Final Report Summary - LIFEMOD (Enhanced material lifing model for rotating welded structures)
The LIFEMOD project has aimed at the study of LASER and EB welds in IN718 sheet material.
The project has focused on two research questions:
- Impact of defects on fatigue life originating from the welding process
- Impact on the crack growth rate due to sustained load (hold time effects) at elevated temperature (T=550 °C)
The design problem can be divided into two main categories:
- Lack of information
- Accuracy of the fatigue model
For an accurate analysis of the fatigue life of a component to be possible very accurate information about loads and geometry is needed. In the case of welding additional information about various defects in the weld is needed. The problem of welding is that many different types of defect are possible (pores, cracks, lack of fusion, inclusions, high stress concentrations, alignment, chain porosities etc). The effect of the defects on fatigue life is very difficult to quantify and therefore to predict by some model
The main concept of LIFEMOD has been to produce welds in IN718 and to use these for study. After manufacture these welds has been tested by both standard and cutting edge NDT methods to quantify the defect density and morphology. Test specimens for fatigue testing have been produced from the welds and have been tested by both traditional HCF (High Cycle Fatigue) methods and creep-fatigue methods with extended hold times at maximum load. The fracture surfaces have after testing all been studied by fractography and the defect status verified in relation to the fracture. A mass of test data has thus be generated. By statistical analysis of the test data the basis, a lifing model has been proposed. The proposed model is based on accounting for competing failure mechanisms in the fatigue analysis.
Based on the results from the study valuable results have been obtained that can be used to design new and more efficient gas turbines. A database of the results has been collated and througly asnalysed. Based on the finding from both testing and post-testing analysis a very flexible probabilistic lifing model for welded structures was developed which can take into account sources of scatter for all parameters in the deterministic model. The probabilistic part is formed by a generic probabilistic tool RAP++. The deterministic model consists of a crack growth tool (NASGRO) and a fatigue analysis tool (Cragro++) and can be easily extended with other tools.
Project Context and Objectives:
Concept and project objectives
The design problem can be divided into two main categories:
- Lack of information
- Accuracy of the fatigue model
For an accurate analysis of the fatigue life of a component to be possible very accurate information about loads and geometry is needed. In the case of welding additional information about various defects in the weld is needed. The problem of welding is that many different types of defect are possible (pores, cracks, lack of fusion, inclusions, high stress concentrations, alignment, chain porosities etc). The effect of the defects on fatigue life is very difficult to quantify and therefore to predict by some model. Typically are two models available:
- Regression type models (e.g. S-N curves)
- Fracture mechanics based models (defect size and location)
To illuminate the modelling problems with welds the following sentences can be considered. The regression type models are difficult to extrapolate from one defect type to another without detailed information from destructive testing. Therefore should all kinds of models be tested (material types, welding types, loading types and various geometries). Further enough test results must be available to get sufficient number of data points for statistical analyses to be meaningful (minimum curves etc). The fracture mechanics models all assume that a defect is equal to a crack (similar to laboratory specimens) and any “grace period” i.e. initiation period is not accounted for. The initial defect size must also be determined. One way to determine the initial defect size is to base it on non-destructuve testing (NDT) capabilitiy The NDT defect size is typically around 2c=1.5 mm. The NDT does however not say what defects are in the material only what potentially could be missed in an inspection. If a welding procedure is defect free, the defect size cannot be reduced via a NDT-logic. To include any improved welding quality in the lifing method, a clear understanding of the defects occurring in the current welding process is necessary i.e. effect of defects on fatigue life. The investigation should include the appropriate physics (relevant temperatures, geometries and loading) to assess the weld life and the possible detrimental effects of weld defects
The main concept of LIFEMOD has been to produce welds in IN718 and to use these for study. After manufacture these welds has been tested by both standard and cutting edge NDT methods to quantify the defect density and morphology. Test specimens for fatigue testing have been produced from the welds and have been tested by both traditional HCF (High Cycle Fatigue) methods and creep-fatigue methods with extended hold times at maximum load. The fracture surfaces have after testing all been studied by fractography and the defect status verified in relation to the fracture. A mass of test data has thus be generated. By statistical analysis of the test data the basis, a lifing model has been proposed. The proposed model is based on accounting for competing failure mechanisms in the fatigue analysis.
State-of-the-art lifing models for HCF tests
Various fatigue crack initiation, fatigue crack growth threshold, fatigue crack growth and sustained load crack growth models have been developed in the past. All these models are more or less of an empirical nature. For crack growth models some mechanistic approaches are applied to model the physical phenomena, such as the Strip-Yield model to better capture the deformation state in front of the crack tip and in the wake of the crack and the more traditional models such as Wheeler and Willenborg that approximate the plastic zone in front of the crack tip. None of these models clearly outperforms the others and as such has been broadly accepted by the industry. Recent attempts to improve fatigue crack growth predictions include the more physically based approached by Pommier et.al. and the block loading approach developed at DSTO in Australia. The latter has a strong focus on highly variable amplitude loading of airframes and the former is still under heavy development and is not widely accepted (yet) in the industry. The objective is therefore to evaluate existing more or less accepted models for their applicability in the life prediction for welded IN718 components. Three other phenomena are important for the life prediction. Firstly, time dependent phenomena related to embrittlement and oxidation around the crack tip, as discussed previously. Secondly, variable amplitude loads, which are commonly included by the mentioned retardation models (Willenborg, Wheeler and Strip-Yield). Finally, depending on the location of the defects, the residual stresses left after welding may accelerate or decelerate the crack initiation and growth. This is often included as a stress ratio correction such as those in the FITNET procedure.
In case of crack growth very limited information is available concerning the stress intensity solution (SIF) for welds, but can nowadays efficiently be generated by advanced finite element analysis tools. The XFEM method herein is a recent development to determine SIF solutions for more complex (realistic) geometries and crack shapes.
A lot of scatter in initiation, threshold, creep and crack growth is often observed. Therefore an improved lifing model should incorporate the relevant scatter sources. This can be done by adding a stochastic layer on top of a deterministic life model. Various sound stochastic models have been developed in the past or more recently for instance Monte-Carlo, Directional Simulation, FORM, SORM, Importance Sampling. Stochastic tools exist by which these models can be easily applied to deterministic models in general.
Inconel 718 in fine grained forms, such as sheets, plates and forgings, are known to be very sensitive to sustained tensile loads at elevated temperatures (T > 450°C), where the crack growth rate increases dramatically due to interactions between the environment and the loading of the crack tip. Such material can, on the other hand, to a very large extent be considered as defect-free which reduces the impact of sustained load cracking as there are virtually no cracks present. Experience has indicated that castings are much less sensitive to sustained load cracking. The sustained load crack growth phenomenon is believed to be a grain boundary related phenomenon and the grain boundaries in forgings and castings are different in nature which may explain the differences in sensitivity to sustained load cracking. Welds are even less investigated than castings and forgings and there is no publically available literature on this topic for superalloys. Furthermore there is a much higher risk of encountering cracks and defects in welds compared to e.g. forgings.
LIFEMOD creep fatigue: an innovative process aimed at the development of a lifing model
The creep-fatigue problem has been addressed by the following approach, aimed at advancing the understanding beyond the current state-of-the art:
• Creep-fatigue and sustained load crack growth tests has been performed on welded IN718 sheet material, where the crack has been initiated at different locations with respect to the weld. The cracking behaviour has been studied under loading scenarios which has been designed to mimic the critical aspects of the open rotor conditions as closely as possible. This has revealed “weak spots” in the welded structure.
• Investigations of the test specimens by high-resolution microscopy and micro-analytical techniques have revealed the effect of local microstructure and the weld, HAZ and base material on the crack propagation.
The overall results has provided the Topic manager with an improved understanding of the creep-fatigue phenomenon in welded IN718 to better estimate the risks and accordingly develop a strategy to mitigate/eliminate such risks in the rotating frame design process
Overall strategy and general description
The overall strategy for the LIFEMOD project has been to perform testing, NDT testing, fractography and analysis as efficient as possible. Thus most activities has started as soon as possible in the project and been performed simultaneously. To handle the project a number of work packages (WP:s) has been set up and each WP has a WP leader responsible for the execution of the work in the work package. A description of the work packages and a summary of the results are given below. A steering (or management) committee was set up to coordinate the work in the work packages. The steering committee has met regularly in both physical meetings and in teleconferences.
Early on in the project a decision was made by the steering committee to divide the number of test specimens into two batches. This to ensure that the welding methods used produced the desired defects.
Project Results:
WP 1 - Specimen manufacture
The objective of this work package has been to receive the welded material from the topic manager and arrange for specimen manufacture at a subcontractor.
As mentioned previously the total number of specimens was divided into two batches and the same subcontractor was used for all manufacturing. The cost for specimen manufacture had been underestimated in the original proposal and both a reduction of the total number of specimens and reallocation of funding between the partners had to be made. The topic manager approved of the alterations to the work plan and the decision was made by the management committee.
WP2 - Non destructive testing (NDT)
In WP2 the WP leader has used the revolutionary new technology of metrotomography and it has been applied to the study of weld defects present in selected specimens before testing. The results have been reported into the LIFEMOD database using a coordinate system that has allowed for comparison with the fractography results from WP3. In this way a defect that has been found using fractograhy of the fracture surface can be traced back to the defect position and size. Moreover the defect distribution statistics has been compared to the fractography statistics in work package 6 and conclusions has been made.
The WP2 leader also provided line scans of all specimens not being metrotomescanned. When welded the blanks for the specimens distorted leading to bent specimens. This bending has an effect on the fatigue properties and needed to be taken into account for the analysis of the results. The management committee and the topic manager decided to linescan all the specimens and the WP2 leader undertook the work. Funding of the linescanning was made available by decreasing the number of specimens to be metrotomescanned. The results from the linescanning was input into the LIFEMOD database and used by WP6 and WP7 for analysis.
WP3 - Fractography of HCF specimens
In WP3 the fracture surfaces from broken WP4 HCF specimens has been studied. For each specimen the crack initiation position has been identified using the same coordinate system as in WP2. The results has been compiled and input into the LIFEMOD database. The results also show the difference in the defect density between batch 1 and batch 2 of the specimens. Batch 1 has contained almost no defects resulting in the crack initiation point being identified as a geometry stress concentration point. In batch 2 both lack-of-fusion and start-stop defects were noted as the initiation point.
WP4 - High cycle fatigue
In WP4 a large number of specimens were tested in high cycle fatigue mode. 4 different specimen geometries with laser weld joints have been tested with respect to fatigue with a stress ratio R=0. The weld joints were oriented parallel as well as perpendicular relatively to the length direction of the specimens. The tests were carried out either at room temperature or at an elevated temperature of 550 ⁰C. The tests were carried out in a frequency interval between 65 to 110 Hz depending on the geometry of the specimen.
The load levels have been chosen with aim on two desired number of cycles to failure – 1∙105 and 2∙106 cycles. In case of 1∙105 cycles the lower and upper limits for changing the load were set to 5∙104 and 2∙105 cycles respectively. For the 2∙106 level the corresponding numbers of cycles were 1∙106 and 5∙106. A load step of 30 MPa was used for a majority of the tests. In some cases a load step of 15 MPa was employed to get as many failures within the desired interval of number of cycles to failure as possible.
The total number of specimens is 284 almost equally divided between batch 1 and batch 2. In batch 2 some specimens was used to calibrate the bending and its influence on the fatigue life. The results from WP 4 were inserted into the LIFEMOD database.
WP5 - Creep fatigue
In WP5 creep fatigue hold time fatigue tests have been carried out by the work package leader. The testing was performed in three different modes, cyclic testing, creep crack propagation testing and dwell holdtime testing. The specimens was planned to be stopped before final fracture, but some of the specimens ruptured before they could be interrupted. After testing the test data was analysed and the equivalent crack radius, a, and the crack stress intensity factor, K, were calculated. A log-log plot of crack growth rate (mm/hour or mm/cycle) versus stress intensity factor was also produced.
Metallographic and fractographic investigations were carried out on tested specimens in order to further examine the mechanisms of crack propagation under different conditions. Representative specimens for each condition was selected, prepared as described in the following section, and investigated using stereo-optical microscopy (SOM) and scanning electron microscopy (SEM).
Both the fracture surfaces and the crack paths were studied. The unbroken specimens were sectioned longitudinally and one half mounted and polished, whereas the other part was cracked open to allow for study of the fracture surface.
WP6 -Statistical analysis
The work in WP6 has been to analyse the HCF data from WP4 and the defect and crack initiation data from WP3 and WP2 respectively and deliver the results to WP7. The work has been divided into two parts, one part analysing the defect distribution in the specimens and one part analysing the high cycle fatigue results. All work is now finished and the report has been produced.
Defect distribution analysis
The damage mechanism in the LIFEMOD project is fatigue crack growth from defects. One can assume that large defects act as initial cracks, and initiate a fatigue crack early in the component usage. As a basis for defect estimation observed defect sizes obtained from WP2 and WP3 results for a number of test specimens has been used. Since the defect contents show a large variation from specimen to specimen it should be suitable to describe the contents by means of a statistical distribution. To make an analysis possible a common definition of defect characterisation was been adopted for WP2, WP3 and WP6.
Typically, the extreme value sampling is performed by defining a control area (or length), and for each such control area the maximum defect is registered. Therefore, this extreme value approach is often called the Control Area Maxima (CAM) method. Another extreme value approach is the so-called the Peak Over Threshold (POT) method, where all defects larger than a given threshold are considered. This method may be used in order to take all “large” observations into account, which can be especially useful in the case when several defects are found in each class object. However, then the set of observations does not directly represent the maximum for the class. In this case the POT-method was judged to be most suitable.
For extreme values like maxima or minima there is a convenient class of analytical statistical distributions, namely the extreme value distributions. It is defined to have the same type when taking maximum of maximum, i.e. it is convenient to relate to different sizes; if the distribution of the maximum defect is known for, say a 100 mm weld length, it is easily transformed to the distribution of the maximum defect for another length, for example, to a 1000 mm weld length.
For Batch 1 it was observed that the inclusions found by fractography were much smaller than the ones from the metrotomography investigation. It should also be noted that most specimens from Batch1 failed due to geometrical defects and not due to internal material defects.
The welds in Batch 2 have an intentional disturbance in the welds. This causes a complication in the analysis in respect with the intensity property which here is for weld length. Since the defects are caused by the deliberately made disturbance, they do not occur randomly along the weld. Instead they appear randomly in conjunction with disturbances during the welding. A better analysis in a real application could be to use some intensity of defects in disturbed welds; disturbances that are either known or are also appearing randomly.
Fatigue test analysis
The fatigue test results are plotted in a log-log diagram. The classical Wöhler curve or SN-curve is a linear approximation in this log-log diagram of the constant amplitude fatigue life N_i as a function of the load level, stress range ΔS_i. The stress is calculated as the applied force over the measured cross section area. The straight line in log-log scale can be expressed by the traditional Basquin equation with parameters k and 〖ΔS〗_0, and where ε_i represents the scatter in fatigue life. The scatter is here assumed to be independent of the load level.
The results analysis gives the fatigue strength at two million cycles, the Wöhler slope with a 95% confidence interval and the standard deviation of log-life. The estimation of the Wöhler curve is performed using the maximum-likelihood method, in order to correctly take run-outs into account. In the estimated Wöhler curve the standard deviation at the 95% prediction band is the interval for a future observation and includes both the estimation uncertainty and the scatter in life. The FAT-value and the lower prediction limit at two million cycles are also marked in the figure.
WP7 - Development of a lifing model
A probabilistic life prediction model for welded (IN718) components which contain defects from which cracks can initiate and grow up to failure has been developed. Considerable scatter is present in the life from component to component, due to scatter in for instance defect size, material properties and loads, where the scatter in defect size is expected to cause most of the scatter in the life of the component. A probabilistic lifing model, taking into account variability due to loads, material properties, initial defect size and other important scatter sources, may improve the component design life, which is the goal of LIFEMOD.
The developed probabilistic lifing model is very flexible and can take into account sources of scatter for all parameters in the deterministic model. The deterministic model consists of a crack growth tool (NASGRO) and a fatigue analysis tool (Cragro++) and can be easily extended with other tools. The probabilistic part is formed by a generic probabilistic tool RAP++.
The probabilistic lifing model allows for a generic set-up of the analysis problem without the need of source code modifications. Typically the analysis consists of definition of the deterministic analysis, using the deterministic model in a probabilistic sensitivity analysis using approximate distributions that model the variability in the various model parameters to determine the most important scatter sources, followed by a reliability analysis to determine the probability of failure, which was demonstrated on an example problem.
The bending and offset present in the high-cycle fatigue (HCF) specimens caused a considerable additional stress field at the weld line during loading, impacting the HCF test results. Both bending and offset varied between the different specimens. A parameterised finite element model of the weld specimen was made to compute this residual stress field for each specimen, based on accurate geometry line-scan measurement data from Carl-Zeiss. The results of the finite element analyses were two-fold: 1) the remote load as applied in the HCF tests can be corrected for the residual stress field present in the specimen, 2) the resulting residual stress field was applied as another source of variability in the life prediction model.
The work in WP7 started early on in the project and has been continuously reworked as new data has been delivered by the other work packages. Progress reports has also been given at most management committee meetings and input and comments from the topic manages has been incorporated in the work. This communication culminated in a workshop and training exercise held by the WP7 leader at the topic managers site in conjunction with the last physical LIFEMOD meeting held at the same time.
WP8 - Project management
The management of the project is handled by the coordinator acting as work package leader for WP8. To assist the management of the project a project management manual was developed and distributed among the partners. The WP8 leader has also kept the records of the project including the LIFEMOD database and the report database.
Project management has been uneventful and all issues that have come up have been able to resolve within the management committee. 23 meetings of the management committee have been conducted. The milestone and delivery tables are also maintained by WP8.
In progress period 1 two major problems has been encountered by the management committee and the project as a whole (the division of the specimens into two batches and the resulting delay in specimen manufacturing and the lack of complete funding for the subcontracted specimens manufacturing). In progress period 2 no major obstacles was encountered with the project and the management thereof.
The consortium has not changed during the project, and the legal status of the partners is not altered.
Potential Impact:
Expected impacts
The development of open rotor, green aircraft engines is very important for the future of the European gas turbine industry. Open rotor engines have the potential of substantially reducing specific fuel consumption and thus increase the efficiency of the engine. The result will be an engine that is both greener and more sustainable to the environment.
Within the open rotor development components has to be manufactured using welds and these welds will be subjected to stresses in an environment that will be aggressive to the component. This project has provided a very extensive database on mechanical testing of welds and a lifing model that will be used in the development of the components.
Furthermore, the non-destructive testing (NDT) performed in this project has provided insights into the defect distributions in a typical weld produced using the best available techniques. Since the NDT used is revolutionary, the results can be used to verify those obtained by standard NDT, and this result in itself will be of great use to the engine manufacturers.
Contribution to European societal objectives
LIFEMOD has contributed to the improvement of the quality of life by means of new and enhanced products in the gas turbine industry, generating wealth and employment. Research activities in the field proposed has provided economic growth and a sustainable development path. The development of a new lifing model to be used by the industry is a new opportunity. European designed products and processes using the new functionality have the potential to improve the quality of life for everybody
Contribution to the European Skills Base
This project has contributed to the preservation of economically important gas turbine industry for Europe, by offering high added value in products, processes and services. The results from the project have increased the competitiveness of the same industry, with the potential of a continuous growth in employment.
European scientists are recognised worldwide for their leading-edge research and the gas turbine industry is a crucial area for European level research. Thus, excellent technical solutions will give the EU as a whole a competitive advantage to develop the leading industries in this field with vast economic potential. This will result in an increased magnetism on young technical professionals, because of high quality jobs. LIFEMOD has contributed to the continuing development of the gas turbine industry.
Dissemination of the results from LIFEMOD
LIFEMOD has disseminated the results of the research to the topic manager who will use it to develop new components. The written part of the dissemination has been performed by periodic reports, work package reports, the result database and a final report. In addition to the written material regular communication with the topic manager on the developments within the project has been held. At the end of the project the leader of work package 7 has held a training workshop of the use of the developed lifing model at the topic managers research site. The workshop had high attendance.
While the dissemination of the work is mainly aimed at the topic manager, but certain parts of the individual work packages might be suitable for dissemination to a wider audience. At the moment several such publications are contemplated, especially regarding the lifing model and the metrotomograpy possibilities.
List of Websites:
no public website