On-line updating Gaussian process measurement model for crack prognosis using the particle filter

Abstract

The particle filter (PF) has shown great potential for on-line fatigue crack growth prognosis by combining crack measurements from structural health monitoring (SHM) techniques. In this method, a key problem is to construct the mapping between the feature extracted from SHM signals and the crack size. However, this mapping may be inaccurate since the data used for establishing the mapping is affected by uncertainties from sources like damage geometries, sensor placements, and boundary conditions. To deal with this problem, this paper proposes an on-line updating Gaussian process (GP) measurement model within the PF based crack prognosis framework. The GP measurement model outputs the mean and variance of the crack length corresponding to the feature of SHM signals, which are input into the PF for evaluating the posterior estimation of the crack length for crack prognosis. Then, this posterior estimation is sequentially appended to the GP dataset for updating the measurement model. Moreover, once crack inspection data is obtained, it is combined with existing SHM data for additional updating of the GP measurement model. Validations are performed on the fatigue test of attachment lug structures, in which the guided wave based SHM technique is applied for crack monitoring. As the cyclic load may cause intricate influences on the guided wave propagation, it is more difficult to quantify the crack length. The validation result shows that the on-line updating GP measurement model can effectively map the feature of SHM signals to the crack length, and result in accurate crack growth prognosis with the PF based method.