Professional Documents
Culture Documents
Steam turbines are a fundamental component in most electrical power generation systems around the world. Their
efficiency and capital cost are dependent on the quality of the blade roots which secure the thermodynamic aerofoil
sections to the high speed rotor. This paper outlines a design analysis for such a root system using classical stress
analysis and the finite element method. Comparisons are made between the results of both and experimental pull tests.
For reasons of simplicity, the analysis considers only the primary centrifugal load on the root and not secondary
steam bending loads.
The M S of this paper Ha5 received at the lnstltution on 13 September 1989 and Axial entry root
acceptedfor publication on 22 Febmary 1990 Firtree root
Now at Westland Heliropters Limited, Yeouil, Somerset
t References are givrn in the Appendix Fig. 1. Current root systems(by kind permission of NEI-Allen Ltd)
following analysis, but the techniques are entirely applic- (i.e., 1% Cr Mo V and Firth Vickers FV520 steels,
able to the other types of root. The proportions of the respectively), rather than rotational tests of complete
root are indicated in Fig. 2. This also shows the primary‘ discs. In this way the tests were an approximation of the
features of the root, including the four load bearing sur- real world, most significantly in the fact that the plate
faces and the leg restraints which prevent the legs from was flat, whereas a real blade root is tapered as a
opening out and sliding over the serrations. segment of an annulus around the rotor disc. However,
the taper is small and the tests have proved to be a
1.1 Notation reasonable basis for the current stressing practice used
by the company. The results of these tests form the basis
A Area of section above plane on which shear stress
of a comparison between theory and reality.
acts
As part of the test procedure, a number of photo-
a Bending moment arm
B Width of section under shear graphs were taken of the root system at various load
increments. Figure 3 shows the situation just before ulti-
d Width of neck of ‘T’ head
E Young’s modulus mate failure. The aerofoil section would normally be
Depth factor attached at the top of the figure and the rest of the disc
e
would be at the bottom. It should be clear from this
h Depth of ‘T’ head
photograph that failure occurs by a shearing of the
I Second moment of area, based on the section m
m Depth of serration wheelhead serrations. The blade serrations show much
P Load applied to serration less deformation because of the superior blade material,
Tensile stress which has a limit of proportionality at room temperature
(T
Tensile strength some 25 per cent higher than the rotor material, Fig. 4.
CYP
t Thickness of section Consequently, to predict the yield strength of the root
Maximum shear stress system analytically, close attention must be paid to the
Tm*x
V Shear load on individual serration yield of the wheelhead serrations.
V Poisson’s ratio The first approximation is to assume that the serra-
Distance from neutral axis to centroid of A tions yield by pure shear, and that the shear stress dis-
Y tribution is uniform. This gives an over-estimate of the
load carrying capacity of the section since, in reality, the
2 THEORETICAL ANALYSIS shear distribution is non-uniform. At the top and bottom
The form of straddle root being considered in this paper of the section the shear stress parallel to the free surface
was tested by NEI-Allen to establish its experimental must be zero since there is no material to carry any
strength. These tests were simple pull tests of plate stress. Therefore, the shear stress rises from zero at the
material, using actual wheelhead and blade materials edges of the section to a maximum at the centre. The
15.4
I c
I 9.1
I
Outer load
bearing serrations
2 Inner load
bearing serrations
i
0
Roo1 Z= 18 wide
(b)
Fig. 2. The straddle root:(a) proportions and features, (b) nominal dimensions(in mm, not to scale)
where
5.6hld + 1
K = 1 + 0.26(:) O.’
x
2.0hld + 1
P = load applied to projection
t = thickness of projection
h = depth of ‘T’ head
d = width of neck of ‘T’
Fig. 3. Yielding of root in pull tests (by kind permission of NEI-Allen
Ltd) Other factors in the equation are specified in Fig. 6.
If the stress concentration factor is ignored, equation
(3) can be used to calculate the bulk yield of the serra-
tions, which for the actual dimensions give a load of 73
maximum transverse shear stress can be obtained follow-
kN. Applying the same dimensions to the calculation of
ing Case (2) which gives
the stress concentration factor gives a value of K = 2.12
VAj for the outer serrations and K = 2.00 for the inner ones,
7msx =- implying that local yield around the fillet radii occurs at
IB
a load of approximately 35 and 37 kN,respectively. Both
The area A is based on the dimension m shown on Fig. the local and bulk yielding loads are illustrated in Fig. 5.
2, with the load carried by the serrations on each side. More recently Fessler and Woods (6)examined tensile
For the maximum shear stress A is equal to m/2 times C. stresses in rectangular ‘T’ heads (c wide, b high) sym-
Based on the Huber-von Mises-Hencky yield criterion, metrically joined to shanks of width a and with smoothly
, , ,z = 0.577 gyp(3),it is possible to calculate the load at blending 90 degrees circular arc fillets of radius r. In
which yield will start. For the dimensions of the root these shapes the maximum tensile stresses were shown to
being considered, the maximum load is 78 kN based on primarily depend on the fillet radius ratio r/a and on the
the load being shared equally between the load bearing positions of the reactive forces on the shoulders, defined
serrations. This load is plotted on Fig. 5 together with by e/a, where e was the distance from the shank outer
the maximum load permitted by the uniform shear stress edge to the position of the resisting load acting under the
assumption. Both are superimposed on a graph of the ‘T’ head. The ratios c/a and b/a became unimportant
pull test for comparison. The extension/load curve across when these ratios were small. The smallest value of stress
the root shown in Fig. 5, was measured over a 50 mm concentration, defined as Kmin,was found to be
length straddling the whole of the root system. Uniform -0.39
shear is shown to give an over-estimate of the load carry-
Kmin= 1.16(:) (4)
ing ability of the root, but the maximum shear stress
JOURNAL OF STRAIN ANALYSIS VOL 25 N O 3 1990 0 IMechE 1990 187
I I I I I I I I
0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8
Strain/% Strain/%
(b)
Fig. 4. Stress-strain curves for (a) blade, and (b) wheelhead material
Applying equation (4) to the outer wheelhead projections 3 ELASTIC FINITE ELEMENT ANALYSIS
(suitably modified to allow for the influence of e/a and In the previous section, a number of hand calculations
b/a ratios as stated in (6)), the stress concentration is were carried out in order to arrive at the stresses in the
higher than the Heywood figure and the root load is turbine blade root. Stress is, however, only half the
estimated to be 43.6 kN. This figure is speculative as the picture. The extension of the root is also important, but
Fessler and Woods results do not consider h/d ratios as for a complex component such as this it is almost impos-
low as that used in the root. For these reasons loads for sible to predict this by hand calculation. Finite element
inner serrations cannot be given as their geometry is far analysis (FEA) can be used to determine the extension of
outside the range considered. In a later paper Fessler and the root and to confirm the stress calculations of the
Woods (7) examined the stress concentrations in 'T' slots, previous section. It can also be used to investigate the
work which is closely related to the geometry of a strad- effects of uneven load carrying by the serrations as a
dle root. As for the 'T' head, the stress concentrations for result of manufacturing tolerances. The particular FE
'T' slots were shown to reach a limiting value. package used for this analysis was PAFEC-FE running
It will be seen from Fig. 6 that the Heywood analysis on Apollo workstations and a VAX 8700 mainframe.
for bulk yield is in good agreement with the pull test
results. It accurately predicts the load at which the curve
turns to bulk yield. When the stress concentration factor 3.1 Themodel
is included, a very low load is predicted which seems The straddle root used by NEI-Allen was modelled
unrealistic in comparison to the observed yield in the using 2D isoparametric elements for plane stress and
pull test and the apparently linear region of the curve. plane strain, although only the plain stress condition was
However, careful inspection of the pull test curve reveals used. These were a combination of 8-noded quadrilateral
a small upward turn starting around 35 kN so that the and 6-noded triangular elements. The model contained
centre section of the curve is not quite linear. Heywood 608 elements with 3633 degrees of freedom.
showed that the highest stresses occur round the fillet The mesh is shown in Fig. 7. From this it will be noted
radius, so that at loads between 35 kN and 73 kN local that the only one half of the root is modelled, the other
plastic deformation occurs. At higher loads, the yielding half being a mirror image. Nodes along the axis of sym-
becomes a bulk condition and the root may be con- metry (x axis) were restrained in the y direction whilst
sidered to have yielded. Bulk yield is a situation where allowing movement in the x direction. Similarly, nodes at
yielding occurs in a section in a manner which cannot be the base of the wheelhead structure (y axis) were
attributed to local stress concentration effects. The finite restrained in the x direction, with movement permitted
element analysis carried out confirmed that local yield- in the y direction. The loading was applied evenly to the
ing would take place. elements across the end of the blade structure and in
188 JOURNAL OF STRAIN ANALYSIS VOL 25 NO 3 1990 0 IMechE 1990
0.20
0.18
&I
I1
-"E l
1 I
2p
31
0.16
I
I
0.14
I
I
E I
.
s
E
.-
0.12
I
n
C
5 0.10 I
I
0.08 I
I
Fig. 6. Geometry for Heywood analysis
0.06
I
I
0.04 I I I and would not be truly 2D. Where the blade root comes
I I I into contact with the wheelhead at the load bearing sur-
faces and at the leg restraint, the FE model included gap
0.02 I I I elements which allow a separation to be defined between
I I I the surfaces. Such a gap may open or close according to
the loads applied and the deflection of the structure.
20 40 60
480- 7 -loob I20
When the gap is closed, compression forces may act
across the gap element. The particular type of element
LoadlkN
used in this problem also takes account of the friction
Fig. 5. Comparison of theoretical results between the two surfaces when they come into contact.
In order to investigate the effects of manufacturing
tolerances, and to make the analysis representative of the
accordance with the normal 1 4 1 load distribution for real case, the interaction of the blade root and wheelhead
parabolic finite elements. The loads were in the x direc- was modelled in two ways, as follows.
tion only and were calculated to give a particular overall
load to the structure. The model was chosen to be a true ( I ) First, the basic structure was created and gap ele-
representation of the pull test specimens, i.e., 2D with ments were positioned across the load bearing sur-
parallel faces. A real blade would have a tapered root faces and across the leg restraint. These elements
forming a segment of an annulus round the turbine disc linked corner nodes and mid-side nodes between the
Fig. 7. FE model
189
JOURNAL OF STRAIN ANALYSIS VOL 25 NO 3 1990 0 IMechE 1990
Downloaded from sdj.sagepub.com at Purdue University Libraries on June 24, 2015
M. D. ROBERTSON AND D. WALTON
II
0.20-
Case (2)
Case(l)
.E
E 0.12-
.-v10
5
2 0.10-
w
0.08 -
0.06-
0.04-
20 40 60 80 I00 120
LoadIkN
surfaces coming into contact. It is impossible to model natural frequency and therefore more difficult to design
this by the finite element technique at this scale. The away from it.
second is when there may be a situation where only one Whilst the closing of the gap is important for vibration
of the load bearing surfaces is actually in contact. The reasons, it is also necessary to consider the effect of a gap
second load bearing surface shows a gap due to manu- on the strength of the root. As long as a gap exists, one
facturing tolerances. The maximum gap allowed within pair of serrations will carry all the applied load, whereas
tolerance is modelled in case (2) as described above, also with no initial gap, the serrations all carry very similar
shown in Fig. 9. In this model 14 load cases were applied loads. As a result, local yielding will occur at approx-
rising in magnitude from 2 to 100 kN. The status of the imately half the load it would normally. Figure 10 shows
gap elements at the outer load bearing surface was a close-up view of the stress contours around the inner
checked for each load, and it was seen that all the gap serrations for case (2) with a load of 17.5 kN, which is the
elements closed together at a load of 18 kN. From Fig. 8 load at which the Heywood analysis predicts local plastic
it is clear that this causes the change of slope of the load deformation will start if only one pair of serrations carry
extension curve. The lower section of the curve shows the the load. The maximum stress contour around the inter-
more flexible situation of only one surface in contact, face between the serrations, I , has a value similar to the
while the upper section shows the stiffer, twin loading limit of proportionality of the wheelhead material, so
case. This loading characteristic need not be a problem, that it is possible to conclude that this load represents
providing that at the normal turbine running speed, both the limit of proportionality foi the root and the start of
surfaces are in contact. If not, the root will be a less rigid local yielding. It should be noted that the gap has not
and poorer platform for the blade, in terms of vibration quite closed at this load. It would be useful to follow
characteristics. through this loading pattern with elastic-plastic material
Vibration of the blades receives much attention in properties, but as will be explained in section 4, this is
steam turbine design because it is quite easy to excite not possible with PAFEC because of the presence of gap
the blade at one of its natural frequencies causing elements. Instead, it is reasonable to suppose that the
catastrophic failure. The problem is overcome by deter- serrations which are initially unloaded will carry a higher
mining the natural frequencies of the blades and ensuring proportion of further loading once the other serrations
that there are no sources of vibration close to these fre- have begun to yield. Ultimate yield failure may be judged
quencies. If the nature of the blade root is variable or to occur when all serrations have started to experience
loose, it would be more difficult to be sure of the blade’s bulk yield. This will be the same, irrespective of the gap,
von Mises
(MW
J = 720
I =644
H = 568
C = 492
F =416
E = 341
D = 265
C = 189
B = 113
A = 37
Max. stress
= 750
Min. stress
= 0.00
Load = 17.5 kN
as long as the pre-loaded serrations do not fail com- head structure will plastically deform and, therefore, the
pletely before unloaded serrations take up some of the extension of the root will be greater than the elastic case
load. Consequently,the ultimate strength of the root will simulated by the FE analysis. There is quite a large
be unaffected by small manufacturing tolerances on the increase in load before the local plastic deformation
serration spacing. becomes bulk plastic deformation. In this region, the
extension will be only slightly greater than the !inear
Linear region elastic case. In section 4, elastic-plastic material proper-
In the linear region of Fig. 9, from 20 to 60 kN, the FE ties will be considered and it will be shown that this local
models have the same slope. This is because the plastic deformation does indeed change the apparent
stiffnesses of the two structures are the same. They are, in slope of the pull test curve.
fact, equivalent once the gap closes. The amount of load
taken by each projection will be slightly different in the 4 ELASTIC-PLASTIC FINITE ELEMENTS
two cases since in case (2) the inner projection has an ANALYSIS
effective pre-load before the outer projection takes any In the previous section it was shown that local plastic
load at all. deformation starts to occur at a load of approximately
The slope of the FE analyses is slightly less than that 35 kN (when there is no gap), yet this design of root has
of the pull test. This may just be due to manufacturing been used successfully up to a load of 80 kN. In this way
tolerances in the root used for the pull tests. If the root the linear finite element analysis cannot predict the bulk
was narrower in any areas, it would be less stiff and failure of the root. Figure 11 shows the stress concentra-
would extend more for any given load. Another, more tions very well, but fails to indicate the stresses in the
important factor, is the local plastic deformation of the bulk material. It is the bulk condition which is more
blade root structure around the fillet radii. Figure 11 important in this type of situation under steady load,
shows the von-Mises stress contours (around the loaded since stress concentrations will yield locally and will not
projections) for a load of 35 kN on the root (case (1)). affect the bulk strength. The authors concluded that the
This is the load at which the Heywood theory predicts only way to investigate the bulk yield condition was to
local plastic deformaton will begin if all the serrations use elastic-plastic analysis, as has been done previously
are equally loaded. The figure shows that the stresses for gas-turbine wheels (8) where wheel stresses exceed
around the radii are approximately the same as the limit yield under start-up conditions due to the combination
of proportionality of the wheelhead material (600 MPa), of steep thermal gradients and centrifugal forces.
suggesting that in the pull tests, local plastic deformation The PAFEC systems solves elastic-plastic problems
starts to occur at this load. At higher loads the wheel- by the initial stress/initial strain method. Detailed dis-
von Mises
(MPa)
J = 713
I =638
H = 563
G = 488
F =412
E = 337
D = 262
C = 187
B = 112
A = 37
Max. stress
= 750
Min. \tress
= 0.00
Load = 35 kN
cussion of the technique would be inappropriate here. close as it is possible to come to the real case without
Instead, reference should be made to PAFEC (9). using gap elements.
As the new model was fundamentally different from
4.1 Alterations to the FE model that used previously, it was decided that a linear analysis
The PAFEC system has limitations when solving elastic- should be performed first. The model extended by 0.122
plastic situations that involve gap elements. The mm under a load of 100 kN. This is shown in comparison
problem is that the solution of the states of the gap ele- to the previous FE work and the pull test results in Fig.
ments requires one level of iteration while the elastic- 12. The new model, without gaps is a little more stiff than
plastic solution requires a further level of iteration. the previous one with gaps. This was expected because
PAFEC has not been designed to solve both problems the solid linking of the structure will inevitably make the
simultaneously and the literature specifically excludes structure more stiff locally. This is reflected in the smaller
models which require both. As a result it was necessary total extension of the root system. As the earlier FE
to alter the model used in section 3 and remove the gap model is a better representation of the real case, the
elements. The new model linked the wheelhead and elastic extension of the earlier model was taken in prefer-
blade directly together at the main load bearing projec- ence to that of this new model when analysing the results,
tions, although it was still possible to specify two i.e., the total root extension was taken as the elastic
material properties for the separate components. In this extension of the gaps model plus the plastic extension of
way the mesh was continuous over the whole model, the new model.
rather than being in two sections. At the leg restraint, the The elastic-plastic analysis required the stress-strain
authors considered it unwise to link the structures solidly characteristics of the two materials in uniaxial tension. In
together as this would significantly increase the stiffness order to provide these, special material tests had to be
of the model and prevent relative motion between root carried out, Fig. 4, because the material data which was
and wheelhead. The photographic evidence from the pull available only quoted the 0.2 per cent proof stress and
tests clearly shows that this exists in the real case (see Fig. ultimate tensile strength. The load for analysis was
3). Instead, nodes on the blade structure were restrained applied in increments to arrive at a total load of 100 kN.
to move only in the x direction, as though the wheelhead The first increment was 40 per cent of the total and was
restraint were rigid with zero friction. This is perhaps as designed to give a small amount of plastic deformation.
0.20 0.2c
0.16 0.16
0.14 0. I4
Gap = Zero
0. I 2 Solid structure 2
. 0.12
z
.
s2
.-
.-0
LI1
C
0.10 2 0.10
u
I I2
L2
0.08 0.08
0.06 0.06
0.04 0.04
0.02 0.02
20 40
,
60 80 100 120 20 40 60 80 100 120
-
LoadlkN LoadlkN
Fig. 12. Load/extension curve for elastic-plastic test case Fig. 13. Comparison of elastic-plastic FE results and pull test
The rest of the load was applied in diminishing illustrate the progression of the plastic region from the
increments of either 10 or 15 per cent so that yielding of stress concentrations to the bulk material. This has been
elements in each increment was kept to a minimum. done in Fig. 14. Initial yielding is very localised around
the contact zones and fillet radii, confirming the theory
4.2 Results that local yield starts at approximately 35 kN. As the
The extension of the root system for each load was load increases, the yielding moves through the load pro-
extracted from the text output file. It has already been jections rather than through the width of the blade or the
shown (Fig. 11) that the elastic stiffness of the elastic- wheelhead. This is in good agreement with the photogra-
plastic model is too great and that the first elastic FE phic evidence from the pull tests, which shows failure by
model was better. In order to take account of this it is shear. The figure also agrees with the photograph in that
necessary to replace the elastic extension of the elastic- it shows that there is more yielding of the wheelhead
plastic model by the elastic extension of the original than the blade. However, more detailed information can
model. At the same time it was possible to take the be gained from the figure regarding the nature of the
opportunity of adding onto the FE results the offset initial yield, which cannot be seen in a pull test because
caused by bedding in and manufacturing tolerances. This of the size of the effect.
is represented by the difference in extension between FE
curve (1) and the pull test curve of Fig. 9 at a load of 20 5 CONCLUSIONS
kN. To summarise, the true extension of the root is cal- The theoretical analysis of section 2 has been shown to
culated as correlate well with the pull test so that its validity would
I
Extension = extension due to bedding appear justified. By using Heywood’s equation it is pos-
of contact surfaces sible to determine the loads at which bulk and local yield
offset
+ extension due to will commence. Future work should use this analysis to
manufacturing tolerance re-proportion the root in such a way that it does not fail
+ elastic extension of case (1) by shearing of the wheelhead serrations only, but by
(section 2) yielding of all areas simultaneously. When such work is
+ plastic extension of undertaken, the steam bending loads, though small,
elastic-plastic model might also be taken into account.
While only one turbine blade root geometry has beer
The results of carrying out this calculation are given in investigated, the value of elastic finite element analysis as
Table 1 and are shown graphically in Fig. 13. In the a design tool has been demonstrated, including the
Figure, the smooth curve shows the pull test result and effects of various manufacturing tolerances. Extending
the points are plotted from the ‘Total’ column of Table 1. this work to an elastic-plastic analysis will show very
The agreement between FE results and pull test is clearly the yielding pattern of the root, and its strength
remarkable. It shows that in future, it might not be neces- may be determined accurately without cutting any metal.
sary to make expensive test pieces and pull tests because A new root form may be fine tuned with the F E analysis
FE analysis can produce most of the information neces- before an actual pull test is finally carried out to confirm
sary to evaluate a particular root design. The only aspect the results.
which cannot easily be found from the FE analysis is the
ultimate strength of the root, which may be of interest as ACKNOWLEDGEMENT
regards the factor of safety. With a ductile material this
should not be a problem since the design practice will be The authors wish to thank Mr P. G. Watson, Turbine
well clear of this level at loading. Technical Manager, NEI-Allen Ltd, for permission to
The text output from PAFEC indicates which ele- publish this paper.
ments have yielded at each load increment. The locations
of these elements may be plotted on the original mesh to APPENDIX
REFERENCES
Table 1
(1) KEARTON, W. J. Steam turbine theory and practice, 1948 (Pitman,
Extension (mm) London).
Load (2) CASE, J. The strength of materials, 1941 (Edward Arnold, London).
(kN) Offset Elastic Plastic Total (3) TIMOSHENKO, S. Strength of materials, 1946 (Van Nostrand,
0.026 - 0.045 New York).
20 0.019
0.039 0.058 (4) LEWIS, W. Investigation of the strength of gear teeth, 1892
30 0.019 -
0.052 0.071 (EngineersClub, Philadelphia).
40 0.019 -
0.065 0.084 (5) HEYWOOD, R. B. Designing by photoelasticity, 1952 (Chapman
50 0.019 -
0.098
and Hall, London).
60 0.019 0.078 0.001
(6) FESSLER, H. and WOODS, P. J. ‘Stress concentrations at axially
65 0.019 0.084 0.002 0.105
loaded projections of flat bars,’ J . Strain Analysis, 1980, 15, 137-
70 0.019 0.091 0.002 0.112
143.
75 0.019 0.097 0.003 0.119
(7) FESSLER, H. and WOODS, P. J. Stress concentrations on axially
80 0.019 0.104 0.005 0.128
loaded, unsymmetrical projections on flat bars, J . Strain Analysis,
85 0.019 0.110 0.009 0.139
1982,17,23-29.
90 0.019 0.117 0.014 0.151
(8) TANG, S. C. and PETROF, R. C. Elasto-plastic analysis of stress
95 0.019 0.123 0.020 0.163
in a gas-turbinewheel, Trans. S A E , 1974,83,144>1460.
100 0.019 0.130 0.027 0.177
(9) P A F E C theory, 1984 (PAFEC, Nottingham).