You are on page 1of 17

Oil Tanker Cargo Hold Structural Optimization

Using Ultimate Limit States


Justin M. Freimuth (AM), Ming Ma (AM)
Advanced Marine Technology Center, DRS Technologies Inc., Stevensville, MD, USA

This paper presents a method to optimize an oil tanker cargo hold’s structural scantlings based on stiffened panel
ultimate limit states. Two different tanker models and their results are presented demonstrating this optimization
procedure using models of varying mesh densities. The full ship finite element models are loaded with multiple load
cases, design bending moments, external hydrostatic pressure, and internal tank pressure. The working stresses of a
stiffened panel, which are used for the panel’s ultimate limit states assessment, are obtained by 3D finite element
analysis. Each stiffened panel is then optimized using multi-objective genetic algorithms for its weight and safety.
The optimization process is performed on two different versions of the same vessel: one with all stiffeners defined
explicitly and one with internal stiffeners (allowing the stiffener layout to be changed during the optimization). The
numerical results show that the proposed method is very useful to perform ultimate strength based ship structural
optimization with multiple objectives, namely minimizing the structural weight and maximizing structural safety.

KEY WORDS: Multi-objective optimization; Ultimate Limit been proposed. In terms of the optimization methods, there are
States; MAESTRO; ALPS/ULSAP basically two main types of optimization algorithms: the
mathematical approaches (deterministic or gradient) and the
INTRODUCTION heuristic or stochastic approaches including concepts inspired
In modern shipbuilding, the emphasis on reliability, by natural biological systems. The main difficulties related to
affordability, and efficiency continues to drive the need for an the use of gradient methods are that they can become stuck in a
optimal ship structural design. During the concept design phase, local optimum and the discrete variables have to be
typically a midship structural design is established which meets approximated as continuous variables. Because the objective
the requirements (e.g., principal dimensions, coefficients of functions and constraints are often highly non-linear, the
form, etc.) of the overall design. In many ships, once a midship problems are usually either sequentially linearized and solved
section structural design is complete, the remaining significant with linear programming (Hughes 1980), or separated into a
portion of the ship’s length can be designed from this section. series of convex problems and solved accordingly with the non-
Since the bending moments and shear loads are greatest between linear gradient-based method (Rigo 2001). Conversely,
the quarter points of the hull, the scantlings toward each end of heuristic methods offer the possibility of handling discrete
the ship need only be given as modifications to the midship variables as well as providing global optimization capabilities.
section. Ships with a fairly uniform midbody, such as an oil The heuristic methods have been growing in popularity over the
tanker, can use the midship section design for the purpose of last few years as more and more researchers discover the
structural weight estimation, rules-based checks, and initial cost benefits of their adaptive search. The Genetic Algorithm (GA) is
estimation. This ship design workflow lends itself to a large one type of popular heuristic methods. Many papers now exist
potential value by performing a midship section structural describing a multitude of different types of genetic algorithms,
optimization. theoretical and practical analyses of GAs and a plethora of
In setting up the optimization problem for a midship applications (Coello 1999, Konak 2006).
structural section, a large number of variables are introduced The genetic algorithm is an optimization technique which
including plating thickness, stiffener properties, and the discrete simulates the natural process of evolution. Solutions to the
number of stiffeners. Additional complications include optimization problem are created in an iterative process
constraints from stiffened panel yielding and buckling under described by generations starting with a random population of
various load conditions and design rules specified by possible solutions. The new generations are developed using
classification societies. Although ship cross section design can methods similar to genetic crossover and mutation. Crossover
be rather complicated, the attempts of structural scantling occurs when more than one parent solution is used to develop
optimization have been plentiful and various strategies have the new solution by crossing over between the parent’s solutions

144
at one or more points. Mutation occurs when the new solution other solutions so results can vary significantly based on the
changes one or more of the parent genes (in this case the genes stopping criterion used.
are representative of optimization variables, such as plate
thickness). At each new generation, or solution, the new LIMIT STATE BASED STRUCTURAL
population’s genes are evaluated using a fitness measure. In the OPTIMIZATION
case of a single objective optimization, the genetic algorithm
stores a single fitness value for each solution in the population.
In the field of ship structures, often designs are based on
For multiple objective problems, it is common practice to use a
simple empirical constraints, such as scantling rules and
weighted sum to combine multiple objectives into a single
allowable stress limits specified by class societies, to size the
objective whose fitness can be measured. For example:
structural scantlings. Today, it is well known that the limit state
is a better basis for the design and assessment of large thin-wall
(1)
structures than allowable working stress. Although the working
stress is an important evaluation parameter, the limit state
where is the normalized objective function and are
analysis checks both individual and combined (i.e., stiffened
the weighting coefficients representing the relative importance
panel) structural members under a combination of loads for
of the n objective functions. It is usually assumed that
particular failure modes. In addition to working stress limits
in order to allow fitness comparisons with single (often relative to the material yield stress), classification
objective optimizations or other multiple objective solutions societies also require that plating, stiffened panels, and primary
with different weighting vectors (each possibly with a different structural members be evaluated against buckling and ultimate
number of weighting functions.) Solving a problem with the strength criteria.
objective function in Eq. 1 for a given weighting vector There are two ways to evaluate the limit states of a
yields a single solution, and if multiple stiffened panel. One is to use full-blown non-linear finite
solutions are desired, the problem must be solved multiple times element analysis methods, and the other is to use semi-analytical
with different weighting coefficient combinations. In future approximate methods. Although general finite element tools are
generations, members of the population with a higher fitness widely available and provide reliable results for structure
level are given a stronger chance of producing offspring in the instability analysis, their application can be prohibitive due to
next generation, which usually provides better solutions. A the computational time. This justifies the interest for more time-
pseudo-code for a simple Random Weight-Based Genetic effective strategies, for which the main idea is to replace the
Algorithm process is given in the following steps: finite element method with approximation techniques. The use
of an analytical or semi-analytical approach results in an
1. Generate a set of random weight vectors
attractive strategy due to its effectiveness in terms of
where
computational time, especially if compared with conventional
2. Generate a random population.
3. Assign a fitness value to each solution as numerical procedures such as the finite element method. This
aspect becomes even more important when dealing with highly
4. Crossover: Generate an offspring population as follows: non-linear analyses and in the context of optimization
Choose two solutions x and y from Sk based on the fitness procedures, in which repeated analyses are required.
values. Using a crossover operator, generate offspring and Several closed-form solutions and semi-analytical
add them to the next generation set Sk+1. approaches can be found in the literature for predicting the limit
5. Mutation: Mutate each solution with a predefined
state of stiffened panels (Paik 2009). Among them, the
mutation rate.
6. Rank the population and update the fittest solution to the ALPS/ULSAP method accounts for a wide range of loads and
non-dominated set M. initial conditions, including combined in-plane, longitudinal,
7. Repeat steps 3~7 until the stopping criterion is satisfied transverse and shear loads, lateral pressure, initial deflection of
8. Repeat steps 1~8 for next set of weighting coefficients. plate and stiffeners, residual stress, structural dents, plate
openings, impact pressure, corrosion, etc. The primary modes of
One of the strengths of the genetic algorithm is its overall failure for a stiffened panel or a grillage under
simplicity, efficiency, and ability to explore a large design space predominately compressive loads can be categorized into the
without performing an exhaustive search of possible solutions. following six types (Paik 2003, Hughes 2010):
However, some of its weaknesses are that it can be difficult to
determine the appropriate weights to scale the optimization ■ Mode I: overall collapse after overall buckling
objectives and that “improved” solutions are only relative to

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 145
■ Mode II: collapse of the plating between stiffeners
without the failure of stiffeners
■ Mode III: beam-column type collapse of a stiffener
with attached plating
■ Mode IV: local buckling of stiffener web (after
buckling collapse of attached plating)
■ Mode V: lateral-torsional buckling (tripping) of a
stiffener
■ Mode VI: gross yielding

To effectively assess limit states of a stiffened panel, the


structural response and working stresses have to be known. One
of the most reliable methods to obtain the structural response
and stress is through finite element analyses. However, due to
the nature of the repeated calculation in optimization, few
approaches have attempted to directly integrate finite element
methods into their optimization schemes. For example, the
optimization program LBR5 uses an analytical method (based
on differential equations of stiffened plates) to compute the
overall response of the hull structure (Rigo 2005). This method Fig. 1 Iterative Optimization Process
is a direct analysis of the stress and strain of the prismatic part
of the ship or a cargo hold. OCTOPUS (Zanic 2009) uses OBJECTIVE FUNCTIONS AND CONTRAINTS
customized super-element finite element method to compute the
structural response of a 2.5D segment. As indicated in the previous section, the multi-objective
In order to leverage widely available generic finite element optimization is executed at the stiffened panel level. A typical
models, Ma, et al. (Ma 2013) proposed a two-step heuristic stiffened panel is shown in Fig. 2.
based approach to optimize ship structures. In this process, the
structural response is calculated using standard finite element
models. Each stiffened panel’s loads are applied from the finite
element analysis and remain constant during the optimization
procedure. ALPS/ULSAP, a semi-analytical ultimate limit state
criteria, is used to assess the structural safety of each panel.
Based on the optimization criteria, each of the panels are
optimized individually using the genetic algorithm and limit
states. Once each panel is optimized, a finite element analysis is
run on the new structural definition of the entire model and the
applied loads for each panel are updated. The optimization then
becomes an iterative process and limits the number of finite Fig. 2 Typical Stiffened Panel Structure
element analysis runs, which can be computationally expensive.
The overall iterative process employed can be illustrated in There are three objective functions for the problem. The
Fig. 1. first is the panel structural weight, W.

(2)

where ρp, ρsx, and ρsy are the plate density, longitudinal stiffener
density, and transverse frame density respectively. B and L are
the panel width and length. tp is the plate thickness. Nx and Ny
are the number of the longitudinal stiffeners and transverse
frames respectively. bf, tf, hw and tw are the stiffener (or frame)

146 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
flange width, flange thickness, web height and web thickness of
the longitudinal stiffener or transverse frame.
The second objective function is the fabrication cost, C,
which in its current implementation, is based on weld length
alone. For a grillage the cost is:

(3)

where cf is the welding cost per length of stiffener. In these


(7)
examples, the cost is not taken into consideration, however the
optimization framework is setup such that more realistic and
The number of the stiffeners can be limited as:
builder-specific cost models can be accounted for.
The third objective function is the safety measure for the
panel or grillage, η, given in Eq. 4.
(8)

Manufacturing related constraints are:


(4)

where Cd and Dd are the characteristic values of capacity and


demand.
The optimization problem is to determine the optimum
values of the variables (tp, bfx, tfx, hwx, twx, Nx, bfy, tfy, hwy, twy, Ny)
which minimize the weight and cost while having a maximum
safety measure. Without loss of generality, the multi-objective
optimization assumes the form:

(9)

(5) The optimization constraints can also be set by using a plate and
stiffener property library, however this typically limits the
design space such that an exhaustive search can be used instead
For genetic algorithm approach, the goal is to maximize the of a genetic algorithm.
fitness on the design space of all possible configurations. The
aggregated fitness function f can be expressed as:
APPLICATION TO CARGO HOLD
STRUCTURAL DESIGN

To illustrate the procedure of the optimization, a cargo hold


(6) of a 200,000 ton double hull oil tanker, shown in Fig. 3(a), is
optimized in this section. The main particulars of the tanker are
where W0 and are the nominal initial design value of a given in Table 1.
stiffened panel weight and safety measure respectively, gi(x) is
the constraint penalty function, and ci is the coefficient of the Table 1. Main Particulars of the Oil Tanker
penalty function. ci is 0 if the design variables satisfy the Length between perpendiculars 264 m
constraints, and is 1 if they violate the constraints. The purpose Breadth 48 m

of the constraint penalty function is to drive the solution away Depth 23.2 m

from violating certain constraints by driving down the fitness Draught 19.2 m

level of that particular solution. Displacement 200,000 ton

Plate and stiffener scantlings can be prescribed by the users Dead weight 169,000 ton

based on their design experience, technological preferences and Block coefficient (Cb) 0.843

the structure’s local safety requirements:

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 147
The finite element model of the tanker has 121,368 nodes The portion of the hull representing the cargo hold is shown
and 513,076 elements. This model was originally created as a in Fig. 5. This portion of the model consists of 13,591 nodes
generic Nastran finite element model, but loaded using and 27,474 elements.
MAESTRO load patterns. The tanker has 6 cargo tanks and 6
ballast tanks, as shown in Fig. 3(c) and 3(d).

(a) Full Ship (b) Finite Element Model

Fig 5. Cargo Hold of Oil Tanker

The stiffened panels in the cargo hold are automatically


found using MAESTRO’s evaluation panel search algorithm,
(c) Cargo Tanks (d) Ballast Tanks and are shown in Fig. 6. The search method looks for a
Fig. 3 Suezmax Class Double Hull Oil Tanker collection of elements representative of stiffened panels in the
finite element model based on changes in geometry (e.g., a deck
For the full load case, all cargo tanks are loaded with a total edge or plate intersection) and user-specified beam elements
deadweight of 169,000 tons. For the ballast load case, all ballast which act as panel boundaries (e.g., transverse frames or
tanks are loaded with a weight of 52,000 tons. Hydrostatic longitudinal girders). MAESTRO retains the critical stiffened
pressure was automatically calculated and applied to the panel boundary conditions and parameters such as the true panel
model’s outer hull. span. Each of the 392 colored groups represent a stiffened panel
whose limit states are evaluated during each FEA run and used
in the optimization procedure.

Fig. 4 Hydrostatic Pressure Distributions for Full Load and


Ballast Conditions

In addition to tank loads and hydrostatic pressure, wave


induced extreme hogging and sagging bending moment design
envelopes, obtained by the CSR empirical formula (IACS, 2012)
were applied to the model:

(10)

(11) Fig 6. Cargo Hold Stiffened Panels

In order to produce a realistic design from the optimization,


where fwave-v is the distribution factor for vertical wave
the stiffened panels are grouped into “design clusters.” The
bending moment along the vessel length, fprob=1, and Cwave purpose of the design cluster is to create larger portions of the
is the wave coefficient:
structure that will have the same scantling properties. This will
prevent changes in plate thickness or stiffener properties at each
panel. The design clusters are defined by the user and can be
used to simplify the manufacturing process. The 56 design
when 150 < L < 300 (12)
clusters for this example are shown in Fig. 7.

148 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
Fig. 7 Cargo Hold Design Clusters

During the optimization cycles, the stiffened panel with the


Fig. 9 Stiffener Design Constraints
worst adequacy parameter from each design cluster is optimized
and the new scantlings are applied to all panels in the cluster
This example is an especially severe test of the optimization
before the next FEA run. The adequacy parameters are a
method presented in this paper. This ship was originally
normalized strength ratio from -1 to 1. Any structure with a
designed using high tensile steel in some regions. The owners
negative adequacy parameter is considered inadequate based on
subsequently decided to use only mild steel. In this
the loads applied and the defined safety factors.
optimization the initial scantlings are those for the mild steel
The safety factors for buckling and yielding are 1.5 and
design, and consequently they are very inadequate. As shown in
1.25 respectively. The overall constraints for each of the design
Fig. 10, 25% of the structure was severely inadequate, with
clusters are shown in Fig. 8. Additional stiffener constraints
adequacy parameters as low as -0.161.
were also created and are shown in Fig. 9.

Fig. 8 Overall Design Constraints

Fig. 10 Optimization Weight and Safety Summary

In spite of this relatively weak initial design, the


optimization converged in 12 design cycles. In addition to

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 149
eliminating all structure with a negative adequacy parameter, the
optimization process was able to achieve a 5.8% weight savings
(from 2401 tons to 2261 tons). The minimum adequacy
parameter was improved from -0.161 to 0.021. The percentage
of failed structure was reduced from 25% to 0%. Fig. 11 shows
the design history for structural weight, minimum adequacy
parameter and percentage of inadequate structure for 12 design
cycles. The optimization automatically stops when there is no
weight improvement within 5 cycles.

(a) Original Design

(b) Optimized Design


Fig. 12 Von Mises Stress Results from Original and Optimized
Design
Fig. 11 Optimization Design Iterations
APPLICATION TO COARSE AND FINE MESH
At the end of the optimization process, the full ship finite CARGO HOLD FE MODELS
element model was run and the Von Mises stress results were
compared for the optimized portion of the structure. Fig. 12
presents the stress distribution from the original design and the A second application of this optimization procedure is used
new optimized design. on a separate full ship oil tanker model to demonstrate the
different results using a coarse mesh and a fine mesh (i.e., all
stiffeners are explicitly defined as beam elements) model of the
same midship section. In the optimization of the coarse mesh
model, the stiffener spacing is now a variable in addition to the
stiffener property whereas the fine mesh model has fixed
stiffener locations. The purpose of this exercise is to investigate
the weight savings comparison between the coarse mesh and
fine mesh optimization. The main particulars of the tanker are
given in Table 2.

150 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
Table 2. Main Particulars of the Second Oil Tanker
Length between perpendiculars 174 m
Breadth 32.2 m
Depth 23.2 m
Draught 11.6 m
Displacement 56,000 ton
Block coefficient (Cb) 0.862

The coarse mesh model consists of 15,000 nodes and 133,679


elements. Stiffeners are not explicitly modeled, however their
property and spacing is stored as an attribute of their
associated shell element. During the finite element analysis,
the extra stiffness is accounted for using orthotropic plate
theory. The coarse mesh model is shown below in Fig. 13. Fig. 15 Second Tanker Midship Fine Mesh Model

The model’s ballast and cargo tanks were loaded and


combined with the structural weight to represent a full load
displacement of 56,000 tons. The model was balanced and the
appropriate hydrostatic pressures were automatically applied to
the outer shell elements. Additional load cases were created
which combined the CSR wave induced extreme hogging and
sagging bending moment design envelopes to the full load
Stillwater condition.
Evaluation patches and design clusters are identical for the
Fig. 13 Second Tanker Coarse Mesh Model coarse and fine mesh models. Fig. 16 shows the evaluation
patches for the two versions of the mesh.
The coarse mesh midship section structure to be optimized is
shown in Fig. 14. The red lines represent the internal
stiffeners.

Fig. 16 Coarse and Fine Mesh Evaluation Patches

The safety factors for buckling and yielding are again 1.5 and
1.25 respectively. The overall constraints for each of the
design clusters are shown in Fig. 17.

Fig. 14 Second Tanker Midship Coarse Mesh Model

The equivalent fine mesh model of this midship section is


shown in Fig. 15. The model is connected to the rest of the
full ship coarse mesh model using rigid spline elements which
impose the coarse mesh displacements onto the fine mesh
boundary nodes.

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 151
Fig. 17 Coarse and Fine Mesh Optimization Constraints Fig. 19 Fine Mesh Optimized Plate Thickness

The coarse mesh optimization introduces a new variable for


stiffener spacing which can range from 100 to 1,000 mm with
50 mm increments.
In this example, the original structure did not have any
failing adequacy parameters, however the structural weight was
reduced in both the coarse mesh and fine mesh. The coarse
mesh model was able to achieve an additional 8% weight
savings relative to the fine mesh model. Figs. 18 through 21
show the optimized plating thickness and stiffener count
comparison for the two models.
In Figs. 20 and 21, the coarse mesh model stiffeners are
shown as red lines within the plate elements and the fine mesh
model stiffeners are represented by the orange beam elements.
It can be seen that in the coarse mesh model, the outer bilge
plating was able to be significantly thinner (8 mm) with the
inclusion of additional stiffeners, whereas the fine mesh was
constrained by the existing stiffener layout and thus required
much thicker plating (32 mm). Fig. 20 Coarse Mesh Stiffener Layout

Fig. 18 Coarse Mesh Optimized Plate Thickness Fig. 21 Fine Mesh Stiffener Layout

152 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
CONCLUSIONS Ma, Ming, Hughes, Owen F., and Paik, Jeom K., Ultimate
Strength Based Stiffened Panel Design Using
In this paper, a Random Weight-Based Genetic Algorithm Multi-Objective Optimization Methods and Its
has successfully been used to optimize the cargo hold structural Application to Ship Structures. PRADS 2013
design of a 200,000 ton tanker and compare the optimization
results of a 56,000 ton tanker using a coarse mesh and fine mesh Paik, Jeom K. and Thayamballi, A.K., Ultimate Limit
model. In both examples, to accurately determine the cargo hold State Design of Steel-Plated Structure. John Wiley
load capacity at each design cycle, standard finite element & Sons, 2003
analysis methods are used. The objectives of the optimization
are the structural weight and the ultimate local panel strength. Paik, Jeom K., Kim, B.J., and Seo, J.K., Methods for
Ultimate Limit State Assessment of Ships and Ship
The paper demonstrated that the cargo hold structural weight
Shaped Offshore Structures - Part III: Hull
can be reduced while also eliminating all local panel failures
Girders. Ocean Engineering 35, (2008): 281-286
using this optimization approach. The second tanker example
demonstrated that a coarse mesh model in which the stiffener
Paik, Jeom K. and Seo, J.K., Nonlinear Finite Element
spacing is not fixed, could potentially provide a greater weight Method Models for Ultimate Strength Analysis of
savings than a fine mesh model, while still maintaining the Steel Stiffened-Plate Structures Under Combined
structural integrity of the design. Based on these examples, the Biaxial Compression and Lateral Pressure
proposed approach is quite capable of generating better cargo Actions - Part II: Stiffened Panels. Thin-Walled
hold designs, or larger portions of the structure, within Structures 47, (2009): 998-1007
reasonable search times. Given their flexibility and ease of
implementation, the proposed multi-objective methods can be Paik, Jeom K., Kim, S.J., Kim, D.H., Kim, D.C., Frieze,
viewed as a valuable and attractive tool for structural P.A., Abbattista, M., Vallascas, M., and Hughes,
optimization. Owen F., Benchmark Study on use of
ALPS/ULSAP Method to Determine Plate and
REFERENCES Stiffened Panel Ultimate Strength. MARSTRUCT
2011: 169-186
Coello, Carlos A. An Updated Survey of Evolutionary
Multi-Objective Optimization Techniques: State of Rigo, P., Differential Equations of Stiffened Panels of Ship
the Art and Future Trends, Int. Conf. Structures and Fourier Series Expansions. Ship
Evolutionary Computation (ICEC’99), 1999 Technology Research, (2005): 82–100

Hughes, Owen F., Mistree, F., and Zanic, Vedran. A Rigo, P. and Fleury, C., Scantlings Optimization Based on
Practical Method for the Rational Design of Ship Convex Linearization and a Dual Approach - Part
Structures. Journal of Ship Research 24/2, (1999): II. Marine Structures 14, (2001): 631-649
101-113
Zanic, Vedran, Andric, J., Prebeg, P., Stipcevic, M.,
Hughes, Owen F. and Paik, Jeom K., Ship Structural Grgic, M., Kitarovic, S., Hadzic, N., Piric, K.,
Analysis and Design. SNAME, 2010 Chirica, I., Giuglea, S., Giuglea, V., Turan, O.,
Khalid, H., Rigo, P. (2009), RoPaX - Structural
IACS, Common Structural Rules for Double Hull Oil Design Aspects. IMPROVE 2009
Tankers. International Association of
Classification Societies, London, 2012

Konak, A., Coit, D.W., and Smith, A.E. (2006), Multi-


Objective Optimization Using Genetic
Algorithms: A Tutorial. Reliability Engineering
and System Safety 91, (2006): 992–1007

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 153
Discussion single solution or Pareto front result from each
Matthew Collette, Member local optimization? The flowchart on page 2
I would like to congratulate the authors on makes reference to a non-dominated set M, but it
another advancement in the world of marine is not clear if this is just an internal feature of the
structural optimization. The problem of both implementation, or a globally-returned Pareto
robustly and efficiently optimizing marine approximation generated by the different
structures in the presence of conflicting weighting schemes. If a front is returned, how is
objectives and numerous loading conditions is the final solution chosen to update the global
both central to naval architecture and highly FEA model? Additionally, the penalty function
challenging to those developing optimization handling in this GA was a little unclear.
algorithms. Here the authors have built on their Equation 6 notes that penalty terms are addressed
two-level sequential optimization approach, with structural adequacy as a parameter using an
using both a global finite element analysis (FEA) external penalty function approach. Were all of
model of the ship and ultimate limit state equations 7, 8, and 9 used as penalty functions or
evaluation at the panel level to guide the local was the GA binary coded to directly include the
optimization. Encouraging results have been range constraints on the variables? While for a
shown for two oil tanker designs, and the relative problem the size of a single stiffened panel, the
complexity of numeric modeling used external penalty function approach is probably
throughout is also notable and reflects current sufficient, both Deb (2001) and Klanac (2009)
best practice for design. have noted that performance gains can be
achieved by tying constraints to objectives
The debate between monolithic optimizers, directly, or changing the selection criteria in the
which attempt to solve the structural design GA to treat constraint violations separately than
problem in one shot, and hierarchal or distributed objective functions.
optimizers like the one presented here, which
partition the problem into simpler sub-problems, In closing, I would like to thank the authors for
continues in the marine research community. an engaging paper that builds on their sequential
While it is clear that partitioning the problem can optimization approach and shows impressive
result in significantly smaller problem sizes and early results.
potential efficiency gains, the challenge with
distributed optimizers is in re-integrating the Additional references
problem in such a way that the optimizer reliably Deb, K. 2001. Multi-Objective Optimization
converges. I noted in figure 11 that the present Using Evolutionary Algorithms. Wiley-
scheme did oscillate after discovering its most Interscience Series in Systems and Optimization.
optimum value. Given that there is no Chichester: John Wiley & Sons.
optimization in the outermost loop of the http://hdl.handle.net/2027/[u]:
optimizer—just an update of the global FEA mdp.39015051290461.
model to ensure load consistency—do the  
authors see a path to a more rigorous Klanac, Alan, and Jasmin Jelovica. 2009.
convergence criteria other than plotting out the “Vectorization and Constraint Grouping to
iterations and determining the optimum by Enhance Optimization of Marine Structures.”
inspection? Would the introduction of artificial Marine Structures 22 (2): 225–45.
inertia terms in the global updating scheme doi:10.1016/j.marstruc.2008.07.001.
reduce the “bounce” seen after the optimum  
point has been found? As far as I could tell in Authors’ Response
the paper, no global section modulus constraint
Thank you for your comments and discussion
was enforced; would additional constraints at the
topics. Your comment is accurate that there is
global level impact this convergence?
no intrinsic criteria to control the outer loop
iteration. Each panel is optimized independently
In terms of integration, the use of the random
and the only interaction between panels is
weight genetic algorithm at the panel level also
through the global load redistribution. We have
was interesting. As the objective function is
not identified a good criteria to “glue” panels
comprised of multiple objective terms—a
together. The sectional modulus wasn’t ideal
weight, strength/constraint, and cost term, and
because the proposed method wasn’t limited to
the weight vector is randomly varied—did a
prismatic longitudinal mid-body structure. Some

154 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
of the panels are transverse bulkheads and deep distribution obtained through optimization
frames. Note that, although the weight objective procedures. This effect has been confirmed in
oscillates relatively significantly in the example, most of the published practical optimization
the safety measure objective converges quickly. examples.

Because of the nature of the two-stage The presented structural optimization technique
optimization, it is not necessary to get a true (MOGA) inbuilt into the two-step iterative
Pareto front in the early stage. As a result, the optimization procedure proved to be a mature
random weight GA becomes more tool. It is applicable in the design process and
computationally efficient. A Pareto front is helps the designer to achieve significant savings
generated for each panel during a local for the shipyard and the shipowner: the increase
optimization. The selection of the local optimal of deadweight; the decrease in the price and
design is hardwired into the code and is weight of construction steel; the decrease of the
automatically selected from the Pareto front production costs; and an increase in safety (and
during the optimization with a greater emphasis robustness). Such mathematical models may be
on reduced weight, while maintaining structural denoted design support systems (DeSS), based
safety. One of the main reasons to use a Pareto on the algorithmic approach with strong
front rather than single solution is to have interactive capabilities. The MAESTRO software
flexibility to select a solution biased toward was a forerunner of those developments (Zanic
safety in the early stages. 2013 a, c), where the obtained results enable the
stakeholders’ direct involvement in the design
The penalty term of equation 6 is process (particularly in the most important
used for simple constraints such as the ratio of concept design phase) and support their educated
plate thickness to stiffener web thickness. The decisions, particularly if the techniques for
smallest adequacy parameter of limit states subjective decision making are used for
(mode I to mode VI) is an objective function. processing generated design variants. The
process stops at the frontier of structural
Equations 7, 8, and 9 define the range of the GA designer’s competence, that is, the Pareto
search space and are not used as penalty frontier (Zanic et al 2013 b), enabling the general
functions. designer to make his/her decisions using
obtained weight, cost, and safety measure values
Vedran Zanic, Visitor in the selection of the preferred design variant.
The authors have presented a well-written paper
on the mathematical models inbuilt into the The design problem characteristics are defined as
MAESTRO software, including FEM analysis follows.
and optimization and their application to tanker ONE: Design problem (DP) identification
structures. The authors begin their paper by (conceptual level) presents a first step of the
introducing the concepts and techniques of design process definition where objectives,
structural optimization starting with the early variables, and safety constraints with adjoining
work of Hughes, Mistreem and Zanic, and of load conditions/cases are determined or selected.
Rigo and Fleury. The authors subsequently • The determined design objectives are
present their comprehensive method for weight, simplified cost, and measure of
performing structural analysis using ultimate design safety/robustness (represented with a
strength concepts and structural optimization simplified safety attribute). Those
based on the selected heuristic method. As the objectives, used to demonstrate the
second objective of the paper, the authors present procedure, can be extended by the
examples demonstrating the design of cargo stakeholders’ expertise to more complex
space amidships for two double-hull tankers to forms.
illustrate the method when applied to practical • Selected design variables among the set of
problems. The first example considers the design descriptors were mainly scantlings,
preliminary design phase (PDP) model and the while topology was the variable only in the
second example is of a tanker for concept design stiffened panels of the coarse mesh model.
problem (CDP) and PDP. CSR rules also were Material type was considered as a fixed
applied. The savings were achieved in weight parameter.
objective with the simultaneous increase in • Selected design criteria functions
safety measure due to more logical material (constraints and attributes) were evaluated

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 155
for the respective load conditions with optimization solvers of the heuristic type are
adjoining load cases (wave loads). In the used in SM.
presentation, a part of the CSR defined loads Summary: FEM models developed (coarse mesh
were used for the set of MAESTRO ultimate for the CDP and fine mesh for the PDP) seem
strength panel criteria (ALPS/ULSAP). adequate. However, CSR require three-hold-
models for each hold to be analyzed and one
Note that the design safety-based objective should be careful to get all the required
function is obtained from constraint functions responses.
(specifying minimum permissible safety level)
and requiring its maximization for the ‘best’ Load model
design (Zanic et al, 2013d). • Implemented loads included all load
components: global forces, bending
TWO: Design Model formulation (algorithmic moments, pressures, accelerations, etc.
level) included the selection of basic • Deterministic definition was based on the
mathematical models: calculation of load component/load patterns.
• MAESTRO design analysis model (AM) for A balancing procedure was inbuilt into the
technical (response, feasibility criteria) and tool.
economical evaluations (cost criteria, • Design pressures (extreme value) are added
safety/risk measure) for the feasibility of calculations.
• Design synthesis model (SM) for objective Summary: The load model is used for the
decision making using the presented demonstration of large-scale analysis needs. The
heuristic optimization model/method. full load set, defined by the CSR, put all
structural elements in their most unfavorable
The presented methods in ßß and SM are general states and requires a large number of load cases
enough to enable the validating of innovative per hold, some of them artificial.
new ship concepts as well as generating
competitive designs for standard tanker types, Response model
based upon requirements for realistic design • Complex FEM response calculations needed
procedure regarding accuracy and timing. The for large structural models are possible, but
authors should present more data on timing, for they affect design procedure via the required
specified applied hardware, to substantiate this accuracy of stress/displacement/fatigue
aspect of the design problem. levels and as time penalty in synthesis
algorithms.
Structural model • MAESTRO FE modeling gives good insight
• The design model was defined via a set of into stresses and deformations and is
design descriptors (fixed parameters and capable, by a direct link to safety criteria, of
design variables are its subsets) and based giving efficient and fast answers regarding
on the midship section design embedded design feasibility in a structural optimization
into the full ship model. It included context.
scantlings of structural elements, whereas Summary: FEM responses, linked to the
geometry of structure and material were analytically defined safety criteria obtained by
treated as parameters. MAESTRO, are very fast and applicable in the
• Ship general arrangement, which defines optimization problems. Non-linear FEM is still
global structural topology, was fixed for the very time consuming on the analysis level and
structural designer by different ship difficult to efficiently include in the optimization
requirements/functions. loop (large number of iterations) except for the
• The variables used were continuous smaller models (local sub-structures).
(scantlings) and discrete (number of
stiffeners). Scantlings given by the available Adequacy/feasibility model (deterministic)
values (thickness of plates, available rolled • Large number of adequacy (for example,
profiles, etc.) can be applied in the safety) criteria checks, including the most
standardization phase. important ultimate strength ones (on the
• Rigid splines were an effective technique for panel level), based on library of various
practical adjustment of mesh size. modes of failure (ALPS/ULSAP), under
• Dimensionality in AM can be high as combined loads are applied.

156 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
• Incremental-iterative procedure (Smith- capable of expansion to novel design
based or similar) for evaluation of hull girder requirements.
ultimate strength is required (CSR) and may Summary: The MAESTRO-based optimization
be used as optimization based safety method seems practical for use in design offices.
objective (Zanic et al 2013d). AM and SM data are easily modified in the
Summary: An excellent library (Paik, Hughes) of interactive mode, opening the road to hybrid
structural safety criteria is inbuilt in MAESTRO solvers combining heuristic and analytical
and, provided that the load set is sufficient, can approaches.
guarantee structural safety for most of the CSR
requirements. Safety factors can be further Subjective decision making/GUI
refined. Automatic failure patch definition seems • Final selection of preferred design is
to be a way ahead in the efficient design forensic influenced by the subjective reasoning of the
investigations. structural and general designer, owners, and
shipyard management (multi-stakeholder
Quality /KPI approach can be needed, Cudina et al 2010).
• The design procedure requires only a • Suitable graphical representation of the
comparison of competing designs, therefore design variants is missing in the GUI.
the authors used relative values in defining Summary: Interesting results were obtained,
the attributes excluding considerations that particularly regarding comparison of results
are the same, or similar, for different obtained via the PDP fine mesh with the CDP
designs. macro element based one. A subjective decision
• Design quality measures, such as simplified making tool should be a part of the interactive
cost and safety measure functions, are part design environment to support DM.
of that group.
Summary: Other key performance indicators THREE: Design Problem solution
should be opened to stakeholders, including (procedural level)
other relevant aspects of design, such as more Application of the design procedure with
elaborate yard-based building cost, maintenance practical implementation of MAESTRO AM
cost, lifecycle cost, ship stability, etc. regarding inbuilt into the heuristic SM should enable
needs of multi-criteria decision making for either efficient synthesis, possibly using surrogate
multi-objective or multi-attribute approach. modeling (Prebeg et al. 2014), due to the
required workload and time available for the
DeSS definition given design phase.
• Flexibility of the design model definition is
one of the first priorities to accommodate Development of parallel processing on modern
the designer’s needs. PDP and CDP model computers also fits very well with presented
meshes were tested. design method where, despite the spiral character
• Convergence of the procedure was of the overall design process, many of the
demonstrated. calculation steps are parallel in nature. The
Summary: MAESTRO is a very flexible tool as increased speed of engineering hardware opens
demonstrated here and in many of its other the possibility of incorporating numerous and
applications. Convergence of optimization complex design criteria into the realistic multi-
procedure was very fast (since 1980 Hughes et objective design procedure.
al. paper to the present example). Equilibrating
(used by the authors) or compatibility approach Finally, the authors succeeded in showing the
both converged efficiently (Zanic, 2013c), potential of structural analysis combining FEM
measured in number of ship (FEM) cycles. and ultimate strength criteria for complex design
tasks. The authors are encouraged to include
Optimization solvers more data on timing and size of problem
• The penalty function approach seems a (dimensionality of design and of criteria
proven way to the fitness calculation for the functions spaces) to further confirm its
constrained problems. applicability for practical design work.
• The design method, as shown in the
examples, was applicable in case studies of Additional references
modifying material and panel topology and Cudina, P., V. Zanic, P. Prebeg: Multiattribute
Decision Making Methodology in the Concept

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 157
Design of Tankers and Bulk-Carriers, PRADS structure. This enabled the structure to receive
2010, Rio de Janeiro, 2010. the global response while only focusing the
optimization on a single cargo hold.
Prebeg, P., Zanic V., Vazic B., Application of a
surrogate modeling to the ship structural design, A significant part of optimization includes the
Ocean Engineering, Volume 84, 1 July 2014, pp. potential cost savings. However, because cost
259-272. models can be very complicated and specific to
each shipbuilder, a cost model was not used for
Zanic, V.: Editorial paper “Methodology and this optimization. In its current state, scantling
Concepts for Multi-criteria Synthesis of Ship libraries can be used to define the design space
Structures,” Ships and Offshore Structures, enabling a user to manually add or remove
(2013a), Volume 8, Special Issue on Analysis structural entities based on their known
and Design of Ship Structures, Number 3-4, manufacturing and application costs. Taking this
pp.225-244. a step further, the optimization could enable
additional penalty functions to be applied to
Zanic, V., Andric, J., Prebeg, P. “Design specific structural entities in a user’s library to
Synthesis of Complex Ship Structures,” Ships encourage more economical designs. This
and Offshore Structures, (2013b), Volume 8, approach also enables a user to allow multiple
Number 3-4, pp.383-403. materials to be used in the design, which can also
have a significant effect on costs. Finally, the
Zanic, V., Andric, J., Hadzic, N. “Optimization current framework supports the addition of a
Procedure for LNG Carrier Structural Design,” more complex cost model, but that particular
Proc.IMechE, part M, Journal of Engineering for data was not available at the time this paper was
the Maritime Environment, (2013c) written.
doi:10.1177/1475090213494195.
We agree that part of the optimization process is
Zanic, V., Piric, K., Kitarovic, S. “Reliability a tool that enables the user to make a subjective
and robustness based design attributes for multi- design on the best design from the optimization
criteria decision making,” Proceedings of ASME results, especially in the case of comparing two
2013, 32nd International Conference on Ocean, models of the same structure. In the current state,
Offshore and Arctic Engineering (OMAE), the optimization results provide a user-defined
Nantes, (2013d). number of viable designs with a summary of
their weight and structural safety. Additionally,
Authors’ Response visual tools exist that enable the user to easily
Thank you for your comments and very thorough present the plating thicknesses, stiffener layouts,
review and discussion of the paper. You make and beam properties. Each of the possible
very good points about the timing and resources designs can be “applied” to the full structural
needed to get the most out of the structural model and used to evaluate things like cross
optimization process. Although the optimization sectional properties and hull girder ultimate
can potentially provide weight reduction and strength; however, this capability is currently
increased structural safety, it also needs to be done through a manual process. We agree that it
performed at a time in the design process that would be beneficial to the user to automate more
makes such structural changes possible. In terms of these capabilities to help assist in the selection
of required computational power, both process.
optimization runs finished in under 3 hours using
a relatively high-powered laptop. You make a Jon Downes, Visitor
good point that much of the calculations could be This paper presents the progress made by the
performed in parallel, potentially saving authors in bringing structural optimization into
significant amounts of time and allowing a user the concept design phase. The method presented
to explore an even larger design space. is a two-step approach based on the behavior of a
stiffened panel with the optimization procedure
In both examples, only a single cargo hold model being undertaken using an analytical-based
was used for the optimization, however the full approach to determine the ultimate limit state.
ship finite element model was run at each Design using ultimate limit states has been
iteration to update the loads for cargo hold around for quite some time and is accepted as
offering significant advantages to the

158 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States
engineering design process. However, such that this would be a very interesting aspect to
processes are not well employed at the concept investigate in future work.
design stages and the efforts of the authors
should be applauded. The coarse mesh model in this study uses a
stiffened plate element, where the stiffeners are
The effects of initial imperfections and residual part of the plate element. For the fine mesh
stresses on the ultimate strength of stiffened model, the stiffeners are explicitly constructed as
panels are known to be significant. It would be beam elements. The main reason that a coarse
of interest to hear from the authors if these mesh model gives better weight saving is
effects were considered within the limit state because the stiffener spacing for the coarse mesh
evaluation and if so, what the relative influence model is a design variable and can be changed
on the resulting optimized panel structures were. without altering the topology of the finite
Similarly, it would be of interest to know if there element model. If the fine mesh model can be
was a significant change in the overall ultimate reconstructed automatically when the stiffener
strength and post-buckling behavior of the panel spacing is changed, the same level of weight
between the pre- and post-optimized versions. saving can be achieved for the fine mesh model.

In consideration of the larger midship sections The focus of this paper was presenting the
investigated, the approach suggests that if the proposed optimization method and the results
optimization is undertaken initially by a coarse using two cargo hold examples. The
mesh model, then more significant changes in optimization was based purely on the local panel
the structure can be achieved than by that ultimate strength and the hull girder strength was
achieved by a fine mesh model. This is not taken into account in this paper. However,
potentially due to the fine mesh model being we authored a paper that was presented at
more highly constrained. It would be interesting COMPIT 2014 in which this optimization
to see whether, if these constraints were lifted, method was used to optimize a single midship
the same effect would be achieved. It would be section, again only focusing on the local panel
of interest to hear from the authors if they ultimate strength. What we found in that case
considered the overall global hull girder ultimate was that, as a result of the optimization, the
strength and if so, how did the original and the structural weight was reduced while also
optimized structure compare? This would give increasing the cross section moment of inertia
further information for assessing if the optimized and hull girder ultimate strength, as shown
structure is a viable solution. below.

The approach suggests that a significant weight


saving might be achieved by using this approach
and that this can be achieved within a reasonable
number of iterations, making the approach less
computationally expensive than other available
techniques. It would be of interest to hear from
the authors if they have undertaken comparative
studies with these alternative techniques to
quantify the differential of this approach and
their resulting views.

Authors’ Response
Thank you for your comments and discussion
topics. You make a very good point regarding
the effects of initial imperfections and residual
stresses on the ultimate strength of panels. In this
particular paper, the panels were modeled as the
idealized structure and did not take into account
imperfections or residual stresses. However, the
ALPS/ULSAP limit states can account for such At this stage in the development of this
imperfections and residual stresses and we agree optimization method, there have not been any
direct comparisons to other optimization

Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States 159
techniques. We agree that it would be of interest
and value to perform a comparison of this
technique with others, both in terms of their
results and required computation time.

160 Oil Tanker Cargo Hold Structural Optimization Using Ultimate Limit States

You might also like