An efficient identification method of the structural parameters of MDOF structures using the wavelet transform and neural networks

Structures undertaking dynamic load store cumulative damages on their structural members. Although these damages are generally estimated by measuring the acceleration, velocity and displacement at several observing points, the health monitoring system based on these measurements could be expensive. Therefore, approaches to the health monitoring system utilizing only the acceleration and decreasing the observing points are growing in importance. Hybridizing wavelet analysis and neural networks gives a powerful method for these approaches. In such a method, wavelet analysis performs efficient time-frequency analysis, which accesses to the information on velocity and displacement from the observed acceleration. On the other hand, neural networks realize a function to estimate the unknown information on points, where the acceleration is not observed, from the data processed by wavelet analysis. As the first step for this approach, LODE (Linear Ordinary Differential Equation) models with piecewise constant coefficients of MDOF (Multi-Degree Of Freedom) structure will be studied here. These LODE models have stiffness and damping matrices of piecewise constant entries which will be identified from the acceleration. This paper presents an efficient identification method of the structural parameters of MDOF structure using the wavelet transform and neural networks. Some simulations will certify the usefulness of this hybrid method.