Structural Health Monitoring for Beam Models using Optimization Methods and Machine Learning

The evolution of Structural Health Monitoring (SHM) has emerged as a vital discipline for safeguarding the integrity of structures, transcending industries from civil engineering to aerospace. Over time, SHM has witnessed a progression from traditional methods to innovative techniques driven by technology and computation. The employment of Artificial Neural Networks (ANNs) and Particle Swarm Optimization (PSO) has significantly reshaped SHM strategies, driven by their capabilities to efficiently process complex datasets and optimize parameters. ANN's journey from its inception in the 1940s, inspired by the brain's neuronal networks, to its application in SHM, has revolutionized damage detection through pattern recognition and prediction. PSO, rooted in social behavior and inspired by flocking birds, has evolved since the 1990s into a robust optimization tool, adept at tackling multi-dimensional search spaces. Despite their merits, both ANN and PSO harbor limitations, including susceptibility to local optima, constrained convergence accuracy, and suboptimal results in certain contexts. This thesis stands as a response to these challenges, aimed at not only leveraging ANN and PSO's strengths but also addressing their limitations. The tripartite structure of this work allows a comprehensive exploration of three distinct methodologies: PSO-YUKI, BOA-ANN, and RSA-ANN. Each approach presents a unique solution to the prevalent issues in SHM. These methodologies are applied across diverse structural scenarios, reflecting their versatility and potential impact. PSO-YUKI, a novel hybrid algorithm, addresses double crack identification in Carbon Fiebers Reinforced Pomymer (CFRP) cantilever beams. Combining Radial Basis Function (RBF) and PSO, YUKI-PSO introduces efficiency in solving fast inverse problems and demonstrates its accuracy through experimental and numerical analysis. In a parallel endeavor, BOA-ANN leverages ANN's potential for crack depth identification in steel beam structures. The integration of Finite Element Method (FEM) and Butterfly Optimization Algorithm (BOA) enhances the accuracy of crack depth prediction. The thesis also introduces RSA-ANN, where the Reptile Search Algorithm (RSA) optimizes ANN training for damage location prediction in steel squared beams. This novel approach is underscored by frequency response function (FRF) estimation, addressing information leakage concerns. The results obtained from these approaches emphasize their accuracy, efficiency, and their potential to revolutionize structural health monitoring practices. The investigation of these approaches spans multiple structural scenarios, culminating in a comprehensive understanding of their efficacy in SHM. By harnessing the strengths of ANN and PSO while mitigating their limitations, this work strives to redefine the landscape of SHM through innovation, precision, and practicality.