Comparison of different life distribution schemes for prediction of crack propagation in an aircraft wing

Estimation of remaining useful life/prognostics of an aircraft structure permits aerospace industry to timely schedule maintenance activities. Accurate planning through prognostics ensures safer flight operation and lower downtimes. The core of any prognostic algorithm is the state transition/degradation model. In the reported research work, Particle Filter (PF) based prognostic algorithm is used to predict crack growth with three different state transition/life distribution models. PF (Bayesian Sequential Monte Carlo) allows using of non-linear state-transition and non-Gaussian/multimodal noise distributions. The typical candidate life distributions for modeling crack growth are Exponential, Weibull and Lognormal distributions. A framework is proposed where effectiveness of the candidate distribution for modeling degradation phenomenon can be adjudged. The algorithm is tested on actual historical NDT data of crack growth on countersunk (CSK) rivet holes on an Airbus A310 aircraft's wing. Historical data is bifurcated into two periods i.e. training and validation periods. The most appropriate distribution based on the comparison of the above mentioned candidate distributions is proposed for prediction of the degradation/ flaw propagation.