Thin-Walled Structures 41 (2003) 329–355

www.elsevier.com/locate/tws

A concise introduction to the idealized

structural unit method for nonlinear analysis of
large plated structures and its application
J.K. Paik a,∗, A.K. Thayamballi b
a
Department of Naval Architecture and Ocean Engineering, Pusan National University, Busan 609-
735, South Korea
b
Chevron Shipping Company LLC, Camino Ramon, San Ramon, CA 94583, USA

Received 10 December 2001; received in revised form 29 October 2002; accepted 4 December 2002

Abstract

The idealized structural unit method (ISUM) has now been widely recognized by researchers
as an efficient and accurate methodology to perform nonlinear analysis of large plated struc-
tures such as ships, offshore platforms, box girder bridges or other steel structures. This paper
presents a summary of pertinent ISUM theory and its application to nonlinear analysis of steel
plated structures. Important concepts for development of various ISUM units which are needed
to analyze nonlinear behavior of steel plated structures are described. Some application
examples are shown, wherein comparisons of ISUM analysis predictions are made with
numerical or experimental results for progressive collapse analysis of general types of steel
plated structures and ship hulls, to illustrate the possible accuracy and versatility of the ISUM
method. The use of ISUM for the analysis of internal collision/grounding mechanics of ships
is also illustrated. This paper is in part an attempt to demystify ISUM and its applications for
the benefit of a designer of steel plates structures (Paik and Thayambali, Ultimate limit state
design of steel plated structures; 2002).
 2003 Elsevier Science Ltd. All rights reserved.

Keywords: Idealized structural unit method (ISUM); Nonlinear finite element method; Steel plated struc-
tures; Nonlinear behavior; Ultimate strength; Internal accident mechanics; Collision; Grounding

∗
Corresponding author. Tel.: +82-51-510-2429; fax: +82-51-512-8836.
E-mail address: jeompaik@pnu.edu (J.K. Paik).

0263-8231/03/$ - see front matter  2003 Elsevier Science Ltd. All rights reserved.
doi:10.1016/S0263-8231(02)00113-1
330 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

1. Features of the idealized structural unit method

Under extreme or accidental loading, steel structures can exhibit highly nonlinear
response associated with yielding, buckling, crushing and sometimes rupture of indi-
vidual structural components. Quite accurate solutions of the nonlinear structural
response in many such cases can be obtained by application of the conventional
finite element method (FEM). However, a weak feature of the conventional FEM is
that it requires enormous modeling effort and computing time for nonlinear analysis
of large sized structures. Reduction in that modeling effort and the associated solution
time while also providing an acceptable level of accuracy in results is the primary
benefit of the idealized structural unit method.
The most obvious way to reduce modeling effort and computing time in FEM
applications is to reduce the number of degrees of freedom so that the number of
unknowns in the finite element stiffness matrix decreases. Modeling the object struc-
ture with very large sized structural units is perhaps the best way to do that. In order
to avoid the related loss of accuracy, ISUM requires the use of special purpose finite
elements. Properly formulated structural units within such an approach can then be
used to efficiently model the actual nonlinear behavior of the corresponding large
parts of structures.
Ueda and Rashed [1,2], who suggested this idea, called it the idealized structural
unit method (ISUM). Their first effort in this regard was to analyze the ultimate
strength of a transverse framed structure of a ship as an assembly of the so-called
deep girder units.
In an almost parallel development to ISUM, Smith [3] suggested a similar
approach to predict the ultimate bending moment of a ship hull. He modeled the
ship hull as an assembly of plate-stiffener combinations, i.e., stiffeners with attached
plating. The load versus end-shortening relationships for these beam–column units
were obtained using conventional nonlinear FE analysis accounting for initial imper-
fections (defections and residual stresses). The behavior of the larger structure was
then constructed. While this method is quite properly called the Smith method, it
may also be classified as a type of ISUM in the particular context for analysis of a
structure made up of repetitive structural elements.
ISUM is a simplified nonlinear FEM. Unlike the conventional nonlinear FEM,
ISUM idealizes a structural component making up the structure as one ISUM unit
with a few nodal points. As a typical example, Fig. 1 compares structural modeling
for the elastic–plastic large deflection analysis of a rectangular plate under a reason-
ably complex load application, as required by the conventional FEM and ISUM for
comparable accuracy in results. In conventional FEM, finer meshes would normally
be used, while on the other hand the plate is modeled using a single ISUM plate
unit for ISUM analysis. It is noted that in this example the entire plate is taken as
the extent of FE analysis because of the unsymmetric characteristics related to defor-
mations.
Of interest, the relative computational efforts required for the nonlinear analysis
of a structure (conventional FEM versus ISUM) are now further illustrated, with
definitions as follows
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 331

Fig. 1. Sample comparison of FEM and ISUM models for elastic–plastic large deflection analysis of a
rectangular plate of a / b = 3 under a complex load application. (a) FEM model: no. of nodes=341, degrees
of freedom=6 per node, no. of unknowns=1874. (b) ISUM model: no. of nodes=4, degrees of freedom=3
per node, no. of unknowns=6.

앫 Ratio of number of unknowns: m = FEM / ISUM.

앫 Ratio of number of iterative loading steps: n = FEM / ISUM.
앫 Ratio of computing time: CPU = FEM / ISUM = n × mc, with c = 1 苲 3.

For the plate shown in Fig. 1, it was found that m⬇300, n = 5 苲 10 and CPU
ratio⬇5 × 103 苲 104. For larger sized structures with more complex geometries, the
relative discrepancy due to the number of elements between the two methods will
become much more significant as well. By applying ISUM, therefore, the size of
numerical computations is much reduced leading to dramatic savings of modeling
and computing times, compared to those of the conventional FEM. One trade-off is
of course the possible loss of generality of the method, in the sense that ISUM must
use specially formulated finite elements that are different, and internally supply more
high level on structural component behavior than the elements typical of conven-
tional FEM.
Among other matters, this paper presents a concise review of the most updated
theory of the ISUM units which are available today for the nonlinear analyses of
steel plated structures subjected to extreme or accidental loading. For a more detailed
description of the theory of ISUM units and applications presented in this paper, the
interested reader is referred to Paik and Thayamballi [4].
332 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

2. ISUM modeling technologies for steel plated structures

A steel plated structure may be modeled as an assembly of many simpler mechan-

ical structural elements or idealized elements, each type of which behaves in a known
manner under given load application, and the assembly of which is constructed to
behave in nearly the same way as the actual structure.
Steel plated structures are typically composed of several different types of struc-
tural members such as support members (or girders), rectangular plates and stiffened
panels. In ISUM modeling, such members are regarded as the ISUM units. It is
important to realize that an identical structure may be modeled in somewhat different
ways by different analysts even within the context of ISUM, but it is of course
always the aim to model so that the idealized structure behaves in nearly the same
way as the actual structure.
Fig. 2 shows some typical examples of ISUM modeling for steel plated structures.
One of the common approaches is to model the structure as an assembly of plate–

Fig. 2. Various types of idealizations for a steel plated structure. (a) A typical steel plated structure. (b)
Structural idealization by an assembly of plate-stiffener combination units. (c) Structural idealization by an
assembly of plate-stiffener separation units. (d) Structural idealization by an assembly of stiffened panels.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 333

stiffener combinations as shown in Fig. 2(b). This type of modeling may for example
not be very appropriate when the stiffeners are relatively weak or unusually strong.
In such cases, other types of modeling could be more relevant. For instance, the
plate–stiffener separation models may be used, where plating between stiffeners is
modeled as one ISUM plate element, while the stiffener without attached plating is
modeled as one ISUM beam–column element, as shown in Fig. 2(c).
An entire panel together with stiffeners in either one or both directions can also
be modeled by one ISUM stiffened panel unit as shown in Fig. 2(d). While a small
stiffener with or without attached plating may be modeled as the beam–column unit,
the girder or support member with deep web may need to be modeled as an assembly
of the plate unit and the beam–column unit, the former being for the web and the
latter being for the flange.
In ISUM analyses, the load and boundary conditions are applied in a similar way
to the conventional FEM. The initial imperfections (initial deformations and residual
stresses) and structural degradation (e.g., effect of corrosion or cracks) as well as
geometric and material properties of the ISUM units can be prescribed as parameters
of influence as well.

3. Procedure for developments of ISUM units

For the nonlinear analysis of complex steel plated structures, it is apparent that
various types of the ISUM units are necessary to make a complete structural model,
and that these units should then be developed in advance. Fig. 3 represents the usual
procedure to develop the ISUM units. The nonlinear behavior of each type of struc-
tural members is idealized and expressed in the form of a set of failure functions
defining the necessary conditions for the failure modes (e.g., yielding, buckling,
cracking) which may take place and are described by the corresponding ISUM unit.

Fig. 3. Procedure for the development of an Idealized Structural Unit.

334 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

A set of stiffness matrices representing the nonlinear relationship between the

nodal force vector and the nodal displacement vector are then formulated until or
after the particular limit state is reached. The post-failure phase is of interest, for
instance, in assessing post-limit state behavior, which one may want to do, say, in
determining the ultimate strength of a complex structure with one or more failed
elements. The ISUM units (elements) so developed are used within the framework
of the nonlinear matrix displacement procedure applying the incremental method,
much like in the case of conventional FEM.
Many types of the ISUM units have so far been developed for the nonlinear analy-
ses of steel plated structures. For the purposes of ultimate strength analyses, the deep
girder unit [1,2], the tubular beam–column unit [5], the I-section beam–column unit
[6], the rectangular plate unit [7–12] and the stiffened panel unit [5,7–9] are available.
For analysis of the internal mechanics in collisions and grounding, the rectangular
plate unit, the stiffened panel unit and the gap/contact unit have been developed [13].
It is important to realize that individual developers of the ISUM units may employ
somewhat different approaches from each other to idealize and to formulate the actual
nonlinear behavior of the structural members. Also, the features of existing ISUM
units may be advanced continually by their developers to accommodate more factors
of influence or to improve the computational accuracy. Even though the geometric
features of the ISUM units are identical, their characteristics or usage in terms of
structural behavior may be different from the purpose of the analyses, e.g., for ulti-
mate strength or accidental mechanics.
ISUM has been successfully applied to the nonlinear analysis of ship structures
in overload situations, e.g., [13–15,27] among others and in collision or grounding
accidents, e.g., [16–19] among others, and also for other marine structures [20,21].
It is also noted that ISUM can be readily applied to the nonlinear analysis of land
based structures such as box-girder bridges or cranes. In the following sections,
ISUM formulations for many types of units that are presently available to a designer
for the nonlinear analyses of steel plated structures are summarized.

4. The beam–column unit

Fig. 4 shows the ISUM beam–column unit which may be used by a designer
with or without attached plating, the former being typically called the plate–stiffener
combination model, i.e., a stiffener with attached plating. This unit has two nodal
points, i.e., node number 1 at the left end and node number 2 at the right end.
A deflected (or buckled) beam–column member is for the purposes of analysis
replaced by an equivalent straight (undeflected) member, but with a reduced level
of axial stiffness due to existence of the lateral deflection. This idealization leads to
a decrease of degrees of freedoms to be considered at each node. As a result,
rotational degrees of freedom at each node are not always needed to represent the
nonlinear behavior of the unit and only three translational degrees of freedom (i.e.,
in the x, y and z directions) at each node are included in the formulation of the
ISUM unit, which accounts for possible failure modes such as yielding and flexural
or lateral-torsional buckling.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 335

Fig. 4. (a) The ISUM beam–column unit (a) with attached plating and (b) without attached plating (쎲:
nodal points).

In the numerical computations, the end conditions of the unit can be applied in
the same way to the conventional FEM. For instance, all of the three translational
degrees of freedom may need to be restrained for a fixed nodal point. The magnitude
of the resulting load effects (e.g., stress, displacement) may thus be different from
that for a simply supported end case, for instance.
It is important to realize that an ISUM unit will behave in accordance with the
insights and knowledge built in by its developer. While one may of course try to
idealize the nonlinear behavior of the unit under more relevant end conditions, the
formulations typically assume that the beam–column unit is simply supported at both
ends. As long as the beam–column members are bounded by relatively strong support
members, this approximation will give practical and adequate results on the pessi-
mistic side.
Since the incremental method is used in ISUM analyses, the nodal force and dis-
placement vectors of the beam–column unit are given in the incremental form, as fol-
lows
{⌬R} ⫽ {⌬Rx1 ⌬Ry1 ⌬Rz1 ⌬Rx2 ⌬Ry2 ⌬Rz2}T (1a)
{⌬U} ⫽ {⌬u1 ⌬v1 ⌬w1 ⌬u2 ⌬v2 ⌬w2}T (1b)
where {⌬R} is the nodal force increment vector, and {⌬U} the nodal displacement
increment vector.
336 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

The relationship between the nodal force increment and the nodal displacement
increment can be expressed by
{⌬R} ⫽ [K]{⌬U} (2)
where [K] is the tangent stiffness matrix which is a nonlinear function of the
nodal displacements.
As the applied loads increase, the beam–column unit can buckle or yield until the
ultimate limit state is reached. The tangent stiffness matrix K can then vary depending
on the condition of the structure with regard to the occurrence of such failures
together with loading details at any point in time.
Fig. 5 represents the idealized behavior of the beam–column unit for the purpose
of ultimate strength analysis. It is important to realize that in contrast to plates, this
beam–column unit would not have reserve strength once buckling occurs, and thus
the unit reaches the ultimate limit state immediately after buckling. The effect of
strain-hardening is neglected in the present formulations of the unit.
The tangent stiffness matrix [K] of Eq. (2) or the condition of failures of the
beam–column unit under combined axial and lateral loads can be evaluated by ana-
lytical approach. The tangent stiffness matrix of the failure-free unit denoted by [K]E
is given by taking into account the effects of large deflection, combined axial and
lateral loads and initial imperfections. The ultimate strength or the gross yield

Fig. 5. Idealized structural behavior of the ISUM beam–column unit for analysis of ultimate strength.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 337

capacity can be expressed in the closed form. The tangent stiffness matrix of the
beam–column unit in the post-ultimate regime denoted by [K]U is also given by a
closed-form expression.

5. The rectangular plate unit for analysis of ultimate strength

Fig. 6 shows the ISUM rectangular plate unit which has four nodal points, i.e.,
one at each corner. A buckled (deflected) plate in this case is in effect replaced by
an imaginary flat (undeflected) plate, but with the corresponding reduced effective
in-plane stiffness. Since the plate configuration always remains flat in the numerical
matrix computations, rotational degrees of freedom at each nodal point are not neces-
sary to represent the nonlinear behavior of the ISUM rectangular plate unit. As a
result, only three translational degrees of freedom at each corner nodal point are
used to formulate the nonlinear response like in the beam–column unit.
The nodal force increment vector {⌬R} and the nodal displacement increment
vector {⌬U} of the ISUM rectangular plate unit are then given by
{⌬R} ⫽ {⌬Rx1 ⌬Ry1 ⌬Rz1···⌬Rx4 ⌬Ry4 ⌬Rz4}T (3a)
{⌬U} ⫽ {⌬u1 ⌬v1 ⌬w1···⌬u4 ⌬v4 ⌬w4} T
(3b)
The relationship between the nodal force increments and the nodal displacement
increments can also be given by Eq. (2). Fig. 7 represents the idealized behavior of
the ISUM rectangular plate unit until and after the ultimate strength is reached. In
this case, the tangent stiffness matrices and the condition of the ultimate limit state
behavior can be evaluated using the theory of plates under combined loads, taking
into account the effects of initial imperfections and other types of structural degra-
dation.
The tangent stiffness matrix of the failure-free unit denoted by [K]E is calculated

Fig. 6. The ISUM rectangular plate unit (쎲: nodal points).

338 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

Fig. 7. Idealized structural behavior of the ISUM rectangular plate unit for analysis of ultimate strength.

by taking into account the effects of both in-plane and out-of-plane large defor-
mations by applying the conventional FE approach [13]. Before the unit reaches the
ultimate limit state, the internal stresses at some nodal points may satisfy the yield
condition and if so, plastic nodes are then inserted into the corresponding nodal
points. In this case, the elastic–plastic stiffness matrix [K]P can be derived by apply-
ing the plastic node method as a function of the stress–strain matrix [D]E [22]. The
ISUM rectangular plate unit reaches the ultimate strength if the ultimate limit state
criterion is satisfied upon substituting the internal stress components. In the post-
ultimate regime of the unit, the stress–strain matrix [D]E should be replaced by the
stress–strain matrix in the post-ultimate regime denoted by [DP]U, while the stiffness
matrix of the collapsed unit denoted by [K]U is still given by [K]E.

6. The rectangular plate unit for analysis of collision and grounding

mechanics

While the collision and grounding mechanics of a plate are also nominally rep-
resented by an ISUM rectangular plate unit with four nodal points, as shown in Fig.
6, the characteristics of the unit for analysis of collision and grounding mechanics
are different from those for analysis of ultimate strength. The relationship between
the nodal force increment vector {⌬R} and the nodal displacement increment vector
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 339

{⌬U} is also given by Eq. (2) where {⌬R} and {⌬U} are defined by Eqs. (3a)
and (3b).
Fig. 8 shows the idealized behavior of the ISUM rectangular plate unit for analysis
of collision and grounding mechanics. As the predominantly compressive load
increases, the unit crushes if a crushing condition is fulfilled. It is assumed that the
internal stresses of the crushed unit are unchanged even if the compressive displace-
ment increases. However, the folding process ends if the compressive strain of the
unit exceeds a critical value. The unit behaves as a rigid body after the folding
process ends. The crushed unit eventually reaches the gross yield condition as the
crushing loads increase further.
On the other hand, with increase in the predominantly tensile loads, some nodes
of the unit will yield if the yield condition is satisfied and in such a case plastic
nodes are inserted into the corresponding nodal points. As long as all of the four
nodes do not yield, the unit may not reach the gross yield condition, but it will show
an elastic–plastic load carrying behavior. If the equivalent tensile strain exceeds the
fracture strain, then ductile fracture takes place. Internal stresses of any fractured
unit must be released from the analysis, with the unit losing its resistive load capacity.
The tangent stiffness matrix of the failure-free unit denoted by [K]E is given by
the conventional FE approach [13,22]. As long as energy absorption capacity is
concerned, the effect of initial deflections may be neglected. The same stress–strain
matrix is thus used until either crushing or yielding occurs. As the applied loads

Fig. 8. Idealized structural behavior of the ISUM rectangular plate unit for analysis of collision and
grounding mechanics.
340 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

increase, the unit begins to be folded if the condition of crushing is satisfied. The
stress–strain matrix of the crushed unit is assumed to become zero, while the tangent
stiffness matrix is still calculated from [K]E.
The folding process of the plate elements under crushing loads may stop if the
folding length reaches a ‘critical’ value which is a function of the geometric and
material properties of the plate. After the folding process ends, the unit behaves as
a rigid body and thus the stress–strain matrix [D]E may become infinite, e.g., with
a very large magnitude of the Young modulus. However, if the internal stress compo-
nents satisfy the yield condition at any node then a plastic node is inserted into the
yielded node of the crushed unit and the elastic–plastic stiffness matrix is applied.
Under predominantly tensile loading, some nodes may yield if the yield condition
is satisfied and the elastic–plastic stiffness matrix is applied. The yielded unit may
eventually rupture if the strain components satisfy the condition of ductile fracture.
The fractured unit releases its internal stresses and takes a zero value in the stress–
strain matrix [D]E.

7. The stiffened panel unit for analysis of ultimate strength

Fig. 9 shows an ISUM stiffened panel unit. The panel with stiffeners in either
one or both directions, bounded by support members, is modeled as a unit by means
of the ISUM stiffened panel unit. When the stiffeners are relatively strong so that
they would not fail prior to plating between stiffeners, the same panel may also be
modeled as an assembly of rectangular plate units and beam–column units, the latter
being without attached plating.
For a panel with relatively weak stiffeners so that the overall panel buckling mode

Fig. 9. The ISUM stiffened panel unit (쎲: nodal points).

 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 341

may be predominant, or in the general case where it can be accommodated, however,

the ISUM stiffened panel unit may be more appropriate.
Like the rectangular plate unit, the stiffened panel unit has four nodal points, i.e.,
each at every corner. The relationship between the nodal force increment vector
{⌬R} and the nodal displacement increment vector {⌬U} is given by Eq. (2) where
{⌬R} and {⌬U} are defined by Eqs. (3a) and (3b).
In formulating the stiffened panel unit, six types of failure modes need to be
considered: overall collapse, biaxiall compressive collapse, beam-column type col-
lapse, stiffener web buckling, tripping and gross yielding [4].

8. The stiffened panel unit for analysis of collision and grounding mechanics

The geometry and dimensions of the ISUM stiffened panel unit for analysis of
collision and grounding mechanics are similar to that of Fig. 9. The unit has four
nodal points, i.e., each at every corner point. Each node has the three translational
degrees of freedom since the deflected panel is treated as an imaginary undeflected
(flat) panel but with reduced in-plane stiffness. The force and displacement vectors
of the unit are again given by Eqs. (3a) and (3b). The idealizations of the ISUM
stiffened panel unit for analysis of the collision and grounding mechanics can be
made in a manner similar to that of the rectangular plate unit as described in Section
6 or Fig. 8.

9. The gap/contact unit

In collision or grounding accidents, the interface between the striking and the
struck bodies will change as the collision or grounding proceeds. In this case, the
gap and contact conditions between the two bodies are modeled by means of the
gap/contact unit.
The gap/contact unit is a special type of the beam–column unit described in Sec-
tion 4. The gap unit has two nodal points and each node has three translational
degrees of freedom. In the structural modeling using the gap/contact unit, one node
may be positioned at the striking body and the other at the struck body. The
gap/contact unit may also be used for connecting two nodes which both are in the
struck body as well.
Two such nodes are then connected by a gap/contact unit with nonlinear character-
istics. As the striking body is pushed into the struck body, the length of the unit in
the gap condition will be reduced without resistance, and eventually both nodes will
come into contact. Coordinates of both nodes are updated at every incremental load-
ing step. It is considered that the unit is in the contact condition if the length of the
unit becomes smaller than a prescribed tolerance.
The gap/contact units have only axial stiffness. In the gap condition, the unit does
not carry any external forces, but under contact it behaves as a rigid body. Therefore,
nearly zero stiffness is assumed for the gap condition, while a very large stiffness
342 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

is assumed for the contact condition. The stiffness equation of the gap/contact unit
may then be given, as follows [13]

再 冎
⌬Rx1
⌬Rx2
⫽ 冋
EgcA 1 ⫺1 ⌬u1
L ⫺1 1 ⌬u2
册再 冎 (4)

where A is the cross-sectional area. It may be assumed that Egc = E×10⫺7 for the
gap condition or Egc = E×107 for the contact condition, where E is the elastic modu-
lus.
In practice, some axial stresses may develop in the gap/contact unit because the
assumed stiffness is neither zero nor infinite, but this effect is generally quite small
and negligible. Also, the node positioned at the striking ship and the other node at
the struck ship may ‘overlap’ if the increment of axial deformation is too large, but
this problem can be overcome by using small load increments.

10. Treatment of dynamic/impact load effects

Ship collision or grounding accidents are dynamic in nature, and this fact will
affect the crushing and rupture response of structures. Three aspects of a dynamic
loading situation are possibly relevant, namely material strain rate sensitivity, inertia
effects and dynamic frictional effects. With increase in the strain rate for steel, the
yield strength of the material increases and the rupture strain may decrease. Due to
inertia effects, deformation patterns may be varied. It is known that as the speed of
dynamic loading increases, the coefficient of friction becomes lower.
When dynamic loads are applied, the crushing response is mainly affected by the
material strain rate sensitivity. During crushing response, the material strain rate
generally varies with displacements. For simplicity, however, the average value of
the strain rate during the dynamic loading phase may be used. The average strain
rate for rectangular plates or stiffened plates can be approximately estimated con-
sidering that the loading speed is linearly reduced to zero.
To estimate the dynamic yield strength of the material, the Cowper–Symonds
equation has been widely used. The crushing effects and yield strength increase as
the collision speed gets faster, while any rupture or tearing of steel occurs earlier.
The dynamic rupture strain is obtained by inverting the Cowper–Symonds equation
for the dynamic yield stress.
The inertia effects may be ignored when the strain rates are less than about 50
s⫺1. Friction effects may play an important role during the grounding process, while
they can be ignored for crushing of structures since the relative velocity between
striking and struck bodies is then normally comparatively small.
In the illustrative ISUM calculations given below of crushing and rupture behavior
taking account of dynamic effects, theoretical developments derived for a quasi-static
condition are employed, but with the use of the dynamic yield stress or the dynamic
fracture strain in place of their static counterparts. Hence certain effects such as
dynamic amplification are not included in the calculation of structural response. For
further details, the interested reader may refer to [4,16].
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 343

11. Application examples

In the following sections, some typical examples for nonlinear analysis of steel
plated structures are illustrated. The ISUM method is used in the calculations. Results
from ISUM are compared against conventional numerical FE analysis or experi-
mental results for the same problem to illustrate the accuracy achieved.
For purposes of these comparative calculations, the ISUM theory we used has
been implemented within a computer program called ALPS/ISUM which stands for
nonlinear Analysis of Large Plated Structures using Idealized Structural Unit Method.
For convenience, ALPS/ISUM is divided into three sub-programs, namely
ALPS/GENERAL, ALPS/HULL and ALPS/SCOL. ALPS/GENERAL is a program
for analysis of progressive collapse behavior of general types of steel plated struc-
tures. ALPS/HULL is a special purpose program for analysis of progressive collapse
behavior of ship hulls. ALPS/SCOL is a program for analysis of internal mechanics
of steel plated structures in collision or grounding.

11.1. Progressive collapse analysis of a cantilever box-girder

Ultimate strength of a cantilever box-girder is analyzed in this case by ISUM and

conventional nonlinear ANSYS FEM [23]. The box-girder is fixed at one end and
free at the other end, and it is subjected to concentrated load at the free end, as
shown in Fig. 10. The cross section of the structure is square, and three transverse
diaphragms are arranged within. The dimensions, material properties and initial
imperfections of individual plate elements are as follows:

앫 plate units: a × b×t = 1000 × 1000 × 15(mm);

앫 Young’s modulus: E = 205.8 GPa;
앫 yield stress: s Y = 352.8 MPa;
앫 Poisson’s ratio: n = 0.3;
앫 initial deflection: wopl / t = 0.15;
앫 welding residual stresses: srcx = srcy = 0.0.

Fig. 10. An internally stiffened cantilever box-girder.

344 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

Figs. 11 and 12 show typical examples of structural modeling for use in this
case by ANSYS and ALPS/GENERAL, respectively. In ISUM modeling, each plate
element is modeled as one ISUM rectangular plate unit, while a number of fine
meshes are used in conventional FE modeling where a half of the box-girder with
respect to the center line is taken as the extent of conventional nonlinear FE analysis
considering the symmetric behavior. Fig. 13 shows the applied load versus deflection
curves at the free end of the structure, as obtained by ANSYS and
ALPS/GENERAL computations.
For reference, the plastic collapse load of a cantilever beam can be predicted when
local buckling is not accounted for [4], as follows:
Mp 1 36
Pc ⫽ ⫽ (10003⫺9703) ⫻ 9.8 ⫽ 1925.56 kN
L 4000 4

Fig. 11. (a) Undeformed shape for the conventional nonlinear FE model for the cantilever box-girder.
(b) Deformed shape for the conventional nonlinear FE model for the cantilever box-girder at the ultimate
limit state.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 345

Fig. 12. ALPS/GENERAL model for the cantilever box-girder.

Fig. 13. The force versus deflection curves at the free end of the cantilever box-girder, as obtained by
ISUM and conventional nonlinear FEM.

Fig. 13 also represents the selected structural failure events. As the applied loads
increase, the compressed flange near the fixed end collapses and the box-girder
reaches the ultimate strength it the transverse floors near the fixed end fail. It is a
benefit of structural designers to have detailed information about the progressive
346 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

collapse behavior until the ultimate strength is reached. It is seen from Fig. 13 that
the ISUM results correlate well with the more refined nonlinear FE analysis. Of
interest, the computing time used was 2 min for the ISUM analysis and 4.5 h for
the FE analysis using a Pentium III personal computer. The structural modeling effort
for ISUM analysis is of course much smaller than that for FE analysis.

11.2. Progressive collapse analysis of a box-column

Ultimate strength of a box column is now analyzed in this example by ISUM and
conventional nonlinear ANSYS FEM [23]. The structure is simply supported at both
ends, as shown in Fig. 14. It has a number of transverse diaphragms (floors). The
dimensions, material properties and initial imperfections of the structure are as fol-
lows:

앫 plate elements: a × b × t = 500 × 500 × 7.5(mm);

앫 Young’s modulus: E = 205.8 GPa;
앫 yield stress:sY = 352.8 MPa;
앫 Poisson’s ratio: n = 0.3;
앫 initial deflection function of plate elements: Ao / t = 0.05;
앫 column type initial deflection function for the whole structure: wo = dosin(px /
L) where do / L = 0.0015;
앫 welding induced residual stresses: srcx = srcy = 0.05 for ISUM analysis and srcx
= srcy = 0.0 for ANSYS analysis.

Figs. 15 and 16 show typical examples of structural modeling for this case, by
conventional ANSYS FEM [23] and ALPS/GENERAL, respectively. In ISUM
modeling, each plate element is modeled as one ISUM rectangular plate unit and
due to the symmetry, a half length of the box column is taken as the extent of
analysis. For conventional nonlinear FE analysis, a quarter of the box column is
subdivided into a number of fine meshes. For convenience related to computational
effort, some coarse meshing was adopted for the conventional nonlinear FE analysis

Fig. 14. A box column with both ends simply supported.

 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 347

Fig. 15. Conventional FE model for a quarter region of the box column.

Fig. 16. ALPS/GENERAL model for a half length of the box column.

such that local plate buckling effect (i.e., local geometric nonlinearity) is not taken
into consideration in the ANSYS analysis, although the influence of plasticity
(material nonlinearity) and column type buckling (global geometric nonlinearity) are
accounted for. On the other hand, the ALPS/GENERAL takes into account both
local and global buckling effects as well as plasticity.
Fig. 17 shows the computed axial compressive load versus mid-span deflection
curves for the box column. The Euler elastic column buckling load which intention-
ally neglects the influence of local plate buckling is also plotted in the figure. When
the structure involves both local and global buckling, the ultimate buckling strength
of the box column is about 67% of the Euler column buckling load. It is evident
that the effect of local geometric nonlinearity as well as global geometric nonlinearity
can be of crucial importance for nonlinear analysis of slender structures (e.g., box
column). Of interest, the computing time used was 3 min for the ISUM analysis and
6 h for the FE analysis using a Pentium III personal computer.
348 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

Fig. 17. The axial compressive load versus deflection curves of the box column, as obtained by ISUM
and conventional nonlinear FEM.

11.3. Progressive collapse analysis of a ship hull

A physical test for investigating the progressive collapse characteristics under a

vertical sagging moment was undertaken on a welded steel frigate ship structure
model with 1/3 scale to the original ship dimensions [24,25]. ALPS/HULL is now
used to analyze the progressive collapse behavior of the Dow test model and the
results are then compared with the experimental results. In the ALPS/HULL compu-
tations, both sagging and hogging cases are considered.
Fig. 18 shows the ALPS/HULL model for the test structure. For simplicity, the

Fig. 18. (a) Mid-section of the Dow frigate test structure. (b) ALPS/HULL model for the progressive
collapse analysis of the Dow frigate test structure.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 349

hull module between two transverse frames is taken as the extent of the present
analysis, while it is not difficult to take the entire structure as the extent of the ISUM
analysis if needed. Plating between stiffeners is modeled using the ISUM rectangular
plate unit and stiffeners without attached plating are modeled using the ISUM beam–
column unit. The webs of deep girders in bottom structures are also modeled using
the ISUM rectangular plate units, while their flanges are modeled using the ISUM
beam–column units.
Fig. 19 shows the progressive collapse behavior of the Dow test structure under
sagging or hogging moment, as obtained by ALPS/HULL. The Dow test result for
sagging is also plotted. In the ALPS/HULL computations, the magnitude of initial
imperfections is varied. It is seen from Fig. 19 that the effect of initial imperfections
on the progressive collapse behavior is of significance. Also, ALPS/HULL provides
reasonably accurate results when compared with the experiment. Of interest, the
computing time used was 2 min for the ISUM analysis using a Pentium III per-
sonal computer.

11.4. Analysis of internal mechanics of a LNG carrier side structure

As a hypothetical collision scenario, it is now considered that the bow of another

LNG carrier with the same size as the object LNG carrier in the full load condition
strikes the mid-side structure of the object LNG carrier. At the initiation of the
collision event, the striking ship is assumed to be at 50% of the full design speed,
while the struck LNG carrier is considered to be at standstill at a pier.
The penetrating depth of the striking ship from the initial contact to the boundary
of the LNG cargo tanks or to the bow position which may cause the fracture of
LNG cargo tanks varies in accordance with the fore-end shape of the striking ship

Fig. 19. Progressive collapse behavior of the Dow test structure under vertical moment.
350 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

Fig. 20. MSC/DYTRAN model for the LNG carrier–LNG carrier collision.

and the initial contact positions and related details for the two ships. The bulbous
bow shape under the sea water level and/or the fore-end shape of the striking ship
determines the initial contact position for the two ships and may also affect the
absorbed energy in the side structure of the struck ship.
Both the striking and struck LNG carriers are assumed to be in the full load
condition. Two computer programs, namely MSC/DYTRAN [26] and ALPS/SCOL,
are used to obtain the collision force–penetration curves as the collision proceeds.
Figs. 20 and 21 represent the structural models used for MSC/DYTRAN and
ALPS/SCOL, respectively.
In the MSC/DYTRAN model, the entire ship structure is included using fine
meshes around the impact location while coarser meshes are used at the other parts

Fig. 21. ALPS/SCOL model for the LNG carrier–LNG carrier collision.
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 351

away from the collision location. In the ALPS/SCOL model, a quarter of the mid-
ship cargo hold of the struck ship including the LNG tank together with the striking
ship bow is taken as the extent of the analysis, using ISUM plate and stiffened panel
units to model the struck structures and gap/contact units to represent the gap and
contact conditions between the struck ship side and the striking ship bow.
In the MSC/DYTRAN computations, the motion of both striking and struck ships
with the surrounding water is taken into consideration, while the ALPS/SCOL com-
putations assume that the struck ship is fixed while the striking ship penetrates with
the constant speed. In both models, it is assumed that the striking ship bow is a rigid
body which does not dissipate any strain energy.
The fracture strain of the structural members in both ship hull and cargo tank
structures is supposed to be 10% in the ALPS/SCOL models which use large sized
elements, while it is assumed to be 20% in the MSC/DYTRAN models which use
fine meshes. This is because the MSC/DYTRAN method using a fine mesh can
automatically handle the localized tearing behavior inside the plate members, while
the ALPS/SCOL method using large sized elements must necessarily account for
such behavior at a macroscopic level. In the failure state of finite elements, the strain
is typically concentrated at the tip of crack or around fractured area. The representa-
tive fracture strain should then become smaller in large sized elements to properly
account for the effect of localized fracture inside the plate unit. Hence in the
ALPS/SCOL model, a fracture strain that is likely to be more characteristic of weld
metal fracture is used. With the larger value of fracture strain, the energy absorption
capability of the structure will of course increase.
The computations were continued until the struck LNG carrier reaches the acciden-
tal limit states in the following two conditions, namely (a) the striking ship bow
penetrates the boundary of the struck ship cargo tank, and (b) the LNG cargo tanks
start to rupture. In reality, the energy dissipation capability of the struck LNG carrier
must be evaluated at the earlier of these two limiting conditions.
Fig. 22 shows the deformed shape of the ALPS/SCOL model. Figs. 23 and 24
show the collision force versus penetration curves and the absorbed energy–pen-
etration curves, respectively, the two curves in each figure being obtained by
MSC/DYTRAN and ALPS/SCOL. Oscillations in the force–penetration curves
obtained by MSC/DYTRAN are primarily due to the simulation being based on time-
variant impact theory. Also, the behavior of localized failure including tearing and
folding is more accurately described by the MSC/DYTRAN method using very fine
meshes, while the ALPS/SCOL method is more concerned with an ‘average’ nature
of the failure response.
As evident, both MSC/DYTRAN and ALPS/ SCOL predictions correlate well with
one another, and thus it may be taken with reasonable certainty that the ISUM com-
puted results for the energy dissipation capability of the struck LNG carrier are
accurate enough for the purposes of design stage safety assessment. For other types
of accident scenarios varying the type of striking ship or loading condition, the inter-
ested reader may refer to [17].
Of interest, the computing time used for the ALPS/SCOL analysis was about 4 h
using a Pentium III personal computer, while it was more than 14.4 h for the
352 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

Fig. 22. Deformed shape of the ALPS/SCOL model immediately after the struck LNG cargo tank starts
to fracture.

Fig. 23. The collision force–penetration curves (SS-F=collision of LNG carrier to LNG carrier in the
full load condition).
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 353

Fig. 24. The absorbed energy–penetration curves (SS-F=collision of LNG carrier to LNG carrier in the
full load condition).

MSC/DYTRAN analysis using the SGI Power Challenge XL Series computer. As

evident, the former will be of a benefit when a quick estimate is more important for
carrying out a series of collision analyses with a variety of accident scenarios.

12. Concluding remarks

While the ISUM has been recognized by researchers as an efficient and accurate
methodology for nonlinear analysis of large plated structures such as ships, offshore
platforms, box girder bridges and other steel structures, the authors are of the opinion
that the number of structural designers who use ISUM are currently quite limited.
This may be partly because an educational effort on details of the basic idea of
ISUM is necessary, and the literature on the theory and application of ISUM is not
sufficient for this purpose at the moment.
The twin aims of the present paper have thus been to summarize the ISUM theory
and to illustrate some examples of its application to nonlinear analysis of steel plated
structures. Important concepts for developments of the various ISUM units which
would be typically needed by a designer to analyze the nonlinear behavior of steel
plated structures are described. Some examples of ISUM application to analysis of
ultimate strength and internal accident mechanics of ships and steel plated structures
are illustrated. The accuracy of ISUM predictions is studied by comparison with the
corresponding results from conventional nonlinear finite element analyses and from
experimental results, as appropriate. The accuracy is illustrated to be good, and per-
354 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355

haps more importantly, it is shown that a level of accuracy comparable to conven-

tional methods can be achieved at a fraction of the cost and effort associated with
the conventional methods.
ISUM technology is now mature, and can clearly be useful to a structural designer
in many contexts, including design assessment based on the ‘true’ ultimate strength
and the general damage tolerant design of structures.

References

[1] Ueda Y, Rashed SMH. An ultimate transverse strength analysis of ship structures. [in Japanese] J
Soc Naval Architects Japan 1974;136:309–24.
[2] Ueda Y, Rashed SMH. The idealized structural unit method and its application to deep girder struc-
tures. Comput Struct 1984;18(2):277–93.
[3] Smith CS. Influence of local compressive failure on ultimate longitudinal strength of ship’s hull.
In: Proceedings of the International Symposium on Practical Design in Shipbuilding, Tokyo. 1977.
p. 73–9.
[4] Paik JK, Thayamballi AK. Ultimate limit state design of steel plated structures. Chichester: John
Wiley & Sons, 2002.
[5] Ueda Y, Rashed SMH, Nakacho K, Sasaki H. Ultimate strength analysis of offshore structures—
application of idealized structural unit method. [in Japanese] J Kansai Soc Naval Architects Japan
1983;190:131–42.
[6] Paik JK. Advanced idealized structural units considering the excessive tension deformation effects.
J Hydrospace Technol, Soc Naval Architects Korea 1995;1(1):125–45.
[7] Ueda Y, Rashed SMH, Paik JK. Plate and stiffened plate units of the idealized structural unit method
(1st report)—under in-plane loading. [in Japanese] J Soc Naval Architects Japan 1984;156:389–400.
[8] Ueda Y, Rashed SMH, Paik JK. Plate and stiffened plate units of the idealized structural unit method
(2nd report)—under in-plane and lateral loading considering initial deflection and residual stress.
[in Japanese] J Soc Naval Architects Japan 1986;160:321–39.
[9] Ueda Y, Rashed SMH, Paik JK, Masaoka K. The idealized structural unit method including global
nonlinearities—idealized rectangular plate and stiffened plate elements. [in Japanese] J Soc Naval
Architects Japan 1986;159:283–93.
[10] Bai Y, Bendiksen E, Pedersen PT. Collapse analysis of ship hulls. Marine Struct 1993;6:485–507.
[11] Masaoka K, Okada H, Ueda Y. A rectangular plate element for ultimate strength analysis. In: Pro-
ceedings of the 2nd International Conference on Thin-Walled Structures; Singapore. 1998. p. 1–8.
[12] Fujikubo M, Kaeding P, Yao T. ISUM rectangular plate element with new lateral shape function—
longitudinal and transverse thrust. J Soc Naval Architects Japan 2000;187:209–19.
[13] Paik JK, Pedersen PT. Modeling of the internal mechanics in ship collisions. Ocean Eng
1996;23(2):107–42.
[14] Paik JK, Thayamballi AK, Che JS. Ultimate strength of ship hulls under combined vertical bending,
horizontal bending, and shearing forces. SNAME Trans, Soc Naval Architects and Marine Engineers,
New Jersey 1996;104:31–59.
[15] Wang G, Chen Y, Zhang H, Shin Y. Residual strength of damaged ship hulls. In: Proceedings
of Ship Structure Symposium on Ship Structures for the New Millennium: Supporting Quality in
Shipbuilding; Arlington, VA; 2000 Jun 13–14.
[16] Paik JK, Chung JY, Choe IH, Thayamballi AK, Pedersen PT, Wang G. On rational design of double
hull tanker structures against collision. SNAME Trans, Soc Naval Architects and Marine Engineers,
New Jersey 1999;107:323–63.
[17] Paik JK, Choe IH, Thayamballi AK. Predicting the resistance of spherical type LNG carrier structures
against ship collisions. Marine Technol, Soc Naval Architects and Marine Engineers, New Jersey
2002;39(2):86–94.
[18] Brown A, Tikka K, Daidola JC, Lutzen M, Choe IH. Structural design and response in collision
 J.K. Paik, A.K. Thayamballi / Thin-Walled Structures 41 (2003) 329–355 355

and grounding. SNAME Trans, Soc Naval Architects and Marine Engineers, New Jersey
2000;108:447–73.
[19] Wang G, Arita H, Liu D. Behavior of a double hull in a variety of stranding or collision scenarios.
Marine Struct 2000;13:147–87.
[20] Ueda Y, Rashed SMH. Advances in the application of ISUM to marine structures. In: Proceedings
of the Second International Conference on Advances in Marine Structures; Dunfermline, UK; May.
1991. p. 628–49.
[21] Moan T, Amdahl J, Engseth AG, Granli T. Collapse behavior of trusswork steel platforms. In:
Proceedings of the 4th International Conference on Behaviour of Offshore Structures (BOSS’85);
Delft, The Netherlands; July. Amsterdam: Elsevier; 1985. p. 255–68.
[22] Paik JK, Thayamballi AK, Lee WH. A numerical investigation of tripping. Marine Struct
1998;11:159–83.
[23] ANSYS. User’s manual (version 5.6). Canonsburg, PA, USA: ANSYS, Inc.; 1999.
[24] Dow RS. Testing and analysis of 1/3-scale welded steel frigate model. In: Proceedings of the Inter-
national Conference on Advances in Marine Structures; Dunfermline, UK. 1991. p. 749–73.
[25] Yao T et al. Ultimate hull girder strength. Report of the ISSC Special Task Committee VI.2, Inter-
national Ship and Offshore Structures Congress, vol. 2. Elsevier; 2000. p. 321–91.
[26] MSC/DYTRAN. User’s manual (version 4.5). Los Angeles, CA, USA: The MacNeal Schwendler
Corp.; 1998.
[27] Paik JK, Wang G, Kim BJ, Thayamballi AK. Ultimate limit state design of ship hulls. SNAME
Transactions, Soc Naval Architects and Marine Engineers, New Jersey 2002:110.