Uncertainty Analysis Of Probabilistic Fatigue Life Prediction On Aeroengine Turbine Disk

Aircraft turbine disk, as a key component of mechanical major equipment, aresubjected to alternating load under a high temperature. It is easy to produce the fatigueand creep damage, and affect the safety of machine. The combination of turbine disk’sfailure mechanism and life prediction model have become one of the most effectiveways to ensure the machine works safely. However, with the development of the fatiguelife prediction methods, the deterministic model cannot accurately predict its service life.According to this issue, a probabilistic based model has been proposed to manage thisproblem. In this research, the Bayes theorem is used to quantify two kinds of uncertainty,namely parameter uncertainty and model uncertainty, by updating with life data.Andtwo kinds of methods are proposed to reduce predictive uncertainty. Generally speaking,this thesis attempts to improve the predictive capability of the model life prediction life,and reduce the influence of two kinds of uncertainty, so it has a strong practicalsignificance.The main content of the thesis and the research results are listed as follows:(1) Choose the PSED(Plastic Strain Energy Density) and GDP(GeneralizedDamage Parameter) model to predict fatigue life of the turbine disk.Engine turbine disk failure mainly in the form of repeated loads of fatigue damageand creep damage under high temperature deformation. This paper argues that theplastic strain energy and failure modes are closely connected. The PSED and GDPmodel assume the plastic strain energy density parameter as the damage parameter. Andthey are suitable for predicting the turbine disc life. According to the turbine disk failuremodes, GH4133are chosen as manufacturing material. Based on the GH4133fatiguetest, the two modes are applied to prediction turbine disk life. Results show that themodels are suitable to predict low cycle fatigue, but there are prediction errors.(2) Build a theoretical framework for quantitative epistemic uncertainty andstochastic uncertainty, and then probability density function describes fatigue lifeprediction uncertainty.According to the characteristics of epistemic uncertainty and stochastic uncertainty and the sources of uncertainty, four types of uncertainty, namely, the measurement error,prediction error, model uncertainty, parameter uncertainty are summarized. Using theGH4133experimental data, historical experience and model update strategy, Bayesianinference quantify these uncertainties. Based on those quantified uncertainties, the useof BF(Bayes Factor) determine the degree of fit of the data to the model, and then usingBMA(Bayesian Model Averaging) or UF(Uncertainty Factor), combined with theMC(Monte Carlo) simulation algorithm describe predicted life uncertainty caused byuncertainty propagation.(3) Propose DRAM-RJMCMC simulation algorithms based on the generalRJMCMC(Reversible Jump Markov Chain Monte Carlo) algorithm, and this advancedalgorithm is applied in probabilistic life prediction of turbine disk.Ordinary RJMCMC algorithm convergence is slow, and the candidate sampleacceptance probability is too high or too low, resulting in a high computational cost orlarge deviations between sampling results and the actual probability density function.DRAM, the advanced MCMC(Markov Chain Monte Carlo) algorithm, has been verifiedmore effective than the ordinary MCMC in the sample generation efficiency andapproximate sampling error. So the combination of DRAM and RJMCMC algorithm,named DRAM-RJMCMC algorithm, can improve effect of RJCMCMC sampling. Thisalgorithm is Applied to approximate the complicated high-dimensional integralsinvolved in the probabilistic life prediction to estimate the probability of model and theparameter probability density function. Results show the DRAM-RJMCMC algorithmconvergence quickly.