.png){kind=icon}
.png){kind=icon}
Return
Data-driven digital twin framework for large-scale dynamic structures based on model reduction and damage regression identification
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engstruct.2025.119688.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
A framework and construction method for data-driven digital twin of large-scale dynamic structures based on model order reduction (MOR) and damage regression identification are proposed. The Krylov subspace order reduction method is used to reduce the orders of the high-fidelity finite element (FE) models corresponding to the possible damaged states of the structure during service, and a reduced-order model library is then set up. Using the models in the library, the dynamic responses of the damaged structure are quickly computed. With the dynamic response dataset, the damage regression identification model of the structure is established by the MLPResNet algorithm and used to update the digital twin following the evolution of the damaged state of the structure. Combining the proper orthogonal decomposition (POD) and deep learning algorithm, a surrogate model for the Krylov subspace projection matrices of the reduced-order models corresponding to the identified damaged states which are not included in the reduced-order model library is established. Using the surrogate model, the projection matrices and the dynamic responses of the damaged structure can be quickly calculated. The efficiency of the digital twin driven by the sensor data is demonstrated by a physical frame structure experimentally and numerically, and the suitability of the method for a large-scale structure is illustrated with the digital twin of a transmission tower. However, the damaged states of a structure during service and the type of sensors and their assignment scheme should be designed specifically in applications.
Keywords:
Digital twin
Large-scale dynamic structure
Krylov subspace order reduction
Damage regression identification model
Surrogate model
Journal
IF:
6.4
Papers:
2.1W
Citations:
8.7W