P. B.

Godbole
Supak Pore
BHEL Corporate R&D Division,
Hyderabad, 500093, India

A New Method for Stress

Categorization on Planes
A new method employing the fitting of a least squares plane to stress distribution obtained from a finite element analysis for evaluating categorized values is presented. The
property of a least squares fit, i.e., the preservation of volume under stress distribution
and the plane of consideration, is used in evaluating various numerical integrals. The
procedure is demonstrated through application to a typical shell nozzle junction.
DOI: 10.1115/1.4001521
Keywords: stress category, least squares fit, membrane and bending stresses

Introduction

The approach of stress categorization provides a systematic

procedure for interpreting stress distribution obtained through a
finite element FE analysis. The first of such an approach dates
back to 1973, when Kroenke 1 put forth the concept and established the procedure to compute the required quantities using
stress results obtained from FE analysis.
Categorization means breaking the given stress distribution into
the average component, the linearly varying component, and the
remaining component. The average component is called membrane M, the linearly varying component is called bending B,
and the remaining component is called peak F. M is compared
with a 2/3 yield stress, M + B is compared with yield, and M + B
+ F is considered for fatigue.
The essence of the philosophy behind resorting to stress categorization is to focus the attention beyond stress at a point. From a
point, Kroenke 1 proposed to consider distribution on a line.
What is naturally obvious is to go beyond a line and consider a
plane. In considering stress at a point, the only unambiguity offered is that there is no second opinion about which point to
consider. The point that shows maximum stress is the obvious
choice. But apart from this simplicity, the stress at a point approach ignores all important considerations regarding the nature
of distribution. The stress distribution could be highly localized
singularity, Fig. 1a, fairly localized stress concentration, Fig.
1b, fairly distributed Fig. 1c, or evenly distributed Fig.
1d. Assessing the severity of these distributions by considering
only the maximum stress reported by the FE analysis is obviously
not correct, and it is natural to conclude that attention must go
beyond the maximum stress alone. Giving due regard to the entire
distribution appears very relevant. The natural choice then is to go
from a point to a line.
If consideration of distribution on a line in the case of plane and
axisymmetric problems is viewed critically, it can be easily appreciated that consideration of a plane is implicit. In both of these
analyses, there is no variation in the third dimension thickness in
the case of plane problems and circumference in the case of axisymmetric problems, and information available on a line implicitly contains information about a plane containing the line. Thus,
even the approach put forth in Ref. 1 has an inbuilt sense of
consideration of a plane.
Once, the need for line or plane is established, there are two
major issues which do not exist in the case of point, which need
to be addressed. The first obvious one is the choice of line or
plane. The second one is that the stresses obtained by the FE

analysis cannot be used directly. In the case of stress at a point, the

numerical value of stress alone is necessary for further interpretation. But in the case of line or plane, the FE results need further
processing. Both of these issues are addressed and being pondered
over through literature 28. There is no definite guideline for
choosing a plane location, orientation, and extent. An analyst has
to make his best judgment in the given situation. The other issue
regarding a computational procedure has been addressed with a
computer program, SOAP, as presented in Ref. 2. The present
paper concerns itself with this, i.e., the computational aspect. A
new method based on fitting a least squares plane to the two
dimensional variation of a stress quantity is presented.

Mathematics of Categorization

The procedure of stress categorization involves mathematical

computations using the particular stress distribution obtained from
the FE analysis. These computations are, chiefly, the evaluation of
various definite integrals.
The plane of a section, where results are to be considered, is the
local x-z plane, and the stress result fx , z is taken as a local y
variable dependent. The plane is bound by a certain curve that is
a function of in plane variables x and z. z = x would represent
this. The stress variation would, in general, be represented by y
= fx , z. In the context of the finite element method, both of these
functions, z = x and y = fx , z, would be described at a discrete
number of points. The various integrals required are presented
below:

x2

I1 =

2x

dxdz = Area of section

1x

x1

x2

I2z =

x1

x2

I2x =

Cgx = I2z/I1,

Cgz = I2x/I1

x2

I3z =

Journal of Pressure Vessel Technology

x2

x1

Copyright 2010 by ASME

first moments of area

zdxdz

1x

x1

I3x =

xdxdz,

1x

2x

x1

Contributed by the Pressure Vessel and Piping Division of ASME for publication
in the JOURNAL OF PRESSURE VESSEL TECHNOLOGY. Manuscript received May 28, 2009;
final manuscript received March 13, 2010; published online October 15, 2010. Assoc.
Editor: William J. Koves.

2x

centroid of area

2x

x2dxdz,

1x

2x

z2dxdz

second moments of area

1x

DECEMBER 2010, Vol. 132 / 061206-1

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

integration. Out of the three, the last one is recommended for its
robustness and mesh insensitivity. The linearized stresses are obtained through the integration of functions representing the distribution of stress components at nodes being considered. The code
recommends the use of only those elements that contribute to a
line under consideration.
Thus, the computations for obtaining categorized stress values
are evaluations of various definite integrals. As the values of various functions to be integrated are available only at a discrete
number of points, the integrations are, without exception, numerical integrations.
In the case of categorization on a line, various integrals result
into area integrals. In the case of planes, these integrals lead to
volume integrals. This makes computations more complicated in
the case of planes. The necessity of going from a point to a line
and from a line to a plane is dictated by the physics of the situation, and the price to be paid for going close to reality is in terms
of having to deal with complicated mathematics. With robust numerical integration schemes available the basis of computations
of finite elements, the computations may not be as complicated as
they appear when put forth in closed forms as above.
Fig. 1

ad Typical stress distributions

x2

I3xz =

Principal moments of inertia:

I3x + I3z

tan 2 = 2I3xz/I3x I3z

x2

I4 =

x1

xzdxdz

1x

x1

Iu,v =

2x

I3x I3z
2

2
2
+ I3xz

orientation of principal axes

f 2x

fx,zdxdz

volume under stress distribution

f 1x

x2

I5z =

x1

x2

I5x =

x1

f 2x

xfx,zdxdz,

f 1x

f 2x

zfx,zdxdz

f 1x

moments of volume of stress distribution

Moments about the principal axes:
M u = I5x cos + I5z sin

Critical Review of Computational Procedures

A method of obtaining categorized values from the FE results is

given in Ref. 2. The computational procedure adopted here is the
Gaussian quadrature technique. A concept of cutting plane and
stress classification plane SCP is put forth. The stress classification plane is a subset of the cutting plane. The Gaussian quadrature is applied to the cutting plane, and hence only those Gaussian
points that fall on the stress classification plane are used for further processing. Thus, the entire FE results available in SCP may
not be used for further processing, and it is quite likely that an
important point may get omitted. Hence, the authors have stated
that near sharp stress changes, the method becomes unstable.
The other important aspect that needs attention is as follows.
These days, FE analyses are carried out by standard packages
available. Hence, the form in which data need to be input for SOAP
and the form in which they are available after analysis may not be
complementary. Nevertheless, this 2 may be the first of its kind,
wherein the actual computational aspect is presented and demonstrated through applications.
Another computational approach that can be used in a limited
sense is available with some packages ANSYS, for example
where the total forces and moments transferred across a section
can be obtained. These can give an average normal stress on a
plane and also two bending stresses about the chosen axes. But
this may not be sufficient to generate the entire information required for the purpose of categorization. Also, the section chosen
must contain a sufficient number of nodes and elements. This
means that a sectioning plan must be available at the time of
creation of the FE model.

M v = I5z cos I5x sin

Membrane stress = I4/I1
Bending stress:
About u =

Mu
v
Iu

About v =

Mv
u
Iv

u and are the coordinates of a point in the principal frame of

reference.
The boiler and pressure vessel code laid down by ASME discusses the computational aspects to achieve the linearized components 9. Three methods are suggested. They are i the stress
integration method, ii the structural stress method based on
nodal forces, and iii the structural stress method based on stress
061206-2 / Vol. 132, DECEMBER 2010

The New Approach

The approach for categorization would obviously mean the

scheme of numerical integration resorted to. The Gaussian quadrature scheme in Ref. 2 or the point by point or node by node
calculations available in some FE softwares are the ones used in
the available literature. A different type of scheme is proposed
here.
The scheme presented here is a logical extension of the numerical method for finding the area under curve fx and a base line
along x 10. The author, there, has shown that the area under the
curve is identical in the engineering sense of the word to that
between the least squares equivalent of fx and the base line.
The logical and rather obvious extension of this important feature of the least squares fit that it preserves the integral is that
the volume formed between a general surface and the base area is
identical to that formed by the equivalent least squares surface and
Transactions of the ASME

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 2 Internal points for calculation of fx , z

the same base area. This postulate is the basis for the new method
being presented here. The proof of the postulate is presented next.

Proof of the Postulate

The mathematical proof for this postulate can be presented in

the similar manner, as given in Ref. 10 for a straight line. But the
complications of the resulting algebra having to deal with the 3
3 matrix and writing an explicit inverse are obvious. Instead of
going for a mathematical proof, an alternate method is resorted to.
As the purpose of this presentation is for engineering applications,
a practical or a numerical proof would suffice.
A base area of a random shape is considered. This area is discretized into some internal points shown in Fig. 2. At each of
these points, ordinate y is calculated using some arbitrary function,
2

Y = 1.5 + 2X + 3Z + X + 1.5Z

With this general surface fx , z and the base area, a volume is

constructed. Using fx , z values and x , z coordinates of points,
the least squares surface a + bx + cz is fitted and values of coefficients a, b, and c are determined using the usual procedure for
solving an overdetermined system of equations. The volume generated using fx , z is shown in Fig. 3a, and the corresponding
volume using a least squares fit is shown in Fig. 3b. The values
of volumes are obtained using solid model attributes. These computations are done for three levels of discretization, i.e., number of
internal points in the base area. The results of these are given in
Table 1. Table 1 shows that the postulation regarding the equality
of volumes is true within the engineering sense.

Application to Categorization

The least squares line 10 fitted to a curve preserves the area

under the curve. This, if interpreted in the context of categorization, has the following meaning. If fx represents stress variation
along a line, then the least squares fit of fx is the membrane
+ bending line, and the ordinate of the least squares line at the
center of the line centroid is the membrane stress. This can be
easily shown for plane problems. The statement is valid for axisymmetric cases also. The only change required is because of the
fact that the centroid of the sectorial section does not fall at the
center of the line. It shifts. Hence, the membrane stress is the
ordinate at the centroid location.
The logical extension of this means that the least squares plane
is the membrane+ bending plane, and the ordinate at the centroid
of the base area is numerically equal to the membrane stress. If xg
and zg are the coordinates of the centroid, then a + bxg + czg is the
membrane stress. The membrane stress is also the average value
of y or fx , z, which is given by the total volume/area. This comparison is shown in Table 2.
The total load represented by the volume under fx , z would, in
general, act eccentric with respect to the centroid of the base area,
Journal of Pressure Vessel Technology

Fig. 3

a Volume with fx , z. b Volume with LSQ plane.

and hence there would be, in addition to direct load, bending

moments about two orthogonal axes. These would result in bending stresses. With the total volume and the offsets between the
volume centroid and the base area centroid, and the principal
properties of the base area being known, bending stresses can be
computed. These are compared with the values obtained using
least squares plane at corner locations shown in Fig. 2. This comparison is shown in Table 3.
Tables 13 clearly demonstrate the use of least squares plane
fitting in obtaining the membrane and membrane+ bending components of the distribution represented by the given fx , z.

Computational Procedure

The concept presented above for obtaining categorized components for a given distribution is condensed into a procedure that is
to be executed in post-processing the results obtained from the FE
analysis. The procedure has the following steps.

Table 1 Comparison of volumes

Volume units
Discretization level

Actual

LSQ

Coarse
Fine
Very fine

0.13128
0.13085
0.13074

0.1270
0.13181
0.13107

Table 2 Comparison of heights i.e., stress

Average value units
Discretization level
Coarse
Fine
Very fine

Volume/area

LSQ

0.6605
0.6583
0.6578

0.6678
0.6631
0.6624

DECEMBER 2010, Vol. 132 / 061206-3

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 4 Comparison of results a test problem

Table 3 Comparison of membrane+ bending stresses

Membrane+ bending
Stress

Point No. in Fig. 2

From first principles

LSQ

1
2
3
4
5
6
7

0.244
0.245
0.583
0.729
1.23
0.694
0.587

0.235
0.213
0.578
0.722
1.247
0.692
0.581

Membrane

1. The plane base area for the results is identified.

2. The stress data along with the coordinates of points with
reference to planelocal xz are stored. Local y is normal to
the plane.
3. The data are used to perform calculations to obtain coefficients a, b, and c of the least squares plane.
4. The centroid of the plane local x , z coordinates is determined.
5. With the values of centroid coordinates and coefficients a, b,
and c, the membrane stress is computed.
6. Using coordinates of points of interest all corner points and
some interior points if desired and coefficients a, b, and c,
membrane+ bending components are determined.
7. The procedure is repeated for three normal and three shear
stresses with reference to the plane.
8. Membrane components corresponding to each stress component are used to compute the membrane of the von Mises
stress.
9. The bending component corresponding to each stress is used
to compute the bending component of the von Mises stress
at each of the required locations.
10. The membrane+ bending component for the von Mises
stress is obtained at each location for each of the individual
stresses.

Examples

The procedure outlined above is applied to sample problems to

demonstrate the utility of this approach in determining the categorized values of the given stress distribution.
The first example chosen is that of a cantilever beam of the
unsymmetrical cross section shown in Fig. 4, subjected to axial

Fig. 4 Cross section of the beam

061206-4 / Vol. 132, DECEMBER 2010

M+B

Location in Fig. 4

From first principles

MPa

Computed
MPa

On the plane
At point 1
At point 2
At point 3
At point 4

1000.0000
2493.6052
1035.0919
5953.6351
2868.1731

999.4644
2492.8620
1034.6600
5952.3730
2866.4310

load and bending moments. This particular problem has known

values applied of membrane and bending components; hence,
the values obtained through the procedure can be easily compared.
Section properties:
Area = 39.5 mm2
Principal moments of inertia:
Iu = 59.556 mm4,

Iv = 303.02 mm4

Loads:
Px = 39,500 N,

Py = 600 N,

Pz = 1000 N

The results are given in Table 4. Table 4 shows that the numerical
accuracy of the procedure is quite adequate, and the overall concepts put forth give expected results.
The second example is of a typical shell nozzle junction. The
FE model is shown in Fig. 5. A typical plane considered for analysis and local axes x-y in plane and z out of plane used to compute stresses are shown in Fig. 6a. This figure also shows the
location of points 18 used for calculations. Stress distribution
corresponding to six stress components three normal and three
shears with respect to the plane is shown in Figs. 6b6g.
Using these values, the categorized values are computed at specified locations shown in Fig. 6a. They are given in Table 5.
Using the results of individual components given in Table 5,
von Mises stresses are computed using a standard relationship.
These stresses are usually used for deciding the acceptance, or
otherwise, of the given distribution. The von Mises stresses for the
chosen plane as computed are given in Table 6.

Physical Significance of Linearized Components

In a general sense, the linearization means obtaining constant,

linearly varying, and remaining parts of given distributions individually, which represent each of the six stress components referred to with respect to a plane under consideration. The reason
behind resorting to this is that the damage potential of each of
these three parts is different. Thus, strictly speaking, these three
quantities are mathematically derived quantities. Only in the case
of stress normal to the plane in question do these three parts,
especially the linearly varying one, have physical significance. It
is from this case that the linearly varying component is termed as

Fig. 5 A typical shell nozzle junction

Transactions of the ASME

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 6 Results at salient points on the chosen plane

Membrane= 83.565 MPa
Point No.
in Fig. 6a

Bending stress
MPa

M+B
MPa

Peak
MPa

Total
MPa

1
2
3
4
5
6
7
8

132.780
132.312
122.478
136.084
155.987
28.504
60.907
71.747

216.345
215.877
206.043
219.649
239.552
112.069
144.472
155.312

57.008
49.497
59.303
177.949
217.129
65.379
46.558
0.23

273.353
265.374
265.346
41.700
22.423
46.690
97.914
155.082

10

Fig. 6 a Local plane, local axes, and location of points. b

g Stress contours corresponding to six stress components.

a bending component. This physical definition does not suit the

other components. But the fact that linearly varying parts whether
they represent physical bending or not should be assessed differently from the average component remains. Hence, breaking all
stress components three normal and three shear under whatever
loading conditions into three parts is justifiable.

Conclusion

A new method of obtaining categorized values of a given stress

distribution is proposed, validated, and demonstrated. The numerical integration required is carried out using least squares plane
fitting for the two dimensional variation of stress data. The complexities of the numerical computations are limited only to the
extent of solving an overdetermined system of algebraic equations. The input required is that of local coordinates of points in a
chosen plane where stress results are available from the FE analysis and the stress values. The entire result available on a plane is
used, and hence the possibility of missing an important input is
avoided.
As the present day analysis is carried out mainly through a
software package, the post-processing required for this categorization would differ from one package to another the form in
which results are available, the conversion of the data into the
format that is required for the above procedure, etc.. A complete
procedure for a particular software ANSYS is developed in the
form of programs.
One significant thing to be noted here is as follows. In general,
the plane considered for analysis would have an odd shape, and
principal axes would be inclined with respect to the centroidal
axes of reference. The computation a + bx + cz x , z coordinates
of point under considerations implicitly covers the alignment
with principal axes and also the use of correct fiber distances for
the computation of the bending stress.

Nomenclature
a, b, c coefficients of the least squares plane
B bending stress linearly varying component
Cgx , Cgz x , z coordinates of the centroid of the base
area
F peak stress
I1 area of the section
I2z first moment of area about the z-axis
I2x first moment of area about the x-axis

Table 5 Categorized results for a chosen plane

Stress
Sx
Sy
Sz normal
Sxy
Syz
Szx

Membrane+ bending
at specified points shown in Fig. 6a MPa

Membrane
MPa

6.460
26.407
98.139
1.328
0.581
0.93

16.43
30.84
237.95
9.63
1.62
3.22

17.61
28.27
236.65
9.61
1.56
3.24

17.67
26.17
225.26
8.98
1.48
3.08

19.07
23.72
45.25
6.14
0.90
0.73

10.99
11.74
69.08
7.85
0.96
1.31

3.03
42.92
84.22
0.11
0.63
0.45

14.50
63.78
72.57
1.06
0.74
0.07

14.89
77.56
147.25
3.11
1.40
0.99

Journal of Pressure Vessel Technology

DECEMBER 2010, Vol. 132 / 061206-5

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

I3z
I3x
I3xz
Iu,v
I4
I5x

I5z
M
M u, M v
P x, P y , P z
S x, S y , S z
Sxy, Sxz, Syz
u, v
u ,

x, y, z

second moment of area about the z-axis

second moment of area about the x-axis
product moment of area
principal moment of inertia
volume under stress distribution
moment of volume of stress distribution about
the x-axis
moment of volume of stress distribution about
the z-axis
membrane stress average component
moments about the principal axes
applied forces along the x, y, and z directions
normal stresses
shear stresses
principal axes
coordinates of a point in the principal frame of
reference
local variables with respect to the plane of
results
orientation angle of the principal axes

References
1 Kroenke, W. C., 1973, Classification of Finite Element Stresses According to

061206-6 / Vol. 132, DECEMBER 2010

3
4

5
6
7
8
9
10

ASME Section. III Stress Categories, ASME Winter Annual Meeting, Detroit,
MI, November, pp. 107140.
Hsu, K. H., and McKinley, D. A., 1990, SOAPA Computer Program for
Classification of Three Dimensional Finite Element Stresses on a Plane,
ASME Pressure Vessel and Piping Conference and Exhibition, Tennessee, PVP
185, pp. 1119.
Hechmer, J. L., and Hollinger, G. L., 1989, Code Evaluation of 3-D Stresses
on a Plane, PVP Am. Soc. Mech. Eng., 161, pp. 3346.
Hechmer, J. L., and Hollinger, G. L., 1987, Three-Dimensional Stress
Criteria-Application of Code Rules, PVP Am. Soc. Mech. Eng., 120, pp.
189195.
Hechmer, J. L., and Hollinger, G. L., 1991, The ASME Code and 3-D Stress
Evaluation, ASME J. Pressure Vessel Technol., 113, pp. 481487.
Hollinger, G. L., and Hechmer, J. L., 1986, Three Dimensional Stress
CriteriaA Weak Link in Vessel Design and Analysis, PVP Am. Soc. Mech.
Eng., 109, pp. 914.
Hechmer, J. L., and Hollinger, G. L., 1988, Considerations in the Calculations
of the Primary Plus Secondary Stress Intensity Range for Code Stress Classification, PVP Am. Soc. Mech. Eng., 136, pp. 1733.
Maekawa, O., Aoki, M., Nishimura, Y., and Nishiguchi, I., 1995, JPVRC
Activities to Establish Stress Classification for 3-D FEA Results, PVP Am.
Soc. Mech. Eng., 313-1, pp. 377392.
ASME Boiler and Pressure Vessel Code, Section VIII, 2007, Linearization of
Stress Results for Stress Classification, Annex 5.A, Division 2Alternative
Rules, The American Society of Mechanical Engineers, New York, NY.
Godbole, P. B., 1989, A Method of Quadrature Using Least Square Straight
Line Fit, Adv. Eng. Software, 113, pp. 156158.

Transactions of the ASME

Downloaded 20 Jan 2011 to 144.177.100.5. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm