You are on page 1of 11

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

net/publication/289077203

Simplified punching shear design method for slab-column connections using


fuzzy learning

Article  in  Aci Structural Journal · July 2007

CITATIONS READS

18 564

3 authors:

Kyoung-Kyu Choi M. M. Reda Taha


Soongsil University University of New Mexico
80 PUBLICATIONS   691 CITATIONS    240 PUBLICATIONS   2,375 CITATIONS   

SEE PROFILE SEE PROFILE

Alaa Sherif
Faculty of Engineering-Mataria
38 PUBLICATIONS   315 CITATIONS   

SEE PROFILE

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

improving structural performance and monitoring of damage in polymer concrete and polymer composites using nanomaterials View project

RC Repair with FRP Composites View project

All content following this page was uploaded by Alaa Sherif on 23 May 2017.

The user has requested enhancement of the downloaded file.


ACI STRUCTURAL JOURNAL TECHNICAL PAPER
Title no. 104-S42

Simplified Punching Shear Design Method for Slab-Column


Connections Using Fuzzy Learning
by Kyoung-Kyu Choi, Mahmoud M. Reda Taha, and Alaa G. Sherif

An alternative approach for predicting the punching shear strength compressive strain at the shear crack. Even though Kinnunen
of concentrically loaded interior slab-column connections using fuzzy and Nylander’s model8 did not provide high accuracy in
learning from examples is presented. A total of 178 experimental punching shear strength predictions, it significantly contributed
datasets obtained from concentric punching shear tests of reinforced
to a better understanding of the failure mechanism of the
concrete slab-column connections from the literature are used in
training and testing of the fuzzy system. The fuzzy-based model is slab-column connections and enabled visualizing a rational
developed to address the interaction between various punching flow of forces in such connections. Alexander and
shear modeling parameters and the uncertainties between them, Simmonds2 proposed a strut-and-tie model with concrete ties
which might not be properly captured in classical modeling to describe the load transfer in the slab-column connections.
approaches. The model is trained using 82 datasets and verified Bažant and Cao9 developed a punching shear strength model
using 96 datasets that are not used in the training process. The considering size effect of concrete based on principles of
punching shear strength predicted by the fuzzy-based model is fracture mechanics. The size-effect model was able to
compared with those predicted by current punching shear strength explain the experimental observations of decreasing
models widely used in the design practice, such as ACI 318-05, punching failure shear stresses of slab-column connections
Eurocode 2, CEB-FIP MC 90, and CSA A23.3-04 codes. It is noted
without punching reinforcement with increasing slab thickness.
that the fuzzy-based model yields a significant enhancement in the
prediction of the punching shear strength of concentrically loaded Numerous models suggested modifications to these general
interior slab-column connections while still respecting the funda- directions outlined previously (flexure, combined stress-
mental failure mechanisms in punching shear of concrete. strength criteria, plasticity, strut and tie, and size effect). A
recent review of such models can be found elsewhere.10 In
Keywords: fuzzy systems; punching shear; slab-column connections. spite of the importance of these models in understanding the
failure mechanism of slab-column connections, there is
INTRODUCTION considerable difficulty in using these models in the daily
design practice. Moreover, the level of complexity encountered
Flat plates consist of slabs directly supported on the
in using these models for design might be difficult to justify
columns without beams. For this simple appearance, flat
given the fact that most of these models do not usually show
plate systems have various economic and functional advantages
high accuracy in the prediction of punching shear strength.11
over other floor systems such as fast construction, low story
height, and irregular column layout. From a viewpoint of To develop simple strength equations, most design codes
structural mechanics, however, flat plates are structures of use the so-called control perimeter approach12-15 depicted in
complex behavior. Moreover, flat plates usually fail in a Fig. 1. The applied punching shear stress is calculated at a
brittle manner by punching at the slab-column connections defined critical perimeter and compared with an allowed
within the discontinuity region known as the D-region.1,2 At value based on the calibration of existing test results. The
these connections, three-dimensional stresses are developed various design codes show significant difference in defining
due to the combined high shear and normal stresses creating the location of the critical section as well as the allowed
a stress state that is complex to analyze accurately.3 punching shear stress. It becomes apparent that the complexity
of the punching problem and the dependence of the punching
For the last three decades, a significant amount of research shear strength on a number of interacting variables necessitate
has been performed to investigate this complex problem of the use of empirical modeling approach to estimate the punching
concentric punching shear of reinforced concrete flat plates shear strength. While classical empirical techniques used by
by using various methods ranging from mechanical models many design codes show limited accuracy, a more robust
up to purely empirical models. In early models including empirical modeling technique that respects fundamental
Yitzhaki4 and Long and Rankin,5 punching shear strength failure mechanisms of the punching shear is needed.
was defined considering the flexural capacity of reinforced
concrete slabs. This was based on the experimental observation
that the punching shear strength was close to the flexural RESEARCH SIGNIFICANCE
capacities of the concrete slabs. Pralong6 and Nielsen7 The present study introduces a new approach for predicting
derived lower bound and upper bound values for punching shear the punching shear strength of concentrically loaded interior
strength based on the theory of plasticity. These formulations slab-column connections using fuzzy learning from examples.
did not consider the effect of flexural reinforcement on the The proposed approach incorporates the control perimeter
punching shear strength. Kinnunen and Nylander8 developed
the first mechanical model for punching shear strength using ACI Structural Journal, V. 104, No. 4, July-August 2007.
failure criteria based on the observation of shear cracks in the MS No. S-2006-214 received May 27, 2006, and reviewed under Institute publication
policies. Copyright © 2007, American Concrete Institute. All rights reserved, including
experiments. In this model, the failure criteria were defined the making of copies unless permission is obtained from the copyright proprietors. Pertinent
discussion including author’s closure, if any, will be published in the May-June 2008
by the inclined radial compressive stress and the tangential ACI Structural Journal if the discussion is received by January 1, 2008.

438 ACI Structural Journal/July-August 2007


ACI member Kyoung-Kyu Choi is a Research Assistant Professor at the University of
New Mexico, Albuquerque, N. Mex. He received his BE, MS, and PhD in architecture
from Seoul National University, Seoul, Korea. He is a member of ACI Committees
440, Fiber Reinforced Polymer Reinforcement; 548, Polymers in Concrete; and Joint
ACI-ASCE Committee 445, Shear and Torsion. His research interests include shear
strength and seismic design of reinforced concrete structures and application of
artificial intelligence in structural engineering.

ACI member Mahmoud M. Reda Taha is an Assistant Professor in the Department


of Civil Engineering at the University of New Mexico. He received BSc and MSc from
Ain Shams University, Cairo, Egypt, and his PhD from the University of Calgary,
Calgary, Alberta, Canada, in 2000. He is a member of ACI Committees 209, Creep
and Shrinkage in Concrete; 235, Knowledge-Based Systems and Mathematical Modeling
for Concrete Materials; 440, Fiber Reinforced Polymer Reinforcement; 548, Polymers
in Concrete; and E803, Faculty Network Coordinating Committee. His research interests
include structural monitoring, using artificial intelligence in structural modeling, and
fiber-reinforced polymers.

ACI member Alaa G. Sherif is an Associate Professor in the Civil Engineering


Department, Helwan University, Mataria-Cairo, Egypt. He received his BSc from
Cairo University, Cairo, Egypt, and his MSc and PhD from the University of Calgary.
He is a member of Joint ACI-ASCE Committee 352, Joints and Connections in Monolithic
Concrete Structures. His research interests include the behavior and serviceability of
reinforced concrete structures and systems for multi-span cable-stayed bridges.

approach and targets predicting the punching shear strength


of the slab-column connections based on various geometric
and material parameters. The proposed fuzzy-based model
presented in a simple form respects the failure mechanics of
punching shear by learning its rules from the experimental
database with the ability to address the interaction between
the modeling variables and the uncertainty in these variables.
The fuzzy-based model shows high accuracy in predicting
punching shear strength. Fig. 1—Current design codes for punching shear.

FUZZY LEARNING OF PUNCHING concrete compressive strength, slab thickness and effective
SHEAR DATABANK depth, span length, column geometry, punching shear
Fuzzy systems have been widely used in the last decade perimeter, and compression and tension reinforcement
for modeling complex engineering systems (for example, ratios. Assuming the geometry of punching shear perimeter
modeling robots16 and in assessing concrete durability17) to be known a priori, the Bayesian analysis showed that for
and their feasibility as universal approximators has been circular and square columns (c1/c2 ratio equals to 1.0), the
proven.18 The capability of the fuzzy systems to model most significant parameters that affect the punching shear
complex systems is attributed to their inherent ability to strength are concrete compressive strength fc′, slab thickness
accommodate a tolerance for uncertainty in the modeling h, and tension reinforcement ratio ρ. The assumption of the
parameters.19,20 While probabilistic empirical models are punching shear perimeter to be known a priori is based on the
limited to random uncertainties, fuzzy systems have the ability fact that the punching shear databank does not include
to consider random and nonrandom types of uncertainties detailed information about the failure pattern and the
that arise due to vagueness and/or ambiguity in the modeling punching shear perimeter. This hinders the ability to learn
parameters/process.18-20 the failure patterns of slab-column connections as part of the
The fundamental concept in modeling complex phenomena new model. It is also noted that the results of Bayesian analysis
using fuzzy systems is to establish a fuzzy rule-base that is showed that the compression reinforcement does not have a
capable of describing the relationship between the input significant effect on the maximum punching shear strength.
parameters and the output parameters while considering This finding is in agreement with the literature8,23 showing
uncertainty bounds.19 This fuzzy rule-base captures individual that the primary effect of compression reinforcement is on
and group relationships that distinguish the internal complex post-punching behavior providing a membrane action.
relations between the system parameters.20 As such, system Hereafter, these three parameters have been used as input
nonlinearity is not recognized by using nonlinear equations parameters to the fuzzy-based model for predicting the
but through establishing a number of fuzzy rules (that could punching shear strength. By considering these three parameters,
use linear relations) such that the fuzzy system becomes the fuzzy-based model considers the major criteria on punching
capable of describing the phenomena to a pre-specified level shear examined by many researchers.24-28 These include shear
of accuracy.18-20 A group of successful techniques to establish strength and cracking capacity conventionally represented
a fuzzy rule-base using exemplar observations was recently by the cubical or square root of the compressive strength,6,24,27
developed.20,21 size effect related to slab thickness,9 and membrane effect28
Here, the use of the fuzzy set theory to model the punching represented by the flexural reinforcement ratio.
shear strength of a slab-column connection is demonstrated. While the ratio of the column dimensions of rectangular
Preliminary investigations using Bayesian analysis of columns and the perimeter-to-depth ratio (bo/d) have been
significance22 have been performed to identify the most reported to affect the punching shear strength of slab-column
primary input parameters that have a significant influence on connections,24,29 the experimental database for rectangular
the punching shear strength. Possible parameters included columns or for slabs with significantly large perimeter-to-

ACI Structural Journal/July-August 2007 439


failure as observed by many researchers.6-9 The choice of the
critical perimeter to be considered at a distance d/2 from the
column face is attributed to the possible use of this location
to estimate the average ultimate shear strength vc for usually
intersecting most plausible failure planes, as shown in Fig. 2,
which is similar to the value (h/2) proposed in Nielsen.7
The modeling is started by defining N fuzzy sets A over the
domain of each input parameter x. This definition˜ provides
each value of the parameter x with N membership values
representing its level of belonging to the N fuzzy set A . The
concept of membership degree of belonging represents ˜ the
18,20,21 The
basis in the formulation of fuzzy set theory.
membership denoted μ A(x) ranges between 0.0 and 1.0.
Fig. 2—Cross section of slab-column connection showing μ A(x) does not express probability
˜ of x but characterizes the
critical section at distance d/2 from column face to intersect ˜
extent to which x belongs to fuzzy set A .20 Several methods
most plausible failure planes (angle θ ranges between 30 for establishing membership functions˜with different levels
and 45 degrees). Choice of d/2 allows obtaining good estimate of complexity exist. While simplified methods can be used
of average ultimate punching shear strength vc. according to expert opinion, complex automated methods
using artificial neural networks or inductive reasoning are
usually considered to be efficient for modeling complex
phenomena.20,30 A technique is adopted herein that is based
on providing an initial definition of the fuzzy sets using
k-means clustering31 followed by the automated update of
the fuzzy sets during the learning process.20,21
The modeling process depends on fuzzifying all three
input domains and constructing a fuzzy rule-base, which
describes the relationship between the fuzzy sets defined on
the input domains and the punching shear strength using a
group of linear equations. Exemplar rule in the fuzzy rule-
base can be defined as

k k k
If f ′c∈ A f , h ∈ A h , and ρ∈ A p , (2)
˜ ˜ ˜
then vi = ai f c′ + bih + ciρ + di
Fig. 3—Pictorial representation of bell-shaped membership
function used to represent fuzzy sets defined over input k k k
domains. where A f , A h , and A p are the k-th fuzzy set (k = 1, 2, … Nj)
defined˜ on ˜the fuzzy˜ domains of compressive strength f ′c,
slab thickness h, and tension reinforcement ratio ρ, respectively.
depth ratio (bo/d > 15) is insufficient to develop the knowledge The value of Nj is the total number of fuzzy sets defined over
rule base that is necessary for the fuzzy-based model to the j-th input parameter. In the present study, ρ is defined
consider both effects on the punching shear strength. There- with respect to effective depth. Equation (2) represents the i-th
fore, first, the fuzzy-based model is trained by using the rule in the fuzzy rule-base. The values ai, bi, ci, and di are
experimental data with square and circular columns only and known as the consequent coefficients that define the output
with perimeter to depth ratio (bo/d) < 15. Based on this fact, side of the i-th rule in the fuzzy rule-base.
prediction of the fuzzy-based model will be modified to A bell-shape membership function is employed to represent
consider the effect of rectangularity of columns or high the fuzzy sets defined on the input domains. The use of other
perimeter-to-depth ratios in excess of that used in the training membership functions (for example, gaussian and triangular) is
(bo/d > 15) as shown in the Results and discussion section. possible, but constrained by having a differentiable membership
In the present study, the punching shear failure load of function.21 The bell-shape membership function to represent
slab-column connections without shear reinforcement Vc is the k-th fuzzy set of the j-th input parameter xj can be
defined as described as μ A(x).
˜
Vc = vcbod (1)
k 1
μ A ( x j ) = -----------------------------------k (3)
˜ k 2q
where Vc equals the punching failure load and bo equals the x j – x cj j
1 + --------------- -
critical perimeter at a distance d/2 from the column face; k
bo = (2c1 + 2c2 + 4d) for a square column and bo = π(D + d) wj
for a circular column. The values c1 and c2 equal the short
and long sizes of a rectangular column, D equals the diameter k , w k, and q k represents the center, the top width,
where x cj j j
of a circular column, and vc represents the average ultimate and the shape parameters of the membership function
punching shear strength, which is defined with respect to defining the k-th fuzzy set defined over the j-th input parameter.
defective depth. Equation (1), although simplified, has been A pictorial representation of the bell-shaped membership
adopted by almost all current design codes and respects the function is shown in Fig. 3. By considering the T-norm
fundamental mechanics governing the slab-column punching (product) operator (Π) to capture the influence of the interaction

440 ACI Structural Journal/July-August 2007


between the input parameters32 on the output, the weight of Table 1—Dimensions and properties of specimens
the i-th rule (λi) in the fuzzy rule-base can be computed as No. of specimens f ′c,
Investigator*† Training Verification MPa h, mm ρ, %
T 1
Π j =1 -----------------------------------k Hallgren and Kinnunen
k 2q (1993a), Hallgren and 79.5 to 239 to 0.6 to
x j – x cj j 3 3
Kinuunen (1993b), 108.8 245 1.2
1 + --------------- - Hallgren (1996)
k
wj
λ i = ------------------------------------------------------------ for i = 1...R (4) Tomaszewicz (1993) 7 6
64.3 to 120 to 1.5 to
119.0 320 2.6
R T 1
Σ i =1 Π j =1 -----------------------------------k Ramdane (1996), 28.9 to 1.0 to
k 2q 4 4 125
x j – x cj j Regan et al. (1993) 74.2 1.3
1 + --------------- - 30.0 to 90 to 0.4 to
k Marzouk and Hussein (1991) 6 8
wj 80.0 150 2.1
Lovrovich and McLean (1990) 2 2 39.3 100 1.7
Factors affecting the choice of the implication operator are 20.1 to 120 to 0.4 to
Tolf (1988) 4 3 25.1 240 0.8
discussed in the following. The value T represents the total
number of input parameters (herein, T = 3). The number of Regan (1986) 11 11
8.4 to 80 to 0.8 to
fuzzy rules R is a function of the number of input variables T and 37.5 250 2.4
the number of fuzzy sets Nj defined over each input domain. 37.4 to 0.6 to
Swamy and Ali (1982) 1 1 125
40.1 0.7
The punching shear strength vc can then be computed as
Marti et al. (1977), 23.1 to 180 to 1.2 to
1 1
Pralong et al. (1979) 30.4 191 1.5
⎛ R ⎞ ⎛ R ⎞ 23.1 to 143 to 0.6 to
vc = ⎜
⎝i = 1

λ i v i⎟ ⁄ ⎜ λ i⎟
⎠ ⎝i = 1 ⎠
∑ (5) Schaefers (1984) 1 1
23.3 200 0.8
Ladner et al. (1977),
24.6 to 110 to 1.2 to
Schaeidt et al. (1970), 2 3
Ladner (1973) 29.5 280 1.8
where vi is the output of the i-th rule in the fuzzy rule-base
and λi represents the weight of the i-th rule in the fuzzy rule- Corley and Hawkins (1968) 1 1 44.4 146
1.0 to
1.5
base as computed using Eq. (4).
The process for learning from example aims at extracting 14.0 to 1.0 to
Bernaert and Puech (1996) 9 9 140
41.4 1.9
a knowledge rule-base from a group of input-output datasets. 24.2 to 0.5 to
This knowledge rule-base can be used later to model the Manterola (1966) 4 4 39.7 125 1.4
behavior of the system (herein the punching shear of slab- 8.6 to 0.7 to
Yitzhaki (1966) 5 6 102
column connections) for input datasets not used in the 19.0 2.0
training process. While other techniques capable of building Moe (1961) 7 7
20.5 to
152
1.1 to
similar learning systems were reported in the literature (for 35.2 2.6
example, artificial neural networks), the advantage of fuzzy Kinnunen and Nylander
6 6
21.6 to 149 to 0.5 to
(1960) 27.7 158 2.1
systems is being able to consider nonrandom uncertainty in
the modeling process and thus yields robust modeling systems. Elstner and Hognestad 9.0 to 1.2 to
8 9 152
(1956) 35.6 3.7
The learning process starts by initializing the premise
25.9 to 138 to 0.77 to
parameters (parameters describing the membership functions Hawkins et al.34 0 6
32.0 142 1.12
x kcj , w kj , and q kj ) using the k-means clustering technique.31 33.0 to
This is followed by computing the consequence coefficients Teng et al.29 0 4
40.2
150 1.24
(ai, bi, ci, and di) using least square techniques33 such that the Criswell35 0 1 35.4 146 1.24
root mean square prediction error E of the punching shear 8.4 to 80 to 0.4 to
strength does not exceed a target root mean square prediction Total 82 96
119.0 320 3.7
error, herein 1.0 × 10–5. The root mean square prediction *Reference to investigators work, unless otherwise noted, can be found in Reference 3.
error E is defined as †
Properties and dimensions of these test specimens were collected from fib Bulletin 12.3
Note: 1 MPa = 0.145 ksi; 1 mm = 0.04 in.

∑ ( vpn – vdbn ) ∂E ( m )
2
k k
w j ( m ) = w j ( m – 1 ) + η ------------------ (8)
n---------------------------------------
=1 ∂w j ( m )
E = - (6)
N
k k ∂E ( m )
where vpn is the predicted punching shear strength for the n-th q j ( m ) = q j ( m – 1 ) + η ----------------- (9)
∂q j ( m )
dataset, vdbn is the punching shear of the n-th dataset from
the database, and N is the total number of training datasets.
As the target mean square prediction error will not be where x kj (m), w kj (m), and q kj (m) are the center, the top width,
achieved from the first learning trial (using the initial fuzzy and the shape of the membership function, respectively,
sets and consequence coefficients), the premise parameters defining the k-th fuzzy set defined over the j-th input param-
describing the fuzzy sets can be updated using the gradient eter in the m-th learning epoch (trial). The values x kj (m – 1),
descent method as w kj (m – 1), and q kj (m – 1) are the center, the top width, and
the shape of the membership function, respectively, defining
the k-th fuzzy set defined over the j-th input parameter in the
k k ∂E ( m ) (m – 1) learning epoch. The value η is the learning rate and
x cj ( m ) = x cj ( m – 1 ) + η ------------------- (7)
∂x cj ( m ) ∂E(m)/∂xj(m), ∂E(m)/∂wj(m), and ∂E(m)/∂qj(m) are components

ACI Structural Journal/July-August 2007 441


or a maximum number of training epochs is reached. The
update process therefore allows the fuzzy-based model to
reduce the root mean square prediction error and thus learn
from examples in a much more robust manner compared with
any other empirical technique.
For training and testing of the fuzzy-based model, 178 test
specimens performed by 21 researchers as reported in the fib
bulletin3 and other reports in the literature29,34,35 were used.
Only specimens that were reported to fail in pure punching
shear (no flexural shear failure) were considered. A specimen
reported by Lovrovich and McLean36 was excluded in this
study because its span length was extremely short (l1/c1 = 2).
Also, six specimens by Yitzchaki,4 Elstner and Hognestad,37
and Tolf38 were also excluded because their tension rein-
forcement ratios were extremely beyond practical design
range (ρ ≥ 6.9%). The specimens had two types of boundary
geometries (circular and rectangular flat plates) and two types of
column shapes (circular and square columns). The dimensions
and properties of the specimens are summarized in Table 1.
The test specimens had a broad range of design parameters:
8.4 ≤ f′c ≤ 119.0 MPa (1.2 ≤ f ′c ≤ 17.3 ksi), 80 ≤ h ≤ 320 mm
(3.1 ≤ h ≤ 15.6 in.), 0.4 ≤ ρ ≤ 3.7%, and 5.5 ≤ bo/d ≤ 24.
These data cover a wide range of the material and geometric
properties of slab-column connections. Eighty-two specimens
were used for training of the fuzzy-based model while 96
specimens were used for testing the model. All specimens used in
Fig. 4—Fuzzy sets used to describe concrete compressive the testing were not used in training the fuzzy-based model.
strength, slab thickness, and tension reinforcement ratio.
Before training (left) and after training (right): MF1 All modeling parameters were normalized to their maximum
(Membership Function 1), MF2 (Membership Function 2), values determined from the database (178 data sets). The
and MF3 (Membership Function 3). normalization process is necessary to avoid the influence of
numerical weights on the learning process.39 The fuzzy rule-
base that achieved the lowest root mean square error during
Table 2—Parameters describing premise training was used for testing and verification of the model capa-
parameters (membership functions)* bility to predict punching shear strength in slab-column connec-
Compressive strength f ′c tions. The optimum number of fuzzy sets for each modeling
xc, MPa (ksi) w, MPa (ksi) q parameter was developed using the k-means clustering tech-
nique.31 The number of membership functions defined on the
1
Af –23.83 (–3.40) 29.9 (4.34) 1.98 domain of any variable x can be used to indicate the sensitivity
˜
2
of the model to this variable x. The higher the sensitivity of the
Af 78.30 (11.40) 78.2 (11.30) 2.02 model to the variable x, the larger the number of membership
˜
Slab thickness h functions used to describe the variable x. It is worth noting,
xc, mm (in.) w, mm (in.) q however, that increasing the number of membership functions
does not guarantee enhancing the model accuracy.20,21
1
Ah 42.05 (1.66) 68.0 (2.68) 1.982 It was found that the best learning represented by the
˜
2 lowest root mean square prediction error was achieved while
Ah 127.07 (5.00) 89.4 (3.52) 2.011
˜ using two fuzzy sets to represent the compressive strength
Ah
3
272.58 (10.73) 126.7 (4.99) 1.994
and the tension reinforcement ratio. Three fuzzy sets were
˜ necessary for describing the slab thickness (N1 = N3 = 2,
Tension reinforcement ratio ρ N2 = 3). The initial and final fuzzy sets, as established by the
xc w q learning algorithm, are shown in Fig. 4 and Table 2. The total
1 number of rules in the rule-base can be computed by
Aρ -0.001 0.012 1.997
˜ multiplying the number of membership functions of the

2
0.035 0.018 2.005
three variables as R = N1N2N3. Thus, 12 rules (R = 12) were
˜ needed to describe the relationship between the input
*For compressive strength f c′ , slab thickness h, and reinforcement ratio ρ. parameters: concrete compressive strength, slab thickness,
tension reinforcement ratio, and the punching shear strength.
of the gradient vector of the mean square prediction error While reduction of the total number of rules in the fuzzy
with respect to the premise parameters of the j-th input rule-base is possible for limiting combinatorial explosion,20
parameter evaluated at the m-th learning epoch. The updated researchers showed that the efficient reduction of the number
premise parameters are then used to recompute a new set of of rules shall be performed considering both accuracy and
consequence parameters and a new root mean square prediction robustness of the model. Exemplar methods for rule reduction
error. The process continues and the fuzzy rule-base parameters in the fuzzy rule-base include the Combs and Andrews40
(premise and consequent parameters) are updated in each method and the method suggested by Lucero41 but are
training epoch until the target root mean square prediction error beyond the scope of this work.

442 ACI Structural Journal/July-August 2007


RESULTS AND DISCUSSION Table 3—Testing to predicted punching shear
The fuzzy-based model was trained using test results with strength ratio using existing design codes and
specific geometrical limits: circular and square columns and fuzzy-based model
slabs with perimeter-to-slab-depth ratio (bo/d) ranging between Fuzzy-
5.8 and 14.9. Therefore, the punching shear strength of any CSA based
slab-column connection within the geometrical limitations ACI 318-05 CEB-FIP Eurocode 2 A23.3-04 model
VTest/ VTest/ VTest/ VTest/ VTest/
listed previously can be computed using Eq. (10) to (12).
Investigator* Vpredicted† Vpredicted† Vpredicted† Vpredicted† Vpredicted†
Equation (10) can be used to compute the weight λ for each
rule in the rule-base using the premise parameters listed in Hallgren and
Kinnunen
Table 1. Equation (11) presents the 12 rules forming the (1993a), Hallgren 0.88 to 0.83 to 0.86 to 0.93 to 0.96 to
fuzzy knowledge rule-base. and Kinuunen 0.98 1.00 0.97 1.02 1.07
(1993b), Hallgren
(1996)
3 1 Tomaszewicz 1.39 to 0.80 to 0.94 to 1.41 to 0.70 to
Π j =1 -----------------------------------k (1993) 1.64 1.17 1.29 1.64 1.23
k 2q
x j – x cj j Ramdane (1996),
1 + --------------- - Regan et al. 1.46 to 1.15 to 1.20 to 1.27 to 1.25 to
k 1.66 1.31 1.37 1.47 1.41
wj (1993)
λ i = ------------------------------------------------------------ for i = 1...12 (10)
12 3 1 Marzouk and 0.71 to 1.13 to 0.97 to 0.63 to 0.91 to
Σ i =1 Π j =1 -----------------------------------k Hussein (1991) 1.61 1.84 1.64 1.40 1.42
k 2q
x j – x cj j Lovrovich and 1.18 to 0.73 to 0.79 to 1.02 to 0.87 to
1 + --------------- - McLean (1990) 1.26 0.78 0.85 1.10 0.94
k
wj 0.88 to 0.92 to 0.82 to 0.77 to 0.91 to
Tolf (1988)
1.21 1.34 1.15 1.05 1.02
1.17 to 0.97 to 1.04 to 1.02 to 0.60 to
Regan (1986)
1 1
R = 1: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ
1 1.78 1.47 1.47 1.54 1.29
˜ ˜ ˜ Swamy and Ali
then v 1 = 0.247f c ′ + 0.008h + 153.7ρ + 4.90 (1982)
1.10 1.19 1.12 0.96 1.00
1 1 2
R = 2: if f c ′ ∈ A , h ∈ A , and ρ ∈ A
f h ρ Marti et al. (1977),
˜ ˜ ˜ Pralong et al. 1.32 0.97 1.00 1.15 0.77
then v 2 = – 0.506f c ′ + 0.026h + 835.4ρ – 11.42
(1979)
1 2 1
R = 3: if f c ′ ∈ A , h ∈ A , and ρ ∈ A
f h ρ Schaefers (1984) 1.19 1.14 1.05 1.04 1.00
˜ ˜ ˜
then v 3 = 0.174f c ′ + 0.028h + 63.9ρ – 8.12 Ladner et al.
1 2
R = 4: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ
2 (1977), Schaeidt 1.48 to 1.22 to 1.26 to 1.29 to 0.89 to
˜ ˜ ˜ et al. (1970), 1.79 1.34 1.47 1.56 1.26
then v 4 = 0.149f c ′ + 0.031h – 136.65ρ – 3.49 Ladner (1973)
1 3 1
R = 5: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ Corley and
0.87 0.85 0.85 0.75 0.72
˜ ˜ ˜ Hawkins (1968)
then v 5 = – 0.248 f c ′ + 0.001h – 236.32ρ + 3.91
1 3 2
Bernaert and 0.88 to 0.80 to 0.81 to 0.76 to 0.70 to
R = 6: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ Puech (1996) 1.93 1.28 1.43 1.68 1.45
˜ ˜ ˜
then v 6 = 0.243f c ′ – 0.006h – 53.35ρ + 3.16 (11) 0.88 to 0.81 to 0.85 to 0.76 to 0.65 to
Manterola (1966)
2 1 1 1.36 0.96 0.98 1.18 0.92
R = 7: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ
˜ ˜ ˜ 1.51 to 1.01 to 1.01 to 1.31 to 0.80 to
then v 7 = – 0.005 f c ′ – 0.031h – 84.38ρ + 3.14 Yitzhaki (1966)
1.98 1.54 1.53 1.72 1.16
2 1 2
R = 8: if f c ′ ∈ A , h ∈ A , and ρ ∈ A
f h ρ 0.70 to 0.83 to 1.07 to 0.68 to
˜ ˜ ˜ Moe (1961) 1.24 to 1.65
then v 8 = 0.006f c ′ + 0.116h – 136.57ρ – 2.67 1.38 1.40 1.43 1.12
2 2 1 Kinnunen and 0.83 to 0.93 to 0.92 to 0.72 to 0.85 to
R = 9: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ Nylander (1960) 1.75 1.23 1.23 1.52 1.36
˜ ˜ ˜
then v 9 = 0.006f c ′ – 0.002h – 30.73ρ + 0.05 Elstner and 1.19 to 0.88 to 1.05 to 1.03 to 0.79 to
2 2
R = 10: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ
2 Hognestad (1956) 2.23 1.20 1.30 1.94 1.27
˜ ˜ ˜ 0.90 to 0.87 to 0.89 to 0.78 to 0.88 to
then v 10 = 0.021f c ′ + 0.043h + 19.19ρ – 7.58 Hawkins et al.34 1.05 1.10 1.05 0.91 1.19
2 3 1
R = 11: if f c ′ ∈ A , h ∈ A , and ρ ∈ A
f h ρ 0.88 to 0.89 to 0.92 to 0.76 to 1.02 to
˜ ˜ ˜ Teng et al.29 1.15 1.15 1.19 1.00 1.49
then v 11 = 0.001f c ′ – 0.006h + 49.86ρ + 1.96
2 3
R = 12: if f c ′ ∈ A f , h ∈ A h, and ρ ∈ A ρ
2 Criswell35 0.94 0.89 0.96 0.82 0.86
˜ ˜ ˜ Mean 1.375 1.098 1.139 1.219 1.019
then v 12 = 0.004f c ′ + 0.018h + 36.19ρ – 6.36
Standard deviation 0.314 0.207 0.198 0.280 0.189
*Reference to investigators work, unless otherwise noted, can be found in Reference 3.
where vi, f ′c, and h are in MPa, MPa, and mm, respectively. †
Strength ratio (= VTest/Vpredicted), where VTest equals actual strengths (test results),
The punching shear strength vcf can be computed using Eq. (11) and Vpredicted equals predicted strengths by current design methods (ACI 318-05,
and (12) CEB-FIP, Eurocode 2, and CSA A23.3-04) or fuzzy-based model, respectively.

⎛ 12 ⎞ ⎛ 12 ⎞ of information, the influence of the fused output on the


v cf = ⎜
⎝i = 1

λ i v i⎟ ⁄ ⎜ λ i⎟
⎠ ⎝i = 1 ⎠
∑ (12) model prediction, and the effect of the fusion method on the
computational efficiency of the learning algorithm. The
product implication was selected herein for three reasons.
It is important to emphasize the fact that several implication First, to perform the fuzzy AND operation as indicated by
operators exist.42 The selection of the implication operator is Eq. (4). Second, the product implication tends to dilute the
governed by three main issues: the needed logical implication influence of joint membership values that are small and

ACI Structural Journal/July-August 2007 443


Fig. 5—Strength prediction by current design method and fuzzy-based model. (Note: 1 MPa =
0.145 ksi; 1 mm = 0.04 in.)

therefore magnify the contribution of the rules associated product operator in artificial neural networks as an efficient
with high membership values in computing the shear Hebbian-type learning algorithm.20 Finally, the choice of the
strength (Eq. (4) and (5)). This fact promoted the use of the product implication was also controlled by the need to

444 ACI Structural Journal/July-August 2007


produce a continuous and differentiable error function (Eq. (6))
to enable efficient computation of the error gradients during
the learning process.
Table 3 presents a summary of punching shear strength of
the specimens predicted by the fuzzy-based model. In the
verification, the 96 specimens, which were not used in the
learning process, were used. Figure 5(a) shows the ratios
between the actual test to the fuzzy-based model predicted
strength (Vtest/Vpredicted) to have a mean value 1.019 and a
standard deviation of 18.9%. Figures 5(b) to (e) show the
ratios between actual to predicted strength (Vtest/Vpredicted)
using the CEB-FIP MC 90,12 the Eurocode 2,13 ACI 318-05,14
and CSA A23.3-04,15 to have mean values of 1.098, 1.139, Fig. 6—Variation of strength-prediction by fuzzy-based
1.375, and 1.219, respectively, with standard deviations of model according to bo/d.
20.7, 19.8, 31.4, and 28.0%, respectively (refer to Table 3).
The results show that the fuzzy-based model can be used to
predict the punching shear strength of slab-column connections
with various slab thicknesses, reinforcement ratios, and
circular and square columns. Moreover, higher prediction
accuracy of the fuzzy-based model can be observed compared
with predication accuracies for all existing design codes.
It is interesting to note that, except for Eurocode 2,13 current
design methods show a considerable scatter represented by
high standard deviations of test-prediction ratios. Moreover,
observing Fig. 5(c), the CEB-FIP MC 90 code underestimates
the punching shear strength of specimens with low tension
reinforcement ratios while it overestimates the punching
shear strength of specimens with high tension reinforcement
ratios. The Eurocode 213 shows good accuracy in predicting Fig. 7—Variation of strength-prediction by fuzzy-based
the punching shear strength at different reinforcement ratios. model according to c2/c1 higher than 1.
Finally, ACI 318-0514 and CSA A23.3-0415 underestimate the
punching shear strength of specimens with high reinforcement n = 1.0 as similar to the ACI equation is used, the model will
ratios while they overestimate the punching shear strength of significantly overestimate the punching slab-column
specimens with low reinforcement ratios. This is attributed connections with rectangular columns and with bo/d higher
to the fact that ACI 318-05 and CSA A23.3-04 codes do not than 15. A mean value and a standard deviation of the
account for the effect of the tension reinforcement ratio on the strength-prediction ratios (Vtest/Vpredicted) of the specimens
punching shear strength. It is also evident from Fig. 5(a) that (Table 3) using n = 1 are 0.977 and 0.193, respectively, while
the fuzzy-based model predicts punching shear strength at those using n = 2 are 1.019 and 0.189. Therefore, the authors
both low and high reinforcement ratios with consistent accu- recommend the use of n = 2. The model prediction with n = 2
racy. It is worth noting that the two parameters in addition to for a wide range of bo/d and for rectangular columns are
the compressive strength found to have a significant influ- shown in Fig. 6 and 7. The choice of n = 2 for the second and
ence on modeling punching shear strength using the fuzzy- third components of Eq. (13) was based on examining each
based model (the slab thickness and the tension reinforce- component separately. It has become evident that refinement
ment ratio) have also been promoted by other researchers in the value of n for each part would not yield any enhancement
before because of their influence on the size effect43 and their in the prediction accuracy of the model.
possible role in developing shear friction.44 Figure 6 demonstrates the fact that the modified fuzzy-
To consider other rectangularity ratios c2/c1 (>1) and high based model using a modification factor (Eq. (13)) can
perimeter to depth ratios bo/d (>15.0), a design approach accurately predict the punching shear strength of slab-
based on the fuzzy-based model is proposed as column connections with various bo/d (5.8 ≤ bo/d ≤ 24.0)
even though the fuzzy-based model (Eq. (12)) was developed
⎧ within the geometrical limits (5.8 ≤ bo/d ≤ 14.9) due to the
⎪ v cf lack of test data. This is attributed to the fact that the fuzzy-
⎪ ⎛ 0.5 + ---- 1⎞ based model was developed by using the average ultimate
⎪ ⎝
- v
n⎠ cf shear strength vc considering bo and d (Eq. (1)). It is evident
v c = min ⎨ βc (13)
⎪ that the modified fuzzy-based model can properly consider
⎪ ⎛ ⎛ 10 -⎞ n⎞ v the interaction between bo/d and vc in its strength equation
⎪ ⎝ 0.5 + ⎝ b----------- ⎠ ⎠ cf
o⁄d
(Eq. (13)). In Fig. 7, the fuzzy-based model also accurately

predicts the punching shear strength of slab-column connections
with rectangular columns (c2/c1 > 1). From this result, it is
where βc = c2/c1, c1 and c2 equal the short and long sizes of noted that the modified fuzzy-based model properly considers
rectangular columns, vcf is the fuzzy-based shear strength the effect of rectangularity of columns in practical design range
estimated using Eq. (12), and n is a power coefficient. (1 ≤ c2/c1 ≤ 5). It is worth noting that, if enough experimental
Equation (13) is modeled in a format similar to that of the data with high bo/d ratios and rectangular columns were
ACI equation for predicting the punching shear strength. If available in the literature, the use for modification factors for

ACI Structural Journal/July-August 2007 445


for a wide range of primary design parameters: 20 ≤ f ′c ≤
100 MPa (2.9 ≤ f ′c ≤ 14.5 ksi), 100 ≤ h ≤ 300 mm (3.9 ≤ h ≤
11.8 in.), and 0.8 ≤ ρ ≤ 2.0%. For space limitations, only four
design charts are developed herein covering the aforemen-
tioned range of parameters. Additional design charts can be
developed using the model equations described previously.
The φ factor of 0.6 corresponds conservatively to the lowest
bound shown in Fig. 5(a). Obtaining a refined shear strength
reduction factor (higher than 0.6) can be done using principles
of load and resistance factor design (LRFD),45 but is beyond
the scope of this study.
It can be observed from Fig. 8(a) to (d) that the punching
shear strength decreases as slab thickness increases, which
respects previous findings of the size effect by Bažant and
Cao9 and Eurocode 2.13 In cases with high reinforcement
ratios, however, this size effect is disturbed by the combined
effect of size and membrane force generated by the tension
reinforcement. As observed in Fig. 8(c) and (d), for high
tension reinforcement ratios and low concrete compressive
strength, the punching shear strength increases as the slab
Fig. 8—Design chart for punching shear strength using thickness increases. This can be attributed to the possibility
fuzzy-based model. (Note: 1 MPa = 0.145 ksi; 1 mm = 0.04 in.) that the increase in the slab thickness with high reinforcement
ratios results in an increase in the axial membrane
force,24,26,27 which contributes to punching shear strength
due to the increase in the shear friction effect.44 This
possible shear friction contribution to the punching shear
strength has been argued by other researchers in classical
shear analysis.44,46
This phenomenon is due to the combined effect of the
primary parameters (compressive strength, slab thickness,
and tension reinforcement ratio) and can be also observed in
previous test results from the punching shear database.3
Figure 9 shows the punching shear strength reported in
Fig. 9—Strength variation according to primary design existing test results. For this study, Elstner and Hognestad,37
parameters.25,37,44-47 Shaeidt el al.,47 Regan,48 Marzouk and Hussein,25 Hallgren
and Kinnunen,49 and Tomaszewicz’s50 specimens were
used. Each data set itself has similar dimension and property.
addressing these issues can be completely omitted. This The dimensions and properties of the specimens are
indicates the fact that a refined fuzzy-based model would summarized in Table 1. As expected, for all data sets with
always be possible to develop, once experimental data high concrete compressive strength, the punching shear
beyond these geometrical limitations becomes available. strength of thick slabs is always less than that of thin slabs
due to the size effect24,38,43 (see Fig. 9(a)). In Fig. 9(b),
PROPOSED DESIGN CHART however, for low concrete compressive strength and high
For design purposes, the direct use of the fuzzy-based reinforcement ratios (ρ ≥ 0.012), the punching shear strength
model as an empirical method using Eq. (10) to (13) and the of thick slabs may be greater than that of thin slabs, which
premise parameters from Table 2 might not be feasible for indicates the trade-off between size effect and shear friction
designers. To avoid such complexity and to make use of the effect. These combined effects can be successfully described
demonstrated ability and relative high accuracy of the fuzzy- by the fuzzy-based model.
based model in design of slab-column connections without
shear reinforcement, the authors suggest a simplified design CONCLUSIONS
model that is developed based on a set of design charts that A new alternative design method and a set of design charts
are developed using the fuzzy-based model. Following a based on fuzzy learning from examples are proposed. The
format similar to that used in ACI 318-05, the design new method can accurately predict the punching shear
strength for punching shear of slab-column connections is strength of simply supported interior slab-column connections
defined as without shear reinforcement. One hundred and seventy eight
test specimens from the punching shear databank were used
φVn = φvcbod (14) for training and testing the proposed model (82 for training
and 96 for testing). The training and testing data sets cover a
where vc is calculated according to Eq. (13) using n = 2, and wide range of the material and geometric properties. The
φ is the strength reduction factor taken equal to 0.6. The testing data set was not used in the training process. Investi-
punching shear strength vcf can be estimated using Fig. 8. gations for developing a model with good accuracy showed
Figures 8(a) to (d) show a group of design charts to estimate that concrete compressive strength, slab thickness, and
the punching shear strength vcf of slab-column connections tension reinforcement ratio are the primary parameters that
using the fuzzy-based model. The design charts are developed dominate the punching behavior of slab-column connections.

446 ACI Structural Journal/July-August 2007


This finding is limited to circular and rectangular columns Connections,” ACI Structural Journal, V. 89, No. 6, Nov.-Dec. 1992,
and slabs with perimeter-to-slab-depth ratios (bo/d) ranging pp. 626-638.
24. Sherif, A. G., and Dilger, W. H., “Critical Review of the CSA A23.3-94
between 5.8 and 24.0 and column size ratios (c2/c1) ranging for Punching Shear Strength Provisions for Interior Columns,” Canadian
between 1.0 and 5.0. The fuzzy-based model demonstrates Journal of Civil Engineering, V. 23, 1996, pp. 998-1011.
higher prediction accuracy compared with all current design 25. Marzouk, H., and Hussein, A., “Experimental Investigation on the
codes including ACI 318-05, Eurocode 2, CEB-FIP MC 90, Behavior of High-Strength Concrete Slabs,” ACI Structural Journal, V. 88,
No. 6, Nov.-Dec. 1991, pp. 701-713.
and CSA A23.3-04 in predicting the punching shear strength
26. Hawkins, N. M., and Mitchell, D., “Progressive Collapse of Flat Plate
of slab-column connections. The proposed model, while Structure,” ACI JOURNAL, Proceedings V. 76, No. 7, July 1979, pp. 775-808.
addressing uncertainty and interactions between modeling 27. Regan, P. E., and Braestrup, M. W., “Punching Shear in Forced
parameters, was shown to respect the fundamental mechanics Concrete: A State of the Art Report,” Bulletin d’information, Comité Euro-
of punching shear as described by many researchers. International du Béton, Lausanne, Switzerland, Jan. 1985.
28. Rankin, G. I. B., and Long, A. E., “Predicting the Enhanced Punching
Strength of Interior Slab-Column Connections,” Proceedings of the Institution
ACKNOWLEDGMENTS of Civil Engineers, V. 82, 1987, pp. 1165-1186.
The financial support by the Defense Threat Reduction Agency (DTRA) 29. Teng, S.; Cheong, H. K.; Kuang, K. L.; and Geng, J. Z., “Punching
University Strategic Partnership to the University of New Mexico is greatly Shear Strength of Slabs with Openings and Supported on Rectangular
appreciated. Columns,” ACI Structural Journal, V. 101, No. 5, Sept.-Oct. 2004, pp. 678-687.
30. Laviolette, M.; Seaman, J. W.; Barrett, J. D.; and Woodall, W. H., “A
REFERENCES Probabilistic and Statistical View of Fuzzy Methods,” Technometrics, V. 37,
1. Schlaich, J.; Schäfer, K.; and Jennewein, M., “Toward a Consistent 1995, pp. 249-261.
Design of Structural Concrete,” Journal of the Prestressed Concrete Institute, 31. Duda, R. O.; Hart, P. E.; and Stork, D. G., Pattern Classification, 2nd
V. 32, No. 3, 1987, pp. 74-150. Edition, John Wiley & Sons, 2001.
2. Alexander, S. D. B., and Simmonds, S. H., “Ultimate Strength of 32. Gupta, M. M., and Qi, J., “Theory of T-Norms and Fuzzy Inference
Slab-Column Connections,” ACI Structural Journal, V. 84, No. 3, May- Methods,” Fuzzy Sets and Systems, V. 40, No. 3, 1991, pp. 431-450.
June 1987, pp. 255-261. 33. Fan, J. Y., and Yuan, Y. X., “On the Convergence of a New Levenberg-
3. CEB-FIP Task Group, “Punching of Structural Concrete Slabs,” fib Marquardt Method,” Technical Report, AMSS, Chinese Academy of Sciences,
12, Lausanne, Switzerland, 2001, 314 pp. 2001, 11 pp.
4. Yitzhaki, D., “Punching Strength of Reinforced Concrete Slabs,” ACI 34. Hawkins, N. M.; Fallsen, H. B.; and Hinojosa, R.C., “Influence of
JOURNAL , Proceedings V. 63, No. 5, May 1966, pp. 527-542. Column Rectangularity on the Behaviour of Flat Plate Structures,” Cracking,
5. Long, A. E., and Rankin, G. I. B., “Prediction of the Punching Deflection and Ultimate Load of Concrete Slab Systems, SP-30, American
Strength of Conventional Slab-Column Specimens,” Proceedings of the Concrete Institute, Farmington Hills, Mich., 1971, pp. 127-146.
Institution of Civil Engineers, V. 82, 1987, pp. 327-345. 35. Criswell, M. E., “Strength and Behaviour of Reinforced Concrete
6. Pralong, J., “Poinçonnement Symétrique des Plachers-Dalles,” IBK- Slab-Column Connections Subjected to Static and Dynamic Loading,”
Bericht No. 131, Insitut für Baustatik und Konstruktion, ETH Zürish, 1982. Technical Report N-70-1, U.S. Army Engineer Waterways Experiment Station,
7. Nielsen, M. P., Limit Analysis and Concrete Plasticity, 2nd Edition, Vicksburg, Va., Dec. 1970.
CRC Press, N. Y., 1999. 36. Lovrovich, J., and McLean, D., “Punching Shear Behaviour of Slab
8. Kinnunen, S., and Nylander, H., “Punching of Concrete Slabs without with Varying Span-Depth Ratios,” ACI Structural Journal, V. 87, No. 5,
Shear Reinforcement,” Transactions No. 158, Royal Institute of Technology, Sept.-Oct. 1990, pp. 507-511.
Stockholm, Sweden, 1960, 112 pp. 37. Elstner, R. C., and Hognestad, E., “Shearing Strength of Reinforced
9. Bažant, Z. P., and Cao, Z., “Size Effect in Punching Shear Failure of Concrete Slabs,” ACI JOURNAL, Proceedings V. 53, No. 7, July 1956,
Slabs,” ACI Structural Journal, V. 84, No. 1, Jan.-Feb. 1987, pp. 44-53. pp. 29-58.
10. Birkle, G., “Flat Slabs: The Influence of the Slab Thickness and the 38. Tolf, P., “Plattjocklekens Inverkan På Betongplattors Hållfasthet vid
Stud Layout,” PhD dissertation, Department of Civil Engineering, The Genomstansning,” Försök med Cikulåra Plattor, TRITA-BST Bull, 146,
University of Calgary, Calgary, Alberta, Canada, 2004. KTH Stockholm, Sweden, 1988, 64 pp.
11. Theodorakopoulos, D. D., and Swamy, R. N., “Ultimate Punching 39. Berenji, H. R., and Khedkar, P., “Learning and Tuning Fuzzy Logic
Shear Strength Analysis of Slab-Column Connections,” Cement and Concrete Controllers Through Reinforcements,” IEEE Transactions on Neural
Composites, V. 24, 2002, pp. 509-521. Networks, V. 3, No. 5, 1992, pp. 724-740.
12. CEB-FIP MC 90, “Design of Concrete Structures,” CEB-FIP Model 40. Combs, W. E., and Andrews, J. E., “Combinatorial Rule Explosion
Code 1990, Thomas Telford, 1993, 437 pp. Eliminated by a Fuzzy Rule Configuration,” IEEE Transactions on Fuzzy
13. EC 2, “Design of Concrete Structures Part I: General Rules and Rules Systems, V. 6, No. 1, pp. 1-11.
for Buildings,” European Committee for Standardization, Brussels, 2002. 41. Lucero, J., “Fuzzy Systems Methods in Structural Engineering,” PhD
14. ACI Committee 318, “Building Code Requirements for Structural dissertation, University of New Mexico, Department of Civil Engineering,
Concrete (ACI 318-05) and Commentary (318R-05),” American Concrete Albuquerque, N. Mex. 2004.
Institute, Farmington Hills, Mich., 2005, 443 pp. 42. Yager, R., “On a General Class of Fuzzy Connectives,” Fuzzy Sets
15. CSA Technical Committee on Reinforced Concrete Design, “A23.3-04 and Systems, V. 4, 1980, pp. 235-242.
Design of Concrete Structures,” Canadian Standards Association, Rexdale, 43. Bažant, Z. P., Fracture and Size Effect in Concrete and Other Quasi
Ontario, Canada, 2004. Brittle Materials, CRC Press, N. Y., 1997.
16. Chatterjee, A., and Watanabe, K., “An Adaptive Fuzzy Strategy for 44. Loov, R. E., “Review of A23.3-94 Simplified Method for Shear
Motion Control of Robot Manipulators,” Soft Computing, V. 9, No. 3, Design and Comparison with Results Using Shear Friction,” Canadian
2005, pp. 185-193. Journal of Civil Engineering, V. 25, No. 3, 1998, pp. 437-450.
17. Zongjin, L.; Chau, C. K.; and Zhou, X., “Accelerated Assessment 45. Nowak, A. S., “Calibration of LRFD Bridge Code,” Journal of
and Fuzzy Evaluation of Concrete Durability,” Journal of Materials in Structural Engineering, ASCE, V. 121, No. 8, 1995, pp. 1245-1251.
Civil Engineering, ASCE, V. 17, No. 3, 2005, pp. 257-263 46. Kani, G. N. J., “The Riddle of Shear Failure and Its Solutions,” ACI
18. Kosko, B., “Fuzzy Systems as Universal Approximators,” IEEE JOURNAL, Proceedings V. 61, No. 4, Apr. 1964, pp. 441-468.
Trans. Comp, V. 43, No. 11, 1993. 47. Schaeidt, W.; Ladner, M.; and Rösli, A., “Berechnung von Flach-
19. Klir, G. J., Uncertainty and Information: Foundations of Generalized decken auf Durchstanzen,” Eidgenössische Materialprüfungs-und Versuch-
Information Theory, John Wiley and Sons, N. J., 2006. sanstalt, Dübendort, 1970.
20. Ross, T. J., Fuzzy Logic with Engineering Applications, 2nd Edition, 48. Regan, P., “Symmetric Punching of Reinforced Concrete Slabs,”
Wiley & Sons, UK, 2004. Magazine of Concrete Research, V. 38, 1986, pp. 115-128.
21. Jang, J. S. R.; Su, C. T.; and Mizutani, E., Neuro-Fuzzy and Soft 49. Hallgren, M., and Kinnunen, S., “Punching Shear Tests on Circular
Computing, A Computational Approach to Learning and Machine Intelligence, High Strength Concrete Slabs,” Utilization of High Strength Concrete,
Prentice Hall, N. J., 1997. Proceedings, Lillehammer, 1993.
22. Carlin, B. P., and Chib, S., “Bayesian Model Choice via Markov 50. Tomaszewicz, A., “High-Strength Concrete: SP2-Plates and Shells—
Chain Monte Carlo Methods,” Journal of the Royal Statistical Society, Report 2.3,” Punching Shear Capacity of Reinforced Concrete Slabs,
Series B, V. 57, No. 3, 1995, pp. 473-484. Report No. STE70 A93082, SINTEF Structures and Concrete, Trondheim,
23. Pan, A. D., and Moehle, J. P., “An Experimental Study of Slab-Column 1993, 36 pp.

ACI Structural Journal/July-August 2007 447


View publication stats

You might also like