Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Application of reduced-order prediction models and digital twins for the prediction of structural mechanical response

Liaskos Nikolaos

Full record

---

Summary

This dissertation focuses on the development and implementation of computational methodologies for accurate and efficient prediction of the mechanical response of complex structures, with emphasis on historical monuments and masonry constructions. The proposed approach combines Reduced-Order Models (ROMs) with the philosophy of Digital Twins (DTs), aiming to bridge the gap between high-fidelity simulations and real-time monitoring and decision-making needs. The framework begins with a comprehensive methodology for data preprocessing and analysis, using information from diverse sources such as sensors (e.g., accelerometers), historical records, and geometric surveys. Emphasis is placed on feature extraction from dynamic time series (e.g., seismic signals, environmental vibrations), using both statistical (mean, standard deviation, skewness, kurtosis) and spectral techniques (Welch analysis, spectral entropy, spectral centroid, bandwidth), enabling efficient quantification of the dynamic behavior under different loading conditions. Next, Reduced-Order Models are applied to significantly lower the computational cost of conventional full-order models (Finite Element Models – FEM), while maintaining adequate accuracy in predicting critical quantities such as displacements, stresses, and strains. Techniques such as projection-based model reduction, using modes or snapshots, are employed to construct compact reduced bases. The integration of Digital Twin architectures forms the cornerstone of the proposed approach, enabling dynamic, two-way communication between the physical structure and its ROM-based digital counterpart. Sensor data (from accelerometers, strain gauges, or other Structural Health Monitoring – SHM – systems) are used to continuously or periodically update the model, improving its predictive capabilities in near real-time. An application example using ARMAX time-series models is presented, where the extracted features serve as exogenous inputs. Results show high accuracy and generalization ability, even in the presence of limited training data. Proper data organization, targeted feature selection, and appropriate optimization and validation techniques (e.g., comparison with experimental data, MAC criterion) significantly enhance ROM reliability, even in nonlinear regimes. Finally, real-world scenarios such as the seismic analysis of historical masonry structures (e.g., the Neoria of Chania) demonstrate that the proposed framework is a powerful, flexible, and computationally efficient tool for monitoring, predictive maintenance, safety assessment, and life-cycle management of critical infrastructure. Keywords: Reduced-Order Models (ROM), Digital Twins, predictive modeling, structural mechanical response, nonlinear FEM analysis, seismic response, historical structures, masonry, SHM, sensor data processing, feature extraction, ARMAX, optimization, model validation, Neoria of Chania.

Available Files

Services