Structural health monitoring by using vibration characteristics of structure and self-organizing map

Abstract

This study applies a self-organizing map (SOM) to a structural health monitoring (SHM). The SOM is a kind of neural network and visualizes complex relationship of multi-dimensional data. The learning by the SOM enables us to convert the nonlinear statistical relationships among high-dimensional data into simple geometric relationships, usually a two dimensional grid of nodes. The SHM is a technique which diagnoses the structural safeness and finds damages from structural response. The present method uses the SOM to discriminate between vibration response of intact and damaged structures. In this method, we estimate the damage by visual change of the SOM. The present method requires only information about output signal of vibration without any prior information or numerical models of structures, allowing to accept the complicated and unknown structure. The effectiveness of present method is evaluated with the Carbon Fiber Reinforced Plastic (CFRP). The obtained results from numerical analysis and experiment revealed that our method is effective in SHM problems.