While monitoring data provides valuable information on the structural performance, simulation models can augment and enhance the measurement data for more comprehensive insights. Continuously updated simulation models in the context of digital twins enable to install virtual sensors in the simulation model, allowing the prediction of quantities that are not easily measurable (e.g., stresses), at positions that are not accessible with the current sensor setup or predict the evolution of condition indicators and prognostics (CIPs) as a basis for decision making and maintenance planning. Establishing a digital twin within the SPP necessitates the integration of multiple simulation models of different subprojects, each focussing on different aspects. The accuracy of these models is critical to the overall reliability of the digital twin. While the models used for the design can often serve as a starting point, this information is often not available for existing structures. Even if complete information is available, in most cases an iterative process to improve the model based on measurement data is required. This improvement involves not only the updating of model parameters, but also revisiting the modeling assumptions, e.g., the level of geometry resolution, the constitutive models, or boundary and initial conditions. As there will always be a discrepancy between experimental data and corresponding model predictions of the digital twin due to both aleatoric and epistemic uncertainties, a reliable simulation model must incorporate these uncertainties into the CIP prognosis. It should also provide metrics to quantify the adequacy of model adaptation, estimate the robustness of predictions and trigger an alarm when anomalies are detected. Furthermore, a key challenge for practical applications of the developed mechanical models is a thermal compensation to transition from constant temperature conditions in the lab to fluctuating environmental conditions.

The project aims to develop a methodology for the iterative improvement of simulation models developed within the SPP based on continuous monitoring data, with a focus on quantifying the uncertainty of the predictions and condition indicators due to noise and model form uncertainty. For that purpose, we aim first at constructing parametrized thermal and thermo-mechanical bridge models in the form of finite element simulation models (WP 1). As their evaluation is computationally expensive, we will also investigate efficient surrogate models. Second, we develop methods for model updating and uncertainty quantification under model form uncertainty based on embedded bias formulations (WP 2), extending it to higher dimensions and with correlations in the embedding. To improve the models, we aim at an explicit identification of parameter dependencies on space, time, and environmental conditions. Third, we define CIPs based on a decomposition of the deformation state, develop metrics of model quality and anomaly detection and estimate the impact of these uncertainties in the embedded parameters on the computed CIPs (WP 3). Finally, these methods are integrated into an open-source software package and applied to improve models from other projects in the SPP including the quantification of the uncertainty. Additionally, these methods are integrated into the SPP demonstrators (Nibelungenbrücke and openLAB). A query-based server will provide simulated full-field temperature data and deformations compensated from thermally-induced strains to partner projects within the SPP (WP 4).