Qualitatively-improved identified parameters of prestressed concrete catenary poles using sensitivity-based Bayesian approach

• Autor(a) principal: Alkam, Feras
• Data de Publicação: 2020
• Outros Autores: Pereira, Isabel, Lahmer, Tom
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/28525
• topic: Prestressed concrete catenary poles; Bayesian inference; Inverse problems; Sensitivity analysis; TMCMC; Vibration test; 3-Point bending test

Prestressed, spun-cast ultrahigh-strength concrete catenary poles have been used widely for electric train systems; for example, thousands of these poles have been installed along high-speed train tracks in Germany. Given the importance of the functionality of train systems, adequate attention has not been paid to catenary poles in research and the literature. Questions regarding the integrity of catenary poles still exist. This study contributes to identify the actual material properties of the poles of interest because the parameter identification is an essential process for any subsequent evaluation of the integrity of catenary poles. Accordingly, a sensitivity-based Bayesian parameter identification approach is developed to estimate the real material properties of the poles using measurements from multiple experiments and numerical models. This approach integrates the sensitivity of time-dependent measurements into the Bayesian inference, which improves the quality of inferred parameters considerably in comparison with classic Bayesian approaches applied in similar case of studies. Furthermore, the proposed approach combines observations of multiple experiments conducted on full-scale poles using a probabilistic uncertainty framework, which provides informative data used in the parameter identification process. Besides, Bayesian inference quantifies the uncertainty of inferred parameters and estimates the hyperparameters, such as the total errors of the observations. The proposed approach utilizes the efficiency of the transitional Markov Chain Monte Carlo algorithm for sampling from the posterior in both levels of Bayesian inference, namely, the unknown parameters and the hyperparameters. The results show the significant influence of the sensitivity concept in improving the quality of the posterior and highlight the importance of identifying the real material properties during the evaluation of the behavior of existing structures, rather than using the characteristic properties from the datasheet. Applying the proposed approach looks very promising when applied to similar applied case studies.