You are on page 1of 8

Design Variation Simulation of

Tliicl(-walled Cylinders
Thick-walled cylinders exposed to high, static internal pressures may experience both
R. J. Eggert elastic and plastic deformation. Primary design considerations include loads, geometry
Mechanical Engineering Department, and material properties. However, variations in geometry and material properties due to
Union College, conventional manufacturing processes, and variations of internal pressure due to actual
Schenectady, NY 12308 usage patterns, propagate through the system resulting in off-design stresses and strains
which may cause failure. These variations can be evaluated using probabilistic methods
which are discussed in this paper Von Mises-distortion energy yield theory is presented
to predict elastic, plastic and residual stresses in thick-walled cylinders. The design
variation simulation method using Monte Carlo simulation and available statistical
information is used to design a pressure vessel for servo-hydraulic experiments. The use
of autofrettage to induce favorable compressive stresses at the inner bore, thereby
improving the margin of safety and overall reliability, is also presented.

Introduction
The design of a product requires knowledge of how the the input random variations which propagate in a cause-and-
product will be used, how it will be manufactured, and of effect fashion according to a functional relationship. For
what materials it is made. Successful design practices also example, the margin of safety is a function of the strength
attempt to compensate for aging processes such as wear and and stress. Variations in strength and or stress propagate
corrosion. The factor-of-safety method of design is one such through to the margin of safety. The mean value and stan-
design practice, where allowable stresses or factors of safety dard deviation of a function of independent random variables
are specified on the basis of experience or recognized design can be estimated by calculating the expected values of a
codes. While such design factors may comfort a design engi- Taylor series expansion of the function (Haugen, 1980; Ka-
neer by providing "rules of thumb," these arbitrary limits pur and Lamberson, 1977; Mischke, 1980; Rao, 1992). The
often lead to overly conservative designs. Probabilistic meth- mean value of the margin of safety and its standard devia-
ods, however, can provide more quantitative and qualitative tion, can therefore be estimated, and thus provide quantita-
design evaluations, leading to higher product functionality, tive measures of central tendency and dispersion of the
quality and overall value. margin of safety. Note that these estimates only depend upon
Probablistic methods are deeply rooted in the fertile fields the means and standard deviations of the input random
of worst case analysis, random variable algebra and Monte variables and are valid regardless of the type of input proba-
Carlo simulation. Worst case analysis is a simple, but effec- bility distribution. The method does however, require the
tive approach to analyzing alternatives with respect to per- assumption that the input random variables be independent
formance criteria. The design engineer merely poses the (i.e., not correlated). Unfortunately correlations sometimes
appropriate what-if questions, such as what if the load is exist, which reduces the generic value of this technique.
doubled, will the machine element fail? What if the actual The likelihood of an event can be determined by matching
yield strength is only 90 percent of the nominal strength? the estimated mean and standard deviation to the mean and
And so on. Assuming that the right what-if questions are variance parameters of a normal (or log-normal) probability
asked, the resulting worst cases reveal the design's weak- distribution. For example, the probability that the margin of
nesses. A more systematic approach couples the worst-case safety is less than zero, assesses the likelihood that the
of all the design variables, thereby describing the impact of machine element fails, and can be calculated from a standard
simultaneous variations (Balling et al., 1986; Parkinson et al., normal distribution table. The random variable approach has
1991; Teng, 1992). been applied to cylindrical, ellipsoidal and toroidal pressure
The random variable algebra approach incorporates statis- vessel analysis (Smith, 1984; Zibdeh, 1990); fatigue design
tics such as the mean value and standard deviation of the (Eggert, 1992; Mischke, 1987); factors of safety (Mischke,
design variables and parameters. The statistics characterize 1986, 1970); stochastic mechanical design (Eggert, 1991,
Mischke, 1989a, 1989b, 1989c, 1989d); optimal design and
reliability of airplane wing structures (Rao, 1986a, 1985,
Contributed by the Reliability, Stress Analysis, and Failure Prevention 1984a, 1984b); gearboxes (Rao, 1986b); geartrains (Rao,
Committee for publication in the JOURNAL O F MECHANICAL DESIGN.
1984c, 1979); machine tool structures (Rao, 1977) and func-
Manuscript received July 1993; revised June 1994. Associate Technical
Editor; T. H. Service. tion generating mechanisms (Rao, 1977). A thorough treatise

Journal of Mechanical Design JUNE 1995, Vol. 117/221

Copyright © 1995 by ASME


Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
of random variable algebra and general applications in me- modified, and a revised candidate design is simulated. The
chanical design has also been presented by Haugen (1980). simulate, analyze, modify, and re-simulate iterations are con-
Monte Carlo simulation produces a large sample of possi- tinued until the design engineer (or computer program) has
ble what-if scenarios based on historical or a priori statistical optimized the controllable design variables.
knowledge of the input design variables and parameters. The The simulations result in a wealth of statistical measures
margin of safety for example, can be assumed to be a such as: (7) minimums and maximums of the response func-
stochastic function of the two random variables, strength and tions, (2) sample average x, (5) sample standard deviation
stress, each characterized by its own mean value and stan- (T^, {4) coefficient of variation C^ = &^/x, (5) coefficient of
dard deviation. One simulation of the margin of safety is skew ^/S], (6) coefficient of kurtosis ^2, (7) and probability
generated by subtracting an instance of stress from an in- of a "failure" event. The average characterizes the sample's
stance of strength. Instances of stress and strength are com- central tendency, while the standard deviation characterizes
puter-generated according to their respective parent proba- dispersion about the average. The coefficients of skew and
bility distribution functions which are known (historically) or kurtosis measure the symmetry and peakedness of the proba-
which can be assumed (a priori). After many instances of the bility density function. The coefficient of variation measures
dependent function are simulated, appropriate sample statis- dispersion relative to the mean value or average. The proba-
tics are calculated. A distinct advantage of this approach is bility of an event, such as strength/stress failure, is directly
that the engineer need not worry about selecting the right obtained by comparing the number of simulated instances
combination of values, as compared to the worst case method. that "fail" to meet a prescribed specification, to the total
Since historical data are used to characterize the input ran- number of simulations.
dom variables, the resulting sample of simulations is a natu- The random variable algebra approach is relatively simple
rally-occurring collection of the possible instances. Most of and fast,;/ the required partial derivatives are readily obtain-
the what-if answers, if not all, are represented in the data able. However, it usually provides first order estimates and
collection, naturally or automatically. Design variation simu- will be invalid if the random variables are correlated. The
lation methods have been applied to a number of engineering design variation simulation method, on the other hand, is
problems including mechanism design (Crawford and Rao, relatively easy to program, capably deals with correlation and
1989), manufacturing and assembly (Early and Thompson, can provide somewhat more accurate results. However, it
1989), brake design (Doepker and Nies, 1989), fatigue design does require historical data or a priori assumptions and it is
(Eggert, 1992), structural wind loading (Shah and Chamis, computationally intensive. These drawbacks may not be that
1991); optimization (Rao, 1978; Siddall, 1983), and finite problematic however. First, more and more information is
element analysis (Thacker et al., 1991; Weber and Penny, becoming available regarding the statistical nature of materi-
1991). A hybrid approach using simulation to "proof-test" als, manufacturing methods, load applications and usages.
optimal designs obtained from random variable algebra solu- And secondly, as will be shown in this paper, existing per-
tions has also been presented (Eggert and Mayne, 1990; sonal computers have sufficient computing memory and pro-
Eggert, 1989). cessing speed. The next sections discuss the elastic break-
down margin of safety for thick-walled cylinders under high
internal pressures, and the application of the design variation
Design Variation Simulation simulation method.
The design variation simulation method goes beyond mere
variation simulation analysis to include the design engineer
in an iterative feedback loop to successively synthesize, ana- Elastic Cylinder Wall Stresses
lyze and optimize the product under consideration. Com- The thick-walled cylinder shown in Fig. 1 has an inner
puter programs can be integrated into the loop as well, such bore radius a, outer wall radius c, is subjected to a static
as in synthesis, to automate the generation of new candidate internal pressure pi and zero external pressure. In the closed
designs, or in the optimization phase, to determine the best case, the cylinder is modelled as being capped at both ends.
feasible design. The resulting, improved design process The resulting axial forces, sustained by the cylinder wall,
thereby makes effective use of historical or assumed statisti- thereby create tensile axial stresses. In contrast, the open
cal knowledge, and broadens the engineer's envelope of case refers to the situation where the axial forces are sus-
understanding. tained by external supports such as bolts, thereby resulting in
As a candidate design is synthesized, specific values for the no axial stresses. Solution of the governing equations is
controllable design variables and parameters are determined. credited to Lame, however a more contemporary derivation
In the case of a thick-walled cylindrical pressure vessel, the is available (Faupel and Fisher, 1981). The resulting nominal
inner and outer radii, internal operating pressure, and mate- stresses are defined as the tangential or hoop stress o-^,, the
rial properties are determined. The candidate design is simu- radial stress o-^, and the axial or longitudinal stress a^^ (open
lated. The simulation results are analyzed, design variables case), (T^^ (closed case). The stresses are given below as

Nomenclature

a = inner bore radius R = radius or diameter ratio (c/a) &^ = standard deviation
b = elastic-plastic boundary r = radius ^/P^ = coefficient of skew
b/c = autofrettage ratio Sy = yield strength jSj = coefficient of kurtosis
COV = coefficient of variation S„ = ultimate strength
C„ = coefficient of variation of X = average of a random variable Subscripts
pressure a-/, = hoop stress o = open case
= outer surface radius o-^ = radial stress c = closed case
m. = margin of safety-elastic o-^j. = axial stress (open case) h = hoop
breakdown a^o = axial stress (closed case) r = radial
= internal pressure a' = Von Mises equivalent stress z = axial

222/Vol. 117, JUNE 1995 Transactions of the ASME

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
/ Von Mises

- ^ ^ / <?, + da,
Hoop, \
\ \

Axial ^ „

U
Radial/

y J
1 2 3
Radial Location r
Fig. 2 Normalized nominal and Von Mises stresses for thick-walled
cylinder; (R = 3; a = 1; c = 3; closed case)

Fig. 1 Thick-walled cylinder under Internal pressure

functions of the radius r, the nondimensional index of wall


thickness R, defined as the ratio of outer to inner wall radii
(or diameters), R = c/a, and the internal pressure p,:
Pi c
(1)
R^ - 1 1 + -^
c
l - - r (2)
R' - 1
0.2 0.4 0.6 0.8 1.2
Location in Wall r/o
(3) Fig. 3 Von Mises stresses for thick-wailed cylinders; (R = 1.5, 2, 3,
' R^-\
and 5; open case)
(T,„ = 0 (4)
These nominal stresses are the principal stresses and can Table 1 Normalized Von Mises stresses for open and closed
be combined into one equivalent Vori Mises stress cr' de- thick-walled cylinders for various diameter ratios
fined as: fl = 5.0 fl=3.0 fl=2.0 fl=1.5 fl=l.l

/I \'^ [o'lpi. 1.8057 1.9526 2.3333 3.2167 11.058

^' = \l\i^k - o>)' + (^. - o-z)' + (o> - o-/,)1 j (5) i.o'lp^. 1.8042 1.9486 2.3094 3.1177 9.980
Ogl O^ 1.0003 1.0021 1.0104 1.0324 1.108
Substituting the Lame solution forCT/,,(T„ a^^, and cr.^. into
the above equation, the equivalent Von Mises stress distribu-
tions are obtained for a function of the dimensionless wall Similarly, all curves are bounded on the right by the maxi-
location parameter r/c: mum value of r/c which is c/c = 1. Note that little decrease
in peak stress is obtained for R values larger than 3. For
1/2
Pi 3 example, the benefit of increasing the wall thickness from
(6) i? = 3 to 5, is to decrease the peak stress from about 2 p, to
R^-l {r/c) 4 1.8 pi or about ten percent. The maximum Von Mises
1/2 stresses at the bore, are obtained by substituting the mini-
Pi 3 mum value of the location parameter into Eqs. (6) and (7):
(7)
R^ - 1 [ir/c)'\
1/2
(37?" + 1)
Figure 2 illustrates the nominal and Von Mises stresses for a Pi (8)
closed cylinder of wall thickness ratio R = 3. The stresses R^ 1
have been normalized by /?, resulting in a convenient multi-
purpose graph. As the radius r increases, a^ and cr,. de- ^/3R^
crease in magnitude while the axial stress o-^^. does not vary Pi (9)
with location. The inner bore undergoes the largest hoop and R^ 1
radial stresses thereby producing the largest Von Mises stress,
approximately twice the value of the internal pressure. Fig- The open and closed Von Mises stresses in (8) and (9) are
ure 3 shows the effects of changing the diameter ratio R on similar for values of R > 2.0 as seen in Table 1. At R - 2.0
the open case Von Mises stress as a function of wall location the open condition stresses are about 1.04 percent larger
r/c for R values 1.5, 2, 3, and 5. Each stress curve is bounded than the closed. At R = 3 and R = 5 the open condition
on the left by the minimum value of r/c which is a/c or \/R. stresses are 0.21 and 0.03 percent larger, respectively.

Journal of Mechanical Design JUNE 1995, Vol. 117/223

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Table 2 Simulation results assessing variations in static pressure
(p = 350 MPa; a = 25 mm; c = 50 mm; S„ = 950.3 MPa)

c, Min. Avg. Max. Std. Skew Kurt.


Dev.
c 0.20
s, all 840.46 950.30 1219.28 48.91 .392 2.859
0% 784.39 816.85 850.74 8.21 .150 3.071
5% 726.95 816.53 920.10 41.77 .019 1.894 0.15
"o
10% 658.97 816.22 990.89 82.20 .013 1.831
15% 590.99 815.91 1062.00 122.91 .011 1.819 0.10
0% 7.46 133.45 412.57 49.57 .370 2.859
m. 5% -49.20 133.76 446.40 64.27 .168 2.745 0.05
10% -119.95 134.07 502.59 95.61 .045 2.351
15% -190.70 134.38 570.01 132.24 .011 2.119
0 100 200 300 100 500 600
Margin of Safety (MPa)
The elastic breakdown failure criterion for thick-walled • 10% • •5% • 0%

cylinders compares the yield strength Sy to the maximum


equivalent Von Mises stress a' at the inner bore. This elastic Fig. 4 Relative frequency diagram of margin of safety {Cp = 0,
5,10, and 15 percent; C^ ^ = 1 percent)
breakdown margin of safety m^, can be restated as:
m„ > 0 (10)
Table 3 Reliability estimates for thicl(-walled cylinder for increas-
ing Cp
Sy- 0-' >0 (11)
Log- Relative Time
Note that the yield strength Sy embodies material properties Normal Normal Frequency
and that the Von Mises stress depends upon the geometry c. (S)
and the internal pressure. In the typical design situation, 0% .9965 .9979 1.0000 38.12
after the operating pressure is characterized, the appropriate 5% .9813 .9818 .9884 38.12
10% .9196 .9163 .9099 38.06
material and geometry are determined to guarantee a satis- 15% .8474 .8089 38.06
.8452
factory margin of safety. The next section discusses how
design variations can be simulated and the design modified
to successively converge to a satisfactory design.
variates a ~ N(25,0.25) mm, and c ~ N(50,0.5) mm, respec-
tively. The material yield strength was simulated as a Weibull
Design Example distribution (Mischke, 1989a) with mean and standard devia-
An experimental program has been undertaken by the tion Sj, ~ W(950.3, 48.9) MPa, (i.e., Weibull distribution pa-
author and a group of graduate and undergraduate students rameters ^0 = 840 MPa, b = 2.4, Q = 965 MPa). Four dif-
to investigate the viability of using an Instron servo-hydraulic ferent coefficients of variation of the internal pressure C
testing machine and specially designed testing fixtures to were examined (0 percent, 5 percent, 10 percent, and 15
assess thick-walled high-pressure cylinder fatigue, fatigue of percent). Each run resulted in 25000 simulated instances of
reciprocating seals subjected to high pressure, computer- the input and output variables. The simulated arrays were
aided fatigue testing and cylinder fatigue resistance improve- then processed by the statistical subroutine. The results are
ment gained by autofrettage processes. The Instron 245 kN shown in Table 2. When the internal pressure is deterministic
(55 kip) load frame will axially load a 20 mm diameter (C = 0 percent), the average margin of safety w^ is 133.45
piston-fluid-cylinder fixture to hydraulically generate internal MPa and the standard deviation 49.57 MPa. This is largely
pressures up to 350 MPa (50.8 ksi). To corroborate prelimi- due to the variation contributed by the yield strength. As C^
nary design calculations, the author prepared a design varia- increases to 15 percent there is little change in m^,, however
tion simulation for the candidate thick-walled cylinder of the standard deviation increases to 132.24 MPa. The total
inner radius 25 mm, outer radius 50 mm, internal pressure variation thus measured, is more than twice that caused by
350 MPa, external pressure 0 MPa (gage), made of 4340 steel the yield strength. The visual impact of static pressure varia-
whose average yield strength and standard deviation have tion on the probability density function (pdf) can be approxi-
been determined as 950 MPa (138 ksi), and 48.9 MPa (7.1 mated by a relative frequency histogram or diagram, as
ksi) respectively (Mischke, 1989a). A Fortran computer pro- shown in Fig. 4. Note how the probability of failure increases
gram was written which includes subroutines to generate as Cp increases. This is seen as the area to the left of the
pseudorandom numbers; generate uniform and Weibull vari- vertical line at w^, = 0. The pdf is slightly skewed to the right
ates using the inverse cumulative distribution method; gener- and its kurtosis decreases, or flattens out with increasing C .
ate normal and log-normal variates using a rational function The cylinder reliabilities were estimated and are shown in
routine; simulate instances of the inner and outer radii, Table 3. If little or no pressure variation exists, the reliability
internal pressure, yield strength, resulting Von Mises stress approximates 100 percent. As variation increases to C^ = 15
and margin of safety; and then calculate desired statistical percent, the reliability decreases to 81 percent. The upper
measures. and lower limits of a uniform distribution are u,l = ;u, + 1.731
The first aspect investigated was the impact of internal &, therefore for C^ = 15 percent, the upper and lower limits
pressure variation on the margin of safety. A steady, or static of pressure would correspond to + 26 percent of the mean
pressure variation results from using the vessel at a pressure pressure. This simple example reinforces the importance of
different from the design pressure. A malfunctioning control using equipment at, or under, the design pressure and to
valve or deliberate constant overload pressure are typical guarantee that control systems do not permit positive pres-
examples. Fluctuating or fatigue loading is different and will sure excursions. Normal and log-normal reliability estimates
be considered in a future paper. The internal pressure was were also obtained using the simulated averages and stan-
simulated from a uniform distribution, arbitrarily chosen, dard deviations. It appears that for C = 15 percent, the
with mean and standard deviation P ~ U(350,35) MPa. The normal and lognormal overestimate reliability by about 4
inner and outer radii were simulated as normally distributed percentage points (85 percent versus 81 percent). Each simu-

224/Vol. 117, JUNE 1995 Transactions of the ASME

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Table 4 Simulation results for elastic-breakdown margin of safety Table 5 Reliability estimates for thiicic-waiied cylinder for increas-
ntg, yield strengtii Sy, and Von MIses stress o-g for different C,,.
values
Log- Relative Time
c«. Min. Avg. Max. Std.Dev. Skew Kurt.
c„ Normal Normal Frequency (S)
Sy all 840.46 950.30 1219.28 48.91 0.392 2.859
0% .9209 .9176 .9118 37.73
0% 675.23 816.03 958.12 81.66 0.007 1.809 5% .8895 .8873 .8791 37.95
O^ 5% 605.55 821.18 1248.93 93.58 0.206 2.448 10% .7635 .7766 .8070 37.95
10% 564.83 840.57 4273.54 145.01 2.471 28.580
0% -103.99 134.26 510.37 95.16 0.049 2.346
m„ 5% -317.33 129.11 511.81 105.48 -0.104 2,650
10% 109.72 542.61 152.91 -2.111 24.137
lastly, autofrettage could be considered. It is a process which
-3364.57
mechanically induces compressive stresses at the inner bore
which offset the stresses caused by the pressure.
0,40'
Autofrettage of Thick-walled Cylinder
0.3S
As the internal pressure is increased, the inner bore
^ 0.30
stresses exceed the elastic breakdown limit. With further
% 0.26
cr
increases, a plastic/elastic interface at radius b, expands
0) from the inner bore, to the outer surface, until the whole
cylinder is in a state of plastic strain. The pressure required
£ 0.20' to cause plastic strain to radius b is (Faupel and Fisher,
1981):
'^ 0.10

-800 ~« «-!-«
-600 -400 -200 0 200
Margin o f Safety (MPa)
400 600
Ml.) = f 1 - + 2 In (12)

The outer portion of the cylinder, from b to c, undergoes


elastic stresses:
Fig. 5 Relative frequency diagram of margin of safety ( C , ,. = 0, 5,
and 10 percent; Cp = 10 percent)
<"'>•• 7f I c 1 + ( ;T (13)
lation run required about 38 seconds on a 16 MHz 386-sx
IBM compatible computer equipped with a math coproces- .('-
sor.
(0>)e. =
\/3 \ c 1 - 7 (14)
A second aspect to be investigated was the impact of
geometric variation as measured by the standard deviations
for the inner and outer radii. For this set of simulations, the (o-.c) (15)
cylinder radii instances were generated as normal variates
with 0 percent, 5 percent, and 10 percent coefficients of (o-zo). = 0 (16)
variation and mean values of ix^ = 25 mm, and fx.^ = 50 mm.
The yield strengths were generated using the same Weibull The inner portion, from a to b, experiences plastic stresses:
distribution and the pressure variates were generated from a
uniform distribution of 350 MPa mean and 35 MPa standard y
deviation. Each run resulted in 25000 simulated instances of ((^h)p + 1 -F 2 In (17)
the input and output variables. The simulated arrays were
then processed by the statistical subroutine. The results are
shown in Table 4. The margin of safety for C^^ = 0 percent - I - H-2 (18)
is 134.26 MPa, or slightly better than the 1 percent COV case (-.). = ^
from Table 2, where w^ = 134.07 MPa. However, larger radii
variations significantly add to the dispersion of the margin of
safety. Note that for C„^ = 10 percent the standard devia- (oic)/. -I + 2 In (19)
tion for m^ is 152.91 MPa, compared to 95.6 MPa (Cp = 10
percent, C^,,. = 1 percent case from Table 2). The relative (20)
(Oio)n = 0
frequency is graphed in Fig. 5. Note that as the geometric
variations increase, the curve skews to the negative side, and The cylinder becomes fully plastic when b = c. Ultimately it
probabilities of failure increase. bursts at the following pressure (Faupel and Fisher, 1981):
The cylinder reliability for C„^ = 0 percent is abut 91.2
percent as measured by the relative frequency and is shown 25^
in Table 5. By relaxing the geometric tolerances to C^ j. = 1 Pbur In R (21)
73
percent, the reliability decreases to 90.99 percent (from Table
3). If the cylinder were dented, or otherwise significantly out The burst pressure is higher than the pressure required to
of tolerance, resulting in 10 percent radii COVs, the reliabil- plastically strain the entire wall. Therefore, during autofret-
ity would drop to 80.7 percent. tage, as the pressure is removed the material elastically
The design variation simulation results have shown that recovers and a residual compressive stress distribution is
the initial candidate design concepts are unsatisfactory. One obtained. This expansion/relaxation process is called au-
improvement can be attained by increasing the diameter tofrettage. For diameter ratios larger than 2.22 (closed) or
ratio to R = 3; resulting in the reliability approaching 100 2.02 (open), the full effect of autofrettage can be attained
percent. Another change could be to restrict excessive pres- with autofrettage ratios b/c less than 100 percent. Under
sure excursions with a relief valve, or other control. And these conditions, higher b/c values cause an elastic recovery

Journal of Mechanical Design JUNE 1995, Vol. 117/225

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
20 25 30 35 40 45 50 56
Radial Distonce ( m m )

-800
20 25 30 35 40 45 50 55
Radial Distance (mm) • Sh • Sr
• VMo • VMo

Sh • • • * - Sr ••••»••• S z
Fig. 7 Total stress distributions for autofrettaged cyiinder (jy c =" 67
VMo VMo percent; b = 33.49 mm; Sy = 950.3 MPa, p = 350 MPa)
Fig. 6 Residual stress distributions of autofrettaged cyiinder (tV c
= 75 percent; b = 37.50 mm; Sy = 950.3 MPa)

1100 \
that exceeds the inner bore yield strength, and re-yielding \\
1000
occurs. \ .>f
900
Autofrettage produces the following residual stresses in 55 \ X
the elastic portion of the cylinder: \ ,.-
,*
..-•'••" ^

b^^ 'v.. +-''"


...-••• /
1+ >
{'^h)re =
(7 /
^

1- I - + 2 1n- (22)
1 300
y'

R^ - 1 c j a 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1


Autofrettage Ratio b/c

b
{<rr)r 350 MPa • • 500 MPa
^ c
Fig. 8 Optimum autofrettage ratios
b\^ b
1- I- + 2 In - (23)
c/ a i^zc)rp-^\\-A +21n
b\^ b
(o,,=i((^r-
^fi\\c] R^ - 1 1- I- + 2 /« -
1 - -c / + 2 In a- (28)
c/ a R^
(24)
(Oio),p = 0 (29)
{'^zo)re = 0 (25)
The residual stress distributions shown in Fig. 6 result from a
The residual stresses in the plastic portion of the cylinder 75 percent autofrettage (i.e., b/c = 75 percent). Note that
the maximum Von Mises stresses occur at the inner bore
are: where the residual stresses are maximum. Figure 7 shows the
final total stress distributions for a cylinder that has been
5„ ilb previously autofrettaged to h/c = 67 percent and now oper-
(^^>- = ^ l7j -^^^^'% ates at a static pressure of 350 MPa. It is obtained by
superposing Eqs. (l)-(4) with Eqs. (22)-(29).
How much autofrettage is enough? For a given operating
b\^ b pressure, as the depth of autofrettage is increased, the maxi-
1+ 1 - I- +lln- (26) mum Von Mises stress location switches from the inner bore
R'-l
cI a to the autofrettage depth b. Therefore in some situations,
such as when the elastic/plastic interface approaches the
Sy ilb'^ outer wall, yield failure begins near or at the outer surface
^^r)rp=^\\-\ - l + 2 1 n - rather than the inner bore. The optimum autofrettage there-
\/3" ^\c
fore, minimizes the maximum Von Mises stresses, and is a
function of a, b, c, S^, and the operating pressure. A coarse,
1 b modified Newton's optimization routine was used to generate
+ 21n - (27) Fig. 8. It shows that for an operating pressure of 200 MPa,
/ ? 2 - l -(7! the optimum autofrettage results at about b/c = 60 percent;

226/Vol. 117, JUNE 1995 Transactions of the ASME

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Table 6 Simulation results for autofrettaged cylinder; (tyc = 67
percent; p = 350 MPa; a = 25 mm; c = 50 mm; S^ = 950.3 MPa)
blc Min. Avg. Max. Std.Dev. Skew Kurt. ovg = 529 MPa

Sy all 840.46 950.30 1219.28 48.91 0.392 2.859


50.0% 658.97 816.22 990.89 82.20 0.013 1.831
o'a 67.0% 530.71 666.07 870.47 44.99 0.026 2.683
80.0% 767.97 823.51 893,08 24.40 0.057 1.986
50.0% -119.95 134.07 502.59 95.61 0.045 2.351
nig 67.0% 83.45 329.06 559.89 62.80 0.067 2.729
80.0% 116.14 285.26 511.24 54.63 0.207 2.810

Table 7 Reliability estimates for autofrettage cylinder

Q B ^1 e - » T " ^ ' i 1 1— -1 1 1 T=^^


Log- Relative Time so 100 150 200 250 300 350 400 450 500 550
b/c Normal Normal Frequency (S) Margin of Safety (MPa)
50.0% 0.9196 0.9163 0.9099 38.12 Fig. 9 Relative frequency for autofrettaged cylinder
67.0% 1.0000 1.0000 1.0000 52.78
80.0% 1.0000 1.0000 1.0000 52.79

The margin of safety relative frequency for the optimally


for 500 MPa, the optimum solution results at about b/c = 70 autofrettaged cylinder is shown in Fig. 9. Its almost symmet-
percent and for 350 MPa, the optimum is about b/c = 70 ric appearance reflects the 0.067 coefficient of skew and the
percent. The 350 MPa case was further optimized using a almost "normal" peakedness is depicted by the 2.729 coeffi-
higher accuracy search resulting in b/c = 61.0 percent and cient of kurtosis versus 3.0 for a normal distribution.
maximum Von Mises stress equal to 609.4 MPa.
The design example is evolving. However, will an autofret-
taged cylinder provide the reliability we require? The next Summary
section discusses the application of the design variation simu- The design of thick-walled cylinders subjected to large,
lation method to an autofrettaged thick-walled cylinder. static internal pressures, includes considerations of geometry,
loading and material properties. Naturally occurring varia-
tions of these items, can be evaluated using the design
Simulation of Autofrettaged Cylinder variation simulation method, in a reasonable amount of time
A computer program was written to calculate the residual on a personal computer. In addition, autofrettaged thick-
stress distribution as a function of the autofrettage radius b, walled cylinders exhibit an optimum autofrettage ratio which
according to Eqs. (22)-(29). Residual stress distributions can be determined using a conventional optimization tech-
were determined for b/c — 67 percent. Then elastic stresses nique. Also, as shown in the example design, an autofret-
as a function of operating pressure were simulated and taged cylinder may provide reliable and safe operation.
superposed to the residual stresses. The internal pressure
was simulated from a uniform distribution with mean and
standard deviation P ~ U(350,35) MPa. The inner and outer References
radii were simulated as normally distributed variables a ~ Balling, R. J., Free, J. C , and Parkinson, A. R., 1986, "Consideration
N(25,0.25) mm, and c ~ N(50,0.5) mm, respectively. And the of Worst-Case Manufacturing Tolerances in Design Optimization," ASME
material yield strength was simulated as a WeibuU distribu- JOURNAL OF MECHANISMS, TRANSMISSIONS, AND AUTOMATION IN DESIGN,
tion with mean and standard deviation S ~ W(950.3, 48.9) Vol. 108, No. 4, Dec.
MPa. Crawford, R., and Rao, S. S., 1987, "Reliability Analysis of Function
Generating Mechanisms Through Monte Carlo Simulation," ASME Ad-
The smallest autofrettage ratio is 50 percent which obtains vances in Design Automation, Vol. 2, pp. 197-202.
an inner bore radius of 25 mm. Similarly, the largest amount Doepker, P. E., and Nies, D., 1989, "Designing Brake Components
of autofrettage corresponds to the outer radius 50 mm which using Variation Simulation Modelling," Failure and Reliability 1989,
obtains a 100 percent autofrettage ratio. Three cases of ASME, DE-Vol. 16, pp. 131-138, Montreal, Canada.
Early, R., and Thompson, J., 1989, "Variation Simulation Modelling-
autofrettage are examined: 50 percent, 67 percent (determin- Variation Analysis using Monte Carlo Simulation," Failure and Reliability
istic optimum) and 80 percent. The 50 percent values shown 1989, ASME, DE-Vol. 16, pp. 139-144, Montreal, Canada.
in Table 6 correspond to the "no-autofrettage" case. There- Eggert, R. J., 1992, "Robust Fatigue Design for Combined Bending
fore data from Table 2 is repeated here for comparison, The and Steady Torsion," ASME Advances in Design Automation, Vol. 2, pp.
67 percent and 80 percent autofrettage results were obtained, 335-342.
Eggert, R. J., 1991, "Quantifying Design Feasibility Using Probabilistic
using 25000 simulations each, and assumed that the_ autofret- Feasibility Analysis," ASME Advances in Design Automation, Vol. 1, pp.
tage depth b is characterized by a mean value of b and a 5 235-239.
percent COV, normally distributed. Eggert, R. J., and Mayne, R. W., 1990, "Probabilistic Optimal Design
As shown for the optimal autofrettage case, the average using Successive Probability Density Functions," Proceedings of the 16th
margin of safety increases to 329.06 MPa as compared to the ASME Design Automation Conference, Vol. DE-Vol. 23-1, pp. 129-136,
Chicago, IL, September.
non-autofrettaged margin of 134 MPa. Note also that the Eggert, R. J., 1989, "Probabilistic Optimization Using Successive Surro-
minimum margin of safety is now 83.45 MPa, compared to gate Probability Density Functions," Ph.D. Dissertation, Department of
— 119.95 MPa or failure. The 80 percent autofrettaged re- Mechanical & Aerospace Engineering, State University of New York at
sults indicate that the margin of safety, although better than Buffalo.
Faupel, J. H., and Fisher, F. E., 1981, Engineering Design: A Synthesis of
the untreated cylinder, are somewhat worse that the opti- Stress Analysis and Materials Engineering, 2nd ed., John Wiley & Sons,
mum, as should be expected from the optimization results. New York, pp. 234-236, 444-452, 743-753, 790-794.
The simulated reliability increases to 100 percent from Haugen, E. B., 1980, Probabilistic Mechanical Design, Wiley, New York.
90.99 percent as shown in Table 7. A reliability of 100 Kapur, K. C , and Lamberson, L. R., 1977, Reliability in Engineering
percent is expected since Table 6 shows the minimum margin Design, John Wiley & Sons.
Mischke, C. R., 1989a, "Stochastic Methods in Mechanical Design:
to be greater than zero. Note that the required computer Part 1: Property Data and Weibull Parameters," Failure and Reliability
time was about 53 seconds. 1989, ASME, DE-Vol. 16, pp. 1-10, Montreal, Quebec, Canada.

Journal of Mechanical Design JUNE1995, Vol. 117/227

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Mischke, C. R., 1989b, "Stochastic Methods in Mechanical Design: Rao, S. S., and Hati, S. K., 1979, " G a m e Theory Approach in Multicri-
Part 2: Fitting the WeibuU Distribution to the Data," Failure and Reliabil- teria Optimization of Function Generating Mechanisms," ASME JOUR-
ity 1989, ASME, DE-Vol. 16, pp. 11-15, Montreal, Quebec, Canada. NAL OF MECHANICAL DESIGN, Vol. 101, pp. 398-406.
MLschke, C, R., 1989c, "Stochastic Methods in Mechanical Design: Part Rao, S. S., 1978, Optimization: Theory and Applications, Wiley.
3: A Methodology," Failure and Reliability 1989, ASME, DE-Vol. 16, pp. Rao, S. S., and Reddy, C. P., 1977, "Reliability Analysis of Machine
17-20, Montreal, Quebec, Canada. Tool Structures," ASME Journal of Engineering for Industry, Vol. 99, Nov.,
Mischke, C. R., 1989d, "Stochastic Methods in Mechanical Design: pp. 882-888.
Part 4: Applications," Failure and Reliability 1989, ASME, DE-Vol. 16, pp.
21-28, Montreal, Quebec, Canada. Rao, S. S., 1974, " A Probabilistic Approach to the Design of Gear
Trains," International Journal of Machine and Tool Design Research, Vol.
Mischke, C. R., 1987, "Prediction of Stochastic Endurance Strength,"
14, pp. 267-278.Engineering for Industry, Vol. 99, Nov., pp. 882-888.
ASME Paper No. 86-WA/DE-lO, pp. 1-10.
Mischke, C. R., 1986, "Some Guidance of Relating Factor of Safety to Shah, A. R., and Chamis, A. C , 1991, "Simulation of Probabilistic
Risk of Failure," ASME Paper No. 86-WA/DE-22, pp. 1-10. Wind Loads and Building Analysis," Reliability, Stress Analysis and Failure
Mischke, C. R., 1980, Mathematical Model Building, 2nd. Rev. Ed., Iowa Prevention, DE-Vol. 30, pp. 5-14, Miami, FL.
State University Press, Ames. Shigley, J. E., and Mischke, C. R., 1989, Mechanical Engineering Design,
Mischke, C. R., 1970, " A Method of Relating Factor of Safety and McGraw-Hill, 5th Ed.
Reliability," ASME Journal of Engineering for Industry, Vol. 92, pp. Siddall, J. N., 1986, "Probabilistic Modehng Design," ASME JOURNAL
537-542. OF MECHANISMS, TRANSMISSIONS, AND AUTOMATION IN DESIGN, Vol. 108,
Parkinson, A., Pourhassan, N., and Sorenson, C , 1991, "Tolerances pp. 330-335.
and Robustness in Engineering Design Optimization: Further Results," Siddall, J. N., 1984, " A New Approach to Probability in Engineering
Design Productivity International Conf, Hawaii, pp. 1-15. Design and Optimization," ASME JOURNAL OF MECHANISMS, TRANSMIS-
Rao, S. S., 1992, Reliability in Design, McGraw-Hill, Inc., New York. SIONS, AND AUTOMATION IN DESIGN, Vol. 106, pp. 5-10.
Rao, S. S., 1986a, "Automated Optimum Design of Wing Structures: A
Siddall, J. N., 1983, Probabilistic Engineering Design: Principles and
Probabilistic Approach," Computers and Structures, Vol. 24, No. 5, 1986,
Applications, Marcel Dekker, Inc., New York.
pp. 799-808.
Rao, S. S., and Eslampour, H. R., 1986b, "Multistage Multiobjective Smith, C. O., 1984, "Probabilistic Design Criteria for Cylinders &
Optimization of Gearboxes," ASME JOURNAL OF MECHANISMS, TRANSMIS- Spheres under Thermal Stresses," ASME Journal of Vibration, Acoustics,
SIONS, AND AUTOMATION IN DESIGN, Vol. 108, D e c , pp. 461-468. Stress, and Reliability in Design, Vol. 106, Oct., pp. 523-528.
Rao, S. S., 1985, "Optimization of Airplane Wing Structures Under Smith, C. O., 1980, "Design of Ellipsoidal and Toroidal Pressure
Gust Loads," Computers and Structures, Vol. 21, pp. 741-749. Vessels to Probabilistic Criteria," ASME JOURNAL OF MECHANICAL D E -
Rao, S. S., 1984a, "Optimization of Airplane Wing Structures Under SIGN, Vol. 102, Oct., pp. 787-792.
Landing Loads," Computers and Structures, Vol. 19, pp. 849-863. Teng, A., 1992, "Robust Optimal Design for Dynamic and Feedback
Rao, S. S., 1984b, "Multiobjective Optimization in Structural Design Control Systems," Ph.D. Dissertation, Brigham Young University, Provo,
With Uncertain Parameters and Stochastic Processes," AIAA Journal, UT.
Vol. 22, pp. 1670-1678. Thacker, B. H., Harren, S. V., and Millwater, H. R., 1991, "Combined
Rao, S. S., and Das, G., 1984c, "Reliability Based Optimum Design Stress and Resistance Modelling with the Nessus Software System," Relia-
Gear Trains," ASME JOURNAL OF MECHANISMS, TRANSMISSIONS, AND A U - bility, Stress Analysis and Failure Prevention, ASME, DE-Vol. 30, pp.
TOMATION IN DESIGN, Vol. 106, pp. 17-22. 49-54, Miami, FL.
Rao, S. S., 1981, "Reliability-Based Optimization Under Random Vi- Weber, M. A., and Penny, R. K., 1991, "Probabilistic Stress Analysis
bration Environment," Computers and Structures, Vol. 14, pp. 345-355. Methods," Reliability, Stress Analysis and Failure Prevention, ASME, D E -
Rao, S. S., 1979, "Reliability Analysis and Design of Epicyclic Gear Vol. 30, pp. 21-28, Miami, FL.
Trains," A S M E JOURNAL O F MECHANICAL DESIGN, Vol. 101, Oct., pp.
625-632. Zibdeh, H. S., 1990, "Reliability of Thermally Loaded Cylinders,"

228/Vol. 117, JUNE 1995 Transactions of the ASME

Downloaded 11 Nov 2008 to 129.5.224.57. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

You might also like