Co-Kriging modeling for structural health monitoring applications

• Ano de defesa: 2024
• Autor(a) principal: Novais, Henrique Cordeiro [UNESP]
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Estadual Paulista (Unesp)
• Palavras-chave em Português: Co-Kriging; Gaussian process regression; Structural health monitoring; Composite laminates; Damage quantification; Regressão por processo gaussiano; Monitoramento de integridade estrutural; Compósitos laminados; Quantificação de danos
• Link de acesso: https://hdl.handle.net/11449/259219

Computational modeling plays a crucial role in applied science and engineering, offering valuable insights into the behavior of structures and mechanical systems. However, the complexity and significant execution time required for these models can limit their applicability in real-world scenarios. To mitigate these challenges, metamodeling, or surrogate modeling, provides a practical solution by substituting the complex model with a simplified function that mimics its behavior, thereby substantially reducing evaluation time. One widely applied method is Gaussian Process Regression (GPR), also known as Kriging, which has proven effective in various structural health monitoring applications. However, achieving accurate predictions for damage-sensitive variables often requires extensive historical data or well-calibrated structural models, which can be costly. Thus, this dissertation proposes applying the co-Kriging method, a multivariate extension of ordinary Kriging that leverages the covariance between two or more related datasets, making it particularly advantageous when the co-variable is more economical to measure than the target variable. This work presents two distinct applications: the first focuses on a concrete bridge, using natural frequencies under different temperature conditions to develop the co-Kriging model; the second application pertains to composite structures, where simulated Lamb wave data is combined with experimental tests to detect and quantify delamination in laminates. In both applications, the co-Kriging method showcased its reliability and superiority over Kriging, offering an efficient means of enhancing prediction accuracy while minimizing data acquisition costs.