Computational intelligence for structural health monitoring

Reliable quantification of the health state of critical infrastructure (such as power plants, high-rise buildings, long-span bridges, dams, airports, tunnels and railway tracks) is a fundamental concern in civil engineering as it can directly affect the national assets and public safety of countries. Concerning this matter, structural health monitoring (SHM) can potentially provide effective solutions to continuously assessing the health state of infrastructure, as it can reduce asset management costs, prolong the structures' operational lifetime and ensure public safety. Therefore, getting access to a robust paradigm to deal with SHM concerns has always been a high priority. The automated condition assessment systems are one of the technologies that have received significant attention in the area of advanced monitoring systems. These systems can interpret large volumes of inspection data to detect and prevent structural failure in early stages by minimising errors to ensure effective risk management while reducing the asset management costs. Although there are various research activities reported in this field, only a few robust methods can determine the adverse condition of a structure effectively, which is the motivation of this research work. Therefore, the main objective of this study is to propose a more robust scheme for automated damage detection systems that can monitor and evaluate the health state of a structure. A non-destructive testing (NDT) method, that has applications in SHM for characterising and assessing the materials and structures, is used to characterise the concrete members (beams). However, after a comprehensive literature review, the mounted Smart Aggregates (SA) based approach is used to monitor cracks in simple concrete and reinforced concrete beams under loading. The collected experimental data is analysed through different proposed algorithms based on signal processing, image processing and artificial intelligence (AI) techniques. These analyses are categorised into four steps of pre-processing and feature extraction, feature selection, classification and visualisation. In the feature extraction step of this detection system, seven sets of features are proposed through statistical parameters, time-frequency analysis, and utilising dynamic and spectral features. However, the time-frequency algorithm which has been proposed based on Hilbert-Huang transform and Wavelet transform is used to perform both feature extraction and crack localisation in structures. Additionally, this thesis proposes two damage indexes (DI), Entropy-based Dispersion (ED) and Entropy-based Beta (EB) for feature extraction, localisation and crack severity purposes. In addition to comparing the damage indexes with the benchmark, experimental investigations are carried out against a conventional load cell system to determine the effectiveness of the proposed DIs. The proposed damage indexes could determine, localise and estimate the cracks' severity status even earlier than the load cell system. However, these damage indexes have also been considered as feature sets which will be used with machine learning approaches. Among the seven investigated feature sets, the hybrid ED-EB could provide the best accuracy and F1-score in determining the damage occurrence through AI techniques. The next step of this system is the feature selection in which the algorithm is proposed based on the Neighbourhood Component Analysis (NCA), Particle Swarm Optimization (PSO) and Support Vector Machine (SVM). The algorithm showed better performance when compared with the traditional approach. The classification step of this research consisted of developing four SVM based algorithms to classify and detect a failure in the structural components. The algorithms are based on the misclassified data, hybrid kernels and hybrid classifiers, respectively. The algorithms outperformed the traditional classifier, and the SVM based on misclassified data could obtain the highest accuracy. The last step of this system is the microwave imaging approach, which is used to perform local visualisation of the damage. This step uses the image processing approaches to enhance the quality of obtained images. To evaluate the performance of these approaches, the original obtained images are compared with the enhanced ones. The experimental results showed that the algorithm could accomplish and enhance the crack visualisation. This thesis provided different algorithms for developing an NDT based damage detection system to detect and localise flexural cracks based on soft computing approaches