You are on page 1of 7

Residual Stress Measurement

in a Ceramic-Metallic Graded
Material
Michael R. Hill1 This paper presents experimental measurements of the through-thickness distribution of
residual stress in a ceramic-metallic functionally graded material (FGM). It further pre-
Wei-Yan Lin2 sents an error analysis and optimization of the residual stress measurement technique.
Measurements are made in a seven-layered plate with a base of commercially pure tita-
Mechanical and Aeronautical nium and successive layers containing an increasing proportion of titanium-boride,
Engineering Department, reaching 85% titanium-boride in the final layer. The compliance method is employed to
University of California, determine residual stress, where a slot is introduced using wire electric-discharge ma-
Davis, CA 95616 chining and strain release is measured as a function of increasing slot depth. Strain
e-mail: mrhill@ucdavis.edu release measurements are used with a back-calculation scheme, based on finite element
simulation, to provide residual stresses in the FGM. The analysis is complicated by the
variation of material properties in the FGM, but tractable due to the flexibility of the finite
element method. The Monte Carlo approach is used for error analysis and a method is
described for optimization of the functional form assumed for the residual stresses. The
magnitude and variation of the resulting residual stress distributions and several aspects
of the error analyses are discussed. 关DOI: 10.1115/1.1446073兴

Keywords: Residual Stress, Functionally Graded Material, Compliance Method, Error


Analysis

1 Introduction shortcomings of the compliance method are also addressed. First,


a new analysis provides error bounds on estimated residual stress
This paper describes an application of the compliance method
resulting from uncertainty in measured strain. Second, the error
to measure residual stress in a titanium 共Ti兲/titanium-boride 共TiB兲
analysis is used as a basis by which to systematically determine
ceramic-metallic functionally graded material 共FGM兲. The recent
the optimal functional representation of the estimated residual
literature contains many works demonstrating the importance of
stress field, where previously the functional form was apparently
residual stress in the successful processing and performance of
left to subjective judgment 关8,9兴. Finally, a method is described
multi-layered or continuously graded material systems. However, for comparing the error-sensitivity of competing experiment de-
the vast majority of these rely on residual stresses found using signs a priori.
analytical or numerical means 关e.g. 关1–3兴兴. Only a very few stud-
ies have presented measurements of residual stress in FGMs.
Three of these employed diffraction methods at a limited number 2 Material and Geometry
of points to compare with numerical residual stress predictions, The material investigated was obtained in plate form, nominally
and therefore did not obtain spatially refined distributions of stress 150 mm square and 17 mm thick, with one surface mostly ce-
关4 – 6兴. Application of the diffraction methods in these materials ramic, one purely metallic, and the property gradient through-
often leads to large uncertainty in residual stress, possibly due to thickness. This material was prepared by Cercom, Inc. using a
the multi-phase nature of the material 关5兴. Mechanical methods do commercially pure titanium metal base plate 关10兴. Tape cast layers
not have a significant sensitivity to the presence of variable mi- composed of various mixtures of Ti and titanium diboride (TiB2 )
crostructure. One recent study employed a mechanical cutting powders were placed on top of this plate and the assembled lami-
method to find residual stress in a hollow, graded cylinder, but the nate was hot pressed at 1578 K at a pressure of 13.8 MPa. In order
method used was only capable of revealing the linear portion of to facilitate densification at this temperature, a proprietary sinter-
the residual stress and did not provide a detailed stress distribution ing aid containing nickel was added to the starting powders. This
关7兴. In contrast to these previous efforts, the present work provides material created a liquid phase at 1215 K that also catalyzed the
an experimentally determined, detailed profile of residual stress reaction of Ti and TiB2 to form TiB with virtually no residual
through the thickness of a seven-layered functionally graded ma- TiB2 . The resulting FGM was composed of seven layers ranging
terial. from pure Ti on one side to 85 volume% TiB on the other. The
To determine the residual stress profile in a graded material, the composition and nominal thickness of each layer are listed in
compliance method is extended. Although the compliance method Table 1 关11兴. Elastic modulus and Poisson’s ratio were determined
has been previously applied to bilayer systems, the application to for each layer by slicing thin beams from the FGM. Specifically,
a more complex, multi-layered system has not been presented. bilayer beams were made from Layers 1 and 2 and from Layers 6
While developing the extension to a multi-layered system, three and 7. Trilayer beams were cut from Layers 3, 4 and 5. Longitu-
dinal and transverse strain gages were bonded to the beams which
1
Corresponding Author. were then loaded in four-point bending. Composite beam theory
2
Formerly Graduate Research Assistant; Presently Structural Design Engineer, was used with the resulting strain data to determine elastic modu-
C&D Aerospace, Huntington Beach, CA. lus and Poisson’s ratio for each layer as shown in Table 1. Ther-
Contributed by the Materials Division for publication in the JOURNAL OF ENGI-
NEERING MATERIALS AND TECHNOLOGY. Manuscript received by the Materials
mal expansion coefficients in Table 1 are interpolated from values
Division March 15, 2001; revised manuscript received August 22, 2001. Associate found by Nelson and Ezis using a rule of mixtures 关10兴. Further
Editor: G. Newaz. details on this material can be found in references 关10–13兴.

Journal of Engineering Materials and Technology APRIL 2002, Vol. 124 Õ 185
Copyright © 2002 by ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Table 1 Nominal thickness and mechanical properties of each
layer in the TiÕTiB composite

Thickness, CTE
Layer %Ti %TiB mm E, GPa ␯ 共ppm/K兲
1 共85 TiB兲 15 85 4.0 274.3 0.170 8.69
2 21 79 1.6 247.6 0.182 8.84
3 38 62 1.6 193.7 0.216 9.28
4 53 47 1.6 162.2 0.246 9.68
5 68 32 1.6 139.4 0.276 10.1 Fig. 2 Coordinates and strain gage locations for the slotting
6 85 15 1.6 120.1 0.310 10.5 experiments
7 共CP Ti兲 100 0 4.0 106.9 0.340 10.9

Table 2 Total thickness and observed layer thickness for slot-


ting locations

Total Observed Layer Thickness 共% of Total Thickness兲


Thickness,
Slot mm 1 共85 TiB兲 2 3 4 5 6 7 共CP Ti兲
8-1 16.5 27.4 8.91 7.52 8.00 9.70 10.2 29.3
8-2 16.9 27.2 8.88 7.46 7.93 9.64 10.1 28.6
7-1 16.3 27.0 8.77 7.42 7.85 9.57 10.0 29.6
7-3 17.4 28.5 9.25 7.82 8.28 10.1 10.6 27.8

gradients near the start of the cut. Placing this gage in such close
proximity to the slot requires careful trimming of the gage back-
ing material. Slotting is performed with wire EDM while the
Fig. 1 „a… Location of samples in larger FGM plate and „b… sample is submerged in de-ionized water. A waterproofing system
measurement locations shown with exaggerated sample ir- is employed to protect the gages, consisting of two coats, a layer
regularity of acrylic followed by a layer of paraffin wax. For the gage near
the start of the slot, adhesive and coatings are carefully masked to
ensure at least 0.5 mm overlap of each coating, which was found
Residual stresses were estimated in two 6.4 mm samples sliced to be crucial for reliable water resistance and accurate strain mea-
from the parent plate using electric-discharge machining 共EDM兲. surement.
The sample locations were parallel to one 150 mm side and near Strain data are gathered at specified increments of slot depth.
the center of the plate, as shown in Fig. 1. Details indicated in Fig. Data are gathered every 0.13 mm 共0.005 in.兲 of depth for the first
1共b兲 are discussed below. These samples allow measurement of 0.64 mm 共0.025 in.兲, and every 0.51 mm 共0.020 in.兲 thereafter.
the component of residual stress normal to the length of the speci- Transduced strain gage signals are produced by typical commer-
men, and its distribution through the plate thickness. Since the cial equipment, and recorded by hand. Cutting is stopped prior to
plate was fabricated by applying nominally uniform heat and pres- making each measurement to avoid stray voltages produced by the
sure perpendicular to the thickness direction, residual stress in the cutting equipment.
plate is likely to be equi-biaxial. Residual stresses in these Geometric variation of the parent plate led to nonuniform ge-
samples, then, provide a good indication of residual stress in the ometry of the removed samples. Two aspects of the sample geom-
plate, along the axis of the samples. etry illustrated in Fig. 1共b兲 will have bearing on the back-
calculation scheme used to estimate residual stress from measured
3 Experimental Technique strain. These are the taper of the specimen thickness from one end
to the other and the skew of the top surface with respect to the
Residual stresses are estimated by relaxation, using the 共crack兲 specimen depth. Table 2 shows the overall and layer thicknesses at
compliance method 关9兴. The general procedure is to gradually cut each of the slot locations. Total thickness varies due to specimen
through the specimen, measure released strain, and use it to com- taper. Layer thicknesses were observed on a polished sample face
pute residual stress perpendicular to the cut. Strain release is mea- near location 8-1 using a scanning electron microscope in back-
sured as a function of depth of cut using metallic foil gages. scatter mode. Layer thicknesses at the other slot locations were
Cutting is performed using EDM. Calculation of pre-cut residual found by assuming the CP titanium base-layer was of uniform size
stress from gathered strain data is performed using a finite element at all locations, and that the other layers changed size in propor-
based scheme, as explained in the following section. tion with the remaining thickness 关14兴. Both taper and skew of the
Experiments are performed at four locations, two on each re- specimens, as shown in Fig. 1共b兲, may have an influence on the
moved sample. The locations are identified first by the removed computation of residual stress from measured strain. Maximum
sample and then by the slot location. The two removed samples skew of the top surface was found to be 0.28 mm at slot location
are identified as 7 and 8. The two samples were removed from 7-3. For the first few cuts, the skew will affect the amount of
adjacent locations near the center of the parent plate, as shown material removed by EDM and, therefore, the amount of strain
schematically in Fig. 1, and so should contain similar levels of change at the top gage.
residual stress. The slots on each sample are identified as 1, 2, or
3 and are located as shown in Fig. 1共b兲. Results are presented for 4 Analytical Technique
locations 8-1, 8-2, 7-1, and 7-3.
Careful attention is paid to several aspects of the experiments. 4.1 General Scheme. The procedure for computing residual
Slot locations are selected to be at least two plate thicknesses from stress from measured strain data assumes elastic behavior and
the sample edges and from other slots, as indicated in Fig. 1共b兲. At relies on the principle of superposition. For a known residual
each location, gages are applied near the start of the slot and near stress in the material, it is rather simple to compute the strain
the end of the slot, as shown in Fig. 2. The small distance between release that would occur due to cutting. For example, a general-
the slot and the top gage 共Gage 1兲 allows good sensitivity to stress purpose, finite element code can be employed. To find an un-

186 Õ Vol. 124, APRIL 2002 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


known residual stress distribution, one adopts a set of basis func- 4.2 Finite Element Modeling. The forward solutions are
tions to represent the variation of the stress field through- preformed using the finite element method 共FEM兲. The analyses
thickness. Finding the unknown stress distribution is then reduced account for residual stress within the body, changing geometry of
to determining a set of amplitudes of the adopted basis 关9兴. In this the body 共i.e., slot-depth兲, and in the present case the variation of
work, we employ a Legendre polynomial basis. material properties within the body. Often, samples employed for
The unknown basis amplitudes are found by inversion of a the compliance method are rectangular and the modeling is sim-
linear system. The strain change due to slotting depends on the plified by using plane strain, symmetric models 关9兴. Specimens
strain gage location, the cut depth, a, and the distribution of pre- removed from the FGM, however, are irregular and their geomet-
existing 共residual兲 stress ␴ (x), where a and x are defined in Fig. ric complexity will have some bearing on the result. The efficacy
2. Using elastic analysis, a ‘‘forward’’ solution is found 关9兴, where of a two-dimensional, symmetric model was investigated in 关14兴
strain, ␧, is computed for a given cut depth, a⫽a i , and residual by estimating residual stress for the same slot using three different
stress corresponding to a given basis function, ␴ (x)⫽ P j (x). That models for the forward solutions and comparing results. The in-
is, the constant C i j is found fluence of specimen taper 共lack of symmetry about the slot兲 was
found to be negligible by comparing results from plane strain
C i j ⫽␧ 共 a⫽a i , ␴ 共 x 兲 ⫽ P j 共 x 兲兲 (1) symmetric and asymmetric models. The effect of surface skew
This analysis is then repeated for all cut-depths and all members was investigated by comparing results from plane strain and three-
of the adopted basis, providing members of the matrix C i j . If the dimensional symmetric models. This comparison revealed a
actual material response during the experiment is elastic, and the change in estimated residual stress, isolated to a small region near
adopted basis is complete, the measured strain release at cut depth the skewed surface. Because the overall effect on the residual
a i , ␧ im , will be approximately equal to a linear combination of stress distribution is small, and the cost of the three-dimensional
the results obtained from the forward solutions computation large, two-dimensional models are used in this study.
Mesh refinement was also investigated, and the models employed
␧ im ⬇ 兺C i j A j ⫽␧
ˆi (2)
herein are adequately converged such that a doubling of mesh
density in all directions results in a negligible change in the com-
puted residual stress. An example plane strain, symmetric mesh is
where ␧ˆ i is referred to as the fitted strain. The unknown amplitude
shown in Fig. 3.
of each basis function, A j , can then be found by inversion of this
Since elastic behavior is assumed, strain release corresponding
system. Nonsingularity of C i j is assured by using an orthogonal
to cutting is simulated by applying traction to the cut faces 关9兴.
basis, and excluding basis functions that cause zero strain. Once
The traction vector at a given point on the cut surface is found
the unknown amplitudes are determined, the stress existing in the
from the residual stress tensor acting at that point prior to cutting,
body prior to cutting is given by
dotted with the inward normal to the surface. It is assumed that
only normal stresses are present relative to the slot face, so that
␴共 x 兲⫽ 兺 A P 共x兲
j j (3) only the normal traction component is nonzero. This assumption
is appropriate for the FGM because the thermal and plastic defor-
In application of the compliance method, Eq. 共2兲 is solved for the mations giving rise to the residual stresses likely varied only
A j ’s in a least-squares sense because the number of measured through the thickness. This assumption is most appropriate away
strains typically exceeds the number of basis functions required to from the free-edges of the parent plate, where the slots are cut.
fit them. The value of the normal traction at a particular point on the cut
In the present case, the estimated residual stress distribution surface depends on its location and the distribution of stress.
should be required to satisfy equilibrium. This requirement fol- Nodal loads consistent with the element interpolation order are
lows from two features of the present work. First, the residual computed from the spatial variation of the traction vector.
stress state in the sample is assumed to be two-dimensional. Sec- The analyses are repeated for all slot depths and basis func-
ond, the residual stress is estimated across the entire sample. Since tions. To handle all combinations, a single analysis is divided into
the residual stress state gives rise to no external force, equilibrium steps, and each step into increments. The slot depth is changed in
requires the net force and moment caused by ␴ (x) acting over the the analysis by removing elements from the model in subsequent
slot plane to be zero. A major advantage of using the Legendre steps. The polynomial basis order increases with increment num-
polynomial basis for such problems is that equilibrium is identi- ber within each step. In this way, tractions corresponding to each
cally satisfied when the zeroth and first order terms in the basis are basis function are applied to models of the specimen with each
excluded 关8兴. Since we exclude two terms in the basis, a stress slot depth. Displacement results at nodes corresponding to strain
polynomial of order n will have (n⫺1) terms and the estimated gage active grid boundaries are post-processed to provide strain
stress will satisfy equilibrium. 关15兴. Arranging the strains for all steps and increments results in
The approach just described is not limited to the use of a single the compliance matrix, C i j , of Eq. 共1兲. For the case of multiple
strain gage. In fact, the use of multiple gages provides a signifi-
cant benefit by improving the numerical condition of the linear
system in Eq. 共2兲. For two gages, as in the present study, a com-
pliance matrix is developed for each location and combined into a
partitioned system

Ci j⫽ 冋 册
C i1j
C i2j
(4)

with corresponding measured strain vector

␧ im ⫽ 冋 册
␧ im,1
␧ im,2
(5)

where superscripts 1 and 2 represent each gage location. As will


be explained in the discussion, the use of one strain gage near the Fig. 3 Finite element mesh for a plane strain, symmetric
start of the slot and another at the end significantly reduces the model for slotting at location 7-3. Slot depth is increased by
uncertainty in estimated residual stress compared with the use of removing elements in the right-most layer in succession from
only a single gage. top to bottom.

Journal of Engineering Materials and Technology APRIL 2002, Vol. 124 Õ 187

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


strain gages, each sub-matrix in the partitioned system of Eq. 共4兲 sponds to a particular slot-depth. The randomized strain vector,
is formed by a particular query of the solution database. ˜␧> m , is computed by adding an uncertainty vector, u> ␧ , to the mea-
Property variations in the FGM are handled by spatially distrib- sured strains
uting elastic material properties. The material variation compli-
˜␧> m ⫽␧> m ⫹u> ␧ (8)
cates the analysis just described, but the overall approach remains
valid; spatially varying the material properties in the forward so- Each member of u> ␧ is selected from a zero-mean normal popula-
lutions is the only change required to handle the FGM. The spatial tion using a random number generator. The standard deviation of
distribution is handled by specifying properties at each integration this population, s u,␧ , is taken to be larger of the maximum abso-
point in the model, rather than for each element 关16兴. This ap- lute value in the fit-error, or the measurement resolution 共taken to
proach has the advantage of allowing the nodal locations in the be 3 ␮␧兲
mesh to follow the actual slot-depths, rather than the layer inter- s u,␧ ⫽max共 兩 e> ␧ 兩 ,3␮ ␧ 兲 (9)
faces. The material properties given in Table 1 and layer thick-
nesses in Table 2 are used in the analyses. Individual models were where the 兩•兩 operator implies the set of absolute values of the
created for each slot to account for variation of the total specimen components of the vector. Therefore, the size of the strain uncer-
thickness at each location. tainty depends on both the quality of the strain fit and the quality
of the strain measurement. The dependence on the strain fit in-
4.3 Stress Determination. Residual stresses are determined creases the sensitivity of the error analysis to the order of stress
from measured strain by inverting Eq. 共2兲. In the present work, the expansion. With the strain uncertainty population defined, Monte
number of polynomial terms in the stress expansion is approxi- Carlo analysis amounts to repeatedly solving Eq. 共6兲 for a sample
mately ten, while the number of measured strains is approximately of random strain vectors, ˜␧> m . Trial analyses with samples of size
fifty. The system in Eq. 共2兲 is therefore over-determined, and a N⫽102 , 103 , and 104 indicated that a sample size of 103 was
least squares solution is pursued adequate to obtain converged estimates of the stress uncertainty.
The Monte Carlo results were reduced to provide both point-
គ T Cគ 兲 ⫺1 Cគ T 兴 ␧> m
A> ⫽ 关共 C (6)
wise and global uncertainties in stress for a range of stress expan-
In this equation, we adopt a matrix notation, with an under-bar sion order. For a given polynomial order, the Monte Carlo solu-
denoting a matrix and an under-tilde denoting a vector. The term tions provide N⫽103 separate sets of expansion coefficients. Each
in square brackets on the right hand side of this equation is called set of coefficients is used to compute residual stress 共using Eq.
the pseudo-inverse, and provides an approximate solution to Eq. 共3兲兲 at forty-one evenly spaced values of x, x i , across the slot
共2兲 with minimum root mean square error between the measured location. The standard deviation of the N stresses at each of the
and fitted strains 关17兴. forty-one points provides a vector of forty-one values of pointwise
Implicit in the solution of Eq. 共6兲 is an assumed number of stress uncertainty, s ␴ (x i ). The root mean square of all pointwise
terms in the stress expansion. Although the order chosen has an uncertainties then provides a measure of the average stress uncer-
impact on the resulting stress distribution, a strategy for selecting tainty for all values of x, s ␴ 关18兴.
a proper expansion order is lacking from the literature. Two com- The error analysis is repeated for a range of stress expansions to
peting influences of the number of terms on the resulting residual identify the best choice for the expansion order. This optimal or-
stress distribution deserve consideration in selecting the expansion der is taken to be that which minimizes the average stress uncer-
order. An increase in the order of the expansion will: 共1兲 improve tainty, s ␴ . Since s ␴ is minimized, residual stress computed from
the fit to measured strain; and, 共2兲 magnify experimental and lack the optimal stress expansion order is the best overall estimate of
of fit errors. The properly selected order will provide the good fit the unknown residual stress field. Further, pointwise uncertainties
to strain while not magnifying errors too greatly. in this residual stress distribution are given by the values of
s ␴ (x i ).

4.4 Error Analysis and Selection of Expansion


Order. The following is the description of a method for order 5 Results
selection that minimizes uncertainty in the resulting residual Strain release for each of the four slot locations is shown in Fig.
stress. First, Eq. 共6兲 is solved for a range of polynomial expan- 4 and optimal residual stress distributions in Fig. 5. The strain
sions with the highest term from second to twelfth order 共since the release at all locations is similar. However, the top strain release
constant and linear terms are neglected, the number of terms for site 7-1 is smaller than for the other three locations. The bot-
ranges from one to eleven兲. Given the set of coefficients for each tom gage data show that location 7-1 and 8-1 are similar to each
polynomial series 共A> in Eq. 共6兲兲, fitted strain, ˜␧> , is computed from other and unique from sites 7-3 and 8-2, which are also similar to
Eq. 共2兲 and an error between measured and fitted strain is com- each other. The optimal fit to measured strain for location 8-1 is
puted from shown in Fig. 6. Figure 5 shows that the residual stress distribu-
e> ␧ ⫽␧>ˆ ⫺␧> m (7) tions at all slot locations are similar. Consistent with the observa-
tions for released strain, there is a similarity in stress at locations
Two global error measures are then defined from e> ␧ , the average 8-1 and 7-1 and at locations 8-2 and 7-3, with the two pairs
and maximum error. The maximum global error is given by the differing slightly from each other. The results are also in general
maximum absolute value among the elements of e> ␧ , and the av- agreement with the results of a simple, elastic cool-down analysis
erage global error by the root mean square of the elements of e> ␧ as discussed below.
关18兴. As might be expected, these analyses reveal decreasing glo- Results of error analyses for location 8-1 are typical of all slot
bal error in strain fit with increasing number of polynomial terms, locations, and suffice to illustrate the general properties of the
as indicated by either global measure 共examples are given in the strain and stress errors. Global error in the strain fit for location
results section兲. Further, the decrease in global fit-error with in- 8-1 is shown in Fig. 7. This plot indicates that the best strain fit is
crease in number of terms plateaus at some number of terms in the obtained for polynomials of tenth or greater order. Uncertainty in
expansion. While the error in the strain fit is an important param- estimated stress is shown in Fig. 8. Figure 8共a兲 shows the average
eter, the influence of the fit-error on the uncertainty in residual and maximum stress uncertainty versus order for location 8-1.
stress is required to determine an optimal stress expansion order. Both global uncertainties indicate that a tenth order stress polyno-
Error analysis, based on Monte Carlo simulation, is used to find mial is optimal 共i.e., minimizes uncertainty兲. Figure 6 shows
the expansion order which minimizes uncertainty in estimated re- graphically that a tenth order stress expansion provides an excel-
sidual stress. In the error analysis, measured strain is converted to lent fit to the measured strain. Pointwise, average, and maximum
a randomized strain vector, where each row in the vector corre- stress uncertainties for a tenth order polynomial are shown in Fig.

188 Õ Vol. 124, APRIL 2002 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


Fig. 6 Measured strain for location 8-1 and strain fit from a
tenth order stress expansion

Similar analyses at the other three measurement locations lead


to the residual stress distributions shown in Fig. 5. Error analyses
like those described for slot 8-1 indicate eight-and tenth-order
polynomials are optimal for particular locations, as noted in the
plot legend. Error bars are shown in Fig. 5 only for location 8-1,
but all error analyses result in pointwise stress uncertainties which
are very similar to those shown in Fig. 8共b兲.

6 Discussion
6.1 Distribution and Magnitude of Residual Stress. The
trends in strain due to slotting give rise to similar trends in the
estimated residual stresses. Aside from the surface stress, all loca-
tions have similar residual stresses over the first 25% of the plate
thickness (x/t⭐0.25). In the remaining region, x/t⬎0.25, the
Fig. 4 Strain release at the four slot locations, „a… Top gage,
measurements fall into distinct pairs, with results for location 8-1
„b… bottom gage
very similar to those for location 7-1, and with location 8-2 simi-
lar to location 7-3. Since slots 8-1 and 7-1 are the closest to the
free edge of the parent plate, it is possible that the two groups of
8共b兲. The pointwise uncertainty is largest near the start of the slot results are caused by the influence of the free edge of the parent
and decreases with increasing distance from the free-surface. In plate during the thermal process used to form the FGM. Even with
fact, the maximum uncertainty is about four times the average the variability between the two groups of results, the trends for all
uncertainty, and all other pointwise uncertainties are less than four locations are in reasonable agreement with one another.
twice the average. The average uncertainty is 10% of the peak The magnitude of the residual stresses are large enough to af-
residual stress. The estimated, tenth-order residual stress distribu- fect material performance. The average peak residual stress is 40
tion for slot 8-1 is shown in Fig. 5. Error bars on the figure MPa, which is only 11% of the tensile strength of CP-Ti and 4.4%
correspond to the pointwise uncertainties shown in Fig. 8共b兲. of the flexural strength of monolithic TiB 关10兴. However, low
fracture resistance of the largely ceramic, TiB-rich layers means
that the residual stress levels can have a significant impact on the

Fig. 5 Residual stress at all locations, error bars for location Fig. 7 Global error in strain fit versus stress expansion order
8-1 for location 8-1

Journal of Engineering Materials and Technology APRIL 2002, Vol. 124 Õ 189

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


The results above also suggest that the optimal order of stress
expansion might be determined without performing the stress er-
ror analyses. For location 8-1, the Monte Carlo error analysis
indicates that a tenth order expansion is optimal for residual
stress. It is noteworthy that this is also the location where the
global strain-fit error plateaus. In fact, examination of the error
trends for the other three locations indicates that the order which
minimizes global stress uncertainty always coincides with the
strain-fit plateau. The four results in this study show that the exact
location of the plateau can be defined as the order where adding
an additional term to the series expansion results in less than a 3
␮␧ change in the maximum strain-fit error. Identifying the optimal
order without performing Monte Carlo error analysis is a signifi-
cant reduction in computational effort. However, a complete error
analysis still provides the best accounting of the uncertainty in
residual stress.
6.3 Use of One Versus Two Strain Gages. Using two
strain gages, one near the start of the slot and one near the end,
greatly reduces the sensitivity of the compliance method to error.
It is possible to compute residual stress from only the back-face
strain gage 共Gage 2 in Fig. 2兲, and several published studies have
used this procedure 关9兴. However, because this gage is insensitive
to stress released for shallow slot depths, larger uncertainty in
estimated stress will result. The increase in uncertainty can be
quantified by repeating the data reduction and error analysis de-
scribed above, but only using data from the back-face strain gage.
While Fig. 9共a兲 shows that it is possible to fit the single strain
gage data to equal accuracy, Fig. 9共b兲 shows that the same strain-

Fig. 8 Uncertainty in estimated stress, location 8-1, „a… Global,


„b… pointwise

effective fracture toughness of flawed geometries. Recently pub-


lished fracture testing results for this material show that residual
stresses altered the measured fracture loads by 34% 关13兴. There-
fore, residual stresses provide important information for integrity
assessment of structures fabricated from this, and probably most,
ceramic-metallic composites.
The residual stress distribution found experimentally is similar
to the results of a simple, thermo-elastic cool-down analysis. Us-
ing properties in Table 1, a finite element analysis was performed
for a unit temperature change, assuming isotropic thermal expan-
sion and elastic response in each layer. The finite element mesh
was similar to that shown in Fig. 3, and the analysis assumed
plane strain. Scaling the resulting stress distribution by ⫺200 K,
to match the peak level of tensile residual stress found in the
experiments, gives the results shown in Fig. 5. The agreement
with the experimental results is surprising since the analysis ig-
nores the variation of CTE with temperature and the effects of
creep and plasticity. The large difference between the 200 K stress
free temperature used for the thermo-elastic analysis and the 1578
K processing temperature used in fabricating the FGM suggests
significant creep and plasticity occurred during cooling.
6.2 Error Analysis and Order Selection. The error analy-
ses provide the dual benefit of estimating likely uncertainty in
residual stress and allowing a quantitative method for selecting
the stress expansion order. For comparing one order of stress ex-
pansion to another, global uncertainty in stress was used, and the
optimal order identified by the minimum uncertainty. The fact that
the average and maximum uncertainty are minimized for the same
order indicates that the selection process is insensitive to the par- Fig. 9 The influence of including the top gage on „a… strain-fit
ticular global uncertainty estimator. error and „b… stress uncertainty

190 Õ Vol. 124, APRIL 2002 Transactions of the ASME

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use


fit error produces four times the uncertainty in residual stress. This Acknowledgments
demonstrates that two strain gages are preferable for through-
The authors wish to thank Michael B. Prime of Los Alamos
thickness residual stress measurement.
National Laboratory for initial consultations regarding the compli-
Numerical analysis can indicate whether one system of strain
ance method, Glaucio Paulino of the University of Illinois at
gages will be less prone to error than another, even without ex-
Urbana-Champaign for providing the FGM samples, and Ryan
perimental data. Error in strain-fit causes uncertainty in stress
Tong of the University of California, Davis for coding the error
through Eq. 共6兲. The numerical condition of the inverted matrix in
analysis routines.
this equation, C T C, indicates the degree to which measurement or
strain-fit errors will be amplified or attenuated in the estimated
References
stress. Numerical condition can be quantified by computing the
ratio of the largest to smallest eigenvalue of the matrix, a quantity 关1兴 Becker, T. L., Cannon, R. M., and Ritchie, R. O., 2000, ‘‘An approximate
method for residual stress calculation in functionally graded materials,’’ Mech.
known as the condition number 关19兴. A larger condition number Mater., 32, No. 2, pp. 85–97.
indicates a more singular system and a larger potential for uncer- 关2兴 Khor, K. A., and Gu, Y. W., 2000, ‘‘Effects of residual stress on the perfor-
tainty in residual stress for a given error in strain. In the present mance of plasma sprayed functionally graded ZrO2 /NiCoCrAlY coatings,’’
Mater. Sci. Eng., A, 277, No. 1–2, pp. 64 –76.
case, C T C has a condition number of 1281 when two gages are 关3兴 Shabana, Y. M., and Noda, N., 2001, ‘‘Thermo-elasto-plastic stresses in func-
used, and 12640 when only a single gage is used. Consistent with tionally graded materials subjected to thermal loading taking residual stresses
the results of the previous paragraph, the condition numbers indi- of the fabrication process into consideration,’’ Composites, Part B, 32, No. 2,
cate that the one-gage experiment is more sensitive to error.3 Be- pp. 111–121.
关4兴 Delfosse, D., Cherradi, N., and Ilschner, B., 1997, ‘‘Numerical and experimen-
cause the determination of the condition number requires no ex- tal determination of residual stresses in graded materials,’’ Composites, Part B,
perimental data, the benefit of the two-gage design can be shown 28, No. 1–2, pp. 127–141.
prior to carrying out the experiments. A similar analysis can in 关5兴 Rabin, B. H., Williamson, R. L. et al., 1998, ‘‘Residual strains in an Al2 O3 -Ni
fact be performed during the planning of any relaxation-based joint bonded with a composite interlayer: Experimental measurements and
FEM analyses,’’ J. Am. Ceram. Soc., 81, No. 6, pp. 1541–1549.
residual stress measurement, and can aid in designing an experi- 关6兴 Bokuchava, G. D., Schreiber, J., Shamsutdinov, N., and Stalder, M., 2000,
ment with acceptable accuracy. ‘‘Residual stress studies in graded W/Cu materials by neutron diffraction
method,’’ Physica B, 276, pp. 884 – 885.
关7兴 Fukui, Y., and Watanabe, Y., 1996, ‘‘Analysis of thermal residual stress in a
thick-walled ring of Duralcan-base Al-SiC functionally graded material,’’ Met-
7 Conclusions all. Mater. Trans. A, 27, No. 12, pp. 4145– 4151.
关8兴 Cheng, W., and Finnie, I., 1990, ‘‘The crack compliance method for residual
This paper presents an application of the compliance method to stress management,’’ Weld. World, 28, No. 516, pp. 103–110.
estimate residual stress in a layered ceramic-metallic composite 关9兴 Prime, M. B., 1999, ‘‘Residual stress measurement by successive extension of
material. The flexibility of the compliance method, using a back- a slot: The crack compliance method,’’ Appl. Mech. Rev., 52, No. 2, pp.
calculation scheme based on finite element analysis, allows the 75–96.
关10兴 Nelson, G., and Ezis, A., 1996, ‘‘Functionally gradient material 共FGM兲 armor
application to a layered material. The residual stresses found ex- in the TiB2/Ti system,’’ Cercom, Incorporated.
perimentally are small compared to the mechanical strength of the 关11兴 Carpenter, R. D., Liang, W. W., Paulino, G. H., Gibeling, J. C., and Munir, Z.
material system, but are large enough to significantly impact the A., 1999, ‘‘Fracture testing and analysis of a layered functionally graded Ti/
fracture behavior of cracked geometries. The residual stress dis- TiB beam in 3-point bending,’’ Mater. Sci. Forum, 308–311, pp. 837– 842.
关12兴 Carpenter, R. D., Paulino, G. H., Munir, Z. A., and Gibeling, J. C., 2000, ‘‘A
tribution found is similar to that obtained from a simple thermo- novel technique to generate sharp cracks in metallic/ceramic functionally
elastic analysis. graded materials by reverse 4-point bending,’’ Scr. Mater., 43, No. 6, pp.
The paper further describes an error analysis and proposes a 547–552.
systematic method for selection of the functional description of 关13兴 Hill, M. R., Carpenter, R. D., Paulino, G. H., Munir, Z. A., and Gibeling, J. C.,
‘‘Fracture testing of a layered functionally graded material,’’ In ASTM STP
the residual stress distribution. The error analysis indicates that 1409 共to appear兲.
typical measurement uncertainties produce acceptable error in es- 关14兴 Lin, W-Y., 1999, ‘‘Measurement of residual stress distribution through the
timated stress. Minimizing the global uncertainty in residual stress thickness of functionally graded material Ti/TiB2 ,’’ MS thesis. University of
provides an optimal functional description for the residual stress California, Davis, 1999.
关15兴 Ritchie, D., and Leggatt, R. H., 1987, ‘‘Measurement of the distribution of
field. In this work, the scope of the optimization was limited to residual stresses through the thickness of a welded joint,’’ Strain, 23, No. 2, pp.
different orders of Legendre polynomials. The error analysis fur- 61–70.
ther provides a basis by which the benefit of the two-gage imple- 关16兴 Giannakopoulos, A. E., Suresh, S., Finot, M., and Olsson, M., 1995, ‘‘Elasto-
mentation of the compliance method can be demonstrated. plastic analysis of thermal cycling: layered materials with compositional gra-
dients.’’ Acta Metall. Mater., 43, No. 4, pp. 1335–1354.
关17兴 Albert, A. E., 1972, Regression and the Moore-Penrose Pseudoinverse, Aca-
3
In fact, the condition numbers can be used to estimate the ratio of uncertainties demic Press, New York.
for each approach. Because the system is solved using least squares, the condition 关18兴 Bevington, P. R., and Robinson, D. K., 1972, Data Reduction and Error Analy-
numbers suggest that the one-gage design will have approximately 冑12640/1281 sis for the Physical Sciences, 2nd ed. McGraw-Hill, New York.
⫽3.1 times more uncertainty that the two-gage design. This estimate is close to the 关19兴 Strang, G. 1986, Introduction to Applied Mathematics, Wellesley-Cambridge
factor of four determined by the full error analysis. Press, Wellesley, MA.

Journal of Engineering Materials and Technology APRIL 2002, Vol. 124 Õ 191

Downloaded From: http://materialstechnology.asmedigitalcollection.asme.org/ on 01/29/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

You might also like