Damage detection of bridge structures under changing environmental and operational conditions

This thesis is focused on developing a vibration-based damage detection method to analyse bridge structures subjected to changing environmental and operational conditions by establishing an effective baseline to represent the undamaged state of the structures. The thesis is divided into two parts. The first part of the thesis is focused on eliminating the effects of outlier measurements affecting a database of damage sensitive features used to construct the baseline of the undamaged structure. Outlier measurements create a larger variation of damage sensitive features in the normal condition, which affect the sensitivity of damage detection methods. Moreover, these outliers affect the baseline by creating a defective model of the undamaged structure, which may lead to false alerts. Therefore, a method is developed in this research work to identify and highlight the outlier measurements before the application of damage detection methods. The proposed method uses the Principal Component Analysis to indicate where the outliers are in relation to the good observations, and the Median Absolute Deviation to separate the good observations from the outliers. The method developed also makes use of Gaussian Mixture Model to indicate the number of environmental and operational mechanisms affecting the features. The benefit of this research work when compared to the traditional damage detection methods which use a database composed of both good and bad measurements for the baseline is that, a cleaning step is implemented before the construction of the baseline. This allows damage detection methods to detect damage at an earlier stage and to reduce the occurrence of false alerts. The second part of this thesis is focused on developing a vibration-based damage detection method that uses natural frequencies obtained at two extreme and opposite temperature conditions as the baseline of the undamaged structure. The proposed method uses extreme and opposite temperature conditions so as to cover all range of values of natural frequencies the structure under consideration may have. The method uses frequencies captured under limited temperature conditions to create the baseline. This is particularly important because damage may occur at an early stage, which may prevent the collection of damage sensitive features from a large range of environmental and operational conditions. Piecewise-linear model is adopted to construct the baseline to represent the different environmental and operational mechanisms affecting the natural frequencies. The proposed method makes use of Principal Component Analysis for data processing to indicate where the monitored observation is in relation to the baseline observations. Outlier analysis is employed to classify between undamaged and damaged states. A numerical beam model, an experimental beam structure and two real-life bridge structures (the Z24 Bridge, in Switzerland, and the Yonghe Bridge, in China), are adopted as case studies in this thesis. All four structures were subjected to changing environmental and operational conditions as well as to damage of structural components. The results obtained highlight the importance of cleaning the database of damage sensitive features before the construction of the baseline so as to alert damage at an earlier stage and to prevent creating a defective model of the undamaged structure. The case studies analysed, especially the Yonghe Bridge case study, show the advantages of using the extreme and opposite conditions as the baseline since damage occurred at an early stage in the structure