Intelligent Load Monitoring in Beam Structures

Abstract

A robust approach for identifying impact load location and impact force history in beam structures is presented in this thesis. Beam strain transients propagating from the impact site can be inverted to yield the impact location and force history. Solving the inverse problem consists of three parts: a transient wave model, an impact location determination, and then an impact history identification. The classical Euler-Bernoulli beam theory (EBT) is used to obtain the dynamic models of a simply-supported beam.

The simulated measured strain data are utilized from the transient wave model derived from finite element method and state variable approach. A wavelet analysis using Gabor basis function is employed to determine the impact load location from strain measurements resulting from a pair of sensors which are located on the opposite side of the impact site. The information on traveling dispersive waves is described by the time-frequency representation of the transient strain data analyzed using wavelet transform. A Radial Basis Function network (RBFN) is then employed to efficiently reconstruct the impact load history from a single strain sensor. The back-propagation algorithm in conjunction with Levenberg-Marquardt method is adopted to update the parameters of RBFN by minimizing the difference between model prediction and the sensor measurements, where the transient wave model is embedded in the RBFN to guide and speed up the training process. The effect of noise on the identification of impact site and loading history is also investigated in detail. Several examples demonstrate the effectiveness of the approach.