Fatigue life prediction on nickel base superalloys
University of Southamtpon, School of Engineering Sciences,
Neural networks have been used extensively in material science with varying success.
It has been demonstrated that they can be very effective at predicting mechanical
properties such as yield strength and ultimate tensile strength. These networks require
large amounts of input data in order to learn the correct data trends. A neural network
modelling process has been developed which includes data collection methodology and
subsequent filtering techniques in conjunction with training of a neural network model.
It has been shown that by using certain techniques to ‘improve’ the input data a network
will not only fit seen and unseen Ultimate Tensile Strength (UTS) and Yield Strength
(YS) data but correctly predict trends consistent with metallurgical understanding.
Using the methods developed with the UTS and YS models, a Low Cycle Fatigue
(LCF) life model has been developed with promising initial results.
Crack initiation at high temperatures has been studied in CMSX4 in both air and
vacuum environments, to elucidate the effect of oxidation on the notch fatigue initiation
process. In air, crack initiation occurred at sub-surface interdendritic pores in all cases.
The sub-surface crack grows initially under vacuum conditions, before breaking out to
the top surface. Lifetime is then dependent on initiating pore size and distance from
the notch root surface. In vacuum conditions, crack initiation has been observed more
consistently from surface or close-to-surface pores - indicating that surface oxidation is
in-filling/”healing” surface pores or providing significant local stress transfer to shift
initiation to sub-surface pores. Complementary work has been carried out using PWA
1484 and Rene N5. Extensive data has been collected on initiating pores for all 3
alloys. A model has been developed to predict fatigue life based upon geometrical
information from the initiating pores. A Paris law approach is used in conjunction with
long crack propagation data. The model shows a good fit with experimental data and
further improvements have been recommended in order to increase the capability of the
||Engineering Mats & Surface Engineerg Gp
||16 Sep 2009
||18 Apr 2017 21:21
|Further Information:||Google Scholar|
Actions (login required)