Вы находитесь на странице: 1из 30

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269602088

Computation of Tensile Strain-Hardening Exponents through the Power-Law


Relationship

Article  in  Journal of Testing and Evaluation · July 2012


DOI: 10.1520/JTE104226

CITATIONS READS
0 458

3 authors:

Ariel Esteban Matusevich Julio Massa


National University of Cordoba, Argentina National University of Cordoba, Argentina
11 PUBLICATIONS   25 CITATIONS    33 PUBLICATIONS   138 CITATIONS   

SEE PROFILE SEE PROFILE

Reinaldo Mancini
Instituto Nacional de Tecnologia Industrial
14 PUBLICATIONS   31 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Diseño estructural mecánico View project

All content following this page was uploaded by Ariel Esteban Matusevich on 19 January 2016.

The user has requested enhancement of the downloaded file.


1 2 3
Ariel E. Matusevich, Julio C. Massa, and Reinaldo A. Mancini

Computation of Tensile Strain-Hardening Exponents through the


Power-Law Relationship

ABSTRACT: Many metals flow in the region of uniform plastic deformation following a power-law
relationship, which states that true stress is proportional to true-plastic strain raised to the power n. The
exponent n, known as tensile strain-hardening exponent, can be determined from a tension test through
appropriate transformations of stress-strain data and least-squares fitting of a straight line. Procedures for
the computation of n have been standardized by ASTM International and ISO. Current ASTM and ISO
standards differ, most notably, in the type of strain used in calculations. The ASTM procedure permits the
use of true strain (true-elastic strain plus true-plastic strain), when true-elastic strain represents less than
ten percent of total strain. On the other hand, the ISO version stipulates the subtraction of true-elastic
strain from true strain, using a formula whose derivation is not publicly available. In this work, we revisit
the expressions that enable the transformation of engineering stress-strain data to true-stress and true-
plastic-strain values. Using eight tension-test curves from several materials, obtained through ASCII files
publicly available at the website of the National Physical Laboratory of the United Kingdom, we compare
n-values obtained via three definitions of strain: (i) true strain, (ii) conventional definition of true-plastic
strain, and (iii) true-plastic strain according to the ISO formula. In addition, we investigate the
dependency of the results on the strain range over which n-values are calculated. To evaluate strain-range
dependency, which arises when metals do not closely follow the power-law relationship, we analyze the
effect of strain intervals of increasing length and study the variation of n-values when the range of interest
is divided in subintervals. To improve the approximation given by the power-law relationship over the
region under analysis, we propose an alternative formulation in which the strength coefficient and the
strain-hardening exponent are functions of true-plastic deformation.

KEYWORDS: tensile strain-hardening exponent, n-values, metallic materials, sheet formability,


Hollomon, true stress, true strain, ASTM E 646, ISO 10275

1
Research Engineer, INTI-Córdoba, and Assistant Professor, Department of Structures, Universidad Nacional de
Córdoba, Av. Vélez Sarsfield 1561, Córdoba, X5000JKC, Argentina, email: ariel.matusevich@gmail.com
2
Professor, Department of Structures, Universidad Nacional de Córdoba, Av. Vélez Sarsfield 1611, Córdoba,
X5016GCA Argentina, email: jmassa@efn.uncor.edu
3
Head of the Materials Division, INTI-Córdoba, and Assistant Professor, Department of Materials and Technology,
Universidad Nacional de Córdoba, Vélez Sarsfield 1561, Córdoba, X5000JKC, Argentina, email:
rmancini@inti.gob.ar

1
Introduction

The following power-law relationship, as presented by Hollomon [1], relates true stress ( σ )

and true-plastic strain (ε p) in the region of uniform plastic deformation:

σ = K ε pn . (1)

In Eq (1), K and n are known as the strength coefficient and the tensile strain-hardening

exponent, respectively. This power-law relationship describes the plastic flow of many metals

and is widely used because of its simplicity. Other equations due to Ludwik, Swift and Voce are

also well known; a study of the applicability of these expressions can be found in [2].

Taking logarithms on both sides of Eq (1)

ln ( σ ) = ln ( K ε p n ) → ln ( σ ) = ln ( K ) + n ln ( ε p ) , (2)

and introducing the following change of variables:

y = ln ( σ ) , x = ln ( ε p ) , (3)

shows that Eq (2) can be represented as a straight line y = b + n x , with ordinate intercept

b = ln( K ) and slope n. Therefore, the strength coefficient, K, and the tensile strain hardening

exponent, n, can be computed by linear regression of x-y data.

True stress and true strain (εT ), which is the sum of true-plastic strain (εp ) and true-elastic

strain (εE ), can be determined through appropriate transformations of stress-strain data recorded

in a uniaxial tension test. To obtain true-plastic strain, true-elastic strain must be subtracted from

true strain; this requires the determination of the modulus of elasticity (E ) of the metal under

analysis [3]. The expressions that enable these transformations are revisited in this work.

2
The tensile strain-hardening exponent is a measure of how rapidly a metal becomes stronger

and harder and represents a very important parameter in the evaluation of sheet-metal formability

[4]. Procedures for the computation of n through the power-law relationship have been

standardized by ASTM International (ASTM) and the International Organization for

Standardization (ISO). Current editions of these standards, ASTM E 646-07 [5] and ISO

10275:2007 [6] differ, most notably, in the type of strain used in calculations. The ASTM

procedure, whose terminology is confusing, allows the use of εT instead of ε p when εE is less than

ten percent of εT; the standard conveys a procedure on how to subtract elastic strain from total

strain through its Note 13 [5]. On the other hand, the use of ε p in the ISO version is mandatory;

however, this standard applies a formula for true-plastic strain proposed by Aegerter et al. [7],

referred to in this work as εɶ p , whose derivation is not publicly available. The previous edition of

the ISO standard [8], similarly to the current ASTM E 646 standard, stipulated the subtraction of

εE from εT when elastic strain represented more than ten percent of total strain; although giving

no further details on how to perform this subtraction.

In this work, we compare the use of εT , εp and εɶ p in the computation of tensile strain-

hardening exponents. In addition, we examine the effect of strain range on the calculation of the

parameters K and n; this source of variability arises from the fact that the metal may not

accurately obey the power-law relationship. To improve the approximation given by Eq (1) over

the region under analysis, we propose an alternative formulation of the power-law relationship in

which K and n are functions of true-plastic deformation.

To carry out numerical comparisons and analyses, we use tension-test curves from several

materials, obtained through ASCII files publicly available at the website of the National Physical

3
Laboratory of the United Kingdom (NPL) [9]. These datasets were originated in the European-

Union-funded project TENSTAND (Tensile Standard), and represent typical tensile

characteristics of a variety of industrially-important materials [10]. The use of TENSTAND files

in this work makes the results presented here readily reproducible.

True Stress and True Strain

In this section, we revisit the expressions that enable the computation of true stress and true-

plastic strain from engineering tension-test data. To derive these expressions, it is assumed that

the volume of the test specimen remains constant and that strain is homogenously distributed

along the gauge length; these hypotheses hold in the region of uniform plastic deformation of the

engineering tension-test curve (excluding the zone of discontinuous yielding, if it is present) until

the onset of necking, when the maximum engineering stress is reached. A thorough analysis of

these formulas can be found in [11].

True Stress

During yielding and the subsequent plastic-flow regime, the material flows with negligible

change in volume; that is, increases in length are compensated by decreases in cross-sectional

area. The constancy of volume implies

A L = A0 Le . (4)

In Eq (4), A 0 is the initial cross-sectional area of the specimen and Le is the initial gauge

length of the extensometer, while A and L represent the instantaneous cross-sectional area and

the instantaneous gauge length, respectively. Expressing L as the sum of Le and elongation ∆L,

the instantaneous cross-sectional area is obtained from Eq (4) as follows:

4
Le Le
A = A0 = A0 . (5)
L Le + ∆L

True stress is the quotient between load (P) and A; using Eq (5), true stress results

P P  Le + ∆L 
σ = =  . (6)
A A0  Le 

As Eq (6) indicates, true stress is related to engineering stress s = P A0 .

True Strain

Ludwik was the first to introduce the concept of true strain, where changes in length are

referred to the instantaneous gauge length instead of the original gauge length [11]. This strain,

also known as natural strain, can be expressed as follows:

L1 − L0 L −L L − L2
εT = + 2 1 + 3 ⋯ (7)
L0 L1 L2

The evaluation of Eq (7) as a definite integral between Le and L, leads to

L
1  L   L + ∆L 
εT = ∫ L
dL = ln( L) − ln( Le ) = ln 
  = ln  e . (8)
Le  Le   Le 

The subscript in εT means “Total”, indicating that true strain equals the addition of true-

plastic strain and true-elastic strain; true strain can be expressed in terms of engineering strain, e,

as follows:

∆L
e= → εT = ln (1 + e ) . (9)
Le

To obtain true-plastic strain, εp, we must subtract true-elastic strain εE from εT

5
P A
ε p = εT − εE → ε p = εT − . (10)
E

To approximate εE in Eq (10), it has been assumed that the metal unloads following a line

whose slope is the initial modulus of elasticity, E. Substituting Eqs (5) and (8) in Eq (10), we

obtain the following expression for εp:

 L + ∆L  P A0  Le + ∆L 
ε p = ln  e −
 Le   . (11)
  E  L e 

Instead of applying Eq (10), ISO 10275:2007 uses the following equation proposed by

Aegerter et al. [7]:

 L + ∆L P A
εɶ p = ln  e − . (12)
 Le E 

Note that in Eq (12) true-elastic strain has been subtracted inside the logarithm; it is not clear

how the integrand in Eq (8) should be modified to yield this result. The article by Aegerter et al.

[7], which is referenced in the ISO standard, lacks the derivation that leads to Eq (12). However,

we can show that this formula is approximately equivalent to Eq (10). Substituting the

instantaneous area A given by Eq (5) into Eq (12)

 L + ∆L L + ∆L P A 0 
εɶ p = ln  e − e  , (13)
 Le L E
 e 

factoring the resulting expression

 L + ∆L  P A 0 
εɶ p = ln  e 1 −  , (14)
 Le  E  

and applying the product property of logarithm, leads to

6
 L + ∆L 
εɶ p = ln  e + ln (1 − β ) ,
 Le 
(15)
 

where β is the engineering-elastic strain

P A0
β = . (16)
E

Expanding ln (1 − β ) in Maclaurin series

1 2 1 3 1 4
ln (1 − β ) = − β − β − β − β ⋯ (17)
2 3 4

and retaining only the first term of the series expansion because β << 1

P A0
ln (1 − β ) ≈ − β → ln (1 − β ) ≈ − , (18)
E

we obtain to the following version of Eq (12):

 L + ∆L  P A 0 P A0
εɶ p ≈ ln  e − → εɶ p ≈ εT −
 Le 
. (19)
  E E

Comparing Eqs (19) and (10), we see that εɶ p ≈ ε p while A 0 ≈ A, which is essentially true in

the range of application of these expressions. Since the logarithm of true-plastic strain is used in

calculations, as Eq (3) indicates, differences in the results between using εɶ p or ε p become less

noticeable.

'umerical comparison

In this section we compare the use of εT, εp and εɶ p in the computation of tensile strain-

hardening exponents. We make use of ASCII data files representing typical tensile

characteristics of several materials, openly available at the web site of NPL [9]. These tension-

7
test curves have agreed values of the designated material properties and enable the validation of

tension-test software [10]. The TENSTAND project that produced these files did not cover the

determination of the tensile strain-hardening exponent; consequently, there are no agreed values

for this parameter. Nevertheless, these curves provide excellent case studies for the comparison

of different procedures.

Table 1 describes eight datasets utilized for the numerical comparison. The values of E

adopted for n computations (third column of Table 1) correspond to the slopes of the elastic part

of these curves. These slopes, which were calculated using an in-house program, are within the

range of agreed values for E [10]. Figure 1 shows the engineering stress-strain plots

corresponding to the datasets. Maxima of engineering stress are indicated with small circles in

Fig. 1; all curves have been truncated beyond their maxima for clarity.

FIG. 1Engineering stress-strain curves of the materials under analysis.

8
Tensile strain-hardening exponents obtained using εT, εp and εɶ p are denoted by n1, n2 and n3,

respectively. Similarly, strength coefficients associated with n1, n2 and n3, are denoted by K1, K2

and K3, respectively. Results for n are listed in Table 2. For each case listed in Table 2, the

portion of the stress-strain curve under analysis (also indicated with square dots in Fig. 1) is

expressed in terms of engineering strain (e); all data points within the specified interval (given in

column 2) were used for the computation of n-values. For the cases analyzed in this work,

differences between using εp or εɶ p in the computation of tensile strain-hardening exponents are

less than 0.12 %. On the other hand, applying εT instead of εp overestimates n by percentages

between 1.0 % and 3.1 %. Since results usually depend on the strain range analyzed, the gap

between n1 and n2 might decrease or increase for other intervals.

FIG. 2Contribution of true-elastic strain in true strain.

To analyze the participation of true-elastic strain in total strain, we plotted the ratio ε E ε T in

the intervals over which n-values were calculated. Figure 2 shows the plots for all datasets, with

the omission of dataset 10, whose plot nearly overlaps the plot of dataset 1. In all cases, true-

elastic strain represents less than ten percent of true strain and the quotient ε E ε T follows a

9
strictly decreasing law.

Influence of E in the Computation of n2 and n3

The evaluation of true-elastic strain involves, according to classical plasticity theory, the

determination of the initial modulus of elasticity of the metal under analysis [3]. This parameter

can be calculated as the slope of the proportional part of the engineering stress-strain curve.

However, standard tension tests are usually performed using single-sided extensometers which

are not suitable for modulus measurement, since any slight misalignment of the test specimen

may cause gross errors in the resulting value of E [12]. It is indeed possible to obtain good

quality tension-test data for modulus determination, but this usually requires the use of high

resolution side-to-side averaging extensometers, in a separate and dedicated test that focuses

only on the initial part of the stress-strain curve [13]. Since measurement of E through a

dedicated test is not practical, the use of handbook values for this parameter is a good alternative.

In fact, ASTM E 646 (through Note 13) suggests the use of nominal values of the modulus of

elasticity for the computation of elastic strain [5].

To investigate the influence of E in the computation of n2 and n3, let us consider dataset 53 as

a case study. For this curve, the results obtained for n2 and E are denoted by n*2 and E *,

respectively; these results are considered as reference values in this analysis (see Tables 1 and 2).

Plots of n1 and n2 versus E are shown in Fig. 3; since n3 almost coincides with n2, its plot has

been omitted. For a wide range of E values, we have found that the difference between n1 and n2

is governed by the following expression:

(
n1 − n2 = n1 − n*2 ) E*
E
. (20)

10
We can prove that Eq (20) is valid in general, although the corresponding mathematical

justification is omitted here. Using this equation, the evaluation of n2 for a value of E different

from E * is straightforward

(
n2 = n1 − n1 − n*2 ) E*
E
. (21)

For instance, an overestimation of E by ten percent reduces the difference between n1 and n2

by 9.1 % (E/E * = 1/1.1 ≈ 0.091) and increases n2 by 0.15 %. On the other hand, a ten percent

reduction of E increases the gap between n1 and n2 by 11.1 % (E/E * = 1/0.9 ≈ 1.11) and

diminishes n2 by 0.18 %. From this analysis, we infer that the evaluation of n2 is not very

sensitive to modulus variations. For this reason, the use of handbook values for this parameter is

fully justified.

FIG. 3Influence of E in the computation of n2.

Influence of strain range in the computation of n-values

The tensile strain-hardening exponent can be determined over the entire uniform plastic

range of the engineering stress-strain curve, or over different portions of this range (which might

be specified in a product specification). Since the hypotheses of constancy of volume and

11
homogenous distribution of strain along the gauge length must hold within the interval over

which n is to be computed, this interval must exclude the range of the engineering stress-strain

curve beyond the maximum engineering stress (where necking takes place) and the zone of

discontinuous yielding, if present. The region of interest can be expressed in terms of different

kinds of strain; ASTM E 646 utilizes engineering strain [5], while the ISO standard specifies

strain ranges in terms of plastic-engineering strain [6], ep, defined as follows:

∆L P A0 P A0
ep = − = e− . (22)
Le E E

Table 3 shows the results for the strain-hardening exponent n2, when the limits of the interval

of interest are expressed in terms of e, ep, εT, and εp, defined in Eqs (9), (22), (8), and (10),

respectively. As Figure 4 illustrates (for the case of material 1), different strain parameters lead

to different strain intervals for the same nominal range (5 % - 20 %); note that plastic-strain

scales, ep and εp, depend on the material under analysis. Since the resulting strain ranges are

different, the values of n2 listed in each row of Table 3 display differences; if these metals

followed the power-law relationship exactly, the parameters of this relationship would not

depend on strain range. In fact, for material 42, variations of n2 over the intervals defined in

Table 3 are almost indistinguishable (between 0.05 % and 0.09 %), while the most noticeable

differences (between 1.86 % and 7.12 %) in the results are displayed by material 10; we may

infer that, in terms of compliance with Eq (1), material 42 is the best from this list while material

10 is the worst. The type of strain used for defining the region of interest must be clearly stated

when informing n-values; otherwise, results might be erroneously interpreted.

12
FIG. 4)ominal interval (5% - 20%) expressed in terms of different strain parameters.

FIG. 5Linear regression of ln(εp ) - ln(σ) data (TE)STA)D file 1).

Let us analyze, for example, the behavior of material 1 (dataset 1). For this material, a log-

log plot of true stress and true-plastic strain does not result in a straight line, as Fig. 5 shows;

consequently, Eq (1) is not strictly valid in the range of analysis. In this section, using dataset 1

as a case study, we investigate the effect of strain range in the computation of the parameters of

the power-law relationship. First, we analyze the influence of strain intervals of increasing length

in the computation of n2. Next, we study the variation of n2 and K2, when the region of interest (5

% ≤ e ≤ 20 %) is partitioned in a succession of consecutive subintervals. Then, we propose an

alternative version of the power-law relationship that improves the approximation of the flow

curve. Finally, we analyze strain-range dependency and accuracy of fit values for all datasets.

13
Variation of n2 over strain intervals of increasing length (dataset 1)

We explore the variation of n2 for two sets of strain intervals: (i) ten intervals sharing the

same initial strain, ei = 5 %, with ends located at ei + ∆e , and (ii) ten strain intervals centered at

the middle of the range of interest, ec = 12.5 %, with boundaries positioned at ec ± ∆e / 2 ; for

both sets the ranges are: ∆e = k eo with eo = 1.5 % and k = 1, 2, … , 10 . As Fig 6 illustrates, for

intervals having a fixed origin, n2 increases when length increases (n2 increases 21 % from the

shortest to the longest interval), while for centered intervals the opposite occurs; n2 decreases

when length increases (n2 decreases 5 % from the shortest to the longest interval). When

∆e = 15 % the resulting range is 5 % - 20 %, and both values of n2 coincide at the result listed in

Table 2 (n2 = 0.39178).

FIG. 6Computation of n2 for strain intervals of increasing length ∆e (TE)STA)D file 1).

Variation of n2 and K2 when the interval of interest is partitioned (dataset 1)

Results for n2 and K2, calculated over 14 consecutive intervals of one percent of true-plastic

strain, are shown in Figs 7 and 8, respectively. In these figures round dots indicate middle points

of each subinterval; the complete strain range extends from εp = 4.5 % to εp = 18.5 %, covering

the region of interest (5 % ≤ e ≤ 20 %), as can be seen in Fig. 4. For increasing true-plastic strain,

14
n2 and K2 increase, reach a maximum at εp ≈ 14 %, and then decrease.

FIG. 7Results for n2, when the range of interest is divided in uniform intervals (Dataset 1).

FIG. 8Results for K2, when the range of interest is divided in uniform intervals (Dataset 1).

Alternative formulation of the power-law relationship

To improve the approximation given by Eq (1) over the region of interest, instead of

performing a linear fit of x-y data as Fig 5 shows, we propose the use of higher-degree

polynomials. From the resulting fit we have

f q ( x ) = a0 + a1 x + a2 x 2 + ⋯ + aq x q , (23)

where q is the degree of the fitting polynomial; when q = 1 Eq (23) reduces to Eq (2), which

results from the log-log transformation of the power-law relationship (a0 = ln(K) and a1= n).

15
Calculating the derivative of Eq (23) with respect to x, we obtain a polynomial function for the

tensile strain-hardening exponent

df q ( x)
n2 ( x ) = = a1 + 2 a2 x + 3a3 x 2 ⋯ + q aq x q −1 . (24)
dx

To determine the corresponding strength-coefficient function, we need to compute the

ordinate intercept, b2(x), of the tangent line to fq (x) at the point x

df q ( x)
b2 ( x ) = f q ( x) − x. (25)
dx

Introducing Eq (24) into Eq (25), we obtain

b2 ( x ) = f q ( x) − n2 ( x ) x → K 2 ( x ) = eb2 ( x ) , (26)

which shows the dependency of K2 (x) on n2 (x).

Using Eqs (24) and (26), the power-law relationship is reformulated as follows:

σ = K 2 ( x) ε p n2 ( x ) , (27)

where K2 and n2 are functions of true-plastic strain because x = ln(εp). When x-y data is

approximated by a straight line (q = 1), Eq (27) reduces to Eq (1).

FIG. 9Results for n2(εp) using several polynomial fits of x-y data (Dataset 1).

16
FIG. 10Results for K2(εp) using several polynomial fits of x-y data (Dataset 1).

Approximating x-y data (TENSTAND file 1) by polynomials of degree two, three and five,

we construct the functions for n2(εp) and K2(εp) that are plotted in Figs. 9 and 10, respectively. As

these figures illustrate, the fifth-degree polynomial approximation of x-y data produces functions

for n2 and K2 that accurately follow the round dots, which represent the middle points of each

consecutive subinterval of one percent of true-plastic strain shown in Figs. 7 and 8. Note that

according to Eq (24), n2(x) is a straight line when q = 2 in Eq (23), but its plot as a function of

true-plastic strain, n2(εp), becomes the curve shown in Fig. 9.

Goodness-of-fit analysis for dataset 1

To evaluate goodness of fit of predictions given by Eq (27), we compute the Root Mean

Squared Error (RMSE), also known as fit standard error

∑(P − σ )
2
i i
RMSE = i =1
, (28)
m

where σi is the observed true-stress value, Pi is the predicted value given by Eq (27), and m is the

number of experimental points. This parameter has the same units of observed values and

enables comparisons between different regression models that fit the same data; when RMSE is

17
closer to zero the accuracy of the regression model is better. To evaluate whether a particular

value of RMSE can be considered as “good”, we define in this work the percent error of fit (PEF)

RMSE
PEF = 100 , (29)
σ

where σ is the mean value of true-stress data in the strain range under analysis. This

normalization of fit standard error also enables goodness-of-fit comparisons between different

datasets. To complement the goodness-of-fit evaluation, we compute the maximum percent error

(MPE)

 P − σi 
MPE = max  i  100 , (30)
 σi 

which indicates the maximum percent deviation of the prediction in the strain interval.

Let us evaluate the compliance of material 1, in the range of interest (5 % ≤ e ≤ 20 %), with

the power-law relationship. For this case, RMSE = 4.068 and PEF = 0.57 % ( σ = 712.0 MPa ) .

The maximum percent error, which occurs at the beginning of the interval, is MPE = 2.09 %

(nearly four times the value of PEF). From these results, we infer that, for material 1, the overall

prediction of the power-law relationship in the range of calculation is good. Note that this

goodness-of-fit analysis has been carried out for true-stress predictions; if we examine the fitting

of x-y data shown in Fig. 5, we see that y = ln(σ) deviates most notably from a straight line at the

beginning of the interval, but the mean prediction of the straight line can also be considered

good.

Strain-range dependency of the parameters of the power-law relationship ought to be

interpreted with care. As this example demonstrates, a strong strain-range dependency, which is

evidenced in Figs 6-10 for material 1, does not imply an overall poor prediction of the flow

18
curve. Strain-range dependency of K2 and n2 means that the resulting estimate is only valid in the

range under analysis and should not be used on other intervals. If the metal accurately followed

the power-law relationship, the prediction would be valid over the complete region of uniform

plastic deformation.

FIG. 11Fitting of x-y data over three different strain intervals (Dataset 1).

To explain the seeming contradiction between strong strain-range variations of n2 and K2 as a

19
function of εp (Figs. 7-10), and the overall good prediction, in terms of x, given by Eq (2) (where

n2 and K2 are constants) shown in Fig. 5, we prepared Fig. 11. This figure illustrates the

approximations for εp = 5 % and εp = 14 % which correspond to the minimum and maximum

points of Figs. 7-10, respectively. Figure 11 includes the following features:

• a table that summarizes results for n2 indicated in Figs. 7 and 9, and values of K2

shown in Figs. 8 and 10,

• x-y data in the range of interest (5 % ≤ e ≤ 20 %),

• the straight line shown in Fig 5, that results from the linear regression of x-y data, and

• two straight lines denoted by y5 and y14 that fit intervals of one percent of εp centered

at εp = 5 % and εp = 14 %, respectively.

Note that the slopes of y5 and y14 differ significantly from each other (n2 = 0.3173 for y5 and n2 =

0.4238 for y14) and also deviate from the slope of the line that fits the complete interval of

interest (n2 = 0.3918). In spite of these variations, y5 and y14 provide excellent estimates in the

ranges 4.5 % ≤ εp ≤ 5.5 % and 13.5 % ≤ εp ≤ 14.5 %, respectively. Since y = ln K 2 + n2 x ,

predictions also depend on the value of K2, which is associated with n2. As Fig. 11 indicates,

ordinate intercepts increase when slopes increase, which is also evidenced by comparing Fig. 7

with Fig. 8. Because both parameters, n2 and K2 vary, they produce a good estimation of ln(σ) in

the interval under analysis.

Goodness of fit can be significantly improved by using higher degrees, q, in Eq (23). In fact,

fit standard error for material 1 almost vanishes (RMSE = 0.015), when q = 5.

Goodness-of-fit comparison for all datasets

In this section we evaluate, for the eight materials analyzed in this work, true-stress

20
predictions given by Eq (27). To carry out goodness-of-fit comparisons between datasets we

analyze Table 4, that lists:

• RMSE values of the predictions given by Eq (27), when q = 1, 2, 3 and 5 in Eq (23)

(columns 3-6),

• percent reductions of RMSE when using Eq (27) with q = 2 and q = 3 (columns 7-8)

instead of Eq (1), and

• percent errors of fit, PEF, (column 10) and maximum percent errors, MPE, (column

11) of the estimates of the power-law relationship.

To complement this analysis, we show in Fig. 12 the variation of n2 as a function of εp for all

datasets.

As Fig. 12 illustrates, material 42 accurately follows the power-law relationship because

results for n2 are almost insensitive to strain-range variations and its accuracy of fit is excellent,

PEF = 0.06 % (see column 10 of Table 4). Similarly, material 53 obeys Eq (1) very well; its plot

in Fig. 12 is quite flat and the corresponding percent error of fit is very low, PEF = 0.13 %.

Stress-strain data for material 10 displays considerable scatter in the region of plastic

deformation (see Fig 1). In addition, the conversion of stress-strain data to σ-εp values is close to

a straight line, so the subsequent log-log transformation of σ-εp data turns this line into a curve.

For these reasons, material 10 exhibits the strongest departure from Eq (1) (PEF = 1.83 %); we

anticipated this conclusion when we analyzed the results listed in Table 3.

The remaining materials, 1, 17, 13, 46 and 50 lie in between the first group (materials 42 and

53) and material 10; their compliance with Eq (1) can be considered as intermediate in this

analysis.

The use of higher-degree polynomials for fitting x-y data improves goodness of true-stress

21
estimates in all cases, as Table 4 indicates. In fact, when we use cubic approximations of x-y

data, column (8) of Table 4 shows that fit standard errors are reduced by percentages between

74.5 % and 97.9 % depending on the datasets, with the exception of material 46 that exhibits

serrated yielding during plastic deformation (see Figs 1 and 2).

FIG. 12Variation of n2 as a function of εp for all datasets.

Concluding Remarks

True-plastic strain should be used in the computation of tensile strain-hardening exponents,

since the power-law relation is expressed in terms of this type of strain. Instead of the alternative

form of true-plastic strain presented in ISO 10275, we suggest the utilization of the conventional

definition presented here, since its derivation is conceptually clear. The computation of true-

plastic strain involves the determination of the modulus of elasticity of the metal under analysis.

We have demonstrated that the use of handbook values for E is a better alternative than the

computation of this parameter as the slope of the proportional part of the engineering-stress-

strain curve; this assertion makes the evaluation of true-plastic strain straightforward.

We have shown that the use of true strain in place of true-plastic strain overestimates n

values; although when results are rounded to two decimal places the differences are hardly

22
distinguishable. However, since both options require the same computational cost, the use of

true-plastic strain in calculations should be mandatory.

When a metal does not accurately obey the power-law relationship, the strength coefficient,

K, and the tensile strain-hardening exponent, n, depend on the strain range used in calculations.

To take into account this situation, we derived an alternative version of the power-law

relationship, in which K and n are functions of true-plastic deformation. The proposed

formulation, which significantly improves goodness of fit of the power-law relationship, can be

used in applications which require an accurate fit of the flow curve.

Acknowledgements

The authors thank Professors Laura Felicia Matusevich and Michael Anshelevich of Texas

A&M University for their help in improving this article. Thanks are also due to the reviewers for

their valuable suggestions and comments.

References

[1] Hollomon, J., “Tensile Deformation,” Trans AIME, Vol. 32, 1945, pp. 268–290.

[2] Kleemola, H. J., Nieminen, M. A., “On the Strain-Hardening Parameters of Metals”,

Metallurgical Transactions, Vol. 5, 1974, pp 1863–1866.

[3] Dieter, G., Mechanical Metallurgy, MacGraw- Hill, Inc., New York, 1986, p. 288.

[4] Gedney, R., “Sheet Formability,” Advances Materials & Processes, Vol. 160, 2002, pp.

33–36.

[5] ASTM Standard E 646-07: Standard Test Method for Tensile Strain-Hardening

Exponents (n -Values) of Metallic Sheet Materials, ASTM International, West Conshohocken,

PA, 2007.

23
[6] International Standard ISO 10275:2007(E): Metallic Materials – Sheet and Strip –

Determination of Tensile Strain Hardening Exponent, International Organization for

Standardization, Geneva, Switzerland, 2007.

[7] Aegerter, J., Keller, S., Wieser, D., “Prüfvorschrift zur Durchführung und Auswertung

des Zugversuches für Al-Werkstoffe (Test Procedure for the Accomplishment and Evaluation of

the Tensile Test for Aluminium and Aluminium Alloys),” conference transcript of the

conference Werkstoffprüfung 2003, Verlag Stahleisen GmbH, Düsseldorf, 2003, pp. 139–150.

[8] International Standard ISO 10275:1993(E): Metallic Materials – Sheet and Strip –

Determination of Tensile Strain Hardening Exponent, International Organization for

Standardization, Geneva, Switzerland, 1993.

[9] Tensile Testing – Standards and TENSTAND, http://www.npl.co.uk/science-

technology/advanced-materials/measurement-techniques/mechanical/tensile-testing-standards-

and-tenstand (Last accessed July 17, 2011).

[10] Lord, J., Loveday, M., Rides, M., McEntaggart, I., “TENSTAND WP2 Final Report:

Digital Tensile Software Evaluation: Computer-Controlled Tensile Testing Machines Validation

of European Standard EN 10000-1,” National Physical Laboratory, Teddington, 2005,

http://resource.npl.co.uk/docs/science_technology/materials/measurement_techniques/tenstand/s

w_validation_eval.pdf (Last accessed July 17, 2011).

[11] Dieter, G., Mechanical Metallurgy, MacGraw- Hill, Inc., New York, 1986, pp. 70–76.

[12] Loveday, M., Gray, T., Aegerter, J., “Tensile Testing of Metallic Materials: A Review,”

http://resource.npl.co.uk/docs/science_technology/materials/measurement_techniques/tenstand/te

st_method_review.pdf (Last accessed July 17, 2011).

24
[13] Lord, J., Rides, T., Loveday, M., “TENSTAND-WP3 Final Report: Modulus

Measurement Methods,” http://www.npl.co.uk/publications/tenstand-wp3-final-report-modulus-

measurement-methods (Last accessed December 13, 2011).

Tables:

TABLE 1—Tension-test curves used for numerical comparison.

TENSTAND E
Material
Dataset [GPa]

1 Nimonic 75, CRM661 210.3

10 13% Mn Steel 181.8

13 S355 Structural steel 222.8

17 316L Stainless Steel 190.0

42 Aluminium Sheet - soft AA1050 68.9

46 Aluminium Sheet - soft AA5182 69.4

50 Sheet steel – DX56 162.1

53 Sheet steel – ZStE 203.9

25
TABLE 2—Results for the tensile strain-hardening exponent. *

Case Strain-Hardening Exponent n Percent Change

Strain  n1 − n2   n2 − n3 
TENSTAND εT − Eq (8) εp − Eq (10) εɶ p − Eq (12)  100  100
interval
*  n1   n2 
Dataset n1 n2 n3
%

1 5 - 20 0.39984 0.39178 0.39137 2.0 0.10

10 5 - 20 0.43003 0.42153 0.42105 2.0 0.11

13 4 - 12 0.18494 0.17915 0.17909 3.1 0.03

17 5 - 20 0.35042 0.34396 0.34370 1.8 0.07

42 5 - 20 0.23030 0.22801 0.22797 1.0 0.02

46 5 - 20 0.30751 0.29842 0.29814 3.0 0.09

50 5 - 20 0.24627 0.24256 0.24248 1.5 0.03

53 5 - 15 0.18055 0.17765 0.17762 1.6 0.02

*
The strain interval used in calculations is expressed in terms of engineering strain.

26
TABLE 3—Results for the Strain-Hardening Exponent n2 when the range of interest is

expressed in terms of different strain definitions.

TENSTAND Strain Strain-Hardening Exponent n2 Percent Change

Dataset range, % e ep εT εp (4) − (3) (5) − (3) (6) − (3)


100 100 100
(3) (3) (3)
(1) (2) (3) (4) (5) (6) (7) (8) (9)

1 5 - 20 0.39178 0.39432 0.39402 0.39688 0.65 0.57 1.30

10 5 - 20 0.42153 0.42939 0.44209 0.45156 1.86 4.88 7.12

13 4 - 12 0.17915 0.17699 0.17645 0.17402 -1.21 -1.51 -2.86

17 5 - 20 0.34396 0.34679 0.34969 0.35252 0.82 1.67 2.49

42 5 - 20 0.22801 0.22789 0.22822 0.22817 -0.05 0.09 0.07

46 5 - 20 0.29842 0.29469 0.29045 0.29384 -1.25 -2.67 -1.53

50 5 - 20 0.24256 0.23922 0.23992 0.23992 -1.38 -1.09 -1.09

53 5 - 15 0.17765 0.17697 0.17624 0.17574 -0.38 -0.79 -1.08

27
TABLE 4—Goodness of true-stress predictions for different polynomial approximations of

x-y data

Stress Root Mean Squared Error Percent reduction of True Error % Maximum

(RMSE) RMSE stress PEF Error

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11)

Dataset Range q=1 q=2 q=3 q=5 1 – (4)/(3) 1 – (5)/(3) σ (3)/(9) MPE (q = 1)

# % [MPa] [MPa] [MPa] [MPa] % % [MPa] % %

1 5 - 20 4.068 0.817 0.283 0.015 79.9 93.1 712.0 0.57 2.09

10 5 - 20 11.468 0.945 0.641 0.629 91.8 94.4 625.5 1.83 6.27

13 4 - 12 1.622 0.061 0.034 0.024 96.2 97.9 587.3 0.28 0.73

17 5 - 20 3.660 0.601 0.125 0.034 83.6 96.6 538.5 0.68 2.49

42 5 - 20 0.056 0.052 0.012 0.012 7.7 78.4 88.5 0.06 0.72

46 5 - 20 2.611 1.850 1.778 1.775 29.1 31.9 289.6 0.90 2.48

50 5 - 20 0.866 0.180 0.126 0.101 79.21 85.5 321.8 0.27 0.58

53 5 - 15 0.431 0.123 0.110 0.109 71.49 74.5 338.6 0.13 0.59

Figure captions:

FIG. 1Engineering stress-strain curves of the materials under analysis.

FIG. 2Contribution of true-elastic strain in true strain.

FIG. 3Influence of E in the computation of n2.

FIG. 4)ominal interval (5% - 20%) expressed in terms of different strain parameters.

FIG. 5Linear regression of ln(εp ) - ln(σ) data (TE)STA)D file 1).

FIG. 6Computation of n2 for strain intervals of increasing length ∆e (TE)STA)D file 1).

FIG. 7Results for n2, when the range of interest is divided in uniform intervals (Dataset 1).

28
FIG. 8Results for K2, when the range of interest is divided in uniform intervals (Dataset 1).

FIG. 9Results for n2(εp) using several polynomial fits of x-y data (Dataset 1).

FIG. 10Results for K2(εp) using several polynomial fits of x-y data (Dataset 1).

FIG. 11Fitting of x-y data over three different strain intervals (Dataset 1).

FIG. 12Variation of n2 as a function of εp for all datasets.

29

View publication stats

Вам также может понравиться