You are on page 1of 16

M A R K O V C H A I N M O D E L FOR C R A C K I N G

BEHAVIOR O F REINFORCED
C O N C R E T E BE AMS

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

By Prakash Desayi1 and K. Balaji Rao 2


ABSTRACT: In this paper, nonhomogeneous Markov chains are proposed for modeling the cracking behavior of reinforced concrete beams subjected to monotonically increasing loads. The model facilitates prediction of the maximum crackwidth
at a given load given the crackwidth at a lower load level, and thus leads to a
better understanding of the cracking phenomenon. To illustrate the methodology
developed, the results of three reinforced concrete beams tested in the laboratory
are analyzed and presented.
INTRODUCTION

The need for probabilistic design has long been felt (Freudenthal 1961).
A number of investigations are available dealing with probabilistic analysis
of reinforced concrete members with respect to different limit states of rupture (Ellingwood 1987; Ellingwood and Galambos 1981; MacGregor 1981;
Mirza and MacGregor 1979, 1982), and limit state of deflection (a serviceability limit state) (Meyers and Benjamin 1972; Ramsay et al. 1979). Based
on these studies load and resistance factors have been suggested to the American Concrete Institute (ACI) ("Building code requirements" 1983). More
recently Israel et al. (1987) have developed partial resistance factors for flexure, compression plus bending and shear limit states.
Crackwidth is one serviceability limit state to be considered in the design
of reinforced concrete beams. Hence it is necessary to understand the phenomenon of cracking of reinforced concrete structures so that the width of
cracks that form could be controlled. Cracking of reinforced concrete beams
is a random phenomenon. The literature on the probabilistic analysis of cracking
of reinforced concrete beams is scanty. Recently the writers performed probabilistic analysis of strains, crack spacings and crackwidths (Desayi and Balaji Rao 1987), at different stages of loading, of three beams tested in the
laboratory. In the probabilistic analysis, the cross-sectional dimensions of
the beams and strengths of steel and concrete were considered as random
variables.
The object of this paper is to develop a stochastic model to represent the
cracking behavior of reinforced concrete beams subjected to monotonically
increasing loads that can be used to predict the future state of the beam given
its present state.
TEST BEAMS

Three singly reinforced concrete beams B l , B2 and B 3 , of rectangular


cross section (200 mm x 350 mm) were tested under symmetrical third point
'Prof., Dept. Civ. Engrg., Indian Inst. Sci., Bangalore 560 012, India.
2
Res. Scholar, Dept. Civ. Engrg., Indian Inst. Sci., Bangalore 560 012, India.
Note. Discussion open until February 1, 1990. To extend the closing date one
month, a written request must be filed with the ASCE Manager of Journals. The
manuscript for this paper was submitted for review and possible publication on July
14, 1988. This paper is part of the Journal of Structural Engineering, Vol. 115,
No. 9, September, 1989. ASCE, ISSN 0733-9445/89/0009-2129/$1.00 + $.15
per page. Paper No. 23834.
2129

J. Struct. Eng. 1989.115:2129-2144.

TABLE 1. Test Data of Beams Used in Present Study


Beam Designation
S1.No

(D
1
2

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

4
5

7
8
9

Description of quantity
(2)
Compressive strength of
concrete (fcu)
Tensile strength of
concrete (/,)
Ultimate bond strength of
concrete (/&,) (obtained
fromCPHO)
Modulus of elasticity of
concrete (Ec)
Depth of neutral axis from
extreme compression
fiber {x)
Moment of inertia of
cracked transformed
section (/)
The effective area of
concrete in tension (Acl)
Cracking moment of the
beam {Mcr)
Ultimate moment of the
beam (Mj/)

B2
(4)

B1

0)

B3
(5)

33.078 N/mm 2

40.417 N/mm 2

22.510 N/mm 2

2.563 N/mm 2

2.836 N/mm 2

2.120 N/mm 2

3.222 N/mm 2

4.080 N/mm 2

2.760 N/mm 2

0.3278 X 105 N/mm 2

0.3624 X 105 N/mm 2

0.2704 X I0 5 N/mm 2

77.420 mm

75.910 mm

92.900 mm

17,093 X 104 mm4

16,213.74 X 104 mm4

23,596.36 X 104 mm4

15,600 mm2

17,840 mm2

18,580 mm2

1.6484 X 104 kN-mm

1.5146 x 104 kN-mm

1.1565 x 104 kN-mm

7.0406 X 104 kN-mm

7.8594 X 104 kN-mm

6.4340 X 104 kN-mm

loading. All three beams were simply supported over an effective span of
4,200 mm. Beam Bl was reinforced with two bars of 16 mm diameter, while
beam B2 was reinforced with two bars of 11.6 mm diameter and two bars
of 12 mm diameter, beam B3 was reinforced with five bars of 11.6 mm
diameter. The effective depths of three beams were 311.0 mm, 305.4 mm
and 303.5 mm, respectively. Thus, the three beams tested had reinforcement
ratios of 0.0065, 0.0072 and 0.0087, respectively. Some test details of these
beams are provided in Table 1.
PROPOSED MODEL

When nominally identical reinforced concrete beams are tested under


monotonically increasing loads, the load-maximum crackwidth response curves
vary from beam to beam, giving rise to different realizations, as shown in
Fig. 1. Hence, the phenomenon of cracking is a stochastic process, which
will be hereafter referred to as a cracking process. The cracking process of
reinforced concrete beam is denoted by (Wmiixq, q Q). For each load q
Q, the maximum crackwidth, Wmaxq, is a random variable (Fig. 1). The set
Q is the index set of the process. The state space of the cracking process is
the set of all possible values the random variables WmMq can assume. Each
state describes the state of the beam, that is, the maximum crackwidth formed
at any level on the surface of beam at a load q. The state space of the
cracking process becomes discrete if each state is considered as corresponding to a finite range of maximum crackwidths as shown in Table 2. The
index set, Q, which is a collection of loads applied on the beam, is assumed
to be discrete and thus the cracking process becomes a discrete load process.
Referring to the beam as the system in the sequel, the state of the system
depends on the maximum crackwidth formed, which in turn depends on the
2130

J. Struct. Eng. 1989.115:2129-2144.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

eolisation

y Realisation 2
y Realisation 1

Load ( q )
FIG. 1. Different Realizations of LoadMaximum Crackwidth Curves of Nominally Identical Beams Tested under Similar Loading Conditions

load applied on the system. Thus, WmlLXq = i indicates that the system is in
state (' when the applied load is q. With the increase of applied load the
system makes a transition from one state to another as described by its transition probabilities (Ang and Tang 1975). In general, the probability of a
future state of the system may depend on its entire history, that is, its conditional probability is,
P[Wm^qk) = )\Wmm(q*-d = i, ., WmM

=1]

(1)

TABLE 2. Classification of States of the System


State of
the system
(1)
l
2

Implication
(2)
The maximum crackwidth is
between w, and w2
The maximum crackwidth is
between vv2 and iv3

Mathematical
representation
(3)

W2 <

Wmm

< H>3

The maximum crackwidth is


between wt and wt+l

w, < Wm wi+l

The maximum crackwidth is


between >v and w+1

w < Wmm < w+1

2131

J. Struct. Eng. 1989.115:2129-2144.

where (qQ,qu...,qk_uqk,...,qm)
= index set Q\ and (1,2,.. .,i,j,...,)
the state space of the cracking process [Wmax(q), q Q].

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

Assumptions
The following assumptions have been made on the nature of the cracking
process:
1. The cracking process is a nonhomogeneous Markov chain. This assumption
is justified since the future crackwidth, more or less, depends on the current
value of the maximum crackwidth. The nonhomogeneity of the process arises
because of dependence of maximum crackwidths on the applied loads.
2. The cracking process has 0?o>?i, ?) a s index set and (1,2,...,) as
state space. The classification of state space of the system is given in Table 2.
As the process is nonhomogeneous Markovian [assumption (1)], Eq. 1 takes
the form,
PiMk-\,qk) = PlWmm(qk) = j\Wmax(q^i) = ;]; qk > qk-

(2)

where Pij(qk-i,qk) = the probability of a system entering a new state j (at


load qk) given that it is currently in state i (at load qk-i).
The transition probability matrix, [Tl(qk-l,qk)], of the system (with n states)
when load on it is increased from qk~l to qk is given by
State entered
1
ITIfe-i,^)] = State
left

...

PiMk-uqk)

Pniqk-uqk)

Pi(ik-i,qk)

P2i(qk-i,qk)

P2ziqk-\,qd

P2,Xik-i,qk)

pi(qk-i,qk)

Pni(qk-i,qk)

pmMk-\,qk)_

0)

Since all states considered (Table 2) are mutually exclusive and collectively
exhaustive, the sum of the probabilities in each row of the above matrix is
equal to 1. That is,
Z, Pu(4k-i,qk) = 1 for all i; k = 1, 2, ..., m; qk> qk-

(4)

With the increase of load on the system, from qk-t to qk, the already formed
crack of maximum width (at load qk-,) can either widen or remain at the
same value (at load qk). In other words, the system cannot make a transition
from a higher state to a lower state. Thus, the transition probability matrix
of the system takes the form,
P\Mk-\,qd

in(qk-.i,qk)] =

Pn(qk-i,1k)

P\(qk-i,qk)

P22<,qk-i,qk)

P2n(qk-i,qk)

(5)

In Eq. 5, state n = an absorbing state since pm(qk-i,qk) ~ ' To determine


[Tk(qk^uqky\, it is enough to obtain the values of pu(qk-i,qk) for j & /.
2132

J. Struct. Eng. 1989.115:2129-2144.

Determination of Transition Probability


Using the information about the classification of states of the system presented in Table 2, Eq. 2 can be written as
Pij(qk-i,1k) = P{[wj < WmJ,qk) < Wj+1]|[w, <

ffjft-i)

< w,+1]}

(6)

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

Using the definition of conditional probability, Eq. 6 can be rewritten as


P{[wj < Wm,x(qk) < wJ+l] n [w, < WmaK(qk-d < w,+1]}
,

(7)
Dr- ^ w
P[wi < Wmax{qk~\) s wl+i]
Frorr Eq. 7, it can be seen that, to determine the transition probability
Pij(qk-Uqk) not only the probability distribution of maximum crackwidth at
load qk-i but also the joint probability distribution of maximum crackwidths
at loads qk_x and qk are needed. Denoting the maximum crackwidths at loads
qk-x and qk, respectively by random variables Yk-X and Yk, and, knowing
their individual and joint density functions, the transition probability can be
obtained as
PtMk-uqk) =

fYk^Yt(yk-uyk)dyk-idyk
'i

Pu(lk-i,qk)

(8)
fYt-^yk-i)dyk-i

Wi

where/yt_,(vt-i) = the probability density function of random variable Yk-u


and fYk^Yk(yk~i,yk) is the joint probability density function of random variables Yk-S and Yk.
For illustration, the expression for pn(qk-i,qk), using Eq. 8, is given by
fyk-,Yt(yk-uyk)dyk-idyk
Pniqk-uQk) =

(9)

7^
frt-iiyk-Ddyt-i

Similarly the expressions for other nonzero transition probabilities in Eq. 5


can be written.
Determination of /n-Step Transition Probability Matrix
The transition probability matrices, [n(g0,<Zi)]> [H(<li,q2J], |TI(9m-i>9m)]
of the system in different load intervals (or increments), namely (q0,qi),
(q\>qz)> , (qm-i,qm) can be determined using Eq. 8. Application of Chapman-Kolmogorov equations (Ang and Tang 1984; Iosifescu 1980) to the
cracking process, which is nonhomogeneous and Markovian, gives the transition probability matrix of the system, for the case when load on it is increased directly from q0 to qm, as
[U(q0,qm)) = [U(q0,gi)] X [Ilfo.ft)]
X [II(92,93)] X X Va.{qm-i,qm)]

(10)

where [H(q0,qm)] is also called the m-step transition probability matrix of


2133

J. Struct. Eng. 1989.115:2129-2144.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

the system. It may be noted that the m-step transition probability matrix is
a n X n matrix.
Determination of Unconditional Probabilities
The transition probability matrix [H(qk-i,qk)] gives the probabilities of
finding the system in any of 1 to n states (at load qk) given the present state
of the system (at load qk-\). Similarly, the transition probability matrix
[H(q0,q,)] gives the probabilities of finding the system in any of 1 to n
states (at load qm) given the present state of the system (at load q0). Thus
the probabilities of [H(qk-i,qk)] and [U(q0,qm)] are conditional and depend
on the initial state of the system from which a transition is taking place. It
may be needed to know the unconditional distribution (or probabilities) of
states of the system, given the initial state distribution of it, when load on
beam is increased from q0 to qm. This can be obtained from,

[nu(q0,qj] = [n(0)] x [iifa>,?i)] x [ntai.fe)] x x [n(<7m-<?,)] (ii)


where [11(0)] = the initial state vector of the system and gives probabilities
of finding the system in different states at load q0; a n d [n(g0><7m)] = the
state vector giving the probabilities (unconditional) of finding the system in
different states at load qm.
Determination of Transition Probability Matrix
under Normality Constraint
In this section, the transition probability matrix of the cracking process
modeled by a nonhomogeneous Markov chain for a reinforced concrete beam
is determined using the concepts and equations developed so far. In addition
to the assumptions made already, it is further assumed that at any applied
load qk (k = 0, 1,2, ..., m), the maximum crackwidth (which is a random
variable) formed at the level of steel follows a normal distribution. Also,
the joint probability density function of maximum crackwidths at loads qk-x
and qk is Gaussian, with a correlation coefficient p.
An extreme value distribution instead of a normal distribution, may be
more appropriate to describe the variations in maximum crackwidth. But this
would increase the mathematical complexity and it would be difficult to obtain a joint probability density function. Moreover, the estimation of parameters of extreme value distribution needs large amounts of test data of maximum crackwidths which may not be available. These difficulties have been
bypassed by making the normal distribution assumption. It may, however,
be noted that the formulation presented (Eqs. 8, 10 and 11) is general, and
can be used in an analysis which does not involve this assumption.
Following the above assumption, the individual and joint probability functions of random variables, Yk-i and Yk can be written as,
/n-,(?*-i) = -7=
,,-</[(--w-oto-iJ1;
,
V2 7r ak-i
fyk(yk) =

e-a/2)[(W-w)M^.

_x

< yk

_, < yt _, < oo

V2TT ak

frk-,rk(yk-i,yk) =

27ra>-.,o>vT- p2
2134

J. Struct. Eng. 1989.115:2129-2144.

(12)
(13)

. e -{l/[2(i-p 2 )]H[(w-i-M-*-i)M-i] 2 -2pKw-i-i 1 *-i)M-iJl(w-w)MJ+L(w-w)M]'}_

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

- c o < yk_i

< oo

- o o < yk < oo

(14)

where JJ^-I, u.,t and (X/t-i, o> = the means and standard deviations, respectively, of the two random variables Yk-i and Yk; and p is the correlation
coefficient. As can be seen from Eqs. 12-14, to define the probability density functions of maximum crackwidths at loads qk^ and qk, the values of
\ik-i, u,*, a*-! and ak should be known. In this paper to compute mean maximum crackwidth at any load, the model proposed by Desayi and Ganesan
(1985) has been used. Accordingly, the maximum crackwidth at the level
of steel is given by,
KtJclAct

e,

(15)

hfbu\-rr) 2Tr<t>
where fct and/ to = tensile and ultimate bond strengths of concrete; Act = the
effective area of concrete in tension; STT(}> = the total perimeter of steel
present in the tension zone; Mcr and M = cracking moment and ultimate
moment of beam, respectively; es is the strain in steel at any stage of loading
and is obtained by assuming a linear strain variation across the depth of the
beam; k kb and 7 are constants whose values are 2/3, 2/3 and 0.33, respectively (Desayi and Ganesan 1985).
Making first order approximation (Ang and Tang 1975), the mean maximum crackwidth at the level of steel at loads qk-x and qk can be obtained
from,

M \\

kbfbu (j-Jcr S ^
and
fy Jet A. c

M* =

TT"^

(17)

hfb

- \ijl^

where the bars on any given quantity = the mean value of that quantity.
Knowing the coefficient of variation, il, of maximum crackwidth, the values
of o v ! and ak can be determined from,
cr,t_i = n ^ _ ,

(18)

ff* = Of*

(19)

SOLUTION FOR BEAM B1

To illustrate the working of the proposed model, steps involved in the


analysis of beam Bl are given (beams B2 and B3 are similarly analyzed and
2135

J. Struct. Eng. 1989.115:2129-2144.

slate 7
6-

0.24

'
1-0-20

'?

0-16

0.28

5.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

UJi

A\
X !

008
0-04

stated

- y/\

-0
20

'

A |

0.12

1
1

// i
1
,1

q,35

q2

i | i

1
ll

1
1

1
1

50 q3

q< 65

'
1

80

Load(kN)
FIG. 2.

LoadMaximum Crackwidth Plot for Beam B1

their final results only are presented). The details of beam Bl are already
given. The load-maximum crackwidth (observed at the level of steel) plot
of this beam is shown in Fig. 2.
Step 1
To determine the transition probability matrix of the cracking process the
index set and state space have to be specified. In the current example the
index set chosen is
1o

q\

qi

q3

q< q5

where Q = (24.525 kN; 33.845 kN; 43.164 kN; 52.484 kN; 61.803 kN;
and 71.123 kN). It may be noted that the choice of the index set is not
governed by any special considerations and could be any finite set satisfying
the constraint that each element should lie between cracking and ultimate
loads.
The state space of the current example is as follows:
State 1:
State 2:

0 < Wmm < 0.05 mm


0.05 mm s Wmax < 0.10 mm

State 7:

Wmi!l > 0.30 mm

All the seven states are mutually exclusive and collectively exhaustive. Thus,
2136

J. Struct. Eng. 1989.115:2129-2144.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

TABLE 3. The Means and Standard Deviations of Maximum CracKwidtns at Different Loads of Beam B1 Assuming ft = 0.25
Standard deviation (mm)
(3)
0.0354
0.0489
0.0624
0.0759
0.0893
0.1028

Mean (mm)
(2)
0.1418
0.1957
0.2495
0.3034
0.3573
0.4111

Load (kN)
(1)
24.525
33.845
43.164
52.484
61.803
71.123

the cracking process is a nonhomogeneous Markov chain with seven possible


states and five load steps (index set).
Step 2
To determine transition probability matrices in different load intervals, it
is assumed that maximum crackwidth (a random variable) at any load qk
follows a normal distribution and also, the joint probability density function
of maximum crackwidths at loads qk-\ and qk is Gaussian with a correlation
coefficient of 0.95 (assumed). Eqs. 12-14 can be used to determine transition probabilities (Eq. 8), provided the parameters of the distributions are
known. While the mean maximum crackwidths at loads qk-x and qk can be
determined using Eqs. 16 and 17, to obtain standard deviations (Eqs. 18 and
19) the coefficient of variation should be known. The estimation of coefficient of variation of maximum crackwidth from the test results requires a
large amount of test data of maximum crackwidths measured at the same
location on the surface of a large number of nominally similar beams at a
given stage of loading. This is time consuming and uneconomical. To overcome this difficulty a simulation technique can be resorted to. Hence, the
writers (Desayi and Balaji Rao 1987), generated data for maximum crackwidths (sample size = 1,000) using Monte Carlo simulation. The statistical
properties of maximum crackwidths were studied at different stages of loading for beams Bl, B2 and B3. From this study it was found that the coefficient of variation of maximum crackwidth varied from 0.15 to 0.25. In the
current example a value of 0.25 is chosen. The means and standard deviations of maximum crackwidths at different stages of loading are given in
Table 3. Now, using Eqs. 12-14 and Eq. 8, the transition probability matrices in different load intervals are obtained. The values of single and double
integrals appearing in Eq. 8 are evaluated using a standard NAG package
(which uses Gaussian quadrature formulas to evaluate integrals) on a DEC
1090 system. The program is written in standard FORTRAN. For illustration
(11070,4,)], [HO?,,42)] and [U(q2,q3)] only are given here.

[11(90,?,)]

0.2659
0
0
0
0
0
0

0.7146
0.1771
0
0
0
0
0

0.0195
0.7425
0.1373
0
0
0
0

0
0.0804
0.6866
0.0743
0
0
0

0
0
0.1757
0.6912
0.0189
0
0

0
0
0.0003
0.2329
0.6801
0.0019
0

2137

J. Struct. Eng. 1989.115:2129-2144.

0
0
0
0.0016
0.3010
0.9980
1.000

(20)

0.6057 0.0288 0.0114


0
0
0
0
0
0.2660 0.6603 0.0674 0.0063
0
0.2071 0.6654 0.1244 0.0031
0
0
0
0.1584 0.6519 0.1875 0.0022
0
0
0
0.1087 0.6420 0.2493
0
0
0
0
0.0620 0.9380
0
0
0
0
0
1.000.

(21)

"~0.3806 0.5452 0.0503 0.0239


0
0
0
0
0
0.3177 0.5852 0.0807 0.0164
0
0
0
0.2650 0.6048 0.1189 0.0113
0
[life,ft)]=
0
0
0
0.2197 0.6097 0.1622 0.0084
0
0
0
0
0.1784 0.6059 0.2157
0
0
0.1393 0.8607
0
0
0
0
0
0
0
0
1.0000.
0

(22)

0.3541

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

in(qq2)] =

0
0
0
0
0
0

Step 3
Knowing the transition probability matrices in different load increments,
namely, [O(^ 0 ,^i)], [H(quq2)], ..., [IIO^,^)], it is possible to determine
the m-step transition probability matrix using Eq. 10. For illustration, the
two-step transition probability matrix, starting from load q0, is [n(q0,q2)]
and is given by
[Tl{q0,q2)] = [IKq0,qi)] X [Jl{quq2)]

(23)

and from Eqs. 20 and 21 is,


0.0942 0.3511
0
0.0471

tn(?o,?2)] ^

0
0

0
0
0

0
0
0
0
0

0.4836 0.0642 0.0069


0
0.2707 0.5187 0.1459 0.0174
0.0284 0.2001 0.4838 0.2420
0
0.0118 0.1236 0.4721
0
0.0021 0.0543
0
0.0001
0
0
0

0
0.0002
0.0457
0,3925
0.9436
0.9999
1.0000

(24)

Similarly, three-step, four-step and five-step transition probability matrices are obtained.
Step 4
All the above probabilities are conditional. To determine the unconditional
probabilities of the beam for an increase of load from q0 to q2, the probability
distribution of initial states should be known. The values of maximum crackwidths experimentally observed at different applied loads are shown in Fig.
2. Also shown in this figure are different states of the system and the loads
q0, .., q5- An estimate of maximum crackwidth in between successive applied loads is obtained by assuming a linear variation. As can be seen from
this figure, the beam is in state 1 when the applied load is q0- Hence, the
initial state probability vector is given by
[11(0)] = ( 1 0 0 0 0 0 0 )

(25)

Knowing [11(0)] and using Eq. 11, the unconditional probabilities of different states of the system, for different load steps, are determined and the
values are presented in Table 4. Also presented in Table 4 are the values of
2138

J. Struct. Eng. 1989.115:2129-2144.

TABLE 4. Unconditional Probabili ies of Different States of Beam B1 for p = 0.95

HI

Increase of
Load

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

III

To

(1)

Ik
(kN)
(2)

Load
step
(m)
(3)

Vector
(4)

1
(5)

2
(6)

(7)

4
(8)

0)

24.525
24.525
24.525
24.525
24.525

33.845
43.164
52.484
61.803
71.123

1
2
3
4
5

WV(q<q,)}
[nv(qa,qi>]
in"(q0,q})]
{nv(q0,qt))
iW (q0,q5)]

0.2659
0.0942
0.0359
0.0135
0.0050

0.7146
0.3511
0.1629
0.0728
0.0311

0.0195
0.4836
0.3384
0.1899
0.0962

0.0000
0.0642
0.3372
0.2916
0.1879

0.0000
0.0069
0.1036
0.2606
0.2469

Unconditional Probabilities of Different States


3

6
(10)

(11)

0.0000
0.0000
0.0201
0.1246
0.2200

0.0000
0.0000
0.0019
0.0470
0.2129

having
greatest Exceptional
probability
of
crackwidth
at load qt
(mm)
(mm)
(13)
(12)
0.05-0.10
0.10-0.15
0.10-0.15
0.15-0.20
0.20-0.25

0.070
0.134
0.178
0.230
0.290

maximum crackwidths obtained from Fig. 2, at different loads, namely qu


q2, . . . , q5.
Interpretation of Results of Beam Bl for p = 0.95
1. The first element of the transition probability matrix (U(q0,qi)] (Eq. 20)
is Pn(go><7i)- This means the probability is 0.2659 the system will be in state 1
at a load 33.845 kN given that the system is in state 1 at a load 24.525 kN. In
otherwords, there is 26.59% probability that the maximum crackwidth at the
level of steel considered will be in the range 0 to 0.05 mm at a load 33.845 kN
given that the maximum crackwidth is in the range 0 to 0.05 mm at a load 24.525
kN. Similarly other elements of [H(qo,q\)] and elements of [H(q,,q2)], . . . ,
[n(<74,g5)] could be explained. In these transition probability matrices, state 7
is the absorbing state.
From Eqs. 20-22, it can be noted that there is a gradual decrease in the values
of probability in all the diagonal elements and in the first super diagonal elements
as the load on the beam is increased from (go><7i) to (qi,q2) and (q2,q^). Thus,
with the increase of load, if the crackwidth is less likely to increase, the beam
considered could be taken to be more resilient than an average beam and one
would expect that the crackwidth is less likely to increase at further load increment.
2. The first element of the two-step transition probability matrix, \Ii{q(),q2y\,
obtained in step three ispn{q0,q2), and is equal to 0.0942. That is, starting from
state 1 at load q0, there is only 9.42% probability that the system will be in state
1 at load q2. In otherwords, there is 9.42% probability that the maximum crackwidth will be in the range 0 to 0.05 mm at load 43.164 kN given that the maximum crackwidth at load 24.525 kN is in the range 0 to 0.05 mm. The m-step
transition probability matrices will be particularly useful in predicting maximum
crackwidths at higher loads given the information about the maximum crackwidths observed at lower loads.
3. Knowing the initial state of the system considered as state 1, the unconditional probabilities of different states of the system for different load increments (steps) are determined in step 4. Thus the probability vector [Hu(qa,q2)],
gives probabilities of finding the system in different states at loads q2, starting
from state 1 at load q0 (Eq. 25).
4. From Table 4 it is clear that the system will eventually be in state 1 with
26.59% probability, in state 2 with 71.46% probability or in state 3 with 1.95%
2139

J. Struct. Eng. 1989.115:2129-2144.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

probability at load 33.845 kN knowing the initial state probability vector of the
system is as given in Eq. 25. Similarly the unconditional probabilities of different
states for other load increments (or steps) can be explained. In this table the
state of each vector \Uu{qk-l,qk)} having maximum probability of occurrence is
underlined. Thus, it is highly likely that the maximum crackwidth at the level
considered will be in the range of 0.05-0.10 mm, after one load step (i.e., when
load on the beam is increased from 24.525 kN to 33.845 kN). The experimentally observed maximum crackwidth at load 33.845 kN is 0.070 mm (obtained
from Fig. 2). Thus, the model predicts satisfactorily the state of the system for
one load step. From Table 4, it can be seen that the model predicts the state of
the system correctly for 1 and 2 loading steps. When the load on the beam is
increased from 24.525 kN to 52.484 kN, it is highly likely (with 33.84% probability) that the beam will be in state 3, i.e., maximum crackwidth will be in
the range of 0.10-0.15 mm. But, experimentally, a maximum crackwidth of
0.178 mm was observed at 52.484 kN. Similarly, while the model predicted that
with higher probabilities the maximum crackwidths would be in the ranges of
0.15-0.20 mm (state 4) and 0.20-0.25 mm (state 5), for load steps 4 and 5,
respectively, maximum crackwidths of 0.230 mm (state 5) and 0.290 mm (state
6), were observed for these load steps experimentally. The model has thus slightly
underestimated the maximum crackwidths for higher loading steps. This could
be due to any of: (1) Normality assumption made in the analysis; (2) assumption
of linear variation of maximum crackwidths as shown in Fig. 2; and (3) a high
value of p assumed.
EFFECT OF p ON TRANSITION PROBABILITIES

In the above illustration a value of 0.95 was assumed for p. Because there
is no information on the possible value of p, it was felt desirable to work
out the solution for other lower values of p also, so that the influence of p
on the solution is determined. For this purpose, three other values, namely,
0.9, 0.85 and 0.8 have been chosen and transition probabilities and unconditional probabilities have been determined for different load steps m using
Eqs. 8, 10, and 11. The unconditional probabilities of different states for p
= 0.80 are presented in Table 5 for the purpose of comparing them with
those given in Table 4. Considering the state with maximum probability of
occurrence for p = 0.95 as the reference state (for any m), it is observed
TABLE 5. Unconditional Probabilities of Different States of Beam B1 for p = 0.80

Increase of
Load
From

To

(kN)

(D

It
(kN)
(2)

Load
step
(rn)
(3)

24.525
24.525
24.525
24.525
24.525

33.845
43.164
52.484
61.803
71.123

1
2
3
4
5

Unconditional Probabilities of Different States


Vector
(4)

1
(5)

2
(6)

4
(8)

(7)

5
(9)

6
(10)

7
(11)

Range of
maximum
crackwidth
having
greatest Exceptional
probability
of
crackwidth
occurrence at load qt
(mm)
(mm)
(13)
(12)

tn"( w ,,)] 0.1446 0.5627 0.2738 0.0189 o.oooo 0.0000 0.0000 0.05-0.10
0.0224
[n u ( ? o,*)] 0.0033
0.0004
in" (q0,qs)] 0.0001
in"(q0,q2)]

0.1598
0.0341
0.0061
0.0009

0.3539
0.1312
0.0335
0.0068

0.3114
0.2520
0.0999
0.0278

0.1186
0.2730
0.1839
0.0728

0.0271
0.1804
0.2253
0.1324

^140

J. Struct. Eng. 1989.115:2129-2144.

0.0068 0.10-0.15
0.1260 0.20-0.25
0.4509
0.30
>0.30
0.7592

0.070
0.134
0.178
0.230
0.290

TABL E 6. Unconditional Pro aabilities of Different States of Beam B2 for p = 0.8O

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

Increase of
Load
From

To

(kN)

Ik
(kN)

(D

(2)

Load
step
(m)
(3)

19.620
19.620
19.620
19.620
19.620

28.449
37.278
46.107
54.936
63.765

1
2
3
4
5

Unconditional Probabilities of Different States


Vector
(4)

1
(5)

in( ? o,?i)] 0.1407


0.0247
in"(q,qi)-\
0.0045
in u too,?4)] 0.008
[n%,<?5)] 0.0001

2
(6)

3
(7)

4
(8)

0)

6
(10)

(11)

0.7870
0.2790
0.0779
0.0191
0.0042

0.0723
0.5496
0.3239
0.1253
0.0386

0.0000
0.1352
0.4008
0.3051
0.1460

0.0000
0.0106
0.1551
0.3144
0.2656

0.0000
0.0009
0.0316
0.1612
0.2636

0.0000
0.0000
0.0068
0.0741
0.2819

Range of
maximum
crackwidth
having
greatest Exceptional
probability maximum
of
crackwidth
at load qk
(mm)
(mm)
(12)
(13)
0.05-0.10
0.10-0.15
0.15-0.20
0.20-0.25
>0.30

0.070
0.124
0.198
0.260
0.284

from Tables 4 and 5 that the probability of finding the system in states higher
than the reference state increases while it decreases for lower states when p
decreases from 0.95 to 0.80. Also, for the lower values of p = 0.80, there
is no overestimation of crackwidths at higher stages of loading.
The analysis has been repeated for beams B2 and B3, and observations
similar to those made in case of Bl were made in these beams also. The
values of \Y\u(qk^uqky\ for different load steps and for p = 0,80 are presented in Tables 6 and 7. From these tables it is inferred that the states of
B2 and B3 are predicted satisfactorily for different load steps. This shows
that a value of p = 0.80 may be more suitable than higher values. Thus,
the cracking of reinforced concrete beams subjected to monotonically increasing loads can be represented by a nonhomogeneous Markov chain, with
the maximum crackwidths at two successive stages of loading (qk~i,qk) following a joint Gaussian distribution with p = 0.80.
Different codes of practice currently suggest suitable detailing procedures
to control the width of crack and additionally some have suggested methods
of computing the crackwidths. Using the method proposed in the paper (which
is based on probabilistic analysisa basis more relevant in studies of cracking behavior), and an acceptable method to compute crackwidths, it is posTABLE 7.

Unconditional Probabilities of Different States of Beam B3 for p = 0.80

Increase of
Load
From

To

(2)

Load
step
(m)
(3)

(kN)

(kN)

Vector
(4)

1
(5)

d)

2
(6)

3
(7)

4
(8)

0)

6
(10)

(11)

19.620
19.620
19.62
19.620
19.620
19.620

25.506
31.392
37.278
43.164
49.050
54.936

1
2
3
4
5
6

in"(qo,q,)]
in"(qo,q2)]
in" (q0,q,)]
in" (qcq,)]
in"(qo,qs)]
mU(q0,qe)]

0.2468
0.0616
0.0154
0.0037
0.0009
0.0002

0.7508
0.5673
0.2853
0.1181
0.0432
0.0143

0.0024
0.3617
0.5525
0.4684
0.2907
0.1472

0.0000
0.0093
0.1291
0.3223
0.4122
0.3581

0.0000
0.0001
0.0169
0.0699
0.1797
0.2863

0.0000
0.0000
0.0008
0.0146
0.0545
0.1256

0.0000
0.0000
0.0000
0.0030
0.0188
0.0683

Unconditional Probabilities of Different States


5

2141

J. Struct. Eng. 1989.115:2129-2144.

Range of
maximum
crackwidth
having
greatest Exceptional
probability
of
crackwidth
at load qt
(mm)
(mm)
(12)
(13)
0.05-0.10
0.05-0.10
0.10-0.15
0.10-0.15
0.15-0.20
0.15-0.20

0.054
0.088
0.118
0.180
0.216
0.236

sible to estimate the probable maximum crackwidth at an expected higher


load knowing the width of crack at a lower level. Thus, it would help in
estimating the cracking state of the beam under possible overloading, i.e.,
loading beyond a level at which the limit state of cracking i s satisfied.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

CONCLUSION

1. The cracking process of a reinforced concrete beam can be modeled by a


nonhomogeneous Markov chain by properly defining the state space and index
set.
2. Expressions derived for transition probabilities are more general and can
be used knowing the distributions of maximum crackwidths.
3. When the value of the correlation coefficient of maximum crackwidths at
two successive loading stages is reduced from 0.95 to 0.80, it is seen that for
any given m, the probability of finding the system in higher states compared to
the reference state increases while the same decreases for lower states.
4. For p = 0.8, the states of the system are predicted satisfactorily. Thus, it
is felt that modeling the cracking process by a nonhomogeneous Markov chain
appears a promising approach to improving the understanding of cracking behavior of reinforced concrete beams.
APPENDIX I.

REFERENCES

Ang, A. H., and Tang, W. H. (1975). Probability concepts in engineering planning


and design. Vol. I. John Wiley and Sons, New York, N.Y.
Ang, A. H., and Tang, W. H. (1984). Probability concepts in engineering planning
and design. Vol. II, John Wiley and Sons, New York, N.Y.
"Building code requirements for reinforced concrete." (1983). ACE Standard 31883, Amer. Concrete Inst., Detroit, Mich.
The code of practice for the structural use of concrete. CP110. (1972). Part 1, British
Standards Institution. London, England.
Desayi, P., and Balaji Rao, K. (1987). "Probabilistic analysis of the cracking of re
beams." Mat. Struct., 20, 408-417.
Desayi, P., and Ganesan, N. (1985). "An investigation on spacing of cracks and
maximum crackwidth in reinforced concrete flexural members." Mat. Struct.,
18(104), 123-133.
Ellingwood, B. (1978). "Reliability basis of load and resistance factors for reinforced
concrete design." NBS Building Science Series 110, Washington, D.C.
Ellingwood, B., and Galambos, T. V. (1981). "General specifications for structural
design loads." Proc, Symp. on Probabilistic Methods in Struct. Engrg., 27-42.
Freudenthal, A. M. (1961). "Safety, reliability and structural design." Proc, ASCE,
J. Struct. Div., 87(3), 1-16.
Israel, B. M., Ellingwood, B., and Corotis, R. B. (1987). "Reliability-based code
formulations for reinforced concrete buildings." J. Struct. Engrg., ASCE, 113,
2235-2252.
MacGregor, J. G. (1981). "Load and resistance factors for concrete design." Proc,
Symp. on Probabilistic Methods in Struct. Engrg., 14-26.
Marius Iosifescu. (1980). Finite Markov processes and their applications. John Wiley
and Sons, Inc., New York, N.Y.
Meyers, B. L., and Benjamin, J. R. (1972). "Variability of deflections of simply
supported reinforced concrete beams." ACI J., 69(1), 29-35.
Mirza, S. A., and MacGregor, J. G. (1979). "Statistical study of shear strength of
reinforced concrete slender members." Proc, American Concrete Inst., 76(11),
1159-1178.
Mirza, S. A., and MacGregor, J. G. (1982). "Probabilistic study of strength of reining

J. Struct. Eng. 1989.115:2129-2144.

forced concrete members." Can. J. Civ. Engrg., 9, 431-448.


Ramsay, R. M., Mirza, S. A., and MacGregor, J. G. (1979). "Monte Carlo study
of short time deflection of reinforced concrete beams." Proc, American Concrete
Inst., 76(8), 897-918.
APPENDIX II.

NOTATION

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

The following symbols are used in this paper:


Aa
Ec
fbu
/(,
fc,
fcl
fcu
fytl(yk~i)
fy^yk)
frt_,Yt(yk-i>yk)

=
=
=
=
=
=
=
=

lct
('
j
kb
k,
M
Mcr
M
Mu
P[-]
Pij(qk-Uqk)

=
=

=
=
=
=
=
=
=
=

Q
q
(<7o, q,,..., qm)
[Wmax(q)]
Wmax
w, < lVmax s w,+ 1
x

=
=
=
=
=
=
=

=
=

IVi

Yk

7
es
(ts)k
(ls)k-s
jLjt

=
=
=
=
=

effective area of concrete in tension;


modulus of elasticity of concrete;
ultimate bond strength of concrete;
mean ultimate bond strength of concrete;
direct tensile strength of concrete;
mean direct tensile strength of concrete;
cube compressive strength of concrete;
probability density function of random variable Yk-1;
probability density function of random variable Yk;
J o m t probability density function of random variables Yk-X and Yk;
moment of inertia of cracked transformed section;
state i of the system;
state j of the system;
constant (= 2/3 in present study);
constant (= 2/3 in present study);
cracking moment of beam;
mean cracking moment of beam;
ultimate moment of resistance;
mean ultimate moment of resistance;
probability of occurrence of event;
probability of transition of system from state i (at
load qk-\) to state j (at load qk);
index set;
load applied;
set of applied loads (index set);
cracking process (stochastic process);
maximum crackwidth which is random variable;
value of Wmax lying between values w, and w,+ 1;
depth of neutral axis from extreme compression fiber;
random variable describing variations in maximum
crackwidth at load #*_,;
random variable describing variations in maximum
crackwidth at load qk;
a constant (= 0.33, in present study);
strain in steel due to applied load;
mean steel strain at applied load of qk;
mean steel strain at applied load of qk-x;
mean maximum crackwidth at level of steel when
applied load is qk;
2143

J. Struct. Eng. 1989.115:2129-2144.

Downloaded from ascelibrary.org by SERC on 03/02/14. Copyright ASCE. For personal use only; all rights reserved.

Vk-i

mean maximum crackwidth at level of steel when


applied load is qk-x;
[11(0)] = initial state probabilities vector of the system;
[IK^-i, qk)] = transition probability matrix of system when load on
it is increased from qk-t to qk;
\Ilu(q0,q^)\
= unconditional probabilities vector of system when
load on it is increased from q0 to qm;
p = correlation coefficient;
XiTcf) = total perimeter of steel bars present in tension zone;
(Tk = standard deviation of maximum crackwidth at level
of steel when applied load is qk;
<rk-i = standard deviation of maximum crackwidth at level
of steel when applied load is qk-t;
<)> = diameter of steel bar;
O = coefficient of variation maximum crackwidth;
( ) = represents a vector of size 1 x n; and
[ ] = represents square matrix of size n.

2144

J. Struct. Eng. 1989.115:2129-2144.

You might also like