Bayesian model calibration techniques are commonly employed in the characterization of nonlinear dynamic systems, as they provide a conceptual and effective framework to deal with model uncertainties, experimental errors and procedure assumptions. This understanding has resulted in the need to introduce a model discrepancy term to account for the differences between model-based predictions and real observations. Indeed, the goal of this work is to investigate model-driven seismic structural health monitoring procedures based on a Bayesian uncertainty quantification framework, and thus make relevant considerations for its use in the seismic structural health monitoring, focusing on masonry structures. Specifically, the Bayesian inference has been applied to the calibration of nonlinear hysteretic systems to both provide: (i) most probable values (MPV) of the parameters following the calibration; and (ii) estimates of the model discrepancy posterior distribution. The effect of the model discrepancy in the calibration is first illustrated recurring to a single degree of freedom using a Bouc–Wen type oscillator as a numerical benchmark. The model discrepancy is then introduced for calibrating a reference nonlinear Bouc–Wen model derived from real data acquired on a monitored masonry building. The main novelty of this study is the application of the framework of uncertainty quantification on models representing data measured directly on masonry structures during seismic events.

Despite the computing power achieved with current technologies, computationally intensive analyses (e.g. optimization problems, reliability or sensitivity analyses) actually remain a time-consuming problem in modern engineering. Traditionally, a widely used approach to deal with this issue based on learning from data, relies on metamodels or surrogate modeling. As a result, the emulator has the attraction of being fast to evaluate. In this context, metamodels based on Kriging (a.k.a. Gaussian Process (GP)) have gained momentum in computational sciences by playing a crucial role in the expansion of machine learning tools. The specific objective of this study is to investigate the use of different Python libraries for Kriging metamodeling purposes, setting out a consistently review of the major frameworks used in the engineering field. In particular, a focus on two primary aims is addressed: (a) to compare the various settings available for each library; (b) to ascertain how they perform and differ under similar assumptions. In this investigation, the aim is to be of value to practitioners wanting to know on the capability and reliability of the multitude of open-source packages available nowadays.

Bayesian model calibration techniques are commonly employed in the characterization of nonlinear dynamic systems, as they provide a conceptual and effective framework to deal with model uncertainties, experimental errors and procedure assumptions. This understanding has resulted in the need to introduce a model discrepancy term to account for the differences between model-based predictions and real observations. Indeed, the goal of this work is to enhance model-driven Structural Health Monitoring procedures by incorporating the posterior uncertainty linked to updated model discrepancy, and thus make relevant considerations for its use in the Structural Health Monitoring. Specifically, the Bayesian inference has been applied to the calibration of nonlinear hysteretic systems to both provide: (i) most probable values (MPV) of the parameters following the calibration, and; (ii) estimates of the model discrepancy posterior distribution. The effect of the model discrepancy in the calibration is first illustrated recurring to a single degree of freedom Bouc-Wen type oscillator, and then applied for calibrating a reference nonlinear Bouc-Wen model, deriving from real data acquired on a monitored masonry building.

Despite the computing power achieved with current technologies, computationally intensive analyses (e.g. optimization problems, reliability or sensitivity analyses) actually remain a time-consuming problem in modern engineering. Traditionally, a widely used approach to deal with this issue based on learning from data, relies on metamodels or surrogate modeling. As a result, the emulator has the attraction of being fast to evaluate. In this context, metamodels based on Kriging (a.k.a. Gaussian Process (GP)) have gained momentum in computational sciences by playing a crucial role in the expansion of machine learning tools. The specific objective of this study is to investigate the use of different Python libraries for Kriging metamodeling purposes, setting out a consistently review of the major frameworks used in the engineering field. In particular, a focus on two primary aims is addressed: (a) to compare the various settings available for each library; (b) to ascertain how they perform and differ under similar assumptions. In this investigation, the aim is to be of value to practitioners wanting to know on the capability and reliability of the multitude of open-source packages available nowadays.