A NEW PROCEDURE FOR THE EVALUATION OF RESIDUAL STRESSES BY THE

HOLE DRILLING METHOD BASED ON NEWTON-RAPHSON TECHNIQUE

G. Petrucci, M. Scafidi

Dipartimento di Meccanica, Universit degli Studi di Palermo,
Viale delle Scienze 90128 Palermo,

ABSTRACT

The hole drilling method is one of the most used semi-destructive techniques for residual stress
analysis of mechanical parts. In the presence of non-uniform residual stresses, the stress field can
be determined from the measured relaxed strains using several methods, but the most used is the
so called integral method. This method is characterized by some simplifications that lead to
approximate results especially when the residual stress varies abruptly. In this paper a new
calculation procedure based on the Newton-Raphson method for the determination of zeroes of
functions is presented.

THE HOLE DRILLING METHOD

The hole drilling method is one of the more used semi-destructive techniques for the analysis of
the residual stresses in mechanical components [1-3]. In the case of stresses not uniform in the
thickness, a hole of radius equal to R is drilled in various steps, up to the maximum depth z
M
and
the relaxed deformations are recorded and subsequently elaborated to calculate the residual
stresses. The relationship between stresses and strains is the following
( ) ( )
0
( , ) , 0 , 0
z
M
p z F Z z P Z dZ z z Z z =

, (1)
being
( ) ( ) 2
3 1
+ = z p , ( ) ( ) 2
3 1
+ = z P , (2,3)
where
1
and
3
are the deformations measured by the grids 1 and 3 of the rosette,
1
and
3
are
the stresses acting in the same directions, z is the current depth of the hole, Z is the abscissa
measured from the surface of the component, F(Z,z), defined in the field Zz, is the influence
function that depends on the geometry of the analyzed component, the hole-rosette assembly and
can be determined through numerical simulations by the finite element method or the boundary
method [4]. Typically, the stresses are evaluated up to a maximum depth z
M
equal to half the
value of the middle radius of the rosette R
m
, i.e. z
M
=R
m
/2.
The determination of the function P(z) through eq.(1) constitutes an inverse problem for whose
resolution different approximate methods have been proposed, like integral method [1], the power
series method [3], the spline method [5], procedures based on the application of methods of
solution of the inverse problem [6-7]. In each method, the P(Z) stresses are approximated by a
function P(z) whose parameters are determined in such a way that the differences between the
measured deformations p(z) and the deformation calculated by eq.(1) introducing the P(z)
stresses p(z)
643 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
( ) ( )
0
( , )
z
p z F Z z P Z dZ =

(4)
are minimized in correspondence to a discreet number of properly chosen depth z
j
(that, in
general don't coincide with all the steps in which the drilling of the hole is divided).
The integral method [1-4] is based on the assumption that the stress acting between two levels z
j-1

and z
j
is equivalent to a uniform stress, i.e. a stepped function P(z) is considered. In the power
series method [3] the approximating function P(z) is a polynomial of proper degree, usually not
higher than 3. In the spline method [5] the P(z) function is a spline whose number of polynomials
and degree is assigned. The coefficients of the spline are determined by solving a linear system
given by the integral equation in correspondence of the extremes of the polynomials of the spline.
As also shown in [5], in each of the proposed methods, the p(z) deformations evaluated by eq.(4)
do not coincide with the experimental p(z) deformations, apart from the levels z
j
in
correspondence of which the parameters of the approximating functions have been evaluated, and
a difference between evaluated deformation and the experimental ones defined as
( ) ( ) ( ) e z p z p z = (5)
can be evaluated at each quote z.
Recently methods of solution of the inverse problem that, using eq.(4), try to correct the stresses
P(z) in such a way that the differences (5) are minimized, have been proposed [6-7]. As an
alternative, in this paper, the iterative Newton-Raphson method is used to correct the P(z)
function so that the difference between the p(z) and p(z) deformations in the whole field (5) is
minimized in all the field.

THE PROPOSED METHOD

In the proposed method, the P(Z) function is approximated by a P(Z) function given by n
polynomials P
i
(Z) of j
M
degree, each representing the function in the field z
i1
Zz
i
, with i=1n:
( )
, 1
0
, , 1
M
j
j
i i j i i
j
P Z a Z z Z z i n

=
= =

. (6)
In this case, the a
i,j
coefficients in (6) are the unknown of the problem represented by eq.(1); in
the proposed method their determination is effected by iterative approximations through the
Newton-Raphson (NR) method [8] in the way described in the following.
In practice the proposed method requires the initial choice of the following parameters:
the number of polynomials n,
the maximum degree of the polynomials j
M
,
the coordinates of the extremes of the field of definition of every polynomial z
i
, with
i=0n.
In the following, some practical suggestions for effecting such choices will be furnished.
Introducing in (4) the polynomials defined in (6) in place of P(Z), the following expression for
the deformations p(z) can be obtained:
( )
1 2
1 1
1, 2, ,
0 0 0
0
( , ) ( , ) ... ( , )
n
M M M
n
z z z j j j
j j j
j j n j
j j j
z z
p z F Z z a Z dZ F Z z a Z dZ F Z z a Z dZ

= = =
= + +


(7)
644 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
that can be synthesized as
( ) ( )
, ,
1 0
M
j
n
k j k j
k j
p z a I z
= =
=

, (8)
where
( )
1
,
( , )
k
k
z
j
k j
z
I z F Z z Z dZ

=

, (9)
being
( , ) 0, F Z z Z z = > . (10)
In general, the abscissas z at which the p(z) function is evaluated lie inside the interval of
existence of one of the polynomials (6), i.e. it is z
k1
zz
k
, with z
k
z
M
. Since the integral (4) is
extended from Z=0 to Z=z, the index of the first summation starts, obviously, from the value k=1,
involving in the calculation of p(z) all the polynomials P
k
(Z) for which the condition z
k
z is
verified. The summation (8) is extended up to k=n, for simplicity of notation and implementation.
This would imply that the integration (4) is always extended up to the maximum depth z
M
>z, so
that, in order to avoid to introduce errors, the function F(Z,z) has to be set to zero for Z>z,
although it has no physical meaning in such field.
Considering eq.(8), the function e(z) defined in (5), can be rewritten as,
( ) ( ) ( )
, ,
1 0
M
j
n
k j k j
k j
e z a I z p z
= =
=

. (11)
For each of the n polynomials that constitute the P(Z) function, the following error function,
whose variables are the a
i,j
coefficients, can be defined
( )
1
2
i
i
z
i
z
E e z dz

=

. (12)
The optimal values of the a
i,j
coefficients are those that minimize the E
i
functions and, as
anticipated, they can be obtained by the NR method [8]. In general, the NR method allows the
determination of the zeros of a function through iterative approximations. In particular, in this
case, the searched values of the a
i,j
coefficients are those that make null the derivative of the E
i

function itself with respect to the a
i,j
variables, in correspondence of which the E
i
function
assumes its minimum value.
The NR method allows to obtain the values of the zeros of the function in iterative way, starting
from first attempt values and bringing opportune corrections at every iteration. Introducing the a
i

vectors that contain the a
i,j
coefficients related to the i-th polynomial, the recursive relationship
for the determination of the zeros is the following:
1
i i i
q q
= + a a a , (13)
in which q is the number of the iteration and a
i
is the vector of the corrections to be assigned to
the values of the zeros that can be obtained by the following relationship:
645 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
( )
1

i i i

= a J H , (14)
where J
i
and H
i
are respectively the Jacobian vector and the Hessian matrix of the E
i
functions,
whose components are given by the partial derivatives of the E
i
functions with respect to the a
i,j

coefficients as follows:
,
i i
j
i j
E
J
a

, i=1,2,..i
M
,
2
,
, ,
i i
j s
i j i s
E
H
a a

=

, j=1, 2,..j
M
; s=1, 2,..j
M
. (15,16)
Remembering eq.(12), eq.(15) can be rewritten as
( )
( )
1
,
2
i
i
z
i
j
i j z
e z
J e z dz
a

. (17)
Referring to eq.(11), the derivatives of the e(z) function can be written as
( )
( ) ( ) ( )
, , ,
1 0 , ,
M
j
n
k j k j i j
k j i j i j
e z
a I z p z I z
a a
= =


= =

(18)
and eq.(17) becomes
( ) ( )
1
,
2
i
i
z
i
j i j
z
J e z I z dz

=

, (19)
being the I
i,j
(z) functions defined in eq.(9).
Operating as in the previous case, the elements of the Hessian matrix can be simply obtained as
( ) ( )
1
, , ,
2
i
i
z
i
j s i j i s
z
H I z I z dz

=

. (20)
It is easy to observe that, once the abscissas of the extremes of the fields of definition of the
polynomials z
i
are fixed, the elements of the Hessian matrix are constants.
In conclusion, fixed the number of the polynomials n, the degree of the polynomials j
M
and the
coordinates of the extremes of the fields of definition of the polynomials z
i
, the procedure for the
calculation of the a
i,j
coefficients consists in the followings steps:
1. the I
i,j
(z) functions are evaluated by eq.(9),
2. the elements of the Hessian matrix are evaluated by eq.(20),
3. first attempt values are assigned to the a
i,j
coefficients (they can be simply set to 0),
4. the iterative procedure for the determination of the a
i,j
coefficients for each polynomial P
i
,
is effected; in particular it consists in the following steps:
4.1. the e(z) and E functions are evaluated by eq.(11) and (12); if the value of E or the
variation of E with respect to the previous step is lower than a pre-defined low value
the iteration is stopped,
4.2. the elements of the Jacobian vector are evaluated by eq.(19),
4.3. the correction to be assigned to the a
i,j
coefficients are evaluated by eq.(14),
4.4.the new values of the a
i,j
coefficients are obtained increasing those of the previous step
by eq.(13).

646 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
NUMERICAL ANALYSIS AND DISCUSSION

The effectiveness of the proposed method has been verified considering simulated residual stress
distributions. In particular, in this work:
1. shot peening [6] and linear residual stress fields P(z) on a mechanical component made of steel
(with E=200 GPa, =0.3) have been simulated, considering a ASTM Type A rosette with mean
diameter D
m
=5.13 mm [10] and hole radius D=0.4 D
m
;
2. by eq.(1), the corresponding theoretical strain distribution p(z) has been evaluated using the
influence function reported in [9];
3. the strain distributions p(z) have been used as experimental strain functions to evaluate the
residual stresses by the proposed method and the integral method;
4. simulated strain errors (having standard deviation similar to that of typical experimental errors
on strains) have been added to the theoretical strains p(z); these noisy strains have been used
to evaluate the residual stresses operating in the same way of point 3;
5. the noisy strains have been fitted by a smoothing spline distribution in order to simulate a
typical elaboration of experimental strains; these smoothed strains have been used to evaluate
the residual stresses operating in the same way of point 3.


Fig.1 p(z) theoretical strains relative to the residual stresses of fig.2


Fig.2 Shot peening residual stresses: theoretical P(z) distribution
and P(z) distributions obtained by the proposed method and the
integral method, using theoretical strains

In the figures the non-dimensional variable h=z/R
m
, defined in the field h=0.00.5, has been used
in place of depth z. In fig.1, the theoretical
strains p(z) obtained from the shot peening
residual stresses distribution shown in fig.2 are
shown. In fig.2 the theoretical stress P(z) and the
residual stress P(z) obtained by the proposed
method and the integral method are shown.
Three polynomials (n=3) of third degree (i
M
=3)
have been used for the proposed method; the
three polynomials have been defined in the fields
whose extremes were z
0
/R
m
=0, z
1
/R
m
=0.075,
z
2
/R
m
=0.15, z
3
/R
m
=0.5. Six optimized steps [11]
have been used for the integral method. In this
case, the residual stresses evaluated with the
Fig.3 Noisy strains and smoothing spline strain distribution
647 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
proposed method coincide with the theoretical stresses, while the results obtained by the integral
method, although more than acceptable, present the typical steps.
The noisy strains obtained by adding random strain errors to the theoretical strains and the
smoothing spline are shown in fig.3. To simulate the effect of the measurement procedure, the
random errors were simulated using a 3% standard deviation from the theoretical values and a
rounding of 0.5 .
The residual stresses P(z) evaluated by the proposed method and the integral method introducing
directly the noisy strains are shown in fig.4a. Practically, the central polynomial of the proposed
method presents various undesired oscillations. In fig.4b the results obtained using a P(z)
function constituted by 3 straight lines, that is 3 polynomials of first degree, is shown: in this
case, having reduced the degree of the polynomials, the results appear satisfactory. As shown in
fig.5, fitting the noisy strains by the smoothing spline shown in fig.3, the results become again
good and the P(Z) stress function compares very well with the theoretical values.
a) b)
Fig.4 P(z) function for the shot peening case obtained using noisy strains. a) n=3, i
M
=3; b) n=3, i
M
=1

Comparing the results obtained using theoretical and noisy strains it is possible to observe that,
similarly to other methods, the proposed method presents a sensitivity to the experimental errors.
The sensitivity increases as the n and i
M
parameters increases and it is different in different zones
of the h field. In particular, the sensitivity is more elevated in the field 0.3<h<0.5. Consequently,
it is opportune: a) to limit the number of polynomials, i.e. considering n4, usually n=3, b) to use
a maximum degree of polynomials i
M
=3, c) to use a properly extended field for the last
polynomial, operating in such a way that it is z
n1
<0.4z
M
, usually 0.25z
M
z
n1
0.35z
M
.
As an example, in the following, results obtained
by the proposed method properly reducing the n
and i
M
parameters are shown.
Letting n=1, the P(z) function is approximated by
a single polynomial. The results obtained using a
single 2
nd
degree polynomial are shown in fig.6;
in particular, the results relative to the shot
peening distribution are shown in fig.6a and that
relative to the linear behavior are shown in fig.6b.
In both cases, the obtained approximation is good
enough, a part from the zone near h=0 for the case
of the shot peening. Fig.5 P(z) function for the shot peening case obtained
using smoothed strains
648 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002

a) b)
Fig.6 P(z) function obtained by a single 2
nd
degree polynomial (n=1 e i
M
=2): a) shot peening case, b) linear stresses distribution

A further possibility of the proposed method is to use a stepped P(z) function as in the case of the
integral method, by using n polynomials of i
M
=0 degree. In this case, the use of z
j
values
corresponding to the optimized steps for the integral method [11] is opportune. The results
obtained are shown in fig.7a (shot peening case) and in fig.7b (linear case). It is easy to see that
the results obtained by proposed method are better than those obtained by the integral method.

a) b)
Fig.7 P(z) function obtained by 0 degree polynomials: a) shot peening case, b) linear stresses distribution

CONCLUSIONS

In this paper a new method for the elaboration of the experimental strains for the evaluation of
the non-uniform residual stresses by the hole drilling method has been introduced. It is based on
the approximation of the residual stress distribution by a proper amount of polynomials of proper
degree, whose unknown coefficients are determined by the Newton-Raphson iterative technique,
minimizing the squared difference between the experimental strains the strains evaluated
introducing the measured stress in the integral equation.
The proposed method is of simpler implementation than the spline method [5] and it is more
precise than the integral method.
Various numerical simulations have been performed to verify the method in comparison with the
well known integral method and in this paper results about the shot peening [6] and the linear
stress fields are presented. The comparisons have been performed using both the theoretical strain
649 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002
distributions, noisy strain distributions obtained adding typical random errors to the theoretical
strains and smoothed spline approximations of the noisy strains.
Using theoretical strains, the proposed method gives practically exact results, in the intrinsic limit
of the integral equation.
Using noisy strains, it is possible to obtain better results than the integral method and than the
other existing methods, principally using a stepped function or a single 2
nd
degree polynomial.
Using the smoothing spline approximation of the noisy strains, the results obtained are
comparable with the ones obtained using theoretical strains.
Further developments, not described in this paper, are actually in study in order to improve the
proposed method in the case of noisy strains; in particular they regard the introduction of a
condition of continuity between the adjacent polynomials that constitutes the P(z) stress function.

REFERENCES

[1] Bijak-Zochoski, M., A semi-destructive method of measuring residual stresses, VDI-
Berichte, 1978, 313, 469-476.
[2] Flaman M.T., Manning B. H., Determination of residual-stress variation with depth by the
hole-drilling method, Experimental mechanics, 1985, 25 (3).
[3] Schajer, G.S., Measurement of non-uniform residual stresses using the hole-drilling method.
Part. I and Part.II, J. of Eng. Mat. and Tech., 1988, 110 (4), 38-343.
[4] Schajer G.S., Application of Finite Element Calculations to Residual Stress Measurement, J.
of Eng. Mat. and Tech., 1981, 103 (2), pp.157-163.
[5] Petrucci G., Zuccarello B., A New Method for Non-uniform Residual Stresses Analysis Using
the Hole Drilling Technique, J. of Strain Analysis, 1997, vol.33, n.1, pp.27-37.
[6] Beghini M., Bertini L., Rosellini W., Applicazione di algoritmi genetici nella misura di
autotensioni variabili nello spessore, 1999, XXVIII Convegno nazionale AIAS (in italian).
[7] Schajer G. S., Prime M. B., Use of inverse solutions for residual stress measurements,
Journal of engineering materials and technology, 2006, 128.
[8] Hoffman J. D., Numerical methods for engineers and scientists, Marcel Dekker Inc, New
York, 2001.
[9] Beghini M., Bertini L, Analytical expression of the influence function for accuracy and
versatility improvement in the hole-drilling method, J. of Strain Anal., 2000, 35.
[10] ASTM E837-01 Standard test method for determining residual stresses by the hole-drilling
strain-gage method, 2001, ASTM International.
[11] Zuccarello B., Optimal calculation steps for the evaluation of residual stress by the
incremental hole-drilling method, Experimental mechanics, 1999, 39 (2).

650 Copyright JCPDS-International Centre for Diffraction Data 2009 ISSN 1097-0002