Академический Документы
Профессиональный Документы
Культура Документы
net/publication/226574043
Generating line spectra from experimental responses. Part II: Storage and loss
functions
CITATIONS READS
84 133
2 authors, including:
Igor Emri
University of Ljubljana
178 PUBLICATIONS 1,989 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Igor Emri on 28 May 2014.
Introduction Theoretical
This paper is a companion to the preceding one The change from the exponential to the half-Lorent-
(Emri and Tschoegl, 1993a). The latter presented a zian and Lorentzian kernels, i.e., from exp ( - t/r) to
general background, briefly reviewed other work in 1/(1 + coat 2) and oor/(1 + (.o2r2), necessitates changes
this area, and provided the general theory of our in the form of the algorithm that we described in the
algorithm for the determination of respondance companion paper (Emri and Tschoegl, 1993a). We
(relaxation or retardation) time distributions in linear begin by discussing the modifications required when
viscoelastic materials. It then discussed the form of adapting the algorithm for use with the storage
the algorithm suitable for obtaining a discrete distri- kernel, 1/(1 + (.o2r2), instead of the transient kernel,
bution of relaxation times from simulated or smooth- exp ( - t/r). Both kernels are monotone non-increas-
ed relaxation modulus data, or of retardation times ing functions of their arguments. The modifications
from simulated or smoothed creep compliance data. to be introduced result primarily from the need to re-
For brevity, we shall refer to this form of the evaluate the limits of the two intervals that we have
algorithm as the transient algorithm. This companion called the Boundary Window, or Window I, and the
paper discusses the storage algorithm and the loss Modeling Window, or Window 2, in the companion
algorithm, i.e., the forms of the algorithm in which it paper. We recall from that paper that Window 1
applies to simulated or smoothed data on the storage determines the maximum allowable width for the
and loss functions. As in the preceding paper, we region from which the algorithm selects datum points
demonstrate the power of the algorithm here using the for the calculation of each successive spectrum line.
standard linear solid and liquid models, and the The respective Windows 1 are compared in Fig. 1
bimodal form of the Kobeko equation, there given by together with the two kernels.
Eq. (29). Again, the use of simulated data allows us to The kernel function is capable of modeling the
avoid complications arising from the presence of ex- storage modulus effectively only in the transition
perimental error. (See the companion paper for region where the first derivative is significantly larger
details). Another paper (Emri and Tschoegl, 1993b) than zero. As can be seen from Fig. 1, this condition
discusses the application of the algorithm in the is well satisfied in the region from about log ~oz~ =
presence of scatter in experimental data. - 0 . 5 to 0.5 which now defines Window 1. The limits
Tschoegl and Emri, Generating line spectra from experimental responses 323
0 3 4
A = 2"k-12"k--2"k-12"k
4 3
(7a)
-2 -1 0 1 2
log (.)'~} B=r4_12"k--2rZ_12"~+22"~_12"~--2"k_,2"4 (7b)
Fig. 3. The Lorentzian kernel at the two Windows 1
C = 23_1 2 2
--22"k_lrk+2rk_12-k--2- k3 (7 c)
Windows 1, as shown in Fig. 3, span the regions from
D = r k - r x_l • (7d)
log 092"= - 1.07 to - 0.07, and from log ~or = 0.07 to
1.07, respectively. These endpoints, as before, limit Into these we introduce the number, r, of spectrum
the region within which the kernel function, when lines per decade via Eq. (4). Equation (6) then yields
plotted semilogarithmically, can be reasonably ap- the lower limits for both Windows 2. The upper limits
proximated by a straight line. In this region the first are obtained in the same way by replacing rk-1 by
derivate is significantly larger than zero. zk+ 1. The values for r = 1 to r = 8 for the right Win-
The two Windows 2 are shown in Fig. 4 along with dow 1 are tabulated in Table 2 below. Those for the
plots of the first logarithmic derivatives of the loss left window are obtained by replacing the positive
kernel, with the negative sign throughout.
We are now ready to discuss the two algorithms
Dl°ss(092")- ddln092"--( - 1 7 - ~ 2 )2"i themselves.
{Ge} = G e when the modulus describes an sidered because the calculated spectral distribution is
arrheodictic 1) material, and that [Ge} = 0 when the at best an approximation to the true spectrum.
material is rheodictic 1). We will use the form having We may reduce the effort (Emri and Tschoegl,
the kernel 1/(1 + ~2r2) because its general character- 1993 a) required to calculate ~'(coj) by beginning the
istics parallel those of the exponential kernel, first summation in Eq. (12) with l_<m = k - 4 n - 1 .
exp ( - t/z). This yields
Normalizing by the difference Gg-{Ge} leads to i=k-1 1 1
g ' ((Dj) = g ' (('OM) -- 2 gi 2~ gk 2~
i=N 1 i=m l +(Dj72i 1+O)jTi
g'(o~) = g g - ~ gi - - , (9)
i= 1 1 + (,02T 2 i=N 1
-- 2 gi 2--~2~-Aj • (13)
where g'(a)) and gg are the normalized storage and i=k+l l q-(-OjTi
glassy modulus, respectively, and the gi's may be
seen to represent the strengths of the delta functions We take four times the number of spectrum lines per
in the normalized discrete relaxation spectrum decade, n, to cut off contributions from lines four
decades downscale where their contribution has
i=N decreased to 10 -s. This value for rn has been chosen
h(z)= ~ giTJi(~('C--'Ci). (10) for convenience. If it is considered to be too large, the
i=1 value of m should be lowered.
The sum of squares of dj within Window 2 is given
The source data are assumed to be available as a
by
discrete set of M datum points {G'(coj)} where j =
J=Sk, u
1,2 . . . . . M. Each of these datum points can be nor-
Ek= 2 A2, (14)
malized by the difference between the largest point, J = Sk, l
G'M, and the smallest point, G~, to yield the set
{~'(coj)}. G~ replaces Gg, and G ~ replaces {Gel. The where sk, l and Sk, u are the first and the last discrete
modulus can then be expressed alternatively by a points in the window that belong to the k th spectrum
discrete set of (normalized) spectrum lines, {~i}. In line. Minimizing the error according to 8Ek/O~k = 0,
terms of these each datum point becomes leads to
i=N 1 j = Sk'u 1
g'(co:) =~'(coM)- ~ ~ i - - 2 2 "
(11) E ~'(COj)--
i=l 1+00j72i j=sk, l 1 -{- (.Dj2 T 2k
gk=j=sk'ull;)2'2 ~ (15)
We intend to determine, from the set of source
data, [~(coj)}, a set of spectrum lines, [~il, which will j=sg,/ l +o)jZ
faithfully reproduce the modulus, G'(co). We split the
sum in Eq. (11) into three parts and write where
i=k-I 1
i=k-1 1 1
¢ ' (coi) = g' (coM) - g' (coj) - 2 2
g'(~°s)=g'(c°M)- i=1
~ g~l 2~ & l
+ (-Oj "Ci
2~
+ (-Oj "Ci i=m l q-(DjTi
i=N 1
i=N 1 - E ~i 2~" (16)
-- 2 gi 2 - " - ' ' " ~ + Z~j , (12) i=k+l I +(DjTi
i=k+l l'[-(.OjTi
zero. The iteration is broken off when the difference G' I~o}
between the previously found and the newly com- .n....
2
since its kernel is identical with that in Eq. (1). For the
normalization, we now use J ~I - J iI .
gO
-10 -8 -6 -4 -2 0 2 4 6 8 10
log ~o - seconds ~, l o g , - seconds
The loss algorithm
Fig. 5. log G'(og), log G"(co), log HB(r), and the line spec-
The algorithm to be applied to the loss modulus, trum obtained from the moduli
i=N
Tschoegl, 1993 a), envelops the line spectrum. This is 3. Tschoegl NW (1989) The phenomenological theory of
shown in Fig. 5. linear viscoelastic behavior. Springer, Berlin
We conclude that our algorithm works just as well 4. Tschoegl NW, Emri I (1992) Generating line spectra
from experimental responses. III. Interconversion be-
with the storage and loss functions as it does with the tween relaxation and retardation behaviour. Intern J
relaxation modulus and the creep compliance. Polym Mater 18:117- 127
Acknowledgements (Received August 24, 1992;
The authors gratefully acknowledge support of this work accepted April 1, 1993)
by the Slovene Ministry of Science under Grant
P2-1131-782/91, and partial support by the California In-
stitute of Technology. Correspondence to:
Prof. Igor Emri
Mechanical Engineering Department
References University of Ljubljana
Murnikova 2
1. Emri I, Tschoegl NW (1993a) Generating line spectra 61000 Ljubljana, Slovenia
from experimental responses. I. Relaxation modulus and
creep compliance. Rheol Acta 32:311-321 Prof. N.W. Tschoegl
2. Emri I, Tschoegl NW (1993b) Generating line spectra California Institute of Technology
from experimental responses. IV. Application to ex- 1201 East California Boulevard
perimental data. To be submitted to Rheol Acta Pasadena, CA 91125, USA