This paper demonstrates a diagnostic-prognostics framework to estimate probabilistic remaining useful life (RUL), in adhesively bonded joints subjected to fatigue loading, by calibrating the predictive model using the diagnostics data and quantifying uncertainty in the model parameters. The matching pursuit algorithm is used as the diagnostic method to measure the crack length and the rapid fatigue damage growth (FDG) simulator is used as the predictive model to estimate the remaining useful life (RUL). In the diagnostic method, Lamb waves are excited in the structure using piezo transducers, and the matching pursuit algorithm is used to quantify the damage from the reflected signal. The proposed diagnostic technique is verified using the signal obtained from finite element simulations and artificial noise is added to mimic the signal from a real structure. The diagnostic method is applied periodically to measure the crack length in the single lap joint (SLJ) subjected to fatigue loading and the crack length data is used to calibrate the parameters of the predictive model, which can estimate the RUL. However, the noise in the signal and assumptions in the diagnostic technique result in errors in the measured crack length. These errors in the crack length contribute to the parameter uncertainties during the predictive model calibration. To quantify the model parameter uncertainties, the Bayesian inference via the Markov chain Monte Carlo method is used, and to expedite the uncertainty quantification problem, the rapid FDG simulator is used as the predictive model. The approach is demonstrated using a fatigue damage growth simulation in the SLJ and promising results were achieved.
|