A reduced-order modelling for real-time identification of damages in multi-layered composite materials

Abstract

This work is focused on the detection of interlayer damages in multi-layer composite materials, which use the inverse analysis approach based on the reduced-order models (ROMs). Direct problems including heat conduction through the solid and heat radiation to the space are solved by finite element method (FEM), and the solutions of transient nonlinear analysis are collected to build the snapshot data matrix. The Proper Orthogonal Decomposition (POD) method coupled with two surrogate models are utilized to construct a low dimensional model which can be used to rebuild physical fields. By minimizing the discrepancy of measured and simulated temperatures using ROMs, the optimal parameter values related to damages can be estimated efficiently. Results prove that an excellent estimation on the position of damages can be obtained in the multi-layer materials. The computational efficiency and solution accuracy of two ROMs are compared, which proves that the POD-Kriging method has faster convergence speed than the POD-RBF method, with the same fitness value. Moreover, noise analysis validates that two ROMs have similar stability, whereas the POD-Kriging method is easier to achieve convergence in this type of problems.