Structural Damage Detection Under Environmental Variability

Structural damage detection (SDD) has become a hotspot and intractable issue in civil engineering. In recent years, many methods have been proposed for structural damage detection, but there are still many challenges and obstacles to be overcome for its application in field, one of which is the environmental and operational variability. Focused on the effect of environmental and operational variability on the structures, the contents of this dissertation are summarized as follows:(1)The vibration-based SDD methods are reviewed. The effects of environmental and operational variability on dynamic characteristic of structures and on the SDD methods are evaluated. The classification framework of vibration-based SDD methods is proposed, in which five ideas forming these methods are classified. They are based on dynamic signatures, model updating, inverse problem, pattern recognition and artificial intelligence and operational modal analysis respectively. The advantages and disadvantages of these types of methods are also discussed.(2)A new algorithm extracting damage sensitive features has been proposed based on the frequency response functions (FRF) and principal component analysis (PCA). This approach is based on the premise that there exist such an eigenspace reflecting the healthy structure. If there were damage in the structure, the newly FRF projections on the eigenspace previously obtained using the healthy FRF would change, thus the damage information can be extracted. Based on the distributions of the principal components (PCs) and the robustness of the sample median values to its extreme values, the median values of the PCs are defined as a damage feature. The robustness of the feature index on the environmental variability has been discussed by some numerical simulations including a simply supported beam, multi-span continuous beam and two-story frame structure, and a series of experimental data from three-story frame structure.(3)An projection pursuit clustering method for the SDD is proposed and further one feature extraction method developed as well. The PCA is explained in view of projection pursuit, and the median values for PCs and kernel PCs of FRF data can be classified based on fuzzy c-means (FCM) clustering. The memberships of each cluster can indicate the probability damage occurs. This method has been successfully applied to the whole state monitoring of three-story frame structure in laboratory. It has many merits, such as lower computational costs, independent on model, sensitive to damage but insensitive to environmental variability, noneed of prior damage information and so on. (4)A new algorithm for structural linear and nonlinear damage detection is proposed based on the time series analysis and the higher statistical moments. The traditional methods, in which the standard deviation of residual error of Autoregressive (AR) model was defined as damage sensitive index, may have the risk of damage information loss.The higher statistical moments of residual error, such as skewness and kurtosis, are then defined as the new damage sensitive index to be a complimentary. A series of experimental data from a three-story building structure with nonlinear damage indexesare used to prove the aforementioned fact. Further six fusion indexes from the standard deviation, skewness and kurtosis are discussed, and the indexes from standard deviation and kurtosis are the best.(5)A series of experiments on a complicated truss bridge combined with a steel bridge plate have been conducted. Damage was simulated by loosening the bolts of joints, and environmental variability were introduced by changing the shaker input level. The modal analysis results show that the dynamic characteristics exhibit 3.28% variation under environmental variability, which is higher than the induced damage less than 3%. Three methods proposed in this thesis are validated again.