You are on page 1of 6

Shrinkage of Butt Welds in Stainless

Steel Pipe

Development and application of a mathematical model

to predict shrinkage


ABSTRACT. This paper presents a data and technique for predicting the rings were measured before and after
general method for developing the axial and diametral shrinkage asso- welding, and the shrinkage was
data to predict the shrinkage of cylin- ciated w i t h the girth welding of stain- determined and prepared statis-
drical weldments as a result of girth less steel cylinders. The resulting tically. The raw data for each coupon
welding. information permits the selection of were t h e n averaged and treated
Gas tungsten-arc welding was weld groove dimensions and welding using a multiple regression analysis
used to butt weld stainless steel parameters w h i c h w i l l result in min- technique.
rings. The resulting shrinkage values imum distortion. The design of the experiment was a
were operated on by a multiple re- Weldment configurations that fall full factorial of three variables at t w o
gression analysis technique to pro- w i t h i n the boundary conditions of this and three levels. One w e l d was made
duce an equation predicting diametral experiment include cylinders and at each of the 18 thickness/groove
and axial shrinkage. spheres 4 in. OD by V4 and Vz in. wall d e s i g n / h e a t input combinations (2 x
thickness. However, the value of this 3 x 3 = 18). In addition, six replicate
Introduction welds w e r e made. The test plan is
study is in the method of applying
The many variables associated statistical techniques to predict axial shown in Fig. 1.
w i t h welding make it generally diffi- and diametral shrinkage of girth
cult to predict the distortion caused by Independent Variables
welds in cylinders.
localized heating and cooling. W e l d - Certain parameters of weld groove The independent variables of this
ing parameters, type and history of size and heat input w e r e selected for study were:
materials, and configuration are each of t w o thicknesses of material. X v material thickness: Vi, V2 in.
some of t h e variables that signif- These independent variables w e r e X 2 , groove w i d t h : 0, 0.075, 0.1 5 0 in.
icantly affect weldment distortion. compared w i t h corresponding shrink- X 3 , heat input: 1 7 . 7 2 , 2 3 . 2 6 ,
For weldments of circular cross sec- ages to determine the likelihood of k J / i n . of weld per pass
tions such as pipe, tubing, and relationships, using Student's t test. The method of measuring X , and
spheres, a knowledge of predicted Using the techniques described X 2 is given under the heading,
axial (end-to-end) and diametral herein, groove w i d t h w a s found di- "Material."
shrinkage is essential for accurate rectly related to shrinkage and heat Note that the data used in this
design and fabrication. The designer input inversely related, Shrinkage study for heat input are in joules per
should be aware of the variables that was minimized by the use of thinner inch of w e l d per pass for the filler
affect distortion and h o w much distor- material, narrower grooves and high- passes, not the total heat input per
tion to expect. er heat input. Quantitative results are joint.
This investigation develops the given. All other factors of this experiment
E. BRANDON, formerly with Rocky Flats were held as constant as possible to
Division, Dow Chemical Company is now Experimental Procedure minimize random error and improve
Senior Welding Engineer, Aerojet Nuclear the validity of the results for the
Company, Idaho Falls, Idaho. Pairs of stainless steel rings were stated conditions.
Paper was presented at the 55th A WS butt welded w i t h zero root opening
Annual Meeting held in Houston during using specific combinations of con- Response Variables
May 6-10, 1974. trolled parameters. The assembled The response variables were:


Y r change in stack height (axial Y = B0 + B , X 1 + B 2 X 2 + B3X3 mensions of the groove were
shrinkage) as a result of the w e l d +
"*" BB44X 1 X 22"'"
+ BB65XX11 X 3 + B 6 X2X 3 maintained w i t h i n 0.005 in. and
filler passes + B7X2= + B8X32 angular dimensions w i t h i n 1 degree.
Y v change in diameter (diametral The filler metal was Type 3 0 8 stain-
shrinkage) as a result of the weld where less steel, 0 . 0 3 0 in. diam.
filler passes Y = calculated shrinkage
The stack height of the coupons B x = constants determined by Welding Procedure
was measured at eight places around regression analysis (B 0 , B,, B 2 , etc.)
the circumference before welding, Xi = w a l l thickness (normalized) The welding procedure is given in
after the root pass, and after the filler X2 = groove w i d t h (normalized) Table 2. The procedure resulted from
passes. The average differences be- X3 = heat input (normalized) preliminary welding to develop suit-
tween the last t w o measurements Different B x values resulted for the able parameters. The root pass
were tabulated as axial shrinkage. t w o types of shrinkage, axial and di- parameters were chosen to produce
The diameter of each of the ametral. complete joint penetration of the
coupons was measured at four places 0.100 in. thick root face. Greater heat
In addition to the calculated shrink-
around the circumference before input w a s required for the B = O
age values, the significant indepen-
welding, after the root pass, and after grooves because of the higher heat
dent variables affecting the t w o re-
the filler passes. The average differ- sinking capacity of the narrow
sponses w e r e determined using Stu-
ences between the last t w o measure- dent's t test. The results of this anal- grooves.
ments w e r e tabulated as diametral ysis readily s h o w the significance of The heat input of the filler passes
shrinkage. the main effects, as well as inter- was chosen such that the low level
The root passes were made at a actions. (17.72 k J / i n . ) was the m i n i m u m heat
heat input necessary to achieve com- The six replicate welds were used that would produce a visually satisfac-
plete penetration. No filler metal was to determine the random error. tory w e l d w h i c h exhibited acceptable
used for the root passes. sidewall fusion, bead contour, and
Note that only the distortion as a Material uniformity. The high level (28.80
result of the filler passes w a s used in Type 3 0 4 stainless steel rings w e r e k J / i n . ) w a s the practical upper limit;
the analysis; the shrinkage resulting machined from a seamless extrusion, greater current caused "burn-
from the root pass is discussed sep- 4 in. OD by 3 in. ID. Mechanical and t h r o u g h " of the B = 0.1 50 joints.
arately under "Discussion." chemical values are given in Table 1.
For purposes of analysis, a quadrat- The rings w e r e machined to the Experimental Results
ic model w a s used to describe the final dimensions s h o w n in Fig. 2.
data. The coefficients were calculated After machining, the coupons were Tabulation of Data
by a multiple regression analysis inspected on a shadowgraph to as-
technique (Ref. 1). By this technique, sure adequate precision of the w e l d The values of the independent vari-
the main and quadratic effects groove d i m e n s i o n s . Linear di- ables and the averages of the corres-
caused by each independent variable ponding response variables are given
could be assessed. The calculated in Table 3. The shrinkage values
shrinkage was t h e n found by applying given in Table 3 are due to the filler
an equation of the form: passes only; shrinkage due to the root
B (Noted) »
2 pass was calculated separately.
1/2 in.-wall
Analysis of Data
•§>/ 44 / 38 1 JNSNM
V S \ N \ M
The data of Table 3 w e r e used to
f determine multiple regression equa-
tions for Y, and Y 2 . The t w o sets of
OS 51 / 30 S *7 / coefficients w h i c h resulted are given
in Table 4.
Groove Width (in. )
As an indicator of degree of correla-
tion between the predicted and actual
1/4 in.-wall thickne s; values, the multiple correlation coeffi-
s/ 41 /
'"X"/ cients w e r e 0.982 for Y, and 0.901
for Y 2 (using nine degrees of

' ~"'/
7JT/ freedom).
The Student's t test was applied to
J9 determine the significance of the vari-
/ 3S /
*/"/ 50 ables in affecting the t w o responses.
0.075 0.150 The results are tabulated in Table 5.
Groove Kidth Ci"-) Fig. 2 — One-haIf of a ring assembly
Of significance here is the magnitude
showing joint design used for both 3 and
Fig. 1 — Test plan showing the three heat 3V2 in. ID rings. For the tightly butted as-
of the t-statistic. The greater the abso-
inputs and three groove width parameters semblies three values of the groove size lute value of the t-statistic, the more
used for each thickness. Numbers in cells parameter B (0, 0.075 and 0.150 in.) were certain that factor is to affect the re-
are weld numbers used for each wall thickness sponse.

Table 1 — Mechanical Properties and Chemical Analysis (wt. %) of Type 3 0 4 Stainless Steel Rings

U.T.S., Y.S., Elong., Hard- C Mn P S Si Ni Cr Mo Cu

ksi ksi % ness, Rb
0.063 1.8 0.027 0.020 0.46 9.52 18.86 0.35 • 0.22
82 34.9 59 84/85

412-s I S E P T E M B E R 1 974
A negative t-statistic for the single
factors indicates an inverse relation- Table 2 — Welding Procedure
ship. A s the welding parameter is i n -
creased, t h e response decreases. A Process Automatic gas tungsten arc
Material Type 304 stainless steel tubing
product of t w o factors indicates an
Configuration Extruded and machined, 2 and 2Vi in. long rings, 4 in. OD
interrelationship. A squared single
*3and3VS in. ID(ViandVS in. wall). Single-U butt joint,
factor simply indicates a second- zero root opening (see Fig. 2)
degree curvilinear relationship be- Position Horizontal rolled
tween the welding parameter and the Filler metal 0.030 in. diam, Type 308 stainless steel, for all except
response. In this case, the positive or root pass
negative sign indicates the direction Electrode Vs in. diam, EWTh-2, truncated to 0.030 in. flat end,
of curvature. 10 deg included angle. Changed for each joint
Groove w i d t h w a s the factor most Torch gas 20 cfh argon
Trailing gas 50 cfh argon, shroud covering about VS the circumference
certain to affect both types of shrink-
No. of passes Determined by heat input and groove size (see below)
age. The relationship was direct, i.e., Preweld clean Wire brush and acetone rinse
an increase in groove w i d t h caused Part speed 6.5 ipm surface speed (11 6 seconds per revolution)
an increase in w e l d m e n t shrinkage. Axial force 50 lb, constant throughout welding
Heat input w a s the second ranking Wire speed See below
significant factor but in an inverse Arc voltage 12 V
relationship w i t h shrinkage. W a l l Arc current per pass:
thickness w a s also a significant Root 220A(B = 0); 115 A(B = 0.075 and 0.150)
factor affecting axial shrinkage but Filler See below:
not diametral shrinkage.
Heat input, Current, Wire speed,
Six replicate w e l d s w e r e made to kJ/in. A ipm
check the reproducibility (random
error) of the experiment. The esti- low 17.72 160 27
mate of random error for axial shrink- med, 23.26 210 65
age w a s 0.0045 and for the diametral high 28.80 260 103
shrinkage w a s 0.001 8.
No. of filler passes:
Graphs of the predicted shrinkage For groove sizes (B in.) of —» 0.075 0.150
values of this experiment are given in Wall thicknesses (in.) of - * Vi Vi V4 VS V* VS
Figs. 3 and 4. Heat input: low 3 10 6 12 8 15
Each of the plots shows shrinkage med. 2 5 4 7 5 10
as a function of heat input for a partic- high 2 3 2 5 3 ,6
ular w a l l thickness. Each line repre-
sents a particular groove width.
W h e r e the lines are relatively hori- Table 3 — Variables of Weldment Shrinkage Study
zontal, shrinkage is insensitive to
heat input. W h e r e the lines have a
substantial slope, shrinkage is strong- Experimental Shrinkage
Thick- Groove Heat
ly affected by heat input. W h e r e the Weld ness, width, input. Axial, Diametral,
lines are widely separated, shrinkage no. in. in. kJ/in. in. in.
is strongly affected by groove w i d t h .
The t-statistic indicated that groove 46 V4 0 17.72 0.049 0.011
48 17.72 .048 .010
width and heat input were signif-
34 23.26 .043 .014
icant. The effects of these parameters 28.80 .049 .009
and the combinations at w h i c h they
are most significant are s h o w n graph- 35 1
/4 0.075 17.72 .105 .023
ically in these plots. 43 23.26 .081 .016
Axial shrinkage (Fig. 3) was 33 28.80 .051 .016
strongly and directly affected by 39 V4 0.150 17.72 .145 .022
groove w i d t h . Shrinkage w a s inverse- 50 17 72 .150 .019
ly related to heat input; this effect 40 23.26 .102 .018
was especially significant at B = 45 28.80 .058 .017
0.075 and 0.150 inch. Shrinkage w a s 52 28.80 .060 .016
slightly greater w i t h 1 /2-inch material. 31 Vl 0 17.72 .080 .014
Diametral shrinkage (Fig. 4) was in- 49 17.72 .089 .018
versely related to heat input. The rela- 36 23.26 .062 .014
tionship between shrinkage and 44 28.80 .047 .012
groove width w a s not linear, although
30 VS 0.075 17.72 .121 .019
the lowest shrinkage occurred w i t h 42 23.26 .096 .016
the narrowest groove (B = O). For B = 38 28.80 .082 .016
0.075 and 0 . 1 5 0 in., the regression
lines are w i t h i n one standard devi- 47 vs 0.150 17.72 .188 .020
ation. The conclusion is that there is 53 17.72 .177 .017
no significant difference in shrinkage 32 23.26 .152 .016
37 28.80 .134 .014
in this range of groove w i d t h . 51 28.80 .131 .016
The values of shrinkage predicted
by the regression equations and the
charts of Fig. 3 and 4 should not be dard deviation of the difference in cor- Diametral: °~A = 0.0011
considered exact. The variability be- responding values. The standard devi- Therefore, 95 % of the observed
tween the predicted and the observed ations were: values would be w i t h i n 0.009 in. of
values can be estimated by the stan- Axial: <rA = 0.0045 the predicted values for axial shrink-


Table 4 — Values of Regression Analysis Table 5 — Significant Factors Contribut-
ing to Responses (Student's t Test Anal-
Regression equation: ysis)(a)
Y = B o + B , X 1 + B 2 X 2 + B3X3 + B 4 X 1 X 2 + B 2 X l X 3 + B 6 X 2 X 3 + B 7 X ; + B X
Axial shrinkage
Regression coefficients: t-
Y, Y2 Variable statistic
Axial Diametral
shrinkage shrinkage x?, groove width 11.809
heat input -7.202
Bo 0.08711108 0.01755555 x„
X , , wall thickness 5.794
B, 0.01488888 0 X?x X 3 -3470
B2 0.03716666 0.002333333 x X2 2.648
B3 -0.02266666 -0.001833333
Diametral shrinkag e
B4 0.008333330 -0.001166667 X , groove width 4.038
BB 0.001666656 0.0001666664 Xa heat input -3.173
Be -0.01337500 -0.0002500005 -2.831
* 2
B, 0.003333347 -0.002833333
B8 0.003333337 0.0001666664 (a) 0.05 level of significance = 2.2622

ndependent variables:
x X2 X3
Material thickness Groove width Heat input
Real, Normal- Real, Normal- Real, Normal- age and w i t h i n 0.002 in. for diametral
in. ized in. ized kj/in. ized shrinkage.
B=0 17.72 -1 The charts s h o w the combinations
VA -1 =0.075 0 23.26 1 of primary variables w h e r e the m i n -
Vi 1 =0.150 1 28.80 1 imum and maximum shrinkages oc-
cur. The regression equation provides
a tool for calculating the amount of
shrinkage. Using these t w o inputs,
1/2-inch wall the predicted m i n i m u m and maxi-
O.20 Hz. thickness 0.020 • 0.075 mum axial and diametral shrinkages
0.150 .^^ 0.150 —.
resulting from welds made w i t h i n the
c n boundary limits of this experiment
0.075 w e r e calculated and a r e given in
M 0.10 . Table 6.
c 0 The axial shrinkage ranged from
1/2-inch vail
thickness 0.039 to 0.189 in. The m i n i m u m axial
0 shrinkage occurred at high heat input
17.72 23 26 28.80
17.72 23.26 28.80 w i t h a narrow groove. Conversely,
Heat Input (kJ/in.) the maximum axial shrinkage oc-
Heat Input (kJ/inO curred at low heat input w i t h a wide
The diametral shrinkage ranged
1/4-inch wall
from 0.010 to 0.021 in. The combina-
thickness tions of parameters resulting in m i n -
imum shrinkage differed, depending
on the thickness of material. For both
thicknesses, m i n i m u m shrinkage oc-
curred at high heat input and max-
1/4-inch wall
thickness imum shrinkage occurred at low heat
input. The net effect of groove w i d t h
.72 23.26 28.80 was quadratic, as confirmed by the t-
17.72 23.26 28.80
statistic of Table 5. Diametral shrink-
Heat Input (kJ/in.)
age was lower at both narrow and
Heat Input (kJ/in.)
wide groove widths.
Fig. 4 — Chart of predicted diametral
Fig. 3 — Chart of predicted axial shrinkage shrinkage

Table 6 — Predicted Minimum and M aximum Shrinkage Values The validity of the regression equa-
tions derived in this experiment w a s
Thick- Groove Heat Resulting tested by w e l d i n g four coupons at
ness, width. input, shrinkage, intermediate parameters. The result-
Direction in. in. kJ/in. in. ing experimental values of shrinkage
Axial Vn. 0 28.80 min. (0.039 w e r e compared w i t h those predicted
Vn 0.150 17.72 max. (0.145) by the regression equations. The stan-
Vi 0 28.80 min. (0.056) dard deviation of the difference be-
Vi 0.150 17.72 max. (0.189) t w e e n the predicted and the experi-
Diametral Vi 0 28.80 min. (0.010) mental values (estimate of error) was
Vi 0.120 17.72 max. (0.021) calculated as a measure of correla-
Vi 0 28.80 min. (0.012) tion. Those values are as follows:
Vi 0.100 17.72 max. (0.020) &(axial)= 0.0029
A (diametral) = 0.0006

414-s I S E P T E M B E R 1 974
The above values can be compared S u m m a r y and C o n c l u s i o n s The validity of the regression equa-
w i t h the standard deviation of the tions derived in this experiment was
raw data from the validity test about tested by welding four coupons at
the regression line (standard error of An experiment was designed and intermediate parameters. The check
estimate). Those values were: conducted to develop the data neces- data w e r e w i t h i n one standard devi-
o- (axial) = 0.0109 sary to quantify the shrinkage result- ation of the predicted values.
o-(diametral) = 0.0020 ing from the girth welding of stain-
The validity test data are w e l l less steel rings. The independent vari-
w i t h i n one standard deviation of the ables were wall thickness of the rings Reference
regression line. Therefore, the equa- (Vi and Vi in.), w e l d groove w i d t h (0, 1. N. R. Draper and H. Smith, Applied
tions are assumed to be valid. 0.075, and 0.150 in. — see Fig. 2), Regression Analysis, John Wiley and
Shrinkage due to the root pass was and welding heat input (1 7.72, 2 3 . 2 6 , Sons, Inc., New York, 1966.
ignored in the preceding analysis. and 28.80 k J / i n ) . The measured re-
The total weldment shrinkage would sponses w e r e axial (end-to-end) Acknowledgments
be calculated by summing the root shrinkage and diametral shrinkage.
I wish to thank my advisor, Dr. S. B.
pass shrinkage and the shrinkage due A multiple regression analysis tech- Thayer, and committee members, Drs. M.
to the filler passes. In this investiga- nique was applied to the experimen- C. Bryson, H. O. Rennant, and F. W. Smith
tion, the root pass caused an average tal data to produce a quadratic equa- for their guidance in the preparation of
axial shrinkage of 0.015 in. and a tion to predict, w i t h i n certain boun- this thesis.
diametral shrinkage of 0.006. Shrink- dary conditions, the t w o types of I am also indebted to Dow Chemical
age due to the filler passes can be shrinkage. U.S.A. and especially to Yvonne Ferris,
determined by reference to Figures 3 The reproducibility of the experi- Lowell Hines, Jerry Kaes, Sue Mette,
and 4 or by the regression equation ment was checked by evaluating the Cindy Morrison, Loren Shuler, and Eldon
given in Table 4. Thus cylinders of Vi Webb for their much appreciated advice
results of six pairs of replicate welds.
and valuable service during this investi-
in. wall and 0 groove w i d t h welded at Good correlation was found; the repli- gation.
28.80 k J / i n . w o u l d show total axial cate values of shrinkage were w i t h i n Work performed under the U.S. Atomic
shrinkage of 0.054 in. and diametral one standard deviation of the pre- Energy Commission Contract AT(29-1)
shrinkage of 0.01 6 in. dicted values. 1106(RFD-2118).

Standard Procedures for

Calibrating Magnetic Instruments
To Measure the Delta Ferrite Content
Of Austenitic Stainless Steel Weld Metal,
AWS A4.274

Ferrite is useful in preventing or minimizing cracking and Assuring

in austenitic stainless steel weld metals. In a few special situations, it
can be detrimental to corrosion resistance and to mechanical properties if
it transforms to sigma phase due to exposure to temperatures above 900 F
(480 C). Within a weld pad, the ferrite content is variable, and it is even
more so from pad to pad or when the welding conditions are changed.
Quantitative measurement of the delta ferrite content of austenitic
stainless steel weld metal is desirable to allow the determination of the
amount of ferrite present. This standard provides instrument calibration
procedures to insure that reasonable reproducibility of such measure-
ments can be obtained from one laboratory to another and from one in-
strument to another. $3.00.
Discounts: 25% to A and B members; 20% to bookstores, public li-
braries and schools; 15% to C and D members.
Send orders to the American Welding Society, 2501 N.W. 7th St.,
Miami, FL 33125. Florida residents add 4% sales tax.


WRC Bulletin
No. 194
May 1974
"Fatigue Behavior of Pressure-Vessel

by J. M. Barsom

The regulations governing the design of pressure vessels are based on expe-
rience gained over many operational years and have evolved, primarily, to
prevent failure under static load conditions. This design philosophy has been suc-
cessful in ensuring adequate service behavior because pressure vessels are not
usually subjected to large numbers of load fluctuations during their expected ser-
vice life. However, the need to effectively utilize materials and to provide the
utmost in safety and reliability has made it imperative to determine the fatigue
behavior of these structures.
The fatigue life of structural components is determined by the initiation of
cracks and their propagation to critical dimensions. This report presents fatigue-
crack-initiation and fatigue-crack-propagation data for pressure-vessel steels oper-
ating in a benign environment and at temperatures below the creep region.
Data obtained by testing pressure vessels and pressure-vessel components,
and the results of surveys of pressure-vessel failures are discussed. It is concluded
that the probability of fatigue failure of properly designed and fabricated
pressure vessels is very low and that the most effective approach to keep this
probability low is to minimize the magnitude of the stress (strain) concentration
factors. This can be accomplished through proper design of details and through
proper fabrication.
This paper was prepared for the Pressure Vessel Research Committee of the
Welding Research Council. The price of WRC Bulletin 194 is $4.50. Orders should
be sent to the Welding Research Council, 345 East 47th Street, New York, N.Y.

WRC Bulletin
No. 196
July 1974

"Electron Beam Welding"

by M. M. Schwartz

The first WRC-sponsored interpretive report on electron-beam welding, published as

WRC Bulletin 100 in 1964, covered the early pioneering stages of the process and the men
who refined theories, described their principles and finally built laboratory equipment. In
addition, the authors covered the materials (steel and aluminum), the limited applications
and the equipment which were available during the process's growing infancy.
This new report, prepared for the Interpretive Reports Committee of the Welding
Research Council, covers the advancements made in understanding the process, new equip-
ment with numerical control, tooling improvements and simplicity, wider use of the process
in no and/or medium vacuum applied to practically all metals and ceramics, and finally
sophisticated and economic applications in a number of industries.
The price of WRC Bulletin 196 is $7.00. Orders should be sent to the Welding Research
Council, United Engineering Center, 345 East 47th Street, New York, N.Y. 10017.

416-s I S E P T E M B E R 1 9 7 4