You are on page 1of 16

Journal of Marine Science and Technology

https://doi.org/10.1007/s00773-021-00804-2

ORIGINAL ARTICLE

Reliability‑based multidisciplinary design optimization


of an underwater vehicle including cost analysis
Hadi Gholinezhad1 · Seyed Hosein Torabi2

Received: 1 March 2019 / Accepted: 5 February 2021


© The Japan Society of Naval Architects and Ocean Engineers (JASNAOE) 2021

Abstract
Today, due to the complexity of systems and interactions among their subsystems, the design optimization of a system is
highly difficult and costly. Multidisciplinary optimization is an approach, in which interactions among different disciplines
are taken into account, and it attempts to optimize all disciplines, simultaneously. In the design process of a system, there
is usually some uncertainty in parameters. This uncertainty creates some challenges in the design process and affects the
systems performance. To cope with the uncertainty, robust design and reliability-based design approaches are developed. In
this paper, a reliability-based multidisciplinary design optimization is presented, in which some of the problem parameters
are uncertain. In this regard, it is assumed that some of the problem parameters are in the form of fuzzy numbers. Moreover,
in this problem cost is considered as one of the design disciplines, due to its importance in engineering problems. To solve
the proposed model, a solution method named the sequential optimization and reliability assessment is presented in which
Genetic Algorithm and Particle Swarm Optimization are used to solve the deterministic problem in each iteration. Finally,
the design of an autonomous underwater vehicle including cost analysis is investigated and two solution methods are applied.
The obtained results from two methods are compared and some conclusion are made. The results show that improving the
reliability between 0.5 and 0.85 is more cost-effective. However, some other factors besides the cost play a role in choosing
the reliability level that must be considered.

Keywords Reliability based multidisciplinary design optimization · Fuzzy numbers · Cost · Sequential optimization and
reliability assessment · Genetic algorithm · Particle swarm optimization

1 Introduction decisions made by other teams. Hence, it is necessary to


be the decision-making process sequentially repeated by
Complex systems consist of different subsystems that affect various teams so that a design accepted by all teams can
each other and the performance of each subsystem is affected be achieved. This design process is time consuming and
by others. Thus, it is necessary to consider the interactions costly and thus is not favorable. Multidisciplinary design
between different subsystems of a system. Traditionally, the optimization is a multivariate optimization approach that is
design of different subsystems of a system is usually car- often used to optimize the whole system. In this approach,
ried out by some teams that have a high degree of expertise variables from different disciplines and interactions between
in a particular discipline. Each team uses the experience them are simultaneously considered and attempt is made
of its members to develop a practical design. However, the to move to the optimal solution or near it in all disciplines.
goals of teams are in contradiction with each other, and the Uncertainty is a phenomenon that there is in many real-
decision taken by a team is usually in contradiction with world problems. There are different categories for uncer-
tainty. Uncertainty is usually divided into two main catego-
* Hadi Gholinezhad ries: parametric uncertainty and nonparametric uncertainty.
gholinezhad.hadi@gonabad.ac.ir Parametric uncertainty relates to the lack of proper recogni-
tion of parameters, that is, whether values considered for
1
Department of Industrial Engineering, University the parameters of the problem are the same as their actual
of Gonabad, Gonabad, Iran
values. Nonparametric uncertainty relates to the correctness
2
Faculty of Aerospace Engineering, Khaje Nasir Toosi of the model, that is, whether the proposed mathematical
University of Technology, Tehran, Iran

13
Vol.:(0123456789)
Journal of Marine Science and Technology

model fully addresses the problem [1]. Parametric uncertain- attempts are made to reduce the probability of constraints
ties are usually expressed in terms of the probability density violation so that they are lower than a desirable amount [6].
function, membership function, or an interval, depending on In robust optimization, the problem formulation is such that
the type of application. The interval display for nondeter- it minimizes the variance of the objective function. How-
ministic parameters is used when the available information ever, in reliability-based optimization, the solution space is
is very raw and primary and only upper and lower limits are divided into two areas: feasible area and infeasible area, and
known. The probability density function is the most com- the goal of optimization is that the final solution sufficiently
plete representation for a nondeterministic parameter. The approaches the feasible area so that the probability of con-
membership function that is used in the fuzzy logic method straints violation is less than a desired value [7]. Uncertainty
lies between the interval display and probability density based design optimization was taken into account a long
function, in terms of details [2]. In a traditional design, one time ago [8, 9], and there has been numerous researches
way to deal with uncertainty is to use a safety factor to take in this field [10, 11]. Ahn and Kwon proposed an efficient
all potential uncertainties into account simultaneously [3]. reliability-based multidisciplinary design optimization
A safety factor is often determined according to the past (RBMDO) strategy to enhance computational efficiency.
experiences and prior knowledge about a system, and so The proposed strategy utilized a single level reliability-based
far, there is no straightforward unitary method for its proper design optimization approach, in which reliability analysis
definition. Considering large values for a safety factor leads and optimization are conducted in a sequential manner by
to obtain highly conservative solutions, which worsens the approximating limit state functions [12]. Youn et al. pre-
cost or other objective functions. While considering small sented an application of the performance measure approach
values for a safety factor reduces the reliability of the sys- and hybrid mean value method for reliability-based design
tem. In addition, the use of past experience to determine optimization in the crashworthiness of a large-scale vehicle
the value of a safety factor for the design of new systems side impact. They showed that the proposed reliability-based
may be inappropriate or impossible. Therefore, it is neces- design optimization approach was very effective in obtaining
sary to develop more advanced and efficient methods to deal a reliability-based optimum design [13]. Zhang and Huang
with uncertainty. These methods are known as "uncertainty proposed formulations of mixed variables MDO (MVMDO),
based design" [4] or "non-deterministic approaches" [5], in which random and fuzzy variables were simultaneously
which help to solve the following two issues: (1) improve used. The MVMDO overcame difficulties caused by insuf-
robustness and reduce design sensitivity to uncertainty; (2) ficient information for uncertainty. They also proposed a
enhance the system reliability and reduce its failure proba- method of MVMDO within the framework of the sequen-
bility. Regarding these two issues, there are two main uncer- tial optimization and reliability assessment (SORA). The
tainty based methods namely "robust design optimization" proposed method enables designers to solve MDO problems
and "reliability based design optimization". The application in the presence of both Aleatory uncertainty and epistemic
of these methods, as shown in Fig. 1, is determined by the uncertainty. Moreover, the proposed method can efficiently
degree of uncertainty and impact of uncertain events. reduce the computational demand [14]. Aleatory uncertainty
Robust optimization is a field of optimization theory, arises because of the natural and unpredictable variety of the
in which outputs are not sensitive to changes in uncertain system performance. Experts’ knowledge is not expected to
parameters. However, in reliability-based optimization, reduce aleatory uncertainty, although it may be useful in
quantifying the uncertainty. Thus, this type of uncertainty
is sometimes referred to as irreducible uncertainty. On the
contrary, epistemic uncertainty is due to a lack of knowledge
about the system’s behavior, which is conceptually resolv-
Performance loss Catastrophe

No engineering Reliability based able. Epistemic uncertainty can, in principle, be eliminated


Impact of event

applications design optimization by sufficient study and therefore, expert judgments may be
helpful in its reduction [15]. Aleatory uncertainty in per-
meability arises from local spatial diversity due to hetero-
geneity in the formation. The epistemic uncertainty arises
Robust design Reliability is from uncertainty about the spatially integrated permeability
optimization not an issue [15]. Aleatory uncertainty is also referred to in the literature
as variability, stochastic uncertainty, inherent uncertainty,
Small perturbation Extreme event and irreducible uncertainty. Epistemic uncertainty is also
Uncertain event referred to in the literature as subjective uncertainty, reduc-
ible uncertainty, and model form uncertainty [16]. Alea-
Fig. 1  Uncertainty based design domains tory uncertainty is used to describe the inherent variation

13
Journal of Marine Science and Technology

associated with a physical system or an environment under problems. They illustrated the implications of the MDO
consideration [17]. Yuan hang Hou et al. presented a mixed problem formulation for the tractability of the resulting
aleatory/epistemic uncertainty analysis for ship hull form design optimization problem by examining a representative
optimization design and the influences of these two types of class of MDO problem formulations known as collaborative
uncertainty were considered. The proposed model was con- optimization. They also discussed an alternative problem
ducted to find a minimum energy efficiency design indicator formulation, distributed analysis optimization that yielded
(EEDI). EEDI, a mandatory regulation, is closely related to a more tractable computational optimization problem [26].
many design parameters of ships’ hulls, which were con- Agarwal and Renaud developed a framework for perform-
ventionally set to be constant. However, it is often the case ing reliability based MDO using approximations. They used
that considerable parameter fluctuations occur during actual response surface approximations of limit state functions to
navigation, so it is more reasonable to state the important estimate the probability of failure. They also incorporated an
parameters as uncertainty variables. In this paper, transfor- outer loop to ensure that the approximate reliability-based
mation methods of uncertainty targets and constraints to design optimization converged to the actual most probable
certainty targets were employed to conduct a mixed uncer- point (MPP) of failure [27]. Chen et al. presented a new
tainty optimization. The proposed uncertainty optimization method for reliability-based optimization which required
method had the effectiveness and superiority, such that the only a modest increase in computational cost over that of
proposed method could obtain the optimum result with deterministic design optimization. The presented method
acceptable constraints and failure plausibility. The results was implemented for comparison with previous methods
showed the more detailed the epistemic uncertainty vari- and appeared to be robust in terms of convergence from
ables, the more accurate the uncertainty distribution descrip- the arbitrarily selected initial design points to the solutions
tion, but this requires longer computing times [18]. The most determined by the previous methods [28]. Meng et al. com-
commonly mathematical representation used for aleatory bined a subset simulation-based reliability analysis (SSRA)
uncertainty is a probability distribution. Propagation of a approach with MDO to improve computational efficiency in
distribution through a modeling and simulation process has reliability-based MDO problems. Furthermore, the SORA
been well described in many researches (see, for example, approach was utilized to decouple a reliability-based MDO
Refs. [19–23]). Epistemic uncertainty is defined as a lack problem into a sequential of deterministic MDO and reliabil-
of knowledge or information in any phase or activity of the ity evaluation problems. The formula of MDO with SSRA
modeling process [17]. An increase in knowledge or infor- within the framework of SORA was proposed to solve a
mation can lead to a reduction in the predicted uncertainty of design optimization problem of a hydraulic transmission
the response of the system. Examples of sources of epistemic mechanism [29]. Yao et al. proposed a reliability-based
uncertainty are limited understanding of complex physical MDO procedure based on combined probability and the evi-
processes, the occurrence of fault sequences or environmen- dence theory to address a reliability-based MDO problem
tal conditions not identified for analysis of the system, and under mixed aleatory and epistemic uncertainties. They used
when there is insignificant or no experimental data for an the sequential optimization and mixed uncertainty analysis
unknown fixed physical parameter [16]. Epistemic uncer- (SOMUA) method to decompose the traditional double-
tainty can be modeled by the possibility theory [14]. level reliability-based optimization problem into separate
Yang et al. investigated the methodology development deterministic optimization and mixed uncertainty analysis
and application of reliability-based multidisciplinary design sub-problems solved sequentially to reduce the computa-
optimization to vehicle crashworthiness under some con- tional load [30]. Batill et al. investigated the use of ana-
straints. They investigated two optimization methodolo- lytic models and numerical simulation in the MDO process.
gies and demonstrated them by applying to a full vehicle They investigated how issues of physical process variability,
design of multiple impact modes with uncertainties taken information uncertainty and use of models and simulations
into consideration [24]. Agarwal et al. investigated how influenced the design decision process [31]. McAllister and
uncertainty could be quantified in multidisciplinary systems Simpson presented a multidisciplinary robust design optimi-
analysis subject to epistemic uncertainty associated with zation formulation to evaluate uncertainty encountered in the
disciplinary design tools. They used the evidence theory design process. The formulation introduced by them was a
to quantify uncertainty in terms of uncertain measures of combination of a bi-level collaborative optimization frame-
belief and plausibility [25]. Alexandrov and Lewis explored work and a multi-objective approach of a compromise deci-
the analytical features of MDO problem formulations that sion support problem. To demonstrate the proposed frame-
had significant practical consequences for the ability of work, they presented the design of a combustion chamber
non-linear programming algorithms to solve the reliably of an internal combustion engine containing two subsystem
and efficiently in the resulting computational optimization analyses [32]. Du and Chen developed a multidisciplinary

13
Journal of Marine Science and Technology

robust design procedure that utilized efficient methods for x an uncertain variable (for example, a fuzzy number)
uncertainty analysis. Different from the previous uncertainty
analysis techniques, the proposed techniques brought the 𝜇A (x) ∈ [0, 1] Membership function of a fuzzy number
[0,1] (𝜇A (x) ∈ [0, 1])
features of the MDO framework into consideration. They
𝛼I An interval variable
developed a system uncertainty analysis method and a con-
𝛼 ,𝛼 The lower and upper bounds of 𝛼 I
current subsystem uncertainty analysis method to estimate
the mean and variance of the system performance subject 𝛼c The mid value of 𝛼 I (𝛼 c = (𝛼 + 𝛼)∕2)
to uncertainties associated with both design parameters and Δ𝛼 The radius of 𝛼 I (Δ𝛼 = (𝛼 − 𝛼)∕2)
design models [33]. 𝛽 A fuzzy number
As previously mentioned, parametric uncertainty can 𝛾 A cut level
be expressed in three ways, one of which is the mem- 𝛽𝛾l The converted fuzzy variable 𝛽 into an interval vari-
able under the cut level γ
bership function that is used in the fuzzy logic method.
𝛽𝛾 ,𝛽𝛾 The lower and upper bounds of 𝛽𝛾l
With the possibility theory, fuzzy variables are utilized
𝛽𝛾c The mid value of the fuzzy variable 𝛽
to represent epistemic uncertainties (uncertainties with
The vector of general design variables
insufficient data) [34]. It has been pointed out that when ds
The vector of local variables of the discipline i
little information is available for input data, the possibility di
The vector of output variables of the discipline i
based method is better as it provides a more conservative yi
f The objective function
design than the probabilistic design that is consistent with
The kth design constraint (k = 1, 2, … , K )
the limited available information [35]. A common goal in gk
Uncertain general variables
designing and optimizing any system can be to gain more xs
Local uncertain variables of the discipline i
profit, and in almost all systems, cost can play an impor- xi
xsc,xic The mid value of xs and xi
tant role in the design of a system. Thus, it is an important
Pr () The reliability of the constraints
factor for designers. Cost is a discipline that has interac-
zk The fuzzy variable used in the kth design constraint
tion with other disciplines [36, 37]. Accordingly, in this
zak The allowed value for zk
paper, cost is considered as a discipline. Moreover, in this
Rk The desired value for the reliability of the kth design
paper, a reliability-based multidisciplinary design opti-
constraint
mization problem is presented for the first time, in which
uncertain parameters are displayed as fuzzy numbers, as
well as the cost is considered. The main contribution of 2.1 Fuzzy numbers
this paper is to use fuzzy numbers to deal with uncertainty
and consider cost as a discipline in reliability-based mul- The fuzzy set theory was introduced by Zadeh [38]. A fuzzy
tidisciplinary design optimization problem. The rest of set is defined by a membership function (𝜇A ∶ X → [0, 1]), in
the paper is organized as follows. In Sect. 2, the problem which the membership grade of an element X to a fuzzy subset
formulation is described. Section 3, the solution method is A is a precise number (𝜇A (x) ∈ [0, 1]).
presented. Design optimization of an autonomous under- Assume that 𝛼 I is an interval variable. Interval variables
water vehicle is investigated as a case study in Sect. 4. The can be described by their lower and upper bounds as follows:
conclusions and future research suggestion are discussed
[ ]
in Sect. 5. 𝛼 l = 𝛼, 𝛼 = 𝛼 c + Δ𝛼𝛿𝛼

(1)

where 𝛿𝛼 = [−1, 1].


2 Problem formulation Moreover, assume that 𝛽 is a fuzzy variable with a mem-


bership function 𝜇(𝛽). A fuzzy variable can be expressed as
In this section, the problem formulation is presented for the interval variables through the level-cut strategy. If 𝛾 is a cut
reliability based multidisciplinary design optimization. As level, then, a fuzzy variable can be decomposed into an inter-
mentioned above, in this paper, the uncertain parameters val variable as Eq. (2).
are displayed as fuzzy numbers. At first, a brief descrip- [ ]
tion of fuzzy numbers is given and then multidisciplinary 𝛽𝛾l = 𝛽𝛾 , 𝛽𝛾 = 𝛽𝛾c + Δ𝛽𝛾 𝛿𝛽𝛾 (2)
design optimization is discussed and the problem formu-
lation is presented. The notations used in this section are In this case, the mid value of the fuzzy variable can be
as follows: defined as Eq. (3).

13
Journal of Marine Science and Technology

If zai and zi are fuzzy variables with a triangular mem-


∫ (3)
𝛽𝛾c = 𝜇(𝛽)d𝛽 bership function as Fig. 2, then, the possibility to satisfy
constraints under different cut levels can be calculated as
Eq. (12).
2.2 Reliability based multidisciplinary design {
optimization ( ) 0 if 𝛾 < 𝛾 ∗
p(𝛾) = poss gi > 0 =
1 if 𝛾 > 𝛾 ∗ (12)
The general formulation of the deterministic multidiscipli-
nary design optimization is as follows:
If 0 and 1 are the failure and safety states under the cut
( ( ))
mind=[ds ,di ] f ds , di , yi ds , di , yi≠j (4) level γ, respectively, then, the reliability of constraints can
be calculated as Eq. (13) [39].
1 𝛾∗ 1
2.2.1 s.t. ( )
∫ ∫ ∫
( ( )) Pr gi > 0 = p(𝛾)d𝛾 = 0d𝛾 + 1d𝛾 = 1 − 𝛾 ∗ (13)
gi ds , di , yi ds , di , yi≠j , i = 1, 2, … , N (5) 0 0 𝛾∗

The equations that describe the behavior of the system


and coupling between disciplines can be written as follows:
( )
yi = Di yj(i≠j) , d , i = 1, 2, … , ny (6) 3 Solution method
Suppose that the design constraints are as follows:
In this paper, the SORA is used as the solution method for
gk = zak − zk > 0 (7) the presented multidisciplinary design optimization model.
SORA was first introduced by Du and Chen [40]. As there
Then, the general formulation of the reliability based is a deterministic optimization problem in each iteration of
multidisciplinary design optimization is as follows: SORA, genetic algorithm (GA) and Particle Swarm Opti-
( ) mization (PSO) are used to find the optimal solution of
min f ds , di , xsc , xic (8)
deterministic problems. In the following, a brief descrip-
s.t. tion of these algorithms is provided. The notations used in
this section are as follows. The other notations not men-
Pr gk = zak − zk > 0 ≥ Rk , k = 1, 2, … , K
( )
(9) tioned here are the same as in the previous section.

In which, zk and the discipline analysis can be as follows: g(X) A limit-state function
( ) ( g(X) = h(X) − c)
zi = Fzi xs , xi , ds , di , yij (10)
X Input variables in the original
design space ( X = {x1 , x2 , ..., xn })
( )
yi = Fyi xs , xi , ds , di , yij (11) U Input variables in the standard
normal space U = {u1 , u2 , ..., un }
Fj (xj ) CDF of the input variable xj
Φ−1 (.) Inverse of standard normal distri-
bution function
𝜃 The shortest distance from the
points on the limit-state surface
to the origin in U space
1 gki The ith limit-state function in kth
iteration
nPop Population size in GA
rC Crossover rate in GA
rM Mutation rate in GA
𝜇 Probability of changing gene val-
ues of the parent chromosome in
a mutation operator in GA
0
MaxIt Maximum number of iterations
in GA
Fig. 2  The safety possibility of a triangular fuzzy system

13
Journal of Marine Science and Technology

( )
g(X) A limit-state function uj = Φ−1 Fj (xj ) ; j = 1, ..., n (15)
( g(X) = h(X) − c)
nf Flock size in PSO where Φ−1 is the inverse of the standard normal distribu-
xi (t) Position of particle i at iteration t tion function. The transformation maintains the CDFs being
in PSO identical both in X space and U space [45]. Thus, the limit-
vi (t) Velocity of particle i at iteration state functions can be rewritten as
t in PSO
w Inertia weight in PSO g(U) = h(U) − c (16)
Hasofer and Lind defined the shortest distance from the
points on the limit-state surface to the origin in U space that
can be obtained by the following minimization problem [42]:
3.1 Sequential optimization and reliability
assessment (SORA) 𝜃 = min ‖U‖
U (17)

One of the main advantages of SORA is that it requires the subject to


solution of a deterministic optimization problem. This fact
h(U) − c = 0 (18)
simplifies the computational implementation of the approach
[41]. In each iteration of SORA, a deterministic optimiza- The optimal solution of this problem is called the MPP. It
tion problem is solved and then the reliability analysis is can be concluded that the joint PDF on the limit-state surface
performed in the same iteration sequentially. But in other has its highest value at the MPP [43]. The MPP has the high-
conventional methods, there is a nondeterministic optimiza- est probability of producing the value of limit-state function
tion problem in each iteration that requires another loop to g(U) in the standard normal space [45]. In other words, it is the
analyze the reliability. In other words, in other methods, a point that contributes the most to the integral for probability
deterministic optimization problem should be solved in two estimation θ [42].
nested loops, but in the SORA, there is a loop in which a In each cycle of SORA, at first, an equivalent deterministic
deterministic optimization problem is solved once in each optimization problem is solved, which is formulated by the
iteration. Therefore, the number of deterministic optimiza- information of the MPP obtained in the last cycle. Then, the
tion problems solved in SORA is much less. design solution is updated, the new MPP is calculated by the
In SORA, a single loop strategy is used, in which the reliability assessment based on the information of the MPP
optimization process and reliability assessment are per- obtained in the last cycle and it is examined whether all the
formed sequentially. In each cycle, optimization and reli- reliability requirements are satisfied. Otherwise, the current
ability analysis are applied separately. Thus, there is no need MPP is used to formulate the constraint for the deterministic
for the reliability analysis during the optimization [40]. The optimization problem in the next cycle. In this problem, the
key concept in the SORA approach is to shift the boundaries constraint boundary is shifted to the feasible region by chang-
of violated constraints to make them feasible and prevent ing the locations of the design variables. When a new MPP is
shifting the boundaries of constraints that are satisfied. In sufficiently close to the optimal point obtained in the previous
each cycle of this method, an MPP is obtained. The MPP cycle, the changes in the transition vector approach zero, caus-
method was originally developed in the field of reliability ing the boundary of the constraints not to be shifted. In this
analysis [42]. approach, the reliability of the constraints improves progres-
In this method, the limit-state functions are defined as sively and the optimal solution for the probabilistic design can
Eq. (14) and then the probability that they are either bigger be found within a few cycles [40]. This process is repeated
or smaller than zero is evaluated approximately [43]. until all probability constraints are satisfied.
The flowchart of the SORA approach is depicted in Fig. 3.
g(X) = h(X) − c (14)
The deterministic multidisciplinary optimization problem in
where c is a constant. this approach (in third block of flowchart) is as Eqs. (19,20),
The MPP is formally defined in terms of an independ- in which k is the algorithm iteration number.
ent and standardized normal vector U = {u1 , u2 , ..., un }. ( )
The input variables in the original design space min f ds , di , xsc , xic (19)
( X = {x1 , x2 , ..., xn }) should be transformed into the standard
normal space. The most commonly used transformation is
s.t.
given by Rosenblatt [44] as follows:
gki = gk−1
i
+ Δk−1
i
> 0, i = 1, 2, … , N (20)

13
Journal of Marine Science and Technology

Fig. 3  The flowchart of the


SORA approach Start

System modeling

Solving multidisciplinary optimization


problem and determining the optimal
values for the design variables

Producing a set of
cutting surfaces γ

Finish
Obtaining converted intervals
and uncertainty analysis

The solution
Calculating probability based on is optimal
cutting surfaces and reliability levels
yes

Calculating the amount of ∆ ≤


constraints displacement

no

3.2 Genetic algorithm (GA) 3.3 Particle swarm optimization (PSO)

GA is a well-known meta-heuristic algorithm that has been PSO is a stochastic global optimization technique inspired
successfully used as a solution method for optimization by the social behavior of bird flocking or fish schooling. PSO
problems. GA was first introduced by Tomassini [46]. In this was first introduced by Kennedy and Eberhart [47, 48]. The
algorithm, each solution is represented as a vector or matrix algorithm starts with an initial flock that is usually produced
called chromosome. In each iteration, there is a fixed number randomly over a searching space where every bird or fish
of chromosomes that form the population. The population is called as a particle. The flock size is considered to be
size is considered to be nPop. GA starts with an initial popu- nf. Every particle flies with a certain velocity. The particles
lation that is usually produced randomly. find the global best position after some iteration. During
To create a new generation in each iteration, new off- the optimization process, the swarm moves towards the best
spring should be produced. These offspring are created by position and the global best position changes at each itera-
crossover and mutation operators. The crossover operator tion. The velocity of each particle changes throughout the
is performed at a rate of rC and the mutation operator is algorithm. Factors influencing the velocity of a particle are
performed at a rate of rM. In a mutation operator, the gene its momentum, the influence of its best position and the best
values for the parent chromosomes are changed with a prede- position of its neighbors. Indeed, each particle can adjust its
termined probability of μ. GA is terminated when a stopping velocity vector based on these factors in each iteration. Then
condition is reached. The proposed GA is terminated when the particle computes a new position that is to fly to.
a predetermined number of iterations which is considered to The velocity and position of the ith particle (­ vi and ­xi) at
be MaxIt is reached. (t + 1)th iteration are updated using the following equations:

13
Journal of Marine Science and Technology

( ) ( ) Table 2  Variables under uncertainty


vi (t + 1) = w.vi (t) + c1 .r1 . pbesti (t) − xi (t) + c2 .r2 . gbest(t) − xi (t)
(21) Variable under uncertainty Symbol Unit Lower limit Upper limit
xi (t + 1) = xi (t) + vi (t + 1) (22) Velocity V m/s 5 10
Sonar horizontal beam- BWh Deg 50 120
where ­pbesti is the best position of the ­ith particle, gbest is width
the best position of the total particle swarm, c1 and c2 are
constants, r1 and r2 are random variable with a uniform dis-
tribution between 0 and 1 and w is inertia weight. ( )
> 0 ≥ R1
This change of velocity and position is repeated in each mT
Pr g1 = 1 − (24)
iteration until the final solution is reached. mTmax

( )
> 0 ≥ R2
Tdet
4 Case study Pr g2 = 0.8 −
TT
(25)

In this section, the multidisciplinary design of an autono-


Pr g3 = 8 − LT > 0 ≥ R3
( )
mous underwater vehicle with six disciplines (subsystems) (26)
is investigated. The disciplines considered in this paper are
sonar, hydrodynamics, structure, power/propulsion, tactics Pr g4 = 7 − 𝛼 > 0 ≥ R4
( )
(27)
and cost. Table 1 shows the design variables for each disci-
pline. Further, the variables under uncertainty are presented
Pr g5 = 7 + 𝛼 > 0 ≥ R5
( )
in Table 2. (28)
The formulation for multidisciplinary design optimization
Pr g6 = 5 − Lb > 0 ≥ R6
of the autonomous underwater vehicle is as follows: ( )
(29)
DR × BWh × V × Tmax
Pr g7 = 1 + Lb > 0 ≥ R7
max pdet = ( ( ))2 (23) ( )
180 × R0 + Vtar Tdet + Tmax (30)

Pr g8 = 1 − Lm > 0 ≥ R8
s.t. ( )
(31)

Pr g9 = Lm − 0.3 > 0 ≥ R9
( )
(32)
Table 1  Design variables for the numerical example

Pr g10 = 0.85D − Ds > 0 ≥ R10


Discipline Variable Symbol ( )
(33)
General variables Sonar horizontal beamwidth ­(BWh) X1

Pr g11 = Ds − 0.15 > 0 ≥ R11


Dynamic viscosity of seawater (­ Lb) X2 ( )
Velocity (V) X3 (34)
Sonar directivity index (DI) X4

Pr g12 = 120 − BWh > 0 ≥ R12


Sonar Circle diameter/Sonar array (Ds) Y1 ( )
Sonar range (DR) Y2 (35)
Hydrodynamics Propulsion force (­ TF) Y3
Pr g13 = BWh − 10 > 0 ≥ R13
Seawater absorption coefficient (α) Y4 ( )
Lift force (­ LF) Y5 (36)
Drag force (­ DF) Y6

Pr g14 = cos t(BWh , V, DI) − BUDGET > 0 ≥ R14


Power/propulsion Energy stored in the battery (E) Y7 ( )
Motor length ­(Lm) Y8 (37)
Power required by the motor (Pow) Y9
The objective function is considered as the maximum
Sizing/structure Total volume of the vehicle (­ VolT) Y10
Total mass of the vehicle (­ MT) Y11 target
( )detection probability and the normalized total
( )mass
, the ratio of detection time to total time Tdet and
mT T
Area of the wetted surface (­ Sb) Y12
m
Total length ­(LT) Y13 Tmax T
the lower and upper values of some variables such as the
Tactic Probability of target detection ­(Pdet) Y14
Target detection time ­(Tdet) Y15 total length ( LT ), the angle of attack (𝛼 ), the battery length
Maximum mission time ­(TT) Y16 ( Lb), the motor length ( Lm), the sonar nose diameter (Ds) and
Range (Rng) Y17 the sonar horizontal beam width ( BWh ) are considered as
Cost Maximum budget (BUDGET) Y18 constraints. ­R1 to R ­ 14 are the reliability level of design

13
Journal of Marine Science and Technology

constraints that the model tries to improve them to a desir- 4.2 Hydrodynamics


able value.
In the following, a brief description of the disciplines is In this discipline, the longitudinal stability equations as well
provided. as the calculation of the drag and lift force is considered. The
drag force for an underwater vehicle usually includes fric-
tion drag and pressure drag. The friction drag is a function
4.1 Sonar of the wetted surface of the device and velocity whereas the
pressure drag is a function of the shape and surface of the
In this discipline, the device searcher to detect the target is
device forehead. The friction drag is created by the shear
considered as a passive sonar. To have a wider visibility and
stress caused by the boundary layer resulting from the flow
beamforming property, an array of acoustic sensors is used
of viscous fluid when passing over the body of the device,
in the passive sonar. The sonar array is as a flat circular array
and it is determined based on the Reynolds number and dis-
with diameter ­Ds that is mounted in the nose of the device.
tribution of the pressure on the body. The pressure drag for
One of the variables in sonar discipline is the beamwidth.
a completely sunken object in a non-viscous fluid is zero.
This design variable is used to define the search area of the
However, there is pressure drag for a viscous fluid due to
sonar. Determining the size of the sonar array (the length,
the presence of the boundary layer and recovery of pressure
height and diameter of the nose) is the first step in the design
at the end of the body. Pressure drag is very small for an
process of the sonar. The dimensions of the sonar array are
object with a high degree of elegance. The Reynolds number
a function of the wavelength, sonar horizontal beamwidth
is calculated as Eq. (44) based on velocity, total length and
and vertical beamwidth that are calculated as the following
dynamic viscosity of seawater.
equations:
V × LT
25.3 × 𝜆 Re = (44)
Lxa = ( ) v
BWh (38)
2
With regard to the velocity and length of the vehicle range
considered in this research, the conditions of the disordered
25.3 × 𝜆
Lya = ( boundary layer are always established and the friction coef-
)
BWv (39) ficient of the crust is calculated as Eq. (45).
2
0.075
√ Cf = (45)
� � (log (Re) − 2)2
Ds = 2 × max Lxa , Lya (40)
The drag body factor can now be calculated according to
In the above equations, 𝜆 is the sonar wavelength which the friction coefficient of the crust as Eq. (46) [49].
is calculated as Eq. (41).
[ ( )3 ( )]
c D L
𝜆= (41)
b
Df = Cf × 1 + 60 + 0.0025 T (46)
Fr LT D
where Fr = 30 kHz and c = 1480 m/s. The directivity index
Moreover, the lift force is calculated as Eq. (47), in which
can also be calculated based on the dimensions of the sonar
𝜌 is the density and assumed to be equal to 1025 and 𝛼 is the
array as Eq. (42).
angle of attack.
( )
Lxa × Lya
DI = 10 × log 4 × (42) LF = 0.5 × 𝜌 × V 2 × LT2 × 𝛼 (47)
𝜆2
The propulsion force can be calculated as follows:
The sonar range can be calculated by Eq. (43).
TF − DF cos(𝛼) − LF sin(𝛼) = 0 (48)
20 log (DR) + α × DR × 10−3 = SL − NL + DI − DT (43)

where SL, NL and 𝛼 are the source level, noise level and
water absorption coefficient, respectively, and are assumed 4.3 Structure
equal to 25, 15 and 0.00006, respectively. The directivity
index and beam width are inversely related. In fact, a sonar The variables in this discipline include the total length,
with a longer range has a narrower beamwidth, while a volume, mass, and area of the wetted surface. The outer
sonar with a wider visibility has a lower directivity index form of the body of the device can be described in terms of
and shorter range.

13
Journal of Marine Science and Technology

radius and longitudinal positions. While the total volume, powm


total mass, and area of the wetted area can be calculated by
pow = powh +
𝜂m (56)
using these two concepts as Eqs. (49–51), in which VOLT ,
MT and Sb are the total volume, total mass and area of the The power source of the device is considered as an elec-
wetted surface, respectively. tric battery with a specific energy density. Energy density
is usually expressed as energy per battery weight unit (el ).
LT
With regard to the proportionality of the battery weight with

VOLT = 𝜋 r(l)2 dl (49) its length, energy per length unit can be considered as the
0 energy density of the battery. In this case, the energy stored
in a battery of length ­Lb is calculated as Eq. (57).
LT
( 2 ) E = Lb × el (57)

MT = 𝜌SH 𝜋 r(l) − ri (l)2 dl + me + mp + mb + mr + mm
0
(50) 4.5 Tactics
LT
The ultimate goal of the design of an underwater vehicle is

Sb = 2𝜋 r(l)dl (51)
to carry out a specific mission in the operational environ-
0
ment. The tactical parameters considered in this paper, by
where 𝜌SH , mb, mr , mm, mp, and me are the aluminum density, which the mission scenario (the tactics for using the device
battery mass, rear part mass, motor mass, payload mass and in the operational environment) is defined, include the vehi-
electronics part mass, respectively. cle range, maximum time of operation (maximum flight
duration), time to reach the target and probability of target
detection. The vehicle range and maximum mission time for
4.4 Power/propulsion the vehicle can be calculated as Eqs. (58, 59) based on the
total power and energy stored in the battery.
In this discipline, the power required for the motor is cal- E×V
culated as Eq. (52) based on the propulsion force, forward Rng =
pow (58)
speed, propeller performance and shaft performance.
TF × V Rng E
powm = (52) TT = = (59)
𝜂p × 𝜂sh V pow

The motor length is calculated as Eq. (53). The following simplifying assumptions are made to cal-
culate the probability of target detection:
powm
Lmotor = × 10−3 (53)
Ca × N × D2m • The distance between the initial position and the target
is estimated to be equal to D ­ 0, and the uncertainty in
where N, Ca, and Dm are propeller rotation velocity, motor
estimation is as a circular region with a radius of ­R0.
power coefficient and motor diameter, respectively. The
• The target speed is equal to ­Vtar, and its target movement
power consumption of the sonar can be estimated as Eq. (54)
direction is not specified along the path of the vehicle
based on the beam width, detectability index and sonar
movement to the target.
power factor.
• The device moves directly to the target.
( )
DI
pows = 𝛽 × BW × 10 20 (54)
According to the two first assumptions, the uncertainty
The internal power of the device, which includes the region of the target presence is a circle with a radius of R
­ 0,
sonar power and electronic part power, is calculated as which is growing at a speed of V ­ tar. Therefore, the target
Eq. (55). detection time can be calculated as Eq. (60).
D0 − DR
powh = pows + powe (55) Tdet = (60)
V + Vtar
Finally, the total power is calculated as Eq. (56), in
which {𝜂}_{m} is the motor performance. And the area of the region where the target is randomly
located ( AUncertainty ) according to time is calculated as
follows:

13
Journal of Marine Science and Technology

( ( ))2 when changing the design variables. Using these estimates,


AUncertainty = 𝜋 × R0 + Vtar Tdet + t (61)
the cost of the system can be estimated for various opera-
The area that is searched by the device at any time to tional scenarios. These costs are entirely notional and are
detect the target is as Eq. (62). based upon the assumption that higher performance com-
( ) ponents, such as the motor and sonar, will translate directly
𝜋 into greater costs [37]. He estimated the cost of the system
Asearch = DR × BWh × V × t (62)
180 based on the velocity, sonar horizontal beamwidth and sonar
The probability of target detection can be written equal directivity index parameters. In the present study, the cost
to the ratio of the area of the target search to the area of has been estimated based on these parameters.
uncertainty as Eq. (63), in which Tmax is the maximum time
to reach the maximum probability of target detection and is 4.7 Computational results
calculated as Eq. (64).
To apply the presented solution method, the parameters of
Asearch DR × BW × V × Tmax GA and PSO are set as Table 3.
pdet = = ( ( ))2 (63)
AUncertainty 180 × R0 + Vtar Tdet + Tmax After applying the solution methods on the numerical
example, the reliability level of design constraints is shown
{ } in Table 4. The trend of moving the algorithms to the opti-
RO mal solution is shown in Fig. 5. As can be observed, by
Tmax = min TT − Tdet , + Tdet (64)
Vtar using both GA and PSO, SORA is convergent. SORA is
terminated after 18 iterations by using GA and after 24 itera-
tions by using PSO. By both algorithms, SORA has reached
4.6 Cost
a 95% reliability level in a relatively low number of itera-
tions, which shows its efficiency. However, SORA by using
The cost is also considered in this paper as a discipline. In
GA is terminated after fewer iterations than by using PSO.
this regard, the cost has been considered as a function of
Since the purpose of providing the solution method is to
some design variables, according to the research conducted
show its ability to obtain the optimal solution and achieve
by Frits [37]. In his PhD thesis, Frits presented a robust
the desired reliability, only the details of reliability are men-
design method for an undersea weapon by using the concept
tioned and unrelated details to our purpose are not provided.
of the probability of success based on the risk, cost, and
efficiency. In this study, no explicit relation was expressed
between the cost and the variables of the problem. How- Table 3  The value of the parameters of GA and PSO for case study
ever, the cost was estimated as a function of some design GA PSO
variables of the problem and it was shown how the cost
Parameter Value Parameter Value
changed with the change of the variables. It is assumed
that there is an exponential price increase due to improv- nPop 50 nf 50
ing the sonar performance, plus a milder price increase for rC 0.6 w 1.5
increasing the motor horsepower. The estimated relationship rM 0.2 MaxIt 100
between the cost and the design variables presented in this μ 0.1
study is as shown in Fig. 4, as the relative cost of the system MaxIt 100

Fig. 4  The estimated system cost [37]

13
Journal of Marine Science and Technology

Table 4  Results obtained by SORA be evaluated well in the SORA method using both GA and
Iteration number Reliability level of design constraints
PSO.
The change in cost while moving to the optimal solution
By using GA By using PSO
is depicted in Figs. 6 and 7 in terms of the reliability level
1 0.484 0.289 achieved in each iteration. The trends of change in cost ver-
2 0.489 0.323 sus reliability in Figs. 6 and 7 are almost the same, with a
3 0.492 0.379 slight difference. At first, the slope of cost increase in SORA
4 0.497 0.422 with GA is steeper than with PSO. But this trend is almost
5 0.519 0.478 vice versa between around 0.5 to 0.6. And from then on, the
6 0.651 0.513 trend is almost the same in both algorithms.
7 0.673 0.581 As can be seen, as reliability increases, the rate of
8 0.716 0.616 increase in cost initially increases with a steeper slope. But
9 0.773 0.662 further, the slope of cost increase decreases significantly.
10 0.839 0.723 This trend continues until around the reliability value of
11 0.842 0.782 0.85 and from there, the slope of cost increase gets steeper
12 0.857 0.801 again. In other words, whatever the reliability gets closer
13 0.883 0.845 to 1, increasing in reliability by a specific amount can be
14 0.886 0.892 achieved by spending more cost. So, it can be said that as the
15 0.913 0.907 reliability gets closer to 1, it is less economical to increase it.
16 0.926 0.918 Therefore, increasing the reliability approximately between
17 0.942 0.924 0.5 and 0.85 is more cost-effective, and by spending a low
18 0.953 0.932 amount of cost, the reliability can be significantly improved.
19 0.932 But by reaching the reliability value to around 0.9, improv-
20 0.937 ing a certain amount of reliability is possible by spending
21 0.940 more cost.
22 0.942 Deciding on the final level of reliability of a system for
23 0.948 its design and construction depends on the type and use of
24 0.952 the system, and some other factors besides the cost can play
a role. Figures 6 and 7 show the behavior of construction
cost versus reliability. Other costs can also affect the choice
The obtained results show that SORA is able to obtain the of reliability level. For example, the cost of failing a system
final solution with the desired reliability level. As can be may be a significant financial or human cost that is separate
observed in Fig. 5, the slope of reliability improvement can from the cost of construction, and so occurrence of a failure

Fig. 5  Trend of moving SORA 1


to the optimal solution 0.95 By using GA
0.9 By using PSO
0.85
0.8
0.75
0.7
Reliability

0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3
0.25
5 10 15 20
Iteration

13
Journal of Marine Science and Technology

Fig. 6  Change in cost while 250


moving to the optimal solution
in SORA by using GA
200

150

Cost (106 $)
100

50

0
0.4 0.5 0.6 0.7 0.8 0.9 1
Reliability

Fig. 7  Change in cost while 250


moving to the optimal solution
in SORA by using GA
200

150
Cost (106 $)

100

50

0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Reliability

can cause a huge cost. As a result, it can be more logical using GA is less than by using PSO, which indicates the
and correct to choose high levels of reliability despite its superiority of GA with respect to this criterion. As well as,
higher cost. This matter may be vice versa for less impor- the average reliability obtained by using GA is less than
tant systems. Therefore, to choose the level of reliability, all by using PSO, which indicates the superiority of PSO with
aspects must be considered and finally, the decision maker respect to this criterion.
can choose the desired level of reliability. To compare the time to reach the optimal solution by
To evaluate the strength of the proposed algorithms, each the two algorithms, the average number of evaluations
of them has been executed five times and the trend of mov- of the objective function by them is compared with each
ing to the optimal solution is depicted in Figs. 8 and 9. As other. The results are shown in Table 6. It should be noted
can be seen, the different execution of the algorithms are that the number of new solutions (chromosomes) gener-
almost the same trend with slightly different, which shows ated in each iteration of GA is nPop × rC + nPop × rM and
the strength of them. the number of new solutions (particles) generated in each
The number of iterations of SORA and the obtained iteration of PSO is equal to nf. As can be seen, the aver-
reliability in each execution are shown in Table 5. As can age total number of the objective function evaluations by
be seen, the average number of iterations of SORA by

13
Journal of Marine Science and Technology

Fig. 8  Trend of moving SORA 0.95


by using GA to the optimal Run #1
0.9
solution in various runs Run #2
0.85
Run #3
0.8
Run #4
0.75
Run #5
0.7

Reliability
0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3
5 10 15 20
Iteration

Fig. 9  Trend of moving SORA 1


by using PSO to the optimal 0.95 Run #1
solution in various runs
0.9 Run #2
0.85 Run #3
0.8 Run #4
0.75 Run #5
0.7
Reliability

0.65
0.6
0.55
0.5
0.45
0.4
0.35
0.3
0.25
5 10 15 20 25
Iteration

Table 5  Number of iterations of SORA and the obtained reliability in using GA is less than by using PSO, which indicates the
executions of the algorithms superiority of GA with respect to this criterion.
By GA By PSO
Number of Obtained Number of Obtained
iterations of reliability iterations of reliability 5 Conclusions and future research
SORA SORA
In most optimization problems, there is a degree of uncer-
Run #1 18 0.953 24 0.952
tainty in variables and parameters. Ignoring uncertainty
Run #2 20 0.954 18 0.953
in the design process leads to a design that is not optimal
Run #3 24 0.951 27 0.951
in the operating conditions. To deal with the uncertainty,
Run #4 21 0.952 24 0.953
there are two approaches called robust design optimization
Run #5 16 0.953 21 0.955
and reliability-based design optimization. In this paper, for
Average 19.8 0.9526 22.8 0.9528
the first time, a reliability-based multidisciplinary design

13
Journal of Marine Science and Technology

Table 6  Average total By using GA By using PSO


number of objective function
evaluations by SORA Number of initial objective function evaluations 50 50
(initial population size)
Number of new solutions generated per iteration 40 50
Number of iterations of the algorithm 100 100
Total number of objective function evaluations in each execution 4050 5050
Average number of iterations of SORA 19.8 22.8
Average total number of objective function evaluations 80,190 115,140

optimization approach is investigated, in which uncertain References


parameters are considered as fuzzy numbers and cost is con-
sidered as a discipline, due to its importance in engineer- 1. Crespo, L. G., & Bushnell, D. M. (2002). Optimization of systems
ing problems. In this regard, the design of an autonomous with uncertainty: initial developments for performance, robustness
and reliability based designs.
underwater vehicle under uncertainty is considered with six 2. Sun, J., Zhang, G., Vlahopoulos, N., & Hong, S. B. (2006). Multi-
disciplines including sonar, hydrodynamics, structure, pro- disciplinary design optimization under uncertainty for thermal
pulsion, tactics, and cost. The SORA approach is used as the protection system applications. In: 11th AIAA/ISSMO Multidis-
solution method, in which a GA and PSO are used to solve ciplinary Analysis and Optimization Conference (p. 7002).
3. Elms DG (2004) Structural safety–issues and progress. Prog Struct
the deterministic optimization problem in each iteration. Mat Eng 6(2):116–126
The reliability level of 95% is considered as the termina- 4. Zang TA, Hemsch MJ, Hilburger MW, Kenny SP, Luckring
tion condition of SORA. The results obtained from solving JM, Maghami P, Stroud WJ (2002) Needs and opportunities for
the model show that SORA achieves the desired reliability uncertainty-based multidisciplinary design methods for aerospace
vehicles.
level in a relatively low iteration number, demonstrating the 5. Oberkampf WL, Helton JC, Joslyn CA, Wojtkiewicz SF, Ferson S
efficiency of this solution method. (2004) Challenge problems: uncertainty in system response given
The change in cost while moving to the optimal solution uncertain parameters. Reliabil Eng Syst Safety 85(1–3):11–19
is also depicted in terms of the reliability level achieved in 6. Choi HJ (2005) A robust design method for model and propa-
gated uncertainty (Doctoral dissertation, Georgia Institute of
each iteration. SORA with GA is terminated in less itera- Technology).
tions than with PSO. Initially, the slope of cost increase in 7. Phadke MS (1989) Quality engineering using robust design:
SORA with GA is steeper than with PSO. But this trend is Robuste Prozesse durch Quality Engineering. German transla-
almost vice versa between around 0.5 to 0.6. And from then tion: G. Liesegang, gfmt-Verlag, Munich.
8. Dantzig GB (1955) Linear programming under uncertainty. Man-
on, the trend is almost the same in both algorithms. age Sci 1(3–4):197–206
In almost both algorithms, as reliability increases, the rate 9. Freund RJ (1956) The introduction of risk into a programming
of increase in cost initially increases with a steep slope, then model. Econometrica 12:253–263
the cost slope decreases significantly (until around 0.85), and 10. Sahinidis NV (2004) Optimization under uncertainty: state-of-
the-art and opportunities. Comput Chem Eng 28(6–7):971–983
finally the cost slope increases again. In other words, what- 11. Schuëller GI, Jensen HA (2008) Computational methods in opti-
ever the reliability gets closer to 1, increasing in reliability mization considering uncertainties—an overview. Comput Meth-
for a specific amount can be achieved by spending more ods Appl Mech Eng 198(1):2–13
cost. So, it can be said that as the reliability gets closer to 1, 12. Ahn J, Kwon JH (2006) An efficient strategy for reliability-based
multidisciplinary design optimization using BLISS. Struct Multi-
it is less economical to increase it. Improving the reliability discipl Optimiz 31(5):363–372
between 0.5 and 0.85 is more cost-effective. But some other 13. Youn BD, Choi KK, Yang RJ, Gu L (2004) Reliability-based
factors besides the cost of construction and design play a design optimization for crashworthiness of vehicle side impact.
role in choosing the reliability level. To choose the level of Struct Multidiscipl Optimiz 26(3–4):272–283
14. Zhang X, Huang HZ (2010) Sequential optimization and reliabil-
reliability, all aspects must be considered and finally, the ity assessment for multidisciplinary design optimization under
decision-maker can choose the desired level of reliability. aleatory and epistemic uncertainties. Struct Multidiscipl Optimiz
For future studies, other types of uncertainty such as ran- 40(1–6):165
dom variables with a probability distribution function are 15. Hora SC (1996) Aleatory and epistemic uncertainty in probability
elicitation with an example from hazardous waste management.
recommended. Moreover, an investigation area would be to Reliabil Eng Syst Safety 54(2–3):217–223
consider other disciplines for the reliability based multidis- 16. Oberkampf W, Helton J, Sentz K (2001) Mathematical representa-
ciplinary design optimization. Moreover, development of a tion of uncertainty. In: 19th AIAA applied aerodynamics confer-
model with another objective function, development of a ence (p. 1645).
17. Oberkampf WL, DeLand SM, Rutherford BM, Diegert KV,
multi-objective problem, and use of other solution methods Alvin KF (2000) Estimation of total uncertainty in computational
can be considered as other options for further studies.

13
Journal of Marine Science and Technology

simulation. Sandia National Laboratories, SAND2000–0824, 35. Choi KK, Du L, Youn BD (2005) Integration of reliability-and
Albuquerque NM. possibility-based design optimizations using performance meas-
18. Hang Hou Y, Jia Li Y, Liang X (2019) Mixed aleatory/epistemic ure approach (No. 2005-01-0342). SAE Technical Paper.
uncertainty analysis and optimization for minimum EEDI hull 36. Khorshidi HA, Gunawan I, Ibrahim Y (2016) A dynamic unreli-
form design. Ocean Eng 172:308–315 ability assessment and optimal maintenance strategies for multi-
19. Cullen AC, Frey HC, Frey CH (1999) Probabilistic techniques in state weighted k-out-of-n: F systems. Appl Stoch Models Bus Ind
exposure assessment: a handbook for dealing with variability and 32(4):485–493
uncertainty in models and inputs. Springer, Berlin 37. Frits AP (2005) Formulation of an integrated robust design and
20. Ang AHS, Tang W (1975) Probability concepts in engineering tactics optimization process for undersea weapon systems (Doc-
planning and design. Wiley, New York toral dissertation, Georgia Institute of Technology).
21. Ditlevsen O (1981) Uncertainty modeling with applications to 38. Zadeh LA (1965) Fuzzy sets. Inf Control 8(3):338–353
multidimensional civil engineering systems, vol 6. McGraw-Hill, 39. Wang C, Qiu Z (2015) Hybrid uncertain analysis for temperature
New York field prediction with random, fuzzy and interval parameters. Int J
22. Neelamkavil F (1987) Computer simulation and modelling. Wiley, Therm Sci 98:124–134
New York 40. Du X, Chen W (2002). Sequential optimization and reliability
23. Haldar A, Mahadevan S (2000) Probability, reliability, and statisti- assessment method for efficient probabilistic design. In: ASME
cal methods in engineering design, vol 1. Wiley, New York 2002 international design engineering technical conferences and
24. Yang RJ, Gu L, Tho CH, Choi KK, Youn B (2002) Reliability- computers and information in engineering conference (pp. 871–
based multidisciplinary design optimization of a full vehicle sys- 880). American Society of Mechanical Engineers.
tem. In: 43rd AIAA/ASME/ASCE/AHS/ASC structures, struc- 41. Torii AJ, Lopez RH, Miguel LF (2014) A generalization of the
tural dynamics, and materials conference (p. 1758). sequential optimization and reliability assessment method for
25. Agarwal H, Renaud JE, Preston EL, Padmanabhan D (2004) RBDO problems. In: Proceedings of the 2nd international sym-
Uncertainty quantification using evidence theory in multidiscipli- posium on uncertainty quantification and stochastic modeling.
nary design optimization. Reliabil Eng Syst Saf 85(1–3):281–294 Rouen, France
26. Alexandrov NM, Lewis RM (2002) Analytical and computational 42. Hasofer AM, Lind NC (1974) Exact and invariant second-moment
aspects of collaborative optimization for multidisciplinary design. code format. J Eng Mech Div 100(1):111–121
AIAA J 40(2):301–309 43. Du X, Chen W (2001) A most probable point-based method
27. Agarwal H, Renaud J (2004) Reliability based design optimiza- for efficient uncertainty analysis. J Design Manufact Automat
tion using response surfaces in application to multidisciplinary 4(1):47–66
systems. Eng Optimiz 36(3):291–311 44. Rosenblatt M (1952) Remarks on a multivariate transformation.
28. Chen X, Hasselman T, Neill D, Chen X, Hasselman T, Neill D Ann Math Stat 23(3):470–472
(1997) Reliability based structural design optimization for prac- 45. Wu YT, Millwater HR, Cruse TA (1990) Advanced probabilistic
tical applications. In: 38th Structures, structural dynamics, and structural analysis method for implicit performance functions.
materials conference (p. 1403). AIAA J 28(9):1663–1669
29. Meng D, Li YF, Huang HZ, Wang Z, Liu Y (2015) Reliability- 46. Tomassini M (1995) A survey of genetic algorithms. In: Annual
based multidisciplinary design optimization using subset simu- reviews of computational physics III (pp. 87–118).
lation analysis and its application in the hydraulic transmission 47. Kennedy J, Eberhart RC. Particle swarm optimization. In: IEEE
mechanism design. J Mech Des 137(5):051402 international conference on neural networks, vol. IV. 1995. p.
30. Yao W, Chen X, Ouyang Q, Van Tooren M (2013) A reliability- 1942–8.
based multidisciplinary design optimization procedure based on 48. Shi Y, Eberhart RC. Parameter selection in particle swarm opti-
combined probability and evidence theory. Struct Multidiscipl mization. In: Evolutionary programming VII: EP 98. New York:
Optimiz 48(2):339–354 Springer; 1998. p. 591–600.
31. Batill S, Renaud J, Gu X (2000) Modeling and simulation uncer- 49. De Barros EA, Pascoal A, De Sa E (2008) Investigation
tainty in multidisciplinary design optimization. In: 8th symposium of a method for predicting AUV derivatives. Ocean Eng
on multidisciplinary analysis and optimization (p. 4803). 35(16):1627–1636
32. McAllister CD, Simpson TW (2003) Multidisciplinary robust
design optimization of an internal combustion engine. J Mech Publisher’s Note Springer Nature remains neutral with regard to
Des 125(1):124–130 jurisdictional claims in published maps and institutional affiliations.
33. Du X, Chen W (2002) Efficient uncertainty analysis methods for
multidisciplinary robust design. AIAA J 40(3):545–552
34. Du L, Choi KK (2008) An inverse analysis method for design
optimization with both statistical and fuzzy uncertainties. Struct
Multidiscipl Optimiz 37(2):107–119

13

You might also like