Ageing and degradation of the infrastructures, especially critical infrastructures (such as power plants, high-rise buildings, long-span bridges, dams, airports, railway tracks, etc.) impose a major concern that affects national assets of each country and endangers public safety. On this point, structural health monitoring (SHM) potentially provides solution to the problem by evaluating the integrity of the infrastructures to determine their current health state. Basically, structural health monitoring deals with early warning on the state of health of infrastructures, localisation and quantification of damage in the structures and prediction of their remaining service life. This, consequently reduces asset management costs, effectively prolongs operational lifetime and ensures public safety. Hence, getting access to a robust paradigm to deal with aforementioned concerns is a major challenge introduced here. Despite high level of research activities in this field, few robust methods of indicating an adverse condition of a structure in service has been demonstrated as effective, which is the motivation for this research work.
The main objective of this study is to investigate a more robust scheme of damage identification, including damage detection and damage localisation, to overcome some of the shortcomings with the current methods. In this regard, firstly, a background on the existing methods is presented in Chapter 1 to evaluate the advantages and limitations with the current methods. According to the literature, frequency response-based damage identification methods are superior to conventional modal-based approaches as they are less error prone and provide abundance of information in a wide range of frequencies; hence, this study starts with developing a more robust frequency response function (FRF) – based damage identification method in Chapter 2 to detect and localise structural damage in single or multiple states. The novelty of the approach is implementation of two-dimensional discrete wavelet transform (2-D DWT) along with the second derivative of the reconstructed operational mode shapes obtained by FRFs to enhance the sensitivity of the approach to damage. Based on the numerical results of this stage, it can be concluded that the method’s performance is quite acceptable once the level of undesirable noise is negligible; however, by increasing the level of uncertainty in the system, the performance of the method deteriorates. Therefore, to overcome this problem, the harmful effects of measurement noise on FRFs are investigated in Chapter 3 and its undesirable effect is suppressed by employing a novel idea using Gaussian Kernel on FRFs. The numerical and experimental damage identification results obtained by the presented noise-suppression approach demonstrate its efficiency to cope with the issue of noise. At this stage, and although, the method is performing well, in terms of its sensitivity to damage as well as its robustness against noise, it still needs to be modified; in real-life applications the source of excitation is random and this important issue has not been taken into account in the methodology described in Chapter 3. Hence, in Chapter 4, the issue of stochastic systems is investigated and a novel spectral-based approach is presented to deal with the issue of random excitation in damage identification process. In this regard, the frequency distribution of the power spectral density of the time responses is analysed by introducing the spectral moments which represent some major statistical properties of a stochastic process. The efficiency of the approach is validated by several numerical case studies.
The method presented in Chapter 2 to Chapter 4 is a frequency-based approach and, therefore, raw time measured responses are first required for transfer from time domain into the frequency domain before further analysis. In some applications, it might be advantageous to deal with directly measured time responses without transferring them into the frequency domain. In this regard, in Chapter 5, a novel time series-based damage identification method is presented based on the idea of symbolic time series analysis. The main idea of the method is to generate the symbol sequences by mapping the time data from the state space into the constructed symbol space and then study any change in the statistical properties of the obtained symbol sequences by developing the probability vectors. This method is very easy to implement, is robust against noise and it has shown a considerable sensitivity to slight structural damage. The efficiency of the method is successfully demonstrated by several numerical and experimental investigations.
The scope of Chapter 1 to Chapter 5 of this thesis is mainly on structural damage identification; however, it might be of great interest to obtain a reliable and representative model of the structure considering the effects of structural damage. This is of considerable demand, as in many applications, it is required to estimate the serviceability and remaining life of a structure after damage occurrence. Because of this, the author explores the issue of finite element model updating in Chapter 6 to investigate how a reliable model of the structure can be obtained. A particular damage case is investigated and the idea of thin layer interface elements is introduced and implemented. It is demonstrated, that by updating the finite element model of the structure, using this technique, the reliability of the model significantly improves.
According to this journey from Chapter 1 to Chapter 6, more robust schemes of damage identification are developed and verified both in time and frequency domains; and, some future works are suggested by the author in Chapter 7 to conclude the work and motivate other researchers to pursue the work further.