tr20 82
tr20 82
tr20 82
Empirical Formulas
Edgar Daniel Rodriguez Velasquez1,2
Olga Kosheleva3 , and Vladik Kreinovich4
1
Department of Civil Engineering
Universidad de Piura in Peru (UDEP)
Av. Ramón Mugica 131, Piura, Peru
edgar.rodriguez@udep.pe
2
Department of Civil Engineering
3
Department of Teacher Education
4
Department of Computer Science
Department of Computer Science
University of Texas at El Paso, 500 W. University
El Paso, TX 79968, USA
edrodriguezvelasquez@miners.utep.edu
olgak@utep.edu, vladik@utep.edu
Abstract
Due to stress, cracks appear in constructions: cracks appear in build-
ings, bridges, pavements, among other structures. In the long run, cracks
need to be repaired. However, our resources are limited, so we need to
decide which cracks are more dangerous. For this, we need to be able to
predict how different cracks will grow. There are several empirical formu-
las describing crack growth. In this paper, we show that by using scale
invariance, we can provide a theoretical explanation for these empirical
formulas.
1
are limited. It is therefore desirable to predict how the current cracks will grow,
so that we will be able to concentrate our limited repair resources on most
potentially harmful cracks.
How cracks grow: a general description. In most cases, stress comes
in cycles: the engine clearly goes through the cycles, the road segment gets
stressed when a vehicle passes through it, etc. Thus, the crack growth is usually
expressed by describing how the length a of the pavement changes during a
stress cycle at which the stress is equal to some value σ. The increase in length
is usually denoted by ∆a. So, to describe how a crack grows, we need to find
out how ∆a depends on a and σ:
∆a = f (σ), (2)
for some function f (σ). Empirical data shows that this dependence is a power
law, i.e., that
∆a = C0 · σ m0 , (3)
for some constants C0 and m0 .
Practical case of reasonable size cracks: Paris law. Very small cracks
are extremely important in critical situations: since there, the goal is to prevent
the cracks from growing. In most other practical viewpoint, small cracks are
usually allowed to grow, so the question is how cracks of reasonable size grow.
Several empirical formulas have been proposed. In 1963, P. C. Paris and
F. Erdogan compared all these formulas with empirical data, and came up with
a new empirical formula that best fits the data:
0
∆a = C · σ m · am . (4)
This formula – known as Paris Law or Paris-Erdogan Law – is still in use; see,
e.g., [3, 6].
Usual case of Paris law. Usually, we have m0 = m/2, in which case the
formula (4) takes the form
√
∆a = C · σ m · am/2 = C · (σ · a )m . (5)
The formula (4) is empirical, but the dependence m0 = m/2 has theoretical
explanations. One of such explanations is that the stress acts randomly at
different parts of the crack. According to statistics, the standard deviation s of
the sum of n independent variables each of which has standard deviation s0 is
2
√
equal to s = s0 · n; see, √
e.g., [10]. So, on average, the effect of n independent
factors is proportional to n. Thus, for a crack of length a, consisting of a/δa
independent parts, the overall effect K of the stress σ is proportional to
√ √
K = σ · n ∼ σ · a. (6)
∆a = C · K m , (4a)
i.e., to m0 = m/2.
Empirical dependence between C and m. In principle, we can have all
possible combinations of C and m. Empirically, however, there is a relation
between C and m:
C = c0 · bm
0 ; (8)
see, e.g., [4, 5] and references therein.
Beyond Paris law. As we have mentioned, Paris law is only valid for rea-
sonably large crack lengths a. It cannot be valid for a = 0, since for a = 0, it
implies that ∆a = 0 and thus, that cracks cannot appear by themselves – but
they do. To describe the dependence (1) for all possible values a, the paper [2]
proposed to use the expression (4) with different values of C, m, and m0 for
different ranges of a. This worked OK, but not perfectly.
The best empirical fit came from the generalization of Paris law proposed
in [8]: γ
∆a = C · σ m · aα + c · σ β . (9)
Empirically, we have α ≈ 1.
What we do in this paper. In this paper, we provide a theoretical explanation
for the empirical formulas (3), (4), and (8), and (9).
Our explanations use the general ideas of scale-invariance, ideas very similar
to what is described in [4].
3
which is λ times smaller, all the numerical values get multiplied by λ: instead
of the original value x, we get a new value x0 = λ · x.
In many physical situations, there is no preferred measuring unit. In such
situations, it makes sense to require that the dependence y = f (x) remain
valid in all possible units. Of course, if we change a unit for x, then we need
to appropriately change the unit for y. So the corresponding scale invariance
requirement takes the following form: for every λ > 0, there exists a value µ(λ)
depending on λ such that, if we have
y = f (x), (10)
y = f (x1 , . . . , xv ), (14)
If we now plug in the expression for y from formula (10) into this formula, we
will conclude that
µ(λ) · f (x) = f (λ · x). (19)
It is known (see, e.g., [1]) that every measurable solution to this functional
equation has the form
y = C · xm , (20)
i.e., the form of a power law.
4
Similarly, for functions of several variables, if we plug in the expressions (15)
and (16) into the formula (17), we get
If we now plug in the expression for y from formula (14) into this formula, we
will conclude that
It is known (see, e.g., [?]) that every measurable solution to this functional
equation has the form
y = C · xm mn
1 · . . . · xn .
1
(23)
5
4 Scale Invariance Explains How C Depends
on m
Idea. Let us show that scale invariance can also the explain the dependence (8)
between the parameters C and m of the Paris law (4a).
Indeed, the fact that the coefficients C and m describing the Paris law are
different for different materials means that, to determine how a specific crack
will grow, it is not sufficient to know its stress intensity K, there must be some
other characteristic z on which ∆a depends:
The logarithm ln(∆a(K, z)) in linear in ln(K) and linear in ln(z), thus it is
a bilinear function of ln(K) and ln(z). A general bilinear function has the form:
By applying exp(t) to both sides of the formula (30), we conclude that the
dependence of ∆a on K has the form
∆a = C · K m , (31)
where
C = exp(a0 + az · ln(z)) (32)
6
and
m = aK + aKz · ln(z). (33)
From (33), we conclude that ln(z) is a linear function of m, namely, that
1 aK
ln(z) = ·m− . (34)
aKz aKz
Substituting this expression for ln(z) into the formula (32), we can conclude
that
aK · az az
C = exp a0 − + ·m , (35)
aKz aKz
i.e., the desired formula (8), C = c0 · bm
0 , with
aK · az
c0 = exp a0 − (36)
aKz
and
az
b0 = exp . (37)
aKz
Thus, the empirical dependence (8) of C on m can also be explained by scale
invariance.
q = F (q1 , q2 ). (38)
7
for all q1 and q2 . In mathematical terms, the combination operation F (q1 , q2 )
should be commutative.
Similarly, if we combine three effects, the result should not depend on the
order in which we combine them, i.e., that we should have
8
Acknowledgments
This work was supported in part by the National Science Foundation grants
1623190 (A Model of Change for Preparing a New Generation for Professional
Practice in Computer Science) and HRD-1242122 (Cyber-ShARE Center of Ex-
cellence).
References
[1] J. Aczel and J. Dhombres, Functional Equations in Several Variables, Cam-
bridge University Press, Cambridge, UK, 2008.
[2] M. Bigerelle and A. Iost, “Bootstrap analysis of FCGR, application to
the Paris relationship and to lifetime prediction”, International Journal of
Fatigue, 1999, Vol. 21, pp. 299–307.
[3] D. Broek, Elementary Engineering Fracture Mechanics, Martinus Lojhoff
Publishers, The Hague, The Netherlands, 1984.
[4] A. Carpinteri and M. Paggi, “Self-similarity and crack growth instability
in the correlation between the Paris’ constants”, Engineering Fracture Me-
chanics, 2007, Vol. 74, pp. 1041–1053.
[5] M. B. Cortie and G. G. Garrett, “On the correlation between the C and m
in the Paris equation for fatigue crack propagation, Journal of Engineering
Fracture Mechanics, 1988, Vol. 30, No. 1, pp. 49–58.
[6] D. Little, D. Allen, and A. Bhasin, Modeling and Design of Flexible Pave-
ments and Materials, Springer, Cha, Switzerland, 2018.
[7] P. C. Paris and F. Erdogan, “A critical analysis of crack propagation laws”,
The American Society of Mechanical Engineers (ASME) Journal of Basic
Engineering, 1963. Vol. 85, No. 4, pp. 528–534.