Fatigue Life Prediction of Steam Turbine Blades during Start-up Operation Using Probabilistic Concepts

Transcription

1 Fatigue Life Prediction of Steam Turbine Blades during Start-up Operation Using Probabilistic Concepts Presenter: C Booysen Academic Mentor: Prof P.S Heyns Date: 2014/07/25

2 Outline Problem statement Fatigue life modelling flow chart Experimental material characterisation Finite element modelling and analysis Probabilistic fatigue life modelling Conclusion Acknowledgements 2

3 Problem statement Fatigue in LP turbine blades has been recognised to be one of the primary causes of steam turbine blade failures worldwide During start-up operation, blades experience vibration and resonances at critical speeds which produce high dynamics stresses resulting in high cycle fatigue damage. Life assessment and prevention of fatigue failures requires the development of a fatigue life model Probabilistic model eliminates overly conservative assumptions and allows for uncertainty in key variables to be accounted Fatigue cracking Fatigue failure 3

4 Fatigue life modelling flow chart Finite element modelling and analysis Telemetry testing Fatigue life analysis Experimental material characterisation 4

5 Experimental materials characterisation Material properties required as inputs for the finite element and fatigue life prediction models 12% chromium martensitic stainless steel X22CrMoV12-1 Prior to fatigue testing, tensile testing to determine monotonic properties Load controlled uniaxial fatigue testing Fatigue cycles tested were fully reversed tensioncompression and tension-tension with a mean stress Tensile test specimen Fatigue specimen MTS fatigue test equipment 5

6 Stress Amplitude, a Stress Amplitude, a Experimental materials characterisation Statistical modelling using linear regression analysis Goodness of fit revealed high coefficient of determination, R 2 Sample run-outs not considered S-N Curve Fitted for Stress Ratio R=0.1 S-N Curve Fitted for Stress Ratio R= R=0.1 S-N Curve Fit R=-1 S-N Curve Fit R 2 = 80.32% 400 R 2 = 96.14% Cycles to Failure, N f [Cycles] Cycles to Failure, N f [Cycles] 6

7 Finite element modelling and analysis Geometrical scanning to determine the blade topology Free-standing last stage LP blade Cyclic symmetric model based on principal of sub-structuring Higher order 3D 20-node solid elements nodes and elements Surface-to-surface contact 7

8 Frequency, f [Hz] Finite element modelling and analysis Modal analysis to characterise blade natural frequencies and mode shapes Pre-stressed mode Campbell diagram identified potential resonance FEM validated against full scale spin test data on LP rotor and OEM results Calculated versus Measured Campbell Diagram Fn1(FEA) Fn2(FEA) Fn3(FEA) Fn1(OEM) Fn2(OEM) Fn1(Meas) Fn2(Meas) Fn3(Meas) Peak stress location Potential resonance Rotor Speed, N s [RPM] 1 st Mode stress distribution 8

9 Normal Stress, y Normal Stress, y Normal Stress, y Normal Stress, y Finite element modelling and analysis Transient stress response assessed at resonance for twenty-four variations in blade damping A pressure excitation was formulated through approximations from the static steam forces during a turbine start-up =0.1% =0.2% Max stress at first serration coincident with crack origin Time, t [s] Time, t [s] =0.3% =0.57% Time, t [s] Time, t [s] 9

10 No. of Cycles, n calc [Cycles] Log Cycles to failure, log N f [Cycles] Probabilistic fatigue life modelling Uncertainty and variability in the material fatigue strength parameters and the stochastic nature of the transient stress in the blade root due to variability in blade damping is investigated by random variable simulations using a statistical analysis algorithm written in MATLAB. Random curve fitting routines were used to ensure the selection of the random variables used in fatigue life calculations is stochastic in nature Random vectors are then chosen from a multivariate normal distribution with mean μ and covariance Σ to create multivariate normal random curves based on the fitted data Data Median fit multivariate normal random curve 95% confidence bounds Data Median fit multivariate normal random curve 95% confidence bounds Peak Stress, peak Log Stress Amplitude, log a 10

11 Probability Density Probabilistic fatigue life modelling Results represent the number of repetitions, which for this case is turbine start-ups, required to initiate a fatigue crack which was defined as the failure point. Fatigue repetitions shown to follow a straight line indicating normality in the data x Damage data Normal Distribution Fit Damage Data Normal Distribution Fit Fatigue Damage Repetitions, B x 10 5 f [Cycles] Fatigue Damage Repetitions, B f [Cycles] x

12 Conclusion The probabilistic fatigue life model was successfully implemented to determine the fatigue life of a blade during a turbine start-up. The method is shown to adequately quantify variability in material strength parameters and varying root stress High dynamic stresses associated to low levels of damping had a low probability of occurrence however in the event they occurred they were shown to have a significant impact on the overall fatigue life The probabilistic results findings reinforced those of the discrete life model but showed that the discrete life solution to be more conservative in predicting the life of the blade. 12

13 Acknowledgements Thanks to the following people for their time and expertise: Prof. Stephan Heyns Academic mentorship Mr Ronnie Scheepers Industrial mentorship Mr Michael Hindley Industrial mentorship Mr Jacques Calitz Industrial mentorship 13

14 Thank you for your time! Eskom Power Plant Engineering Institute 14 University of Pretoria