Detecting defects in composite beams and plates using Bayesian inference
The topic of this paper is an inverse problem of identifying defects in composite beams and plates. The physical representation of defects is parametrized. Assuming Gaussian errors in measurements, the Bayesian inference is performed for those unknown parameters, and the most probable physical representations of detects are estimated. A composite beam/plate is usually made up of several layers, and there may be some defect in the bonding process or a defect may develop later. We use the natural frequencies of the beam/plate to estimate the position and the size of the defects. We propose that the bonding within the beam and plate can be modelled as added rigidity, which can be incorporated as an extra energy to the conventional strain energy. Standard Monte-Carlo simulation will then give the probabilistic properties of the natural frequencies of the beam/plate. The more prior information about the defects is limited, and thus we estimate the posterior distribution using the trans-dimensional Bayesian method, which lets us make an inference of different types of defects.