| Abstrakt: |
One of the most important concepts in the context of structural health monitoring (SHM) is to find a method that would be able to detect the occurrence of damage, find its location, and measure its severity. It is possible to consider SHM as an inverse problem. By doing so, optimization algorithms can be implemented to perform the damage detection process. However, there are mainly two problems associated with employing such methods. The growth in the number of variables as the components of a structure increase, and the sensitivity of optimization algorithms, especially in Nelder-Mead (NM), to the selection of initial points. In order to address these problems and achieve the desired SHM goal as mentioned, a collaboration of artificial neural networks with Nelder-Mead and Particle Swarm Optimization (PSO) has been investigated in this paper. The ultimate aim of this research is to illustrate that employing ANN in collaboration with optimization algorithms whether classical such as NM or metaheuristic such as PSO helps boost speed and accuracy. In order to verify and evaluate their performance, two Finite element truss models, a 10-bar 2D truss and a 54-bar 3D truss, have been implemented and their modal features are introduced as the damage-sensitive features. It is observed that implementing ANN in collaboration with optimization algorithms boosts their performance, increases their accuracy, and reduces their convergence time. Finally, based on several criteria, the performance of the proposed hybrid methods (ANN-NM and ANN-PSO) is discussed and compared with singular forms of ANN, NM, and PSO. It is observed that ANN-NM outperformed other methods. |
Full text from SpringerLink