Stress Calculation Error Analysis For Incremental Hole-Drilling Residual Stress Measurements

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Stress Calculation Error Analysis

for Incremental Hole-Drilling


G. S. Schajer^
Residual Stress Measurements
E. Altus The incremental hole-drilling method for measuring non-uniform residual stresses
gives stress results that are very sensitive to errors in the measured data. The resulting
Department of Mectianical Engineering, stress errors can easily become large enough to compromise seriously the usefulness
Technion - israei Institute of Tectinology, of the calculated stress results. This paper describes a straightforward method for
Haifa 32000, Israel calculating the stress range that has a specified probability of containing the actual
residual stresses. Knowledge of this range allows informed interpretations of the
stress results to be made. The four measurement error sources considered are: strain
errors, hole depth errors, uniform hole diameter errors, and material constant errors.
Both the Integral and Power Series stress calculation methods are investigated, and
their different responses to measurement errors are described.

Introduction Two methods are commonly used to determine non-uniform


The hole-drilling method [1-4] is a popular technique for residual stresses, the Integral Method and the Power Series
measuring residual stresses near the surface of a specimen. It Method. In preparation for the error sensitivity analysis, this
involves drilling a small hole and using a strain gage rosette of section briefly reviews the two stress calculation methods.
the type shown in Fig. 1 to measure the resulting strain reliefs The Integral Method for calculating non-uniform residual
in the surrounding material. In the most common application, stresses is a direct generalization of the uniform stress calcula-
the residual stresses are assumed to be uniform with depth from tion method. In the uniform stress case, the strain measured by
the specimen surface. The associated stress results calculated a typical element of the strain gage rosette shown in Figure 1
from the measured strain data are stable and repeatable, espe- is
cially when strain data averaging [5] is included.
The hole-drilling method can also evaluate residual stresses + Cmin)
that vary with depth from the specimen surface [ 6 - 1 1 ] . In this
case, an incremental technique is used in which relieved strains
are measured during a series of small hole depth increments. ami)cos2/3 (1)
In contrast to the uniform stress case, non-uniform stress calcu- + -5(a.
lations are extremely sensitive to errors in the measured data
[5, 6, 8, 11]. The resulting stress evaluation errors can easily where
grow to be similar in size to the stresses being measured. This
error sensitivity occurs because of an inherent physical limita- e = measured strain
tion of the hole-drilling method. The strains are measured at o'max, Cmin = priucipal strcsscs
the specimen surface, but the desired non-uniform stresses are /3 = angle to maximum principal stress direction
deep in the interior. St. Venant's Principle indicates that the E = Young's modulus
surface strain response rapidly becomes insensitive to the effects v = Poisson's ratio
of interior stresses existing at increasing distances from the a, b = calibration constants.
measurement surface.
Clearly, the possible presence of large errors can seriously The calibration constants S'and b are almost material indepen-
compromise the usefulness of the calculated stress results. A dent for a wide range of Poisson's ratio [8]. These constants
meaningful interpretation of the stress results can therefore only include the effects of strain averaging over the finite area of the
be made with a knowledge of the probable range of the stress strain gage grid [5^]. Apart from small changes in the numerical
evaluation errors. Without such knowledge, it is impossible to values of a and b, Eq. (1) applies both to a blind hole in a
judge whether a given stress result is real or just a calculation thick plate and to a through hole in a thin plate.
artifact. Similar equations apply for each of the three strain gages in
Although high error sensitivity was observed even in the the rosette, the equations differing only in the assignment of
earliest studies of incremental hole-drilling [ 6 ] , no systematic angle f3. For the uniform stress case, a single set of three rosette
quantitative investigation of the resulting stress errors has yet strain measurements is sufficient to calculate the three principal
been reported. This study considers four major error sources in quantities a,nax, Cmin and^. In this uniform stress calculation, a
hole-drilling measurements, and presents a practical method for relative error in e, a, or b causes similar relative errors in a^^^^
determining the probable range of their influence on calculated and a^in- There is no unusually high error sensitivity.
residual stress values. Non-uniform stress evaluations require additional strain mea-
surements. Such evaluations can be made by incremental hole-
Stress Calculation IVIethods drilling, where strain measurements are made during a series
The effects of measurement errors on calculated residual of hole depth increments. Equation (1) then becomes
stresses depend on the method used to evaluate the stresses.
NEERING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials
' On leave from tlie University of Britisli Columbia, Vancouver, Canada. Division July 17, 1994; revised itianuscript received December 10, 1994. Associ-
Contributed by the Materials Division for Publication in the JOURNAL OF ENGI- ate Technical Editor: K. S. Chan.

120 / Vol. 118, JANUARY 1996 Transactions of the ASME


Copyright 1996 by ASME
Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
Factors Affecting Stress Calculation Accuracy
The matrix character of Eqs. (3) and (4) introduces a sub-
stantial new influence on stress calculation accuracy in addition
to the direct influence of relative errors. In a scalar equation
such as Eq. (1), a relative error in a given quantity, say e,
typically causes a comparable relative error in a calculated quan-
tity, say (Tniax' This feature is not true for a matrix equation. For
example, relative errors in (p ) in Eq. (3) can cause significantly
greater relative errors \n{P]. The maximum ratio of the relative
errors in the solution vector compared with those in the given
vector is described by the condition number, A', of the matrix
[12]. A well-conditioned matrix for experimental work has a
condition number close to unity, and gives solutions with rela-
tive errors similar to the data relative errors. A poorly condi-
tioned matrix has a condition number much greater than unity,
and gives solutions subject to much larger relative errors.
The condition number of a lower triangular matrix of the
type considered here depends on the size of its determinant
relative to its norm. A smaller determinant corresponds to a
Fig. 1 Typical Piole-drilling rosette. Adapted from [2]. higher condition number. In the extreme case, a zero determi-
nant corresponds to a singular matrix, which has infinite error
sensitivity and is unsolvable. For a triangular matrix such as
[a] or [S], the determinant is simply the product of the diagonal
(1 + ^ ) r - n , . elements. Unfortunately, these elements rapidly get smaller as
L l I (O'max + CTniin )}
2E hole depth increases, and can become zero at hole depths equal
to about half of the rosette mean radius [11]. Smaller hole depth
ai)cos2^1 (2)
increments also reduce the size of the diagonal elements. Thus,
^^mio the greater the stress spatial resolution and measurement depth
that are sought, the worse will be the possible stress calculation
where the braces indicate vector quantities and the brackets errors. High spatial resolution and large measurement depth
matrices. The vector {e} represents the set of strains measured cannot be achieved at the same time as high stress calculation
after each hole depth increment, and (a^.^, a mm} the principal accuracy. This characteristic, which derives from St. Venant's
stresses within each of the increments. Matrices [a] and [fe] are Principle, can be considered as a stress measurement analogy
lower triangular. Calculation of the stresses within each depth of the Heisenberg Uncertainty Principle of atomic physics.
increment involves solving matrix equation (2) and similar The Power Series Method achieves a superior stress calcula-
equations for the other two gages in the rosette. The form of tion stability by limiting the spatial resolution to a small number
equation (2) is non-linear because of the cos ip term. However, of power series polynomial terms. The two terms in Eq. (4)
Eqs. (2) for the three rosette gages can be linearized by rewrit- represent the maximum practical limit. If one attempts greater
ing them as three uncoupled matrix equations [11], one involv- stress spatial resolution by increasing the number of terms, the
ing only matrix [a] and the other two only matrix [5]. The first condition number of the resulting larger matrix in Eq. (4) rap-
of these three uncoupled equations is idly increases, and the superior stress calculation stability is
lost.
(1 +v)
m{p] (3)
Measurement Error Sources and Effects
where stress P = (ff;, + a,)!! and strain p = (ej -(- e^)l2. Measurement errors can originate from five main sources:
Equation (3J| and the two other similar uncoupled equations
involving [b] can be solved using standard matrix meth- 1. Strain measurement errors. These include instrumenta-
ods [12]. tion errors, thermally induced apparent strains, and addi-
In the Power Series Method, the residual stress variation with tional residual stresses induced by the hole drilling pro-
depth is assumed to be expressible in the form of a power cess.
series polynomial. In practice, only the first two terms of the 2. Hole depth measurement errors. These include possible
polynomial' can be used effectively. The first term represents errors in measuring hole depth, and any non-flatness of
uniform stresses and the second term stresses varying linearly the bottom of the hole.
with depth from the specimen surface. The coefficients of the 3. Hole diameter measurement errors. These include diame-
two terms are determined by a least-squares procedure [11]. ter measurement errors, tapering of the hole, and devia-
Again, it is convenient to linearize the least-squares equations tion from roundness.
by rewriting them as three uncoupled equations, the first involv- 4. Material constant estimation errors. These include uncer-
ing only a terms and the other two only 5 terms. The first of tainties in the given elastic constants of the specimen
these equations is material.
5. Hole eccentricity errors. These include the effects of pos-
(1 +^) aoUo S oai' [Pol ' Z ap sible eccentricity of the hole from the center of the strain
(4)
Oiao X aittt _ . P i . -Xi/' gage rosette.

where ob and Oi represent the measured strain responses of a The various sources produce errors of different types. Cate-
hole drilled into unit residual stress fields that are respectively gory 1 strain measurement errors only affect the right sides of
uniform with depth and proportional to depth. The solution Eqs. (3) and (4). Error categories 2 - 5 mostly affect the left
vector contains the coefficients of the least-squares best-fit lin- sides.
ear stress profile, P{h) = PQ + P\h, where h is the distance The part of the matrix equation affected by a particular error
from the specimen surface. source determines the nature of the response to that error source.

Journal of Engineering l\^aterials and Technology JANUARY 1996, Vol. 1 1 8 / 1 2 1

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


For example, category 1 errors, affecting the right side, produce Table 1 Example hole-drilling measurements
stress calculation errors that are independent of the stresses that
are present. The effect is an additive one, and can exist even Rosette type = MM 062-RE
Mean rosette radius = 2.57 mm
in the absence of residual stresses. In contrast, category 2 - 5 Young's modulus = 207 GPa
errors, affecting the left side, produce stress calculation errors Poisson's ratio = 0.3
that are proportional to the residual stresses. Here, the effect is Hole diameter = 2.0 mm
a relative one, and can exist only when stresses are present. For
Std. strain error = 3 /.it
simplicity, the two general error types are called here * 'right- Std. hole depth error = 0.01 mm
side" and "left-side" errors. Std. diameter error = 0.02 mm
This study concentrates on first four error sources. Hole ec- Std. modulus error = 2%
centricity effects are not included here because Eqs. ( l ) - ( 4 )
do not explicitly refer to this type of error. Hole eccentricity Hole depth, Strain 1 Strain 2 Strain 3
effects have been studied for the uniform stress case [13, 14], mm ixe tJ.e IJ.e
and have been shown to be significant. It is believed that typical
stress evaluation errors due to hole eccentricity are less signifi- 0.00 0 0 0
cant for non-uniform stress evaluations. The effects are esti- 0.25 -77 -32 13
0.50 -163 -66 30
mated to be 2 - 3 times the size of the (very small) effects of 0.75 -226 -89 46
hole diameter measurement errors. 1.00 -264 -102 58

Stress Calculation Probability Bounds


To simplify this analysis, it is assumed that all measurement surements. This study uses a 90 percent probability level in its
errors have a zero mean, a normal distribution, and are indepen- example calculations. However, the method described is gen-
dent of each other. Further, it is assumed that each measurement eral, and any other probabiUty level can be used instead.
error is linearly related to the resulting errors in the calculated
stress results. This second assumption is always true for "right-
side" errors. However, for "left-side" errors, it is valid only Strain Measurement Errors
when the errors are small. For the sizes of "left side" errors Strain measurement errors are "right-side" errors. Say that
encountered in practice, the linearity assumption is an excellent there is a strain perturbation {8p ] and that the resulting calcu-
approximation. lated stress perturbation is {SP}. Then, for the Integral Method
The error in a calculated stress value is the sum of the stress
errors caused by each of the measurement errors (1 +'^)
[a][P + 6P] = {p + 6p] (7)
(/ ^ Cj [ (5)
Subtracting Eq. (3) gives
where e,- are the various measurement errors, and the index " / "
ranges over the total number of those errors. The factors c, ^^^misp} = {6p] (8)
relating each measurement error to its effect on calculated stress
error directly define the error sensitivity of the stress calculation.
The larger these factors are, the more sensitive is the calculation and in inverted form
to measurement errors. The factors c, depend on hole diameter
and depth, and hole depth increment size. In the case of ' 'left- \5P] = [ar'{6p] (9)
side' ' errors, they also depend on the stresses that are present. (1 + ; . )
This stress dependence creates an interaction with the "right-
where [ a ] " ' is a lower triangular matrix whose principal diago-
side" errors because the "right-side" errors affect the calcu-
nal elements are the recipocals of the corresponding elements of
lated stresses. Strictly, this interaction violates the initial as-
[fl]. The inverse equation (9) is shown here only for conceptual
sumption of independence among the error sources. Fortunately,
clarity. In practice, [6P] is determined from {6p} by solving
the interaction is weak, and can be handled satisfactorily by
Eq, (8) using back-substitution, rather than explicitly forming
evaluating the "left-side" errors after the "right-side" errors,
the matrix inverse required for Eq. (9).
as discussed in the next sections.
If the strain measurement errors are expressed in terms of
In general, measurement errors e, and factors c, can have strains p, then from Eq. (6) the resulting standard deviation in
positive or negative signs. Thus, some cancellation of stress the error in calculated P stress in hole depth increment " / " is
error is likely in Eq. (5). In the present case, where the measure-
ment errors are assumed to have normal distributions and be
independent, the stress errors also have normal distributions Spi E (,7'5)^ (10)
with standard deviation (1 + J^)
In the case where the standard deviations of the strain mea-
s. = ^'Lic,s,r (6)
surement errors are all equal, Sp, depends on the root sum of
the squares of the elements in row " / " of matrix [S]"'. Equa-
where s, are the standard deviations of the individual measure- tions (11) and (12) show the matrices [a] and [a]' for the
ment errors [15]. When many observations of a given measure- example experimental measurements summarized in Table 1,
ment error are made, the standard deviation approximately where a hole is drilled to a depth of 1 mm in four equal incre-
equals the root mean square error. ments. The root sum of squares for each row of [ a ] " ' approxi-
Given that s has a normal distribution, a range of stress error mately doubles from one row to the next, indicating a rapidly
can be specified that has a given probability of containing the increasing strain error sensitivity with hole depth.
actual error. For example, the stress error will lie within a range
of 1.64 standard deviations with a probability of 90 percent. 0448
In this study, the limits of the corresponding calculated stress 0614 -.0374
[a] = (11)
range are called "probabiUty bounds." The objective here is 0691 -.0475 -.0235
to determine these probability bounds for practical stress mea- 0728 -.0513 -.0296 -.0118

122 / Vol. 118, JANUARY 1996 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


pected from St. Venant's Principle). A depth of 0.4 r^ (1.0 mm
here), where r, is the mean radius of the strain gage rosette,
is a practical working limit [11]. At larger depths, there exists
the possibility of having a zero on the diagonal of [a], thus
making the matrix uninvertible.
In the Power Series Method, the calculated stresses errors
are also linearly related to the measured strain errors. Thus, the
same procedure for finding the probability bounds described
above can be used. In the Power Series calculations shown in
Figure 2 (ft), the width of the 90 percent probability bounds also
increases with depth from the surface. However, the probability
bounds do not widen monotonically. Instead, they have an X-
shape, corresponding to a "pivoting" of the calculated stress
0.00 0.25 line about a point approximately one-third within the available
hole depth data. This ' 'pivot point'' does not occur at the mid-
300
point of the hole depth data because the stresses near the speci-
men surface more strongly influence the measured strains than
the stresses farther away. The width of the "neck" approxi-
mately equals the probability bound width that would be calcu-
lated for a uniform stress calculation using the Averaging
I Method [5].
The calculated stress probability bounds are narrower for the
I Power Series Method than for the Integral Method. The data
averaging implicit in Eq. (4) causes this increased stability of
the Power Series Method. However, there is a price to pay in
terms of reduced stress spatial resolution capability. In the ex-
ample case, the Power Series Method was a good choice because
0.00 0.25 0.50 0.75 1.00
the particular residual stress profile actually does vary linearly.
Depth from Surface [mm] If this were not true, for example a U-shaped stress profile, the
Fig. 2 90 percent probability bounds for stresses caiculated from mea-
Power Series Method would be unable to represent the shape
sured strains containing errors witli a standard deviation of S/jte. (a) adequately. The actual residual stress profile could then lie out-
Integrai Method, {b) Power Series Method. side the calculated probability bounds.
The trade-off between stress calculation accuracy and spatial
resolution can be used to adjust the way in which the stress
probability bounds vary with depth. With the Integral Method,
root sum a strategy for reducing error sensitivity remote from the surface
of squares is to give up some spatial resolution by progressively increasing
-22.3 22.3 the sizes of the hole depth increments. This procedure increases
36.7 -26.8 45.4 the size of the elements in the lower rows of matrix [a] relative
[a]-' = (12) to the upper rows, and causes the diagonal elements to have
-8.5 54.1 -42.6 69.4
more uniform sizes. This change also causes the diagonal ele-
-0.5 -19.2 106.4 -84.5 137.2
ments and root sum of row squares of ["]"' to become more
Equations (8) and (9) describe the relationship between the similar in size. An empirical sequence of four hole depth incre-
error in " p " strains and the resulting effects of calculated " p" ments that has uniform a^ probability bounds is 0.13, 0.34, 0.55
stresses. However, in practice, interest centers on the effects of and 1.00 mm. The associated 90 percent probability range of
errors in the measured strains ci, 2 and e^ on the calculated 35 MPa is an average of the individual ranges shown in Fig.
Cartesian stresses a^, a, and r^y Given the linearity of the 2 ( a ) . The corresponding [a] and [o]^' matrices are
measured strain versus calculated stress relationship, these ef-
fects can be evaluated simply by doing conventional hole-dril- -.0197
ling stress calculations using the error strains alone. This proce- -.0299 -.0351
(13)
dure finds the coefficients c, in equation (5). Some economy -.0359 -.0467 -.0312
can be achieved in this calculation by noting that the rosette in L-.0403 -.0536 -.0419 -.0297.
Figure 1 is symmetrical about the gage 2 axial direction. The
root sum
Ox and o-j, stress calculation errors produced by gage 1 strain
errors are the same as theCTJ,and a^ errors produced from gage of squares
3. This feature reduces the number of needed calculations from -50.7 50.7
3n to In, where n is the number of hole depth increments. [a]-' = 43.2 -28.5 51.7 (14)
-6.4 42.8 -32.1 53.9
Figure 2 shows the results of an example calculation for hole -0.3 - 8 . 8 45.2 -33.6 57.0
drilling into a stress profile that varies linearly with depth from
the specimen surface. Table 1 summarizes the associated experi- In equation (14), the root sum of squares of each row of
mental data. In this calculation, the standard deviation of the matrix [a]^' have similar sizes, reflecting the similar width
strain measurement error is 3//e, and all other error sources are stress probability bounds for each depth increment. The root
assumed zero. For the Integral Method calculations in Fig. 2 ( a ) , sums are not exactly equal because both [a] and [5] enter the
the probability bounds for the calculated stresses become expo- calculation of a^^, and the two matrices have slightly different
nentially wider with distance from the specimen surface, ap- spatial characteristics.
proximately doubling with each hole depth increment. Mathe- Adjustment of the hole depth increment size is only effective
matically, this behavior reflects the approximate doubling of for controlling the probability bound width when using the Inte-
the root sum squares of the matrix rows in Eq. (12). Physically, gral Method. For the Power Series Method, the only result is a
it reflects the diminishing capability of the hole-drilling method slight movement of the center of the X-shape towards a greater
to measure residual stresses remote from the surface (as ex- depth. The two stress calculation methods also have different

Journal of Engineering Materials and Technology JANUARY 1996, Vol. 118/123

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


responses to changes in the number of hole depth increments
used. If the number of depth increments is increased by making
each increment smaller, then the width of the stress probabiHty
bounds for the Power Series Method decreases in inverse pro-
portion to the number of increments. The reverse is true for the
Integral Method. The width of the probability bounds increases
in direct proportion to the number of hole depth increments.
Thus, using more hole depth increments improves the stress
calculation accuracy of the Power Series Method, but using
fewer hole depth increments improves the accuracy of the Inte-
gral Method.
The probabiUty bounds in Fig. 2 represent the likely range
of the calculated stress at any particular depth. However, the
graphs do not imply that it is possible for all the calculated 0.75 1.00
stresses to lie entirely along either probability bound. This result
can be seen graphically for the Power Series Method because
the stress probability bounds are curved lines. In contrast, the
calculated stresses must lie on a straight line. For the Integral
Method, the stresses along either probability bound in Fig. 2(a) S.
could be calculated simultaneously only if the columns of [ s ] " '
each has uniform sign, either positive or negative. This would
correspond to a particular sequence of positive and negative
strain errors {Sp}. However, Eq. (12) shows that the columns
of [ a ] " ' do not each have uniform sign. The matrix components
along the upper three diagonals have alternating signs (empiri-
cally true in practice, but not consistently true for the lower
diagonals). The fact that the upper three diagonals also contain
I
the largest values suggests that the most adverse combination 0.00 0,26 0.50 0.75 1.00
of strain errors for the Integral Method occurs when the errors
Deptli from Surface [mm]
for successive hole depth increments have alternating signs.
Then the resulting stress errors also have alternating signs in Fig. 3 90 percent probability bounds for stresses calculated from mea-
successive increments, and jump from one probability bound sured hole depths containing errors with a standard deviation of 0.01
to the other. In contrast, alternating sign strain errors have the mm. (a) Integral Method, (b) Power Series Method.
least influence on Power Series calculations because the calcula-
tion method tends to average out such evenly distributed errors.
Power Series Method, where perturbations in [a] affect both
In the particular example considered here, the stress probabil-
sides of Eq. (4).
ity bounds are fairly wide. This behavior occurs because the
given hole diameter is less than the maximum allowed size, For practical purposes, it is desirable to work in terms of
(hole diameter = 2.0 mm, maximum allowed diameter = 2.5 measured strains ej, 62 and 3 and Cartesian stresses a^, Oy and
mm). Since strain sensitivity is proportional to the square of r,j, rather than p and P. Equations similar to (15) and (16)
the hole diameter [5], the largest possible hole diameter is apply, and no new concepts are involved. The resulting propor-
desirable. In this example, a maximum size hole would give tionality constants between the hole depth errors and calculated
1.5 times greater strain sensitivity, and 1.5 times narrower stress stress errors give a set of constants c,- which can be used with
probability bounds for strain-induced errors. Eqs. (5) and (6) to determine calculated stress probability
bounds. To be conservative, the constants c, are determined
using negative hole depth errors (shallower holes) because this
Hole Depth Measurement Errors choice gives slightly larger values.
Figure 3 shows the stress calculation probability bounds cal-
Hole depth measurement errors are "left-side" errors. Say culated using the example data in Table 1 for hole depth mea-
that there is a perturbation [6a] in equation (3) and that the surement errors with standard deviation 0.01 mm. To simplify
resulting calculated stress perturbation is [SP], Then, for the interpretation, all other measurement errors, including strains,
Integral Method are assumed zero. The calculated stress probability bounds for
each calculation method have similar shapes to those in Fig. 2.
(I + u)
[a + 6a][P + SP] = {p] (15) However, the probability bound width does not grow as rapidly
with increasing depth from the specimen surface. This behavior
occurs because hole depth errors change the diagonal elements
where stress perturbations {6P} have different numerical values of Eq. (11) in an approximately proportional manner. The
than in equation (7). The stress perturbation {6P ] can be found smaller elements lower down are not disproportionally changed.
by subtracting the inverse of Eq. (3) from the inverse of equa- In this particular example, the probability bound width is also
tion (15). reduced by the lower stress levels remote from the surface.

\8P\ = [[s-+5fl]-' - lar'\{p\ (16) Hole Diameter and Material Constant Errors
(1 ^v)
In this analysis, the hole is assumed to have a uniform cross-
Unfortunately, the relationship between {dP) and \biX\ is not section at all depths from the specimen surface. The diameter
exactly linear. A stress perturbation of slightly different size measurement error refers to the uncertainty in measuring that
occurs depending on whether a hole depth perturbation is posi- one uniform diameter. For mechanical hole-drilling using a den-
tive or negative (deeper or shallower). However, for small tal burr or small milling cutter, diameter uniformity is a reason-
perturbations \ba\, the effect on {bP\ is approximately linear. able assumption. Hole diameter errors are of the "left-side"
To simplify the analysis, it is assumed here that the relationship type because they perturb [S] in Eq. (3). As with hole depth
actually is linear. The linearity assumption also applies to the errors, the relationship between the hole diameter errors and

124 / Vol. 118, JANUARY 1996 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


A hole eccentricity error is expected to have a somewhat
similar effect on the calculated stresses as a hole diameter error.
The eccentricity would be the same at all hole depths, and it
would thus also involve only one coefficient c,. The size of that
coefficient could be 2 - 3 times larger because hole eccentricity
causes opposite sign strain errors in the gages of the hole-
driUing rosette, while hole diameter causes the same sign strain
errors.
g 100

i All Error Sources Present Simultaneously


The example results in Figs. 2 - 4 show the probability bounds
resulting from single types of measurement errors, either ' 'right-
0.2S O.BO side" or "left-side." A complication arises when both error
types exist at the same time. The "left-side" errors and their
associated coefficients Ci in Eqs. (5) and (6) depend on the
stresses that are present. However, in the presence of "right-
side" errors, these stresses are only known within the probabil-
ity bounds shown in Fig. 2. Strictly, the resulting interactions
I among error sources violate the initial assumption of error inde-

I
pendence. However, for the small error sources of interest here,
an approximate calculation can be made by evaluating the "left-
side" coefficients c, twice, once for each stress profile along

I the "right-side" probability bounds in Fig. 2. The resulting


combined "right and left-side" probability bounds are slightly
unsymmetrical because each bound derives from a different
stress profile. The upper and lower combined probability bounds
0.00 0.25 0.50 0.75 1.00 derive from the different upper and lower ' 'right-side'' proba-
Depth from Surface [mm]
bility bounds. The stress dependency of the "left-side" errors
causes the combined probability bound describing higher abso-
Fig. 4 90 percent probability bounds for stresses calcuiated from mea- lute stresses to be further away from the "zero error" result.
sured fioie diameters and Young's moduius containing errors with stan-
dard deviations of 0.02 mm and 2 percent respectively, (a) Integral
When doing the above calculations for the stress profile along
Method, (b) Power Series Method. a "right-side" probability bound, it is necessary to know the
strains corresponding to those stresses. When using the Integral
Method, the strains can be found directly by matrix multiplica-
tion using Eq. (3). When using the Power Series Method, a
the resulting stress errors is not exactly linear. Errors of slightly complication arises when doing the strain calculation. Equation
different sizes occur depending on whether a hole diameter (4) cannot be used directly because the stress probability
error is positive or negative (larger or smaller hole). However, bounds in Fig. 2(b) are not straight lines. Thus, they are not
for the small hole diameter errors encountered in practice, the exactly expressible by a two-term power series polynomial. This
relationship is approximately linear. To simplify the analysis, complication is handled approximately by doing the two strain
it is assumed here that the relationship actually is linear. To be calculations using the two extreme straight lines possible within
conservative, the approximate proportionality constants relating the X-shaped pattern in Fig. 2(b). The combined "right and
hole diameter errors and stress calculation errors (the coeffi- left" probability bound calculations are done using those por-
cients c,) are determined using positive hole diameter errors tions of the straight lines that are closest to the corresponding
(larger holes) because this choice gives slightly larger C/. "right-side" probability bounds.
Young's modulus appears as a factor in the denominator of Figure 5 shows the calculated stress probability bounds for
Eqs. (3) and (4). Thus, errors in Young's modulus are also of the example data in Table I, when all four error sources are
the "left-side" type. A simple way of including the effects of present simultaneously. As with previous graphs. Fig. 5 shows
hole diameter and material constant errors is to combine their that the probability bounds for the Integral Method rapidly get
constants c, with those from the hole depth errors as additional wider with depth from the surface. In this case, one extreme
members of t h e ' 'left side'' group. Figure 4 shows the calculated probability bound stress profile is almost uniform while the
stress probability bounds using the example data in Table 1, for other reduces towards zero at 1 mm depth. Thus, the stress
hole diameter and Young's modulus standard deviations of 0.02 results have to be interpreted with some caution. The probability
mm and 2 percent respectively. To simplify interpretation, all bounds for the Power Series Method are generally much nar-
other measurement errors, including strains and hole depths, are rower, but at the expense of reduced spatial resolution capabil-
assumed zero. In contrast to previous figures, the calculated ity. These stress results are only acceptable if the stress profile
stress probability bounds are quite narrow, indicating that hole actually is approximately linear.
diameter and Young's modulus errors are not major stress error Figure 5 has a very similar shape to Fig. 2, indicating that
sources. In this example, the stress probability bounds get nar- strain measurement error is the dominant error source. Thus, if
rower with depth from the surface, reflecting the smaller stresses narrower stress probability bounds are required, the two primary
at those depths. factors to address are reduced strain measurement error and
Hole diameter and material constant errors have much smaller increased hole-drilling strain sensitivity. Improved experimental
effects on the calculated stresses than do hole depth errors be- technique and larger hole size are the main areas for attention.
cause they are each single items involving just one coefficient
c, in Eqs. (5) and (6). In contrast, the hole depth errors involve
a combination of the coefficients c, for all the various hole Conclusions
depth increments. These coefficients are relatively large because The incremental hole-drilling method for measuring non-uni-
they include a multiplication by the condition numbers of the form residual stresses gives stress results that are very sensitive
corresponding matrices [a]. to the effects of measurement errors. This error sensitivity oc-

Journal of Engineering Materials and Teclinology JANUARY 1996, Vol. 1 1 8 / 1 2 5

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


constant gradient, the Power Series Method with many small
hole depth increments is preferable. However, if the stress pro-
\ file has large changes in stress gradient, the Integral Method
with a few hole depth increments is the better choice.
l_
I
1 1 Acknowledgments
M
1
1
1 This work was supported by grants from the Natural Sciences
and Engineering Research Council of Canada (NSERC) and
from a UBC Killam Faculty Research Fellowship. Dr. Sheldon
(a) Green, Mr. Mohammad Tootoonian, and Mr. Jianping Shen
kindly reviewed the manuscript. Ms. Arian Amirkeyvan skill-
fully prepared the figures.
1 1 1

Postscript
300
After the review and acceptance of this paper, another paper
addressing some similar issues was published in this journal
(D. Vangi, JEMT, Vol. 116, No. 4, 1994. pp. 561-566.) Dr.
I Vangi uses a different approach to discuss the effects on calcu-
lated stresses of strain and hole diameter measurement errors.
He does not include the effects of hole depth or material con-
stant errors. Although the approaches taken by Dr. Vangi and
J the present authors are different, both studies show that the
calculated residual stresses are very sensitive to measurement
errors, and that this error sensitivity rapidly increases with depth
from the surface.
0.00 0.25 0.50 0.75 1,00
Depth from Surface [mm] References
1 ASTM, 1992, "Determining Residual Stresses by the Hole-Drilling Strain-
Fig. 5 90 percent probability bounds for stresses calculated from mea- Gage Method," ASTM Standard E837-92, American Society for Testing and
sured data containing ail four error sources present simultaneously, (a) Materials, Philadelphia.
Integral Method, {b) Power Series Method. 2 Measurements Group, 1993, "Measurement of Residual Stresses by the
Hole-Drilling Strain-Gage Method," Tech Note TN-503-4, Measurements Group,
Inc., Raleigh, NC, 16 pp.
curs because the strains are measured at the specimen surface, 3 Procter, E., and Beaney, E. M., 1987, "The Trepan or Ring Core Method,
Centre-Hole Method, Sach's Method, Blind Hole Methods, Deep Hole Tech-
remote from the desired non-uniform stresses. Most of stresses nique," Advances in Surface Treatments, ed. A. Niku-Lari, Pergammon, Oxford,
are deep in the interior, The resulting stress evaluation errors Vol. 4, pp. 165-198.
can therefore easily become large enough to compromise seri- 4 Rendler, N. J., and Vigness, I., 1966, "Hole-drilling Strain-gage Method
ously the usefulness of the calculated stresses. The error sensi- of Measuring Residual Stresses," Experimental Mechanics, Vol, 6, No. 12, pp,
577-586,
tivity becomes more severe with distance from the measurement
5 Schajer, G, S., 1991, "Strain Data Averaging for the Hole-Drilling
surface. A straightforward method is presented for estimating Method," Experimental Techniques, Vol, 15, No, 2, pp. 25-28,
the probability bounds for calculated stress profiles. These prob- 6 Kelsey, R. A., 1956, "Measuring Non-Uniform Residual Stresses by the
ability bounds allow informed interpretations to be made of the Hole Drilling Method," Proceedings SESA, Vol. 14, No. 1, pp. 181-194,
calculated stress results. 7 Bijak-Zochowski, M., 1978, "A Semidestructive Method of Measuring
Residual Stresses," VDI-Berichte, Vol, 313, pp. 469-476.
Four major measurement error sources are considered here: 8 Schajer, G. S., 1981, "Application of Finite Element Calculations to Resid-
strain errors, hole depth errors, uniform hole diameter errors, ual Stress Measurements," ASME JOURNAL OF ENGINEERING MATERIALS AND
and material constant estimation errors. In the example calcula- TECHNOLOGY, Vol, 103, No, 2, pp, 157-163,
9 Niku-Lari, A,, Lu, J,, and Flavenot, J, F,, 1985, "Measurement of Residual
tions, strain measurement errors were identified as the major Stress Distribution by the Incremental Hole-Drilling Method," Experimental Me-
error source. Thus, for improved stress calculation accuracy, chanics, Vol. 25, No. 6, pp. 175-185.
the main areas for attention are reduced strain measurement 10 Flaman, M, T,, and Manning, B, H,, 1985, "Determination of Residual
experimental error and increased hole-drilling strain sensitivity, Stress Variation widi Depth by the Hole-Drilling Method," Experimental Mechan-
ics, Vol. 25, No, 9, pp. 205-207,
for example, by using a larger hole diameter.
11 Schajer, G, S,, 1988, "Measurement of Non-Uniform Residual Stresses
Non-uniform hole-drilling residual stress calculations require Using the Hole-Drilling Method," ASME JOURNAL OF ENGINEERING MATERIALS
a compromise to be made between stress spatial resolution and AND TECHNOLOGY, Vol, 110, No, 4, Part I: pp, 338-343, Part II: pp. 344-349,
stress calculation error sensitivity. An improvement in spatial 12 Watkins, D, S., 1991, Fundamentals of Matrix Computations, Wiley, New
York,
resolution causes a deterioration in stress calculation errors, and 13 Ajovalasit, A,, 1979, "Measurement of Residual Stresses by the Hole-
vice versa. Of the two stress calculation methods commonly Drilling Method: Influence of Hole Eccentricity," Journal of Strain Analysis.
used, the Integral Method favors stress spatial resolution, Vol. 14, No. 4, pp. 171-178.
whereas the Power Series Method favors reduced error sensitiv- 14 Wang, J,-Y., 1988, "Measurement of Residual Stress by the Hole-Drilling
ity. Error sensitivity is proportional to the number of hole depth Method: General Stress-Strain Relationship and ifs Solution," Experimental Me-
chanics, Vol. 28, No, 4, pp, 355-358,
increments for the Integral Method, but inversely proportional 15 Peebles, P, Z,, 1993, Probability, Random Variables, and Random Signal
for the Power Series Method. If the stress profile has a relatively Principles, 3rd, ed,, McGraw-Hill, New York,

126 / Vol. 118, JANUARY 1996 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like