed researched. Then the computing methods of damage threshold and damage probability are systematically proposed. A simply supported beam is regarded as the research object, and its acceleration response data of different damage conditions are obtained via numerical experimentation method. The proposed time series-based structural damage probability computing method is used to compute the damage probability of different damage cases. Then from the damage probability point of view, the differences of sensibility and anti-noise capability among three damage indexs have been analyzed and compared, so as to validate the effectiveness of the proposed method.(4) The general situation of an aluminum simply supported beam model and the vibration testing system are introduced and then the measured acceleration response data of the model are obtained via model experiment method. The index based on residual which has the best damage identification result is selected to conduct the study on actual structural damage identification so as to further validate the validity and practicality of the proposed method.(5) The general information and health monitoring system situation of a long-span arch bridge——Guangzhou Xinguang Bridge with a 428 m main span length are introduced. Then the damage probability analysis of a boom of the bridge is performed using the proposed method, so as to validate the prospect of proposed methods on the engineering application.The main conclusions of this dissertation are as follows:(1) The proposed time series analysis-based damage identification method needs neither complicated structural model nor the structural response data’s teansformation of time and frequency domains. It can take the impact of environmental factor, measurement noise and modeling error on damage identification result into account. Furthermore, it can provide a probability measurement for structural damage, which is fit for the continuous online monitoring of the structural component’s damage identification.(2) The damage identification results of numerical example show that the proposed method can effectively distinguish the structural damage. From the damage probability point of view, in term of sensibility, the index based on residual behaved the best among the three indexs, the index based on first three model parameters takes second place, and the index based on mahalanobis distance is the worst. In term of anti-noise capability, the index based on residual also behaved the best, the index based on mahalanobis distance takes second place, and the index based on first three model parameters is the worst. As far as comprehensive damage identification effect, the index based on residual behaved the best.(3) The damage identification results of the simply supported beam model experiment show that the proposed time series-based structural damage probability computing method can effectively distinguish between normal structure and damaged structure. At the same time, the successful application of the proposed method on the damage probability analysis of Xinguang Bridge’s boom shows that the proposed method has a certain extent prospect of engineering application.