Resistance and Seakeeping Optimization Model (Function)

Kivanc Ali ANIL

In this study, a ship design optimization model (function) was created in order to
improve a preliminary ship design prototype. The model was based on two performance
criteria: Resistance and Seakeeping.

The Resistance Performance is evaluated using Fung’s Resistance Prediction

Algorithm. The Fung’s Bare Hull Residuary Resistance Coefficient was stated as the
sum of 10 different residuary resistance coefficient components functions of
“Displacement-Length Ratio”, “Beam-Draft Ratio”, “Prismatic Coefficient”, “Maximum
Section Area Coefficient”, “Half Entrance Angle”, “Transom Area Ratio”, “Transom
Width Ratio”, “Transom Depth Ratio”, “Bow Area Ratio”, and “Wetted Surface
Coefficient”. These components were tabulated for different “Speed-Length Ratios” [1].
Linear interpolation (and extrapolation if necessary) is performed using those tables to
assess the values of residuary resistance coefficient components.

The second criterion, Seakeeping Performance is assessed using Bales Formula

(Bales Seakeeping Rank). The Bales Seakeeping Rank Equation has 6 components.
These components are “Waterplane Area Coefficient Forward of Amidships”,
“Waterplane Area Coefficient Aft of Amidships”, “Draft-To-Length Ratio”, “Cut-Up
Ratio”, “Vertical Prismatic Coefficient Forward of Amidships”, and “Vertical Prismatic
Coefficient Aft of Amidships”. [2]

MATHEMATICAL MODEL

Mathematical formulation of the ship design optimization model is arranged

based on the Parameter Space Investigation (PSI) technique definitions. In this method, it
is assumed that “the system is based on r design variables, which form a point in r-
dimensional space, i.e., a vector of design variables (the design variable vector)”. Each
design variable has its own reasonable boundary, i.e. the “design variable constraint”.
The design variables with their corresponding design variable constraints constitute the

1
“design variable space” (П) in the r-dimensional space. The “functional relations”, which
are functions of the design variables, outline the characteristics of the design model. The
functional relations are only subject to the “functional constraints”. They are not
optimized. The “performance criteria (design criteria)”, which have to be either
minimized or maximized, are the principle characteristics of the design model. In other
words, the performance criteria determine the efficiency of the design. The performance
criteria are subject to the “criteria constraints”. Unlike the functional constraints, the
criteria constraints are selected during the analysis of the problem. They are not rigid, i.e.,
the designer can continually determine and revise them. [3 & 4]

It is very important to understand these definitions in order to avoid confusion.

For example, if “block coefficient” is a design variable in a design model, its acceptable
boundaries become the design variable constraints. In another design model, “block
coefficient” might be a function of the design variables (functional relation). In that case,
its tolerable limits turn out to be the functional constraints. Furthermore, “block
coefficient” might be a performance criteria in a case where it have to be optimized. At
that time, its margins form the criteria constraints.

There are 2 Constants 15 Design Parameters, 10 Functional Relations, and 2

Design Criteria in our design optimization model. Note that during the explanation of the
model, subscripts of some variables were written in normal text to conform to
programming language notation.

CONSTANTS
RHO = 1.025 : Seawater Density, metric ton/m2
VK = 15 : Sample Speed, knots

DESIGN PARAMETERS
p1 L : Length of Design Waterline (assumed to be equal to Length Between
Perpendiculars), m
p2 B : Beam of Design Waterline (assumed to be equal to Beam amidships), m
p3 T : Draft (assumed to be equal to Draft Amidships), m
p4 c : Distance from the Station 0 (FP) to the cut-up point, m

2
p5 AWF : Waterplane area forward of amidships, m2
p6 AWA : Waterplane area aft of amidships, m2
p7 VF : Displaced volume forward of amidships, m3
p8 VA : Displaced volume aft of amidships, m3
p9 Cp : Prismatic Coefficient
p10 A_T : Projected Transom Area, m2
p11 B_T : Projected Transom Width, m
p12 T_T : Projected Transom Depth, m
p13 IE : Half Entrance Angle, degrees
p14 A_B : Bow Area, m2
p15 LCB : Longitudinal Center of Buoyancy from FP, m

FUNCTIONAL RELATIONS
f1 DISP : Displacement, metric ton
f2 CB : Block Coefficient
f3 Cx : Maximum Section Area Coefficient (assumed to be equal to midship
section coefficient Cm)
f4 Cwp : Waterplane Area Coefficient
f5 CWF : Waterplane Area Coefficient forward of amidships
f6 CWA : Waterplane Area Coefficient aft of amidships
f7 TL : Draft-to-length ratio
f8 cL : Cut-up ratio
f9 CVPF : Vertical prismatic coefficient forward of amidships
f10 CVPA : Vertical prismatic coefficient aft of amidships

CRITERIA
c1 R : Seakeeping Rank (BALES), maximize
c2 CR : Residuary Resistance Coefficient (FUNG), minimize

There are 5 Functional Constraints and 2 Design Variable Constraints in the

design optimization model.

3
FUNCTIONAL CONSTRAINTS
f1 DISP ≤ 2100
f2 CB ≤ 0.51
f3 Cx ≥ 0.77
f4 Cwp ≤ 0.84
f4 Cwp ≥ 0.80

DESIGN VARIABLE CONSTRAINTS

p9 Cp ≥ 0.62
p15 A_B = 0 (CONSTANT)

The last six of the functional relations are the components of the Bales
Seakeeping Rank Equation. The coefficients of these components are tabulated in Table
1. The components with positive coefficients have to be maximized to maximize the
Seakeeping Rank. On the other hand, the components with negative coefficients have to
be minimized to maximize the Seakeeping Rank. Therefore, these functional relations
may be assumed to be “pseudo-criteria”.

CWF CWA T/L c/L CVPF CVPA

a0 a1 a2 a3 a4 a5 a6
8.422 45.104 10.078 -378.465 1.273 -23.501 -15.875
maximize maximize minimize maximize minimize minimize

Table 1 Coefficients of the Bales Seakeeping Rank Equation Components (Ref. [2].).

PSEUDO-CRITERIA (BALES)
f5 CWF : Pseudo-criteria, maximize
f6 CWA : Pseudo-criteria, maximize
f7 TL : Pseudo-criteria, minimize
f8 cL : Pseudo-criteria, maximize

4
f9 CVPF : Pseudo-criteria, minimize
f10 CVPA : Pseudo-criteria, minimize
In reality, some of these pseudo-criteria are in some way restricted by the
functional constraints because of the following formulation [2].

CWF = 2×AWF/(L×B) : Waterplane Area Coefficient forward of amidships

CWA = 2×AWA/(L×B) : Waterplane Area Coefficient aft of amidships
TL =T/L : Draft-to-length ratio
cL =c/L : Cut-up ratio
CVPF = VF / (AWF×T) : Vertical prismatic coefficient forward of amidships.
CVPA = VA / (AWA×T) : Vertical prismatic coefficient aft of amidships.

DISP = (VF+VA)×RHO : Displacement, metric ton

CB = (VF+VA)/(L×B×T) : Block Coefficient
Cx = CB/Cp : Maximum Section Area Coefficient
Cwp = (CWF+CWA)/2 : Waterplane Area Coefficient

The “Waterplane Area Coefficient forward of amidships” and “Waterplane Area

Coefficient aft of amidships” are bounded by the functional constraints of the
“Waterplane Area Coefficient”. Same argument is valid for the “Vertical prismatic
coefficient forward of amidships” and “Vertical prismatic coefficient aft of amidships”,
since the design variables: “Displaced volume forward of amidships” and “Displaced
volume aft of amidships” are used in their formulation. The functional constraints of
“Displacement”, “Block Coefficient”, and “Maximum Section Area Coefficient”
indirectly limit these two prismatic coefficients.

5
 REFERENCES

[1] Fung S.C.(AM), “Resistance and Powering Prediction for Transom Stern Hull
Forms During Early Stage Ship Design”, Department of the Navy, NAVSEA,
Washington, DC, Sname Transactions, Vol 99, 1991, pp. 29-84
[2] Bales N.K., “Optimizing the Seakeeping Performance of Destroyer-Type Hulls”,
Proceedings, 13th ONR Symposium, October 1980
[3] Statnikov R.B., Matusov J.B., Multicriteria Analysis in Engineering,
Dordrecht/Boston/London, Kluwer Academic Publishers, 2002.
[4] Anil, K.A., Multi-Criteria Analysis in Naval Ship Design, Master’s Thesis, Naval
Postgraduate School, Monterey, California, March 2005.

6
function [c1,c2,f1,f2,f3,f4,f5,f6,f7,f8,f9,f10] = ShipM(p1,p2,p3,p4,p5,p6,p7,p8,p9,p10,p11,p12,p13,p14,p15,p16)

% Kivanc Ali ANIL, 2005

%----------------------------------------------------------------------
% CONSTANTS
RHO = 1.025 ; % Seawater Density (SU YOGUNLUGU), metric ton/m^2
VK = 15 ; % Sample Speed (SURAT), knot

%----------------------------------------------------------------------
% DESIGN PARAMETERS (COMMON)
L = p1 ; % Length of Design Waterline (SU HATTI BOYU), m
% (assumed to be equal to Length Between
% Perpendiculars)
B = p2 ; % Beam of Design Waterline (SU HATTI GENISLIGI),m
% (assumed to be equal to Beam Amidships)
T = p3 ; % Draft ,m
% (assumed to be equal to Draft Amidships)
%----------------------------------------------------------------------
% DESIGN PARAMETERS (BALES)
c = p4 ; % Distance from the Station 0 (forward
% perpendicular) to the cut-up point, m
AWF = p5 ; % Waterplane area forward of amidships, m^2
AWA = p6 ; % Waterplane area aft of amidships, m^2
VF = p7 ; % Displaced volume forward of amidships, m^3
VA = p8 ; % Displaced volume aft of amidships, m^3
%----------------------------------------------------------------------
% DESIGN PARAMETERS (FUNG)
Cp = p9 ; % Prismatic Coefficient (PRIZMATIK KATSAYI)
SS = p10 ; % Wetted Surface Area (ISLAK YUZEY ALANI), m^2
A_T = p11 ; % Projected Transom Area (KIC AYNALIK ALANI), m^2
B_T = p12 ; % Projected Transom Width (KIC AYNALIK GENISLIGI), m
T_T = p13 ; % Projected Transom Depth (KIC AYNALIK DRAFTI), m
IE = p14 ; % Half Entrance Angle (GIRIS ACISI), degrees
A_B = p15 ; % Bow Area (BALB KESIT ALANI), m^2
LCB = p16 ; % Longitudinal Center of Buoyancy from FP
% (KIC DIKMEDEN BOYUNA SEPHIYE MRK.), m
%----------------------------------------------------------------------
% BALES
CWF = 2*AWF/(L*B) ; % Waterplane Area Coefficient forward of amidships
CWA = 2*AWA/(L*B) ; % Waterplane Area Coefficient aft of amidships
TL = T / L ; % Draft-to-length ratio
cL = c / L ; % Cut-up ratio
CVPF = VF / (AWF*T); % Vertical prismatic coefficient forward of amidships.
CVPA = VA / (AWA*T); % Vertical prismatic coefficient aft of amidships.
% SEAKEEPING RANK:
R =8.422+45.104*CWF+10.078*CWA-378.465*TL+1.273*cL-23.501*CVPF-15.875*CVPA;
%----------------------------------------------------------------------
DISP = (VF+VA)*RHO ; % Displacement (DEPLASMAN TONAJI), metric ton
CB = (VF+VA)/(L*B*T); % Block Coefficient
Cx = CB/Cp ; % Maximum Section Area Coefficient (MAX.KESIT ALAN KATS)
% (assumed to be equal to midship section coefficient Cm)
Cwp = (CWF+CWA)/2 ; % Waterplane Area Coefficient (SU HATTI ALAN KATS)
%----------------------------------------------------------------------
% FUNG
%----------------------------------------------------------------------
% VL : Speed-Length Ratio (HIZ-BOY ORANI)
% DL : Displacement-Length Ratio (DEPLASMAN-BOY ORANI)
% Bx/Tx : Beam-Draft Ratio (GENISLIK-DRAFT ORANI)
% Cp : Prismatic Coefficient (PRIZMATIK KATSAYI)
% Cx : Maximum Section Area Coefficient (MAX.KESIT ALAN KATS)
% IE : Half Entrance Angle (GIRIS ACISI)
% TA : Transom Area Ratio (AYNALIK ALAN ORANI)
% TW : Transom Width Ratio (AYNALIK GENIS.ORANI)
% TT : Transom Depth Ratio (AYNALIK DRAFT ORANI)
% BA : Bow Area Ratio (BALB KES.ALAN ORANI)
% CWS : Wetted Surface Coefficient (ISLAK ALAN KATS)
%----------------------------------------------------------------------
VL = VK/(L/0.3048)^.5;
DL = Cp*Cx/((L/B)^2*(B/T))*28570;
BxTx = B/T;
Cp ;
Cx ;
IE ;
TA = A_T/(Cx*B*T);
TW = B_T/B ;
TT = T_T/T ;
BA = A_B/(Cx*B*T);
CWS = -6.2263-.0094*DL+16.0209/(L/B)+...
.9207*(B/T)+5.4630/(B/T)+...
9.8528*Cx+7.1592/Cx+.0201*IE+...
2.1857/IE+1.0359*TW+...
2.4925*BA-3.7181*LCB/L;
%----------------------------------------------------------------------
VL_tab = .60:.10:2.30;
DL_tab = [20:5:100, 110:10:250];
BxTx_tab = 2.2:.2:5.2;
Cp_tab = .52:.01:.7;
Cx_tab = .62:.02:1;
IE_tab = 2:1:20;
TA_tab = 0:.02:.40;
TW_tab = 0:.05:.85;
TT_tab = 0:.02:.42;
BA_tab = 0:.01:.10;
CWS_tab = [14.90:.05:16.20, 16.30:.1:18.20];
%----------------------------------------------------------------------
% Table 6A Residuary Resistance Coefficient Component CR1 f:(DL: Displacement-Length Ratio)
CR1_tab = [...
-00.217 -00.681 -00.950 -01.041 -01.213 -01.080 -00.754 -00.693 -00.476 000.601 001.346 001.815 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.160 -00.533 -00.740 -00.796 -00.908 -00.768 -00.437 -00.294 000.001 001.035 001.737 002.163 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.119 -00.432 -00.597 -00.627 -00.693 -00.543 -00.195 000.020 000.389 001.427 002.120 002.527 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.088 -00.357 -00.492 -00.500 -00.530 -00.366 000.004 000.286 000.727 001.795 002.498 002.899 003.082 003.499 002.896 002.494 002.988 002.690;...
-00.063 -00.300 -00.410 -00.401 -00.399 -00.221 000.176 000.521 001.033 002.148 002.874 003.278 003.449 003.821 003.216 002.797 003.256 002.944;...
-00.042 -00.254 -00.344 -00.318 -00.289 -00.096 000.330 000.736 001.319 002.491 003.247 003.660 003.823 004.161 003.546 003.105 003.541 003.212;...
-00.023 -00.216 -00.289 -00.249 -00.195 000.015 000.472 000.937 001.590 002.827 003.619 004.044 004.203 004.513 003.883 003.417 003.837 003.489;...
-00.006 -00.183 -00.242 -00.188 -00.111 000.115 000.605 001.128 001.850 003.158 003.990 004.431 004.586 004.876 004.225 003.731 004.142 003.773;...
000.009 -00.155 -00.201 -00.134 -00.035 000.208 000.731 001.311 002.102 003.485 004.361 004.819 004.973 005.245 004.571 004.048 004.453 004.062;...
000.023 -00.130 -00.164 -00.085 000.034 000.294 000.852 001.488 002.347 003.809 004.731 005.208 005.361 005.620 004.919 004.367 004.769 004.355;...
000.036 -00.108 -00.132 -00.041 000.098 000.376 000.969 001.661 002.588 004.130 005.100 005.598 005.752 005.999 005.270 004.687 005.089 004.651;...
000.048 -00.087 -00.102 000.001 000.158 000.453 001.083 001.830 002.825 004.450 005.469 005.988 006.143 006.382 005.622 005.008 005.411 004.950;...
000.060 -00.069 -00.074 000.039 000.216 000.528 001.193 001.996 003.059 004.769 005.838 006.379 006.536 006.767 005.976 005.330 005.736 005.251;...
000.071 -00.052 -00.049 000.076 000.270 000.600 001.302 002.159 003.291 005.086 006.206 006.771 006.930 007.155 006.331 005.652 006.064 005.553;...
000.082 -00.036 -00.025 000.110 000.322 000.671 001.409 002.320 003.520 005.402 006.575 007.163 007.325 007.544 006.687 005.975 006.392 005.857;...
000.092 -00.020 -00.003 000.143 000.373 000.739 001.514 002.480 003.747 005.717 006.943 007.555 007.720 007.935 007.044 006.298 006.723 006.162;...
000.102 -00.006 000.019 000.175 000.421 000.806 001.618 002.638 003.973 006.032 007.311 007.948 008.116 008.328 007.402 006.622 007.054 006.468;...
000.122 000.020 000.059 000.235 000.515 000.936 001.823 002.950 004.421 006.659 008.046 008.734 000.000 000.000 000.000 000.000 000.000 000.000;...
000.141 000.044 000.095 000.292 000.605 001.062 002.024 003.258 004.865 007.285 008.782 009.520 000.000 000.000 000.000 000.000 000.000 000.000;...
000.159 000.067 000.130 000.347 000.691 001.185 002.223 003.564 005.306 007.908 009.517 010.307 000.000 000.000 000.000 000.000 000.000 000.000;...
000.177 000.088 000.163 000.399 000.775 001.305 002.419 003.867 005.744 008.531 010.251 011.095 000.000 000.000 000.000 000.000 000.000 000.000;...
000.195 000.109 000.194 000.449 000.857 001.424 002.614 004.168 006.180 009.153 010.986 011.883 000.000 000.000 000.000 000.000 000.000 000.000;...
000.212 000.128 000.224 000.499 000.938 001.541 002.808 004.468 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.229 000.147 000.253 000.547 001.017 001.657 003.000 004.767 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.246 000.165 000.281 000.594 001.095 001.772 003.192 005.064 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.263 000.183 000.309 000.641 001.172 001.886 003.383 005.361 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.279 000.200 000.336 000.686 001.248 001.999 003.573 005.657 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.295 000.217 000.362 000.732 001.324 002.112 003.763 005.952 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.312 000.233 000.388 000.776 001.399 002.224 003.952 006.247 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.328 000.250 000.414 000.821 001.473 002.336 004.140 006.541 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.344 000.266 000.439 000.865 001.547 002.447 004.329 006.835 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.360 000.282 000.464 000.908 001.621 002.558 004.517 007.128 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000];
% Table 6B Residuary Resistance Coefficient Component CR2 f:(Bx/Tx: Beam-Draft Ratio)
CR2_tab = [...
-00.224 001.297 001.271 001.688 002.037 002.800 002.780 002.605 002.405 001.736 001.214 000.922 000.744 001.632 000.724 000.957 001.366 001.404;...
-00.216 001.272 001.244 001.648 001.986 002.733 002.720 002.547 002.354 001.706 001.205 000.924 000.750 001.605 000.728 000.964 001.343 001.400;...
-00.210 001.258 001.227 001.622 001.953 002.689 002.683 002.510 002.324 001.690 001.204 000.931 000.760 001.592 000.737 000.976 001.332 001.405;...
-00.206 001.252 001.218 001.607 001.934 002.664 002.664 002.491 002.309 001.685 001.211 000.944 000.774 001.589 000.750 000.994 001.329 001.418;...
-00.203 001.252 001.216 001.602 001.925 002.654 002.659 002.485 002.306 001.688 001.223 000.960 000.790 001.593 000.765 001.015 001.333 001.437;...
-00.201 001.258 001.219 001.603 001.925 002.656 002.666 002.490 002.314 001.699 001.239 000.979 000.809 001.605 000.783 001.039 001.343 001.461;...
-00.200 001.268 001.227 001.611 001.933 002.668 002.683 002.505 002.329 001.715 001.260 001.001 000.830 001.621 000.803 001.065 001.356 001.489;...
-00.200 001.282 001.238 001.624 001.946 002.688 002.708 002.527 002.352 001.737 001.283 001.025 000.852 001.643 000.824 001.094 001.374 001.521;...
-00.200 001.299 001.253 001.640 001.965 002.715 002.740 002.555 002.381 001.762 001.309 001.051 000.876 001.668 000.846 001.124 001.395 001.555;...
-00.201 001.319 001.270 001.661 001.988 002.748 002.778 002.589 002.414 001.791 001.337 001.078 000.901 001.696 000.870 001.156 001.419 001.592;...
-00.203 001.341 001.289 001.684 002.015 002.786 002.820 002.627 002.452 001.822 001.367 001.107 000.927 001.727 000.895 001.189 001.444 001.631;...
-00.204 001.365 001.311 001.710 002.045 002.829 002.867 002.670 002.494 001.857 001.399 001.136 000.954 001.760 000.920 001.223 001.472 001.672;...
-00.206 001.391 001.334 001.739 002.078 002.876 002.917 002.716 002.538 001.893 001.432 001.167 000.981 001.796 000.947 001.258 001.502 001.714;...
-00.209 001.418 001.358 001.769 002.113 002.925 002.971 002.765 002.586 001.932 001.466 001.199 001.009 001.833 000.973 001.294 001.533 001.758;...
-00.211 001.446 001.384 001.801 002.151 002.978 003.028 002.817 002.636 001.972 001.501 001.231 001.038 001.872 001.001 001.331 001.565 001.803;...
-00.214 001.476 001.411 001.835 002.190 003.033 003.087 002.871 002.688 002.013 001.538 001.264 001.067 001.912 001.029 001.368 001.599 001.849];
% Table 6C Residuary Resistance Coefficient Component CR3 f:(Cp: Prismatic Coefficient)
CR3_tab = [...
002.926 000.540 -00.194 -00.618 -02.396 -02.998 -03.808 -08.631 -10.642 -12.234 -11.651 -11.481 000.000 000.000 000.000 000.000 000.000 000.000;...
002.918 000.540 -00.187 -00.596 -02.329 -02.924 -03.757 -08.597 -10.627 -12.221 -11.646 -11.470 000.000 000.000 000.000 000.000 000.000 000.000;...
002.912 000.539 -00.179 -00.575 -02.264 -02.853 -03.709 -08.568 -10.616 -12.213 -11.644 -11.464 -08.303 000.000 000.000 000.000 000.000 000.000;...
002.906 000.539 -00.171 -00.555 -02.201 -02.784 -03.662 -08.541 -10.609 -12.209 -11.647 -11.461 -08.302 -04.504 -06.063 -07.007 -08.848 -09.879;...
002.902 000.539 -00.164 -00.535 -02.140 -02.717 -03.618 -08.519 -10.605 -12.209 -11.653 -11.463 -08.304 -04.526 -06.072 -07.010 -08.847 -09.885;...
002.899 000.539 -00.157 -00.516 -02.081 -02.652 -03.576 -08.499 -10.605 -12.213 -11.663 -11.467 -08.308 -04.548 -06.083 -07.016 -08.849 -09.894;...
002.897 000.540 -00.150 -00.497 -02.023 -02.590 -03.535 -08.482 -10.609 -12.221 -11.676 -11.476 -08.315 -04.572 -06.096 -07.023 -08.853 -09.906;...
002.895 000.540 -00.143 -00.478 -01.967 -02.529 -03.496 -08.468 -10.615 -12.232 -11.692 -11.488 -08.324 -04.596 -06.110 -07.033 -08.860 -09.920;...
002.895 000.541 -00.137 -00.460 -01.912 -02.470 -03.459 -08.456 -10.624 -12.246 -11.712 -11.502 -08.336 -04.622 -06.126 -07.044 -08.869 -09.938;...
002.895 000.542 -00.130 -00.443 -01.859 -02.412 -03.423 -08.447 -10.636 -12.264 -11.734 -11.520 -08.349 -04.648 -06.143 -07.058 -08.880 -09.957;...
002.896 000.543 -00.124 -00.426 -01.807 -02.356 -03.389 -08.441 -10.651 -12.284 -11.759 -11.540 -08.365 -04.675 -06.161 -07.072 -08.894 -09.979;...
002.898 000.544 -00.118 -00.409 -01.757 -02.302 -03.357 -08.437 -10.668 -12.307 -11.787 -11.563 -08.382 -04.703 -06.181 -07.089 -08.910 -10.003;...
002.900 000.545 -00.112 -00.392 -01.707 -02.249 -03.325 -08.434 -10.687 -12.333 -11.817 -11.589 -08.401 -04.731 -06.202 -07.107 -08.928 -10.030;...
002.903 000.546 -00.106 -00.376 -01.659 -02.197 -03.395 -08.434 -10.709 -12.362 -11.850 -11.617 -08.422 -04.760 -06.224 -07.126 -08.947 -10.058;...
002.907 000.548 -00.100 -00.361 -01.612 -02.147 -03.266 -08.436 -10.733 -12.393 -11.885 -11.647 -08.445 -04.790 -06.247 -07.147 -08.969 -10.088;...
002.911 000.549 -00.095 -00.345 -01.566 -02.098 -03.239 -08.440 -10.759 -12.426 -11.922 -11.680 -08.469 -04.821 -06.271 -07.169 -08.992 -10.120;...
002.916 000.551 -00.089 -00.330 -01.521 -02.050 -03.212 -08.446 -10.788 -12.462 -11.961 -11.714 -08.495 -04.852 -06.297 -07.192 -09.017 -10.154;...
002.921 000.553 -00.083 -00.316 -01.477 -02.003 -03.187 -08.453 -10.818 -12.500 -12.002 -11.751 -08.522 -04.884 -06.323 -07.217 -09.043 -10.190;...
002.927 000.554 -00.078 -00.301 -01.435 -01.957 -03.162 -08.462 -10.849 -12.540 -12.045 -11.790 -08.551 -04.916 -06.350 -07.243 -09.071 -10.227];
% Table 6D Residuary Resistance Coefficient Component CR4 f:(Cx: Maximum Section Area Coefficient)
CR4_tab = [...
-02.051 -00.710 000.220 000.797 002.754 002.722 004.147 009.755 012.824 014.832 014.618 014.490 000.000 000.000 000.000 000.000 000.000 000.000;...
-02.031 -00.709 000.208 000.773 002.721 002.687 004.107 009.697 012.752 014.753 014.539 014.409 000.000 000.000 000.000 000.000 000.000 000.000;...
-02.013 -00.710 000.196 000.750 002.691 002.656 004.072 009.649 012.695 014.692 014.476 014.345 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.997 -00.711 000.185 000.729 002.665 002.628 004.042 009.612 012.651 014.645 014.428 014.295 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.984 -00.712 000.174 000.708 002.642 002.603 004.016 009.584 012.619 014.612 014.394 014.259 010.916 007.072 007.991 009.507 010.110 012.514;...
-01.973 -00.714 000.163 000.689 002.621 002.581 003.994 009.565 012.598 014.592 014.372 014.236 010.890 007.054 007.971 009.484 010.099 012.495;...
-01.963 -00.717 000.153 000.671 002.604 002.561 003.976 009.553 012.587 014.583 014.362 014.224 010.873 007.043 007.957 009.468 010.097 012.487;...
-01.956 -00.720 000.143 000.654 002.588 002.544 003.961 009.548 012.585 014.585 014.362 014.223 010.864 007.036 007.949 009.460 010.101 012.488;...
-01.949 -00.723 000.134 000.638 002.575 002.529 003.949 009.551 012.592 014.598 014.372 014.232 010.862 007.035 007.947 009.458 010.113 012.497;...
-01.945 -00.727 000.125 000.623 002.563 002.516 003.940 009.559 012.607 014.619 014.392 014.249 010.868 007.038 007.950 009.463 010.131 012.515;...
-01.941 -00.731 000.116 000.608 002.554 002.505 003.934 009.573 012.630 014.649 014.420 014.275 010.881 007.045 007.957 009.473 010.154 012.539;...
-01.939 -00.735 000.108 000.594 002.546 002.496 003.930 009.592 012.659 014.686 014.455 014.309 010.899 007.057 007.970 009.489 010.183 012.571;...
-01.938 -00.740 000.100 000.581 002.540 002.488 003.928 009.617 012.695 014.731 014.498 014.349 010.923 007.072 007.986 009.509 010.218 012.608;...
-01.938 -00.745 000.092 000.569 002.535 002.482 003.929 009.645 012.737 014.783 014.548 014.397 010.953 007.090 008.006 009.535 010.256 012.652;...
-01.939 -00.750 000.084 000.557 002.532 002.477 003.931 009.679 012.784 014.841 014.604 014.451 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.940 -00.756 000.077 000.545 002.530 002.473 003.936 009.716 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.943 -00.762 000.069 000.534 002.530 002.471 003.942 009.757 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.947 -00.768 000.062 000.524 002.530 002.470 003.949 009.801 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.951 -00.774 000.055 000.514 002.532 002.470 003.959 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-01.956 -00.780 000.048 000.504 002.534 002.471 003.970 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000];
% Table 6E Residuary Resistance Coefficient Component CR5 f:(IE: Half Entrance Angle)
CR5_tab = [...
-00.263 000.360 000.550 000.762 000.969 001.223 001.250 001.051 000.780 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.171 000.297 000.437 000.594 000.741 000.928 000.934 000.774 000.565 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.123 000.283 000.403 000.535 000.656 000.814 000.806 000.657 000.471 000.140 000.022 -00.006 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.094 000.288 000.399 000.521 000.628 000.773 000.753 000.605 000.426 000.117 000.011 -00.015 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.073 000.302 000.410 000.528 000.628 000.768 000.738 000.585 000.404 000.102 000.002 -00.023 000.005 -01.189 -00.258 -00.222 -00.729 -00.515;...
-00.058 000.323 000.431 000.548 000.645 000.784 000.745 000.583 000.397 000.093 -00.005 -00.030 -00.007 -01.114 -00.244 -00.213 -00.692 -00.499;...
-00.046 000.346 000.457 000.576 000.672 000.812 000.765 000.593 000.398 000.086 -00.011 -00.037 -00.017 -01.071 -00.235 -00.209 -00.673 -00.495;...
-00.036 000.372 000.486 000.609 000.705 000.850 000.794 000.610 000.405 000.081 -00.017 -00.043 -00.025 -01.048 -00.232 -00.208 -00.666 -00.499;...
-00.027 000.400 000.519 000.646 000.743 000.893 000.829 000.633 000.416 000.078 -00.022 -00.049 -00.034 -01.041 -00.231 -00.210 -00.668 -00.509;...
-00.020 000.429 000.553 000.685 000.785 000.941 000.868 000.659 000.429 000.076 -00.028 -00.056 -00.041 -01.044 -00.233 -00.214 -00.675 -00.522;...
-00.013 000.459 000.588 000.726 000.830 000.992 000.912 000.688 000.445 000.075 -00.032 -00.062 -00.049 -01.055 -00.236 -00.219 -00.688 -00.538;...
-00.007 000.489 000.625 000.769 000.876 001.045 000.957 000.720 000.463 000.074 -00.037 -00.068 -00.056 -01.072 -00.240 -00.225 -00.704 -00.557;...
-00.001 000.520 000.662 000.814 000.924 001.101 001.005 000.753 000.482 000.074 -00.042 -00.074 -00.063 -01.094 -00.246 -00.232 -00.723 -00.577;...
000.004 000.551 000.701 000.859 000.973 001.158 001.055 000.788 000.502 000.074 -00.046 -00.080 -00.070 -01.120 -00.253 -00.240 -00.744 -00.599;...
000.009 000.583 000.739 000.904 001.023 001.217 001.106 000.824 000.523 000.075 -00.051 -00.085 -00.077 -01.149 -00.260 -00.248 -00.767 -00.622;...
000.013 000.615 000.779 000.951 001.074 001.276 001.158 000.861 000.545 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.018 000.647 000.818 000.998 001.126 001.337 001.211 000.899 000.567 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.022 000.680 000.858 001.046 001.178 001.398 001.264 000.937 000.590 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.026 000.712 000.898 001.094 001.231 001.460 001.319 000.976 000.613 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000];
% Table 6F Residuary Resistance Coefficient Component CR6 f:(TA: Transom Area Ratio)
CR6_tab = [...
000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.013 000.006 000.007 000.013 000.012 000.011 -00.006 -00.013 000.006 -00.011 -00.003 -00.013 -00.027 -00.007 -00.033 -00.008 -00.015 000.004;...
000.029 000.016 000.017 000.028 000.026 000.025 -00.008 -00.022 000.012 -00.020 -00.006 -00.026 -00.052 -00.014 -00.064 -00.015 -00.030 000.005;...
000.047 000.029 000.031 000.046 000.043 000.041 -00.006 -00.029 000.019 -00.029 -00.009 -00.039 -00.077 -00.019 -00.093 -00.023 -00.046 000.005;...
000.067 000.046 000.049 000.065 000.061 000.059 -00.002 -00.033 000.027 -00.036 -00.011 -00.051 -00.101 -00.024 -00.119 -00.030 -00.061 000.001;...
000.090 000.067 000.069 000.087 000.083 000.079 000.006 -00.035 000.036 -00.043 -00.014 -00.062 -00.123 -00.028 -00.144 -00.037 -00.076 -00.004;...
000.116 000.091 000.093 000.111 000.106 000.101 000.018 -00.033 000.045 -00.048 -00.017 -00.074 -00.145 -00.030 -00.167 -00.044 -00.092 -00.012;...
000.143 000.118 000.121 000.137 000.132 000.126 000.033 -00.028 000.055 -00.053 -00.019 -00.084 -00.166 -00.032 -00.187 -00.050 -00.107 -00.023;...
000.173 000.150 000.152 000.166 000.160 000.153 000.051 -00.021 000.066 -00.056 -00.021 -00.095 -00.186 -00.033 -00.206 -00.057 -00.123 -00.036;...
000.206 000.185 000.187 000.196 000.191 000.183 000.073 -00.011 000.078 -00.058 -00.024 -00.105 -00.205 -00.033 -00.222 -00.063 -00.139 -00.051;...
000.241 000.223 000.225 000.229 000.224 000.215 000.098 000.003 000.091 -00.060 -00.026 -00.114 -00.223 -00.032 -00.237 -00.069 -00.155 -00.069;...
000.278 000.265 000.266 000.264 000.259 000.248 000.126 000.019 000.104 -00.060 -00.028 -00.123 -00.240 -00.030 -00.249 -00.075 -00.170 -00.089;...
000.318 000.311 000.311 000.301 000.297 000.285 000.158 000.038 000.119 -00.060 -00.030 -00.132 -00.256 -00.027 -00.260 -00.081 -00.186 -00.111;...
000.360 000.360 000.359 000.340 000.336 000.323 000.193 000.059 000.134 -00.058 -00.032 -00.140 -00.271 -00.023 -00.268 -00.087 -00.202 -00.136;...
000.405 000.412 000.411 000.382 000.379 000.364 000.232 000.084 000.149 -00.055 -00.034 -00.148 -00.285 -00.019 -00.275 -00.092 -00.218 -00.163;...
000.452 000.469 000.466 000.426 000.423 000.407 000.274 000.112 000.166 -00.051 -00.036 -00.155 -00.298 -00.013 -00.279 -00.098 -00.235 -00.193;...
000.501 000.529 000.525 000.472 000.470 000.452 000.320 000.142 000.183 -00.047 -00.037 -00.162 -00.310 -00.007 -00.281 -00.103 -00.251 -00.224;...
000.553 000.592 000.587 000.520 000.520 000.500 000.386 000.175 000.201 -00.041 -00.039 -00.169 -00.321 000.001 -00.281 -00.108 -00.267 -00.259;...
000.607 000.659 000.652 000.570 000.571 000.550 000.421 000.212 000.220 -00.034 -00.040 -00.175 -00.331 000.009 -00.280 -00.113 -00.283 -00.296;...
000.664 000.730 000.721 000.622 000.625 000.602 000.476 000.251 000.240 -00.026 -00.042 -00.181 -00.341 000.019 -00.276 -00.117 -00.300 -00.335;...
000.723 000.804 000.794 000.677 000.682 000.656 000.536 000.293 000.260 -00.017 -00.043 -00.186 -00.349 000.029 -00.270 -00.122 -00.316 -00.376];
% Table 6G Residuary Resistance Coefficient Component CR7 f:(TW: Transom Width Ratio)
CR7_tab = [...
000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
-00.028 -00.022 -00.018 -00.016 -00.005 -00.006 -00.014 -00.024 -00.026 -00.027 -00.031 -00.033 -00.024 000.000 000.000 000.000 000.000 000.000;...
-00.051 -00.039 -00.033 -00.028 -00.009 -00.012 -00.027 -00.048 -00.051 -00.053 -00.062 -00.065 -00.048 000.000 000.000 000.000 000.000 000.000;...
-00.070 -00.053 -00.045 -00.038 -00.011 -00.016 -00.039 -00.069 -00.075 -00.078 -00.091 -00.097 -00.071 000.000 000.000 000.000 000.000 000.000;...
-00.083 -00.063 -00.053 -00.044 -00.012 -00.021 -00.050 -00.089 -00.099 -00.103 -00.120 -00.127 -00.095 000.000 000.000 000.000 000.000 000.000;...
-00.092 -00.069 -00.057 -00.048 -00.012 -00.024 -00.059 -00.108 -00.122 -00.127 -00.149 -00.157 -00.118 000.000 000.000 000.000 000.000 000.000;...
-00.096 -00.071 -00.058 -00.048 -00.010 -00.027 -00.067 -00.126 -00.144 -00.151 -00.176 -00.186 -00.141 000.000 000.000 000.000 000.000 000.000;...
-00.096 -00.069 -00.056 -00.046 -00.006 -00.029 -00.074 -00.141 -00.165 -00.174 -00.203 -00.213 -00.164 000.000 000.000 000.000 000.000 000.000;...
-00.090 -00.063 -00.050 -00.040 -00.002 -00.030 -00.079 -00.156 -00.186 -00.196 -00.228 -00.241 -00.187 -00.280 -00.259 -00.169 -00.221 -00.299;...
-00.080 -00.053 -00.040 -00.031 000.004 -00.030 -00.083 -00.169 -00.207 -00.218 -00.254 -00.267 -00.209 -00.319 -00.293 -00.199 -00.253 -00.334;...
-00.065 -00.039 -00.028 -00.020 000.012 -00.030 -00.086 -00.181 -00.226 -00.239 -00.278 -00.292 -00.231 -00.358 -00.328 -00.231 -00.286 -00.369;...
-00.045 -00.022 -00.011 -00.005 000.020 -00.029 -00.088 -00.191 -00.245 -00.260 -00.301 -00.317 -00.253 -00.399 -00.362 -00.265 -00.320 -00.403;...
-00.020 000.000 000.009 000.013 000.031 -00.028 -00.088 -00.200 -00.263 -00.280 -00.324 -00.341 -00.275 -00.440 -00.398 -00.301 -00.356 -00.436;...
000.009 000.026 000.032 000.034 000.042 -00.026 -00.087 -00.207 -00.281 -00.299 -00.346 -00.364 -00.297 -00.482 -00.433 -00.339 -00.392 -00.470;...
000.043 000.055 000.059 000.057 000.055 -00.023 -00.085 -00.213 -00.297 -00.318 -00.367 -00.386 -00.318 -00.524 -00.469 -00.379 -00.429 -00.503;...
000.082 000.089 000.089 000.084 000.069 -00.019 -00.082 -00.218 -00.314 -00.336 -00.388 -00.407 -00.340 -00.568 -00.506 -00.421 -00.467 -00.535;...
000.126 000.126 000.123 000.114 000.085 -00.015 -00.077 -00.221 -00.329 -00.354 -00.408 -00.427 -00.361 -00.612 -00.542 -00.464 -00.506 -00.567;...
000.174 000.167 000.160 000.147 000.102 -00.010 -00.071 -00.223 -00.344 -00.371 -00.426 -00.447 -00.382 -00.657 -00.579 -00.510 -00.546 -00.598];
% Table 6H Residuary Resistance Coefficient Component CR8 f:(TT: Transom Depth Ratio)
CR8_tab = [...
000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.005 000.007 000.009 000.008 -00.003 -00.003 000.002 000.009 000.005 000.009 000.007 000.010 000.015 000.000 000.000 000.000 000.000 000.000;...
000.010 000.014 000.017 000.016 -00.005 -00.006 000.005 000.017 000.010 000.018 000.015 000.020 000.029 000.000 000.000 000.000 000.000 000.000;...
000.015 000.020 000.026 000.024 -00.008 -00.009 000.007 000.026 000.015 000.027 000.022 000.029 000.044 000.000 000.000 000.000 000.000 000.000;...
000.020 000.027 000.035 000.032 -00.010 -00.012 000.009 000.035 000.020 000.037 000.030 000.039 000.058 000.000 000.000 000.000 000.000 000.000;...
000.026 000.034 000.043 000.040 -00.013 -00.016 000.012 000.044 000.026 000.046 000.037 000.049 000.073 000.035 000.066 000.019 000.056 000.045;...
000.031 000.041 000.052 000.048 -00.015 -00.019 000.014 000.052 000.031 000.055 000.045 000.059 000.088 000.042 000.079 000.023 000.067 000.054;...
000.036 000.048 000.061 000.056 -00.018 -00.022 000.016 000.061 000.036 000.064 000.052 000.069 000.102 000.049 000.092 000.026 000.079 000.062;...
000.041 000.055 000.069 000.064 -00.020 -00.025 000.019 000.070 000.041 000.073 000.060 000.078 000.117 000.056 000.105 000.030 000.090 000.071;...
000.046 000.061 000.078 000.072 -00.023 -00.028 000.021 000.079 000.046 000.082 000.067 000.088 000.131 000.063 000.118 000.034 000.101 000.080;...
000.051 000.068 000.087 000.081 -00.026 -00.031 000.023 000.087 000.051 000.091 000.075 000.098 000.146 000.070 000.131 000.038 000.112 000.089;...
000.056 000.075 000.095 000.089 -00.028 -00.034 000.026 000.096 000.056 000.101 000.082 000.108 000.161 000.077 000.144 000.041 000.124 000.098;...
000.061 000.082 000.104 000.097 -00.031 -00.037 000.028 000.105 000.061 000.110 000.090 000.117 000.175 000.084 000.157 000.045 000.135 000.107;...
000.066 000.089 000.113 000.105 -00.033 -00.041 000.030 000.114 000.067 000.119 000.097 000.127 000.190 000.091 000.171 000.049 000.146 000.116;...
000.072 000.096 000.121 000.113 -00.036 -00.044 000.033 000.122 000.072 000.128 000.104 000.137 000.204 000.098 000.184 000.053 000.157 000.125;...
000.077 000.102 000.130 000.121 -00.038 -00.047 000.035 000.131 000.077 000.137 000.112 000.147 000.219 000.105 000.197 000.057 000.169 000.134;...
000.082 000.109 000.139 000.129 -00.041 -00.050 000.037 000.140 000.082 000.146 000.119 000.157 000.234 000.112 000.210 000.060 000.180 000.143;...
000.087 000.116 000.148 000.137 -00.044 -00.053 000.040 000.149 000.087 000.155 000.127 000.166 000.248 000.119 000.223 000.064 000.191 000.152;...
000.092 000.123 000.156 000.145 -00.046 -00.056 000.042 000.157 000.092 000.165 000.134 000.176 000.263 000.126 000.236 000.068 000.202 000.161;...
000.097 000.130 000.165 000.153 -00.049 -00.059 000.044 000.166 000.097 000.174 000.142 000.186 000.277 000.133 000.249 000.072 000.213 000.170;...
000.102 000.137 000.174 000.161 -00.051 -00.062 000.047 000.175 000.102 000.183 000.149 000.196 000.292 000.140 000.262 000.075 000.225 000.178;...
000.107 000.143 000.182 000.169 -00.054 -00.066 000.049 000.184 000.107 000.192 000.157 000.206 000.307 000.147 000.276 000.079 000.236 000.187];
% Table 6I Residuary Resistance Coefficient Component CR9 f:(BA: Bow Area Ratio)
CR9_tab = [...
000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000 000.000;...
000.015 000.013 000.008 000.002 -00.003 -00.006 -00.008 -00.016 -00.019 -00.022 -00.023 -00.020 -00.027 000.000 000.000 000.000 000.000 000.000;...
000.030 000.026 000.016 000.004 -00.007 -00.011 -00.016 -00.031 -00.039 -00.045 -00.046 -00.039 -00.053 000.000 000.000 000.000 000.000 000.000;...
000.045 000.038 000.024 000.006 -00.010 -00.017 -00.025 -00.047 -00.058 -00.067 -00.069 -00.059 -00.080 000.000 000.000 000.000 000.000 000.000;...
000.060 000.051 000.032 000.008 -00.014 -00.023 -00.033 -00.062 -00.078 -00.090 -00.091 -00.078 -00.107 000.000 000.000 000.000 000.000 000.000;...
000.075 000.064 000.040 000.009 -00.017 -00.028 -00.041 -00.078 -00.097 -00.112 -00.114 -00.098 -00.133 000.000 000.000 000.000 000.000 000.000;...
000.090 000.077 000.048 000.011 -00.021 -00.034 -00.049 -00.094 -00.117 -00.134 -00.137 -00.117 -00.160 000.000 000.000 000.000 000.000 000.000;...
000.105 000.089 000.056 000.013 -00.024 -00.040 -00.057 -00.109 -00.136 -00.157 -00.160 -00.137 000.000 000.000 000.000 000.000 000.000 000.000;...
000.120 000.102 000.064 000.015 -00.027 -00.045 -00.065 -00.125 -00.156 -00.179 -00.183 -00.157 000.000 000.000 000.000 000.000 000.000 000.000;...
000.135 000.115 000.072 000.017 -00.031 -00.051 -00.074 -00.141 -00.175 -00.202 -00.206 -00.176 000.000 000.000 000.000 000.000 000.000 000.000;...
000.150 000.128 000.080 000.019 -00.034 -00.056 -00.082 -00.156 -00.195 -00.224 -00.229 -00.196 000.000 000.000 000.000 000.000 000.000 000.000];
% Table 6J Residuary Resistance Coefficient Component CR10 f:(CWS: Wetted Surface Coefficient)
CR10_tab = [...
-00.230 -00.666 -00.744 -01.302 -01.997 -02.319 -02.856 -02.955 -03.436 -03.647 -03.928 -04.056 -03.971 -03.864 -02.862 -03.449 -02.707 -03.886;...
-00.231 -00.668 -00.747 -01.307 -02.003 -02.327 -02.866 -02.965 -03.448 -03.659 -03.941 -04.070 -03.984 -03.877 -02.872 -03.460 -02.716 -03.899;...
-00.232 -00.670 -00.749 -01.311 -02.010 -02.335 -02.875 -02.975 -03.459 -03.671 -03.955 -04.083 -03.998 -03.890 -02.881 -03.472 -02.726 -03.912;...
-00.232 -00.673 -00.752 -01.315 -02.017 -02.343 -02.885 -02.985 -03.471 -03.683 -03.968 -04.097 -04.011 -03.903 -02.891 -03.483 -02.735 -03.925;...
-00.233 -00.675 -00.754 -01.320 -02.023 -02.351 -02.895 -02.995 -03.482 -03.695 -03.981 -04.111 -04.024 -03.916 -02.901 -03.495 -02.744 -03.938;...
-00.234 -00.677 -00.757 -01.324 -02.030 -02.358 -02.904 -03.005 -03.494 -03.708 -03.994 -04.124 -04.038 -03.929 -02.910 -03.506 -02.753 -03.951;...
-00.235 -00.679 -00.759 -01.328 -02.037 -02.366 -02.914 -03.014 -03.506 -03.720 -04.007 -04.138 -04.051 -03.942 -02.920 -03.518 -02.762 -03.964;...
-00.236 -00.681 -00.762 -01.333 -02.044 -02.374 -02.923 -03.024 -03.517 -03.732 -04.021 -04.151 -04.064 -03.955 -02.929 -03.530 -02.771 -03.977;...
-00.236 -00.684 -00.764 -01.337 -02.050 -02.382 -02.933 -03.034 -03.529 -03.744 -04.034 -04.165 -04.078 -03.968 -02.939 -03.541 -02.780 -03.990;...
-00.237 -00.686 -00.767 -01.342 -02.057 -02.390 -02.942 -03.044 -03.540 -03.757 -04.047 -04.179 -04.091 -03.981 -02.949 -03.553 -02.789 -04.003;...
-00.238 -00.688 -00.769 -01.346 -02.064 -02.397 -02.952 -03.054 -03.552 -03.769 -04.060 -04.192 -04.104 -03.993 -02.958 -03.564 -02.798 -04.016;...
-00.239 -00.690 -00.772 -01.350 -02.070 -02.405 -02.962 -03.064 -03.563 -03.781 -04.073 -04.206 -04.118 -04.006 -02.968 -03.576 -02.807 -04.029;...
-00.239 -00.693 -00.774 -01.355 -02.077 -02.413 -02.971 -03.074 -03.575 -03.793 -04.086 -04.219 -04.131 -04.019 -02.977 -03.587 -02.816 -04.042;...
-00.240 -00.695 -00.777 -01.359 -02.084 -02.421 -02.981 -03.084 -03.586 -03.806 -04.100 -04.233 -04.144 -04.032 -02.987 -03.599 -02.825 -04.056;...
-00.241 -00.697 -00.779 -01.363 -02.090 -02.428 -02.990 -03.094 -03.598 -03.818 -04.113 -04.247 -04.158 -04.045 -02.997 -03.611 -02.835 -04.069;...
-00.242 -00.699 -00.782 -01.368 -02.097 -02.436 -03.000 -03.104 -03.609 -03.830 -04.126 -04.260 -04.171 -04.058 -03.006 -03.622 -02.844 -04.082;...
-00.242 -00.702 -00.784 -01.372 -02.104 -02.444 -03.010 -03.114 -03.621 -03.842 -04.139 -04.274 -04.184 -04.071 -03.016 -03.634 -02.853 -04.095;...
-00.243 -00.704 -00.787 -01.377 -02.111 -02.452 -03.019 -03.124 -03.632 -03.855 -04.152 -04.288 -04.198 -04.084 -03.025 -03.645 -02.862 -04.108;...
-00.244 -00.706 -00.789 -01.381 -02.117 -02.460 -03.029 -03.133 -03.644 -03.867 -04.166 -04.301 -04.211 -04.097 -03.035 -03.657 -02.871 -04.121;...
-00.245 -00.708 -00.792 -01.385 -02.124 -02.467 -03.038 -03.143 -03.655 -03.879 -04.179 -04.315 -04.224 -04.110 -03.045 -03.668 -02.880 -04.134;...
-00.246 -00.711 -00.794 -01.390 -02.131 -02.475 -03.048 -03.153 -03.667 -03.891 -04.192 -04.328 -04.238 -04.123 -03.054 -03.680 -02.889 -04.147;...
-00.246 -00.713 -00.797 -01.394 -02.137 -02.483 -03.057 -03.163 -03.678 -03.904 -04.205 -04.342 -04.251 -04.136 -03.064 -03.692 -02.898 -04.160;...
-00.247 -00.715 -00.799 -01.398 -02.144 -02.491 -03.067 -03.173 -03.690 -03.916 -04.218 -04.356 -04.264 -04.149 -03.073 -03.703 -02.907 -04.173;...
-00.248 -00.717 -00.802 -01.403 -02.151 -02.499 -03.077 -03.183 -03.702 -03.928 -04.232 -04.369 -04.278 -04.162 -03.083 -03.715 -02.916 -04.186;...
-00.249 -00.719 -00.804 -01.407 -02.157 -02.506 -03.086 -03.193 -03.713 -03.940 -04.245 -04.383 -04.291 -04.175 -03.093 -03.726 -02.925 -04.199;...
-00.249 -00.722 -00.807 -01.412 -02.164 -02.514 -03.096 -03.203 -03.725 -03.952 -04.258 -04.396 -04.304 -04.188 -03.102 -03.738 -02.934 -04.212;...
-00.250 -00.724 -00.809 -01.416 -02.171 -02.522 -03.105 -03.213 -03.736 -03.965 -04.271 -04.410 -04.318 -04.201 -03.112 -03.749 -02.944 -04.225;...
-00.252 -00.728 -00.814 -01.425 -02.184 -02.537 -03.125 -03.233 -03.759 -03.989 -04.297 -04.437 -04.344 -04.227 -03.131 -03.773 -02.962 -04.251;...
-00.253 -00.733 -00.819 -01.433 -02.198 -02.553 -03.144 -03.252 -03.782 -04.014 -04.324 -04.464 -04.371 -04.253 -03.150 -03.796 -02.980 -04.277;...
-00.255 -00.737 -00.824 -01.442 -02.211 -02.569 -03.163 -03.272 -03.805 -04.038 -04.350 -04.492 -04.397 -04.279 -03.170 -03.819 -02.998 -04.303;...
-00.256 -00.742 -00.829 -01.451 -02.224 -02.584 -03.182 -03.292 -03.828 -04.063 -04.377 -04.519 -04.424 -04.305 -03.189 -03.842 -03.016 -04.329;...
-00.258 -00.746 -00.834 -01.460 -02.238 -02.600 -03.201 -03.312 -03.851 -04.087 -04.403 -04.546 -04.451 -04.331 -03.208 -03.865 -03.034 -04.355;...
-00.259 -00.751 -00.839 -01.468 -02.251 -02.615 -03.220 -03.332 -03.875 -04.112 -04.429 -04.573 -04.477 -04.357 -03.227 -03.888 -03.053 -04.382;...
-00.261 -00.755 -00.844 -01.477 -02.265 -02.631 -03.240 -03.352 -03.898 -04.136 -04.456 -04.601 -04.504 -04.382 -03.246 -03.912 -03.071 -04.408;...
-00.263 -00.760 -00.849 -01.486 -02.278 -02.646 -03.259 -03.371 -03.921 -04.160 -04.482 -04.628 -04.531 -04.408 -03.266 -03.935 -03.089 -04.434;...
-00.264 -00.764 -00.854 -01.495 -02.291 -02.662 -03.278 -03.391 -03.944 -04.185 -04.508 -04.655 -04.557 -04.434 -03.285 -03.958 -03.107 -04.460;...
-00.266 -00.769 -00.859 -01.503 -02.305 -02.678 -03.297 -03.411 -03.967 -04.209 -04.535 -04.682 -04.584 -04.460 -03.304 -03.981 -03.125 -04.486;...
-00.267 -00.773 -00.864 -01.512 -02.318 -02.693 -03.316 -03.431 -03.990 -04.234 -04.561 -04.709 -04.611 -04.486 -03.323 -04.004 -03.143 -04.512;...
-00.269 -00.778 -00.869 -01.521 -02.332 -02.709 -03.335 -03.451 -04.013 -04.258 -04.587 -04.737 -04.637 -04.512 -03.342 -04.027 -03.162 -04.538;...
-00.270 -00.782 -00.874 -01.530 -02.345 -02.724 -03.355 -03.471 -04.036 -04.283 -04.614 -04.764 -04.664 -04.538 -03.362 -04.050 -03.180 -04.564;...
-00.272 -00.786 -00.879 -01.538 -02.358 -02.740 -03.374 -03.490 -04.059 -04.307 -04.640 -04.791 -04.691 -04.564 -03.381 -04.074 -03.198 -04.590;...
-00.273 -00.791 -00.884 -01.547 -02.372 -02.755 -03.393 -03.510 -04.082 -04.332 -04.667 -04.818 -04.717 -04.590 -03.400 -04.097 -03.216 -04.616;...
-00.275 -00.795 -00.889 -01.556 -02.385 -02.771 -03.412 -03.530 -04.105 -04.356 -04.693 -04.846 -04.744 -04.616 -03.419 -04.120 -03.234 -04.642;...
-00.276 -00.800 -00.894 -01.564 -02.399 -02.787 -03.431 -03.550 -04.128 -04.381 -04.719 -04.873 -04.771 -04.642 -03.438 -04.143 -03.252 -04.668;...
-00.278 -00.804 -00.899 -01.573 -02.412 -02.802 -03.450 -03.570 -04.151 -04.405 -04.746 -04.900 -04.797 -04.668 -03.458 -04.166 -03.271 -04.694;...
-00.280 -00.809 -00.904 -01.582 -02.425 -02.818 -03.470 -03.590 -04.174 -04.430 -04.772 -04.927 -04.824 -04.694 -03.477 -04.189 -03.289 -04.721;...
-00.281 -00.813 -00.909 -01.591 -02.439 -02.833 -03.489 -03.609 -04.197 -04.454 -04.798 -04.954 -04.851 -04.720 -03.496 -04.212 -03.307 -04.747];
%----------------------------------------------------------------------
if VL <= 1.30
VL1_tab = VL_tab(1:8);
CR1_tab = CR1_tab(:,1:8);
elseif VL <= 1.70 & VL > 1.30
VL1_tab = VL_tab(1:12);
DL_tab = DL_tab(1:22);
CR1_tab = CR1_tab(1:22,1:12);
elseif VL > 1.70
VL1_tab = VL_tab;
DL_tab = DL_tab(4:17);
CR1_tab = CR1_tab(4:17,:);
end
%----------------------------------------------------------------------
VL2_tab = VL_tab;
%----------------------------------------------------------------------
if VL <= 1.70
VL3_tab = VL_tab(1:12);
CR3_tab = CR3_tab(:,1:12);
elseif VL <= 1.80 & VL > 1.70
VL3_tab = VL_tab(1:13);
Cp_tab = Cp_tab(3:19);
CR3_tab = CR3_tab(3:19,1:13);
elseif VL > 1.80
VL3_tab = VL_tab;
Cp_tab = Cp_tab(4:19);
CR3_tab = CR3_tab(4:19,:);
end
%----------------------------------------------------------------------
if VL <= 1.20
VL4_tab = VL_tab(1:7);
CR4_tab = CR4_tab(:,1:7);
elseif VL <= 1.30 & VL > 1.20
VL4_tab = VL_tab(1:8);
Cx_tab = Cx_tab(1:18);
CR4_tab = CR4_tab(1:18,1:8);
elseif VL <= 1.70 & VL > 1.30
VL4_tab = VL_tab(1:12);
Cx_tab = Cx_tab(1:15);
CR4_tab = CR4_tab(1:15,1:12);
elseif VL > 1.70
VL4_tab = VL_tab;
Cx_tab = Cx_tab(5:14);
CR4_tab = CR4_tab(5:14,:);
end
%----------------------------------------------------------------------
if VL <= 1.40
VL5_tab = VL_tab(1:9);
CR5_tab = CR5_tab(:,1:9);
elseif VL <= 1.70 & VL > 1.40
VL5_tab = VL_tab(1:12);
IE_tab = IE_tab(3:15);
CR5_tab = CR5_tab(3:15,1:12);
elseif VL > 1.70
VL5_tab = VL_tab;
IE_tab = IE_tab(5:15);
CR5_tab = CR5_tab(5:15,:);
end
%----------------------------------------------------------------------
VL6_tab = VL_tab;
%----------------------------------------------------------------------
if VL <= 1.80
VL7_tab = VL_tab(1:13);
CR7_tab = CR7_tab(:,1:13);
elseif VL > 1.80
VL7_tab = V7_tab;
TW_tab = TW_tab(9:18);
CR7_tab = CR7_tab(9:18,:);
end
%----------------------------------------------------------------------
if VL <= 1.80
VL8_tab = VL_tab(1:13);
CR8_tab = CR8_tab(:,1:13);
elseif VL > 1.80
VL8_tab = VL_tab;
TT_tab = TT_tab(6:22);
CR8_tab = CR8_tab(6:22,:);
end
%----------------------------------------------------------------------
if VL <= 1.70
VL9_tab = VL_tab(1:12);
CR9_tab = CR9_tab(:,1:12);
elseif VL <= 1.80 & VL > 1.70
VL9_tab = VL_tab(1:13);
BA_tab = BA_tab(1:7);
CR9_tab = CR9_tab(1:7,1:13);
elseif VL > 1.80
VL9_tab = VL_tab;
CR9_tab = CR9_tab;
end
%----------------------------------------------------------------------
VL10_tab = VL_tab;
%----------------------------------------------------------------------
CR1 = interp1(DL_tab, CR1_tab, DL, 'linear','extrap');
CR1 = interp1(VL1_tab, CR1,VL, 'linear','extrap')/1000;
CR2 = interp1(BxTx_tab,CR2_tab, BxTx,'linear','extrap');
CR2 = interp1(VL2_tab, CR2,VL, 'linear','extrap')/1000;
CR3 = interp1(Cp_tab, CR3_tab, Cp, 'linear','extrap');
CR3 = interp1(VL3_tab, CR3,VL, 'linear','extrap')/1000;
CR4 = interp1(Cx_tab, CR4_tab, Cx, 'linear','extrap');
CR4 = interp1(VL4_tab, CR4,VL, 'linear','extrap')/1000;
CR5 = interp1(IE_tab, CR5_tab, IE, 'linear','extrap');
CR5 = interp1(VL5_tab, CR5,VL, 'linear','extrap')/1000;
CR6 = interp1(TA_tab, CR6_tab, TA, 'linear','extrap');
CR6 = interp1(VL6_tab, CR6,VL, 'linear','extrap')/1000;
CR7 = interp1(TW_tab, CR7_tab, TW, 'linear','extrap');
CR7 = interp1(VL7_tab, CR7,VL, 'linear','extrap')/1000;
CR8 = interp1(TT_tab, CR8_tab, TT, 'linear','extrap');
CR8 = interp1(VL8_tab, CR8,VL, 'linear','extrap')/1000;
CR9 = interp1(BA_tab, CR9_tab, BA, 'linear','extrap');
CR9 = interp1(VL9_tab, CR9,VL, 'linear','extrap')/1000;
CR10 = interp1(CWS_tab, CR10_tab, CWS, 'linear','extrap');
CR10 = interp1(VL10_tab,CR10,VL, 'linear','extrap')/1000;
%----------------------------------------------------------------------
% CR1 : f(DL)
% CR2 : f(Bx/Tx)
% CR3 : f(Cp)
% CR4 : f(Cx)
% CR5 : f(IE)
% CR6 : f(TA)
% CR7 : f(TW)
% CR8 : f(TT)
% CR9 : f(BA)
% CR10 : f(CWS)
%----------------------------------------------------------------------
% Residuary Resistance Coefficient
CR = CR1+CR2+CR3+CR4+CR5+CR6+CR7+CR8+CR9+CR10;
%----------------------------------------------------------------------
% Criteria
c1 = R; % Seakeeping Rank (maximize)
c2 = 1000*CR; % Residuary Resistance Coefficient (minimize)

%----------------------------------------------------------------------
% Functional Relations:
f1 = DISP; % Displacement (DEPLASMAN TONAJI), metric ton
f2 = CB; % Block Coefficient
f3 = Cx; % Maximum Section Area Coefficient (MAX.KESIT ALAN KATS)
% (assumed to be equal to midship section coefficient Cm)
f4 = Cwp; % Waterplane Area Coefficient (SU HATTI ALAN KATS)

f5 = CWF; % Waterplane Area Coefficient forward of amidships

% a1 = 45.104 (pseudo-criteria, maximize)
f6 = CWA; % Waterplane Area Coefficient aft of amidships
% a2 = 10.078 (pseudo-criteria, maximize)
f7 = TL; % Draft-to-length ratio
% a3 = -378.465 (pseudo-criteria, minimize)
f8 = cL; % Cut-up ratio
% a4 = 1.273 (pseudo-criteria, maximize)
f9 = CVPF; % Vertical prismatic coefficient forward of amidships
% a5 = -23.501 (pseudo-criteria, minimize)
f10 = CVPA; % Vertical prismatic coefficient aft of amidships
% a6 = -15.875 (pseudo-criteria, minimize)

%----------------------------------------------------------------------
% REFERENCES:
% [1] FUNG S.C.(AM),"Resistance and Powering Prediction for Transom
% Stern Hull Forms During Early Stage Ship Design", Department
% of the Navy, NAVSEA, Washington, DC, Sname Transactions,
% Vol 99, 1999, pp. 29-84
% [2] BALES N.K.,"Optimizing the Seakeeping Performance of Destroyer-
% Type Hulls", Proceedings, 13th ONR Symposium, October 1980
% [3] movi_oscillator.m (MATLAB m-file) (STATNIKOV R.B.)
% [4] STATNIKOV R.B. 2003, MOVI 1.3 Software Package User's Manual.
JOURNAL OF MULTI-CRITERIA DECISION ANALYSIS
J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
Published online in Wiley InterScience
(www.interscience.wiley.com). DOI: 10.1002/mcda.424

Visualization Approaches for the Prototype Improvement Problem

ROMAN STATNIKOV a,b, KIVANC ALI ANILc, ALEX BORDETSKYa and ALEXANDER STATNIKOVd,
a
Department of Information Sciences, Naval Postgraduate School, Monterey, California, USA
b
Mechanical Engineering Research Institute, Russian Academy of Sciences, Moscow, Russia
c
Naval Architecture and Ocean Engineering Faculty, IstanbulTechnical University, Maslak, Istanbul,Turkey
d
Department of Biomedical Informatics,Vanderbilt University, Nashville,Tennessee, USA

ABSTRACT
One of the basic engineering optimization problems is improvement of the prototype. This problem is often
encountered by industrial and academic organizations that produce and design various objects (e.g. motor vehicles,
machine tools, ships, and aircrafts). This paper presents an approach for improving the prototype by constructing
the feasible and Pareto sets while performing multicriteria analysis. We introduce visualization methods that
facilitate construction of the feasible and Pareto sets. Using these techniques developed on the basis of Parameter
Space Investigation method, an expert can correctly state and solve the problem under consideration in a series of
dialogues with the computer. Finally, we present a case study of improving the ship prototype. Copyright r 2008
John Wiley & Sons, Ltd.

KEY WORDS: Parameter Space Investigation (PSI) method; multicriteria analysis; prototype improvement; feasible
solution set; Pareto optimal set

1. INTRODUCTION et al., 2000), space shuttle, nuclear reactor, unmanned

vehicle configuration, airplane (Statnikov, 1999;
The prototype improvement (or operational Stadler and Dauer, 1992), ship (Anil, 2005;
development) problem can be defined as follows: Statnikov and Matusov, 1995), truck frame, multi-
There exists a real object (e.g. car, airplane, stage axial flow compressor for an aircraft engine,
machine tool, or ship) that we call a ‘prototype’. robot, machine tools and their units, rear axle
The prototype is described by k performance housing for a truck (Statnikov and Matusov, 2002),
criteria that take values Fp ¼ ðFp1 ; . . . ; Fpk ). It is a parafoil-load delivery system (Yakimenko and
necessary to improve all or the most significant Statnikov, 2005), a controllable descending system
criteria of the prototype. The problem of opera- (Dobrokhodov et al., 2003), vibration machines
tional development of a prototype is currently one (Sobol’ and Statnikov, 2006), air bearing (Barrans
the most pressing and complex design problems. and Bhat, 2003), and so on.
This problem is often encountered in the produc- To solve the above problem, the prototype’s
tion of machine tools, cars, ships, and aircrafts, operational development requires two stages. In
where significant resources are allocated for the the first stage, based on the tests one must identify
operational development of the existing object. the mathematical model of the object and
Furthermore, making the operational development determine its parameters. To this end, one can
of the prototype as quick as possible is solve an identification problem by working with
highly desirable. Prior studies provide examples of particular adequacy (proximity) criteria. By
real-life multicriteria problems of improving the adequacy criteria we mean the discrepancies
prototype of a car, valve gear, metal-cutting between the experimental and computed data,
machine tools, gear units, flexible manufacturing the latter being determined on the basis of
systems (Statnikov and Matusov, 1995; Gobbi mathematical model. The number of adequacy
criteria can reach many dozens if not hundreds.
Multicriteria identification and adequacy of the
*Correspondence to: Department of Biomedical Infor- mathematical models are discussed in detail in
matics, Vanderbilt University, Nashville, TN 37232, Statnikov and Matusov (2002). In the second
USA. E-mail: [email protected] stage, an expert formulates and solves the

Copyright r 2008 John Wiley & Sons, Ltd.

46 R. STATNIKOV ET AL.

multicriteria optimization problem using the (Statnikov and Matusov, 2002) and the Pareto
performance criteria and the mathematical model optimal set often contains only a few solutions
whose adequacy has been established in the first due to stringent constraints.
stage.
Sometimes we face another situation: The Therefore, constructing the feasible solution set
mathematical model of the prototype is given is the most important step in formulating and
and we know the values of criteria of the prototype solving the prototype improvement problems.
(Anil, 2005). In this case, we proceed with analysis In the present paper, solution of the prototype
that resembles the second stage of the operational improvement problem is based on construction of
development of the prototype. the feasible solution set via the Parameter Space
The present work is devoted to approaches for Investigation (PSI) method that is widely used in
improving the basic performance criteria by chan- various fields of industry, science, and technology
ging the design variable, functional, and criteria (Sobol’ and Statnikov, 2006; Statnikov and
constraints while constructing the feasible and Matusov, 1995, 2002; Statnikov et al., 2006). We
Pareto optimal solution sets. These approaches introduce novel visualization procedures that
allow one to determine the potential of improving guide experts in analysis of such problems. In
the prototype. some sense, the proposed visualization procedures
In general, the prototype improvement problem are ‘diagnostic tools’ for the experts.
possesses several distinctive features outlined During the last 30 years, significant research
below. The methodology proposed in the has been done for visualization of two
present work is devoted to analysis of such dimensional projections of criterion, design, and
problems. mixed vectors (e.g. see Cleveland, 1985; Lotov
et al., 2004; Meisel, 1973). It is also recognized in
1. The problem is essentially multicriteria, and the the field that these projections by them-
criteria are usually contradictory. For this selves cannot give a comprehensive idea about
reason there are difficulties in defining the localization of the feasible solution set in the
criteria constraints correctly. multidimensional design variable and criteria
2. The initial constraints on the design variables, spaces. The novelty of visualization techniques
criteria, and functional dependencies may result proposed in this work is that they are used
in an empty or very sparse feasible solution set. together with the multicriteria test tables obtained
Therefore, it is necessary to correct the initial by the PSI method during construction of the
statement of the problem. Thus, the formula- feasible solution set. These tools are an important
tion and solution of the problem have to addition to the PSI method but they cannot
comprise a single process (Sobol’ and Statni- substitute for it. Thus, the main thesis of the
kov, 2006; Statnikov and Matusov, 2002). present paper is that in order to construct the
3. The feasible solution set can be multiply feasible solution set, it is necessary to consider
connected, and its volume may be several both multicriteria test tables and various
orders of magnitude smaller than that of visualization tools.
the domain where the optimal solutions are This paper is organized as follows: Section 2
sought. Generally, both the feasible solu- discusses the formulation and solution of the
tion set and the Pareto optimal set are prototype improvement problem by optimizing
nonconvex. performance criteria. In this section, we also
4. Mathematical models are usually complex sys- briefly review the PSI method that allows to
tems of equations (including differential equa- construct the feasible and Pareto sets. The visua-
tions) that may be nonlinear, deterministic or lization tools are discussed in Section 3. The
stochastic, with distributed or lumped para- geometric interpretation of the prototype
meters. Information about the smoothness of improvement problem is provided in Section 4.
goal functions is usually not available. Finally, to illustrate the usefulness of the
5. The analysis of Pareto optimal set to determine techniques presented in this paper, Sections 5
the most preferable solution does not pose a and 6 contain a case study that solves the
challenge to the expert (Ozernoy, 1988; Statni- problem of improving the pretsaliminary ship
kov, 1999). This is because experts have a design prototype by using the PSI method and
sufficiently well-defined system of preferences various visualization tools.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 47

2. FORMULATION AND SOLUTION OF THE that FðaÞoFðbÞ if for all n ¼ 1; . . . ; k, Fn ðaÞpFðbÞ

PROTOTYPE IMPROVEMENT PROBLEM and for at least one n0 2 f1; . . . ; kg, Fn0 ðaÞoFn0 ðbÞ.
Finally, an expert determines a vector of design
2.1. Generalized formulation of multicriteria opti- variables aO 2 P that is the most preferred among
mization problems the vectors belonging to set P.
We assume that a prototype depends on r design
variables a1 ; . . . ; ar representing a point a ¼ ða1 ; 2.2. Parameter space investigation (PSI) method
. . . ; ar Þ in the r-dimensional space. In the general Now we proceed by describing the PSI method
case, one has to take into account design variable, that allows to determine F

n and, hence, the
functional, and criteria constraints. feasible solution set correctly. The PSI method is
The design variable constraints have the form based on the investigation of the parallelepiped P
a
j paj pa



j , j ¼ 1; . . . ; r. Constraints aj and aj



with points of uniformly distributed sequences
define a parallelepiped P in the r-dimensional (e.g. LPt sequences), see Stadler and Dauer (1992),
design variable space. From the expert’s Statnikov and Matusov (1995, 2002) and Statni-
perspective, the values of design variable kov et al. (2006) for details. The method consists
constraints can be modified, if that leads to of three stages:
improvement of the basic criteria. Stage 1. Compilation of test tables via computer.
The functional constraints can be written as First, one chooses N test points a1 ; . . . ; aN that
follows: Cl
pfl ðaÞpCl

, l ¼ 1; . . . ; t, where fl ðaÞ is satisfy the functional constraints. Then all the
a functional relation, Cl
and Cl

are some particular criteria Fn ðai Þ are calculated at each of
constants. Functional relations together with the points ai ; for each of the criteria a test table is
constraints are some requirements of the object compiled so that the values of Fn ða1 Þ; . . . ; Fn ðaN Þ
that sometimes an expert can successively revise are arranged in increasing order, i.e.
in order to improve the basic performance
Fn ðai1 ÞpFn ðai2 Þp    pFn ðaiN Þ; n ¼ 1; . . . ; k ð2Þ
criteria. These can be norms, standards, and
other requirements such as mass, overall dimen- where i1 ; i2 ; . . . ; iN are the numbers of tests (a
sions, allowable stress in structural elements, and separate set for each n). Taken together, the k
so on. tables form a complete test table.
The operation of a prototype is described by the Stage 2. Preliminary selection of criterion
particular performance criteria Fn ðaÞ, n ¼ 1; . . . ; k. constraints. This stage includes interaction with
All other things being equal, it is desired that these an expert. By analysing inequalities (2), an expert
criteria are optimized. For simplicity, we assume specifies the criteria constraints F

n . Actually, an
that functions Fn ðaÞ are to be minimized. To avoid expert has to consider one criterion at a time and
situations in which the expert regards the values of specify the respective constraints. One analyses a
some criteria as unacceptable, we introduce test table and imposes the criterion constraint.
criteria constraints in the form Fn ðaÞpF

n , Then one proceeds to the next table, and so on.
n ¼ 1; . . . ; k, where F

n is the worst value of Note that the revision of the criteria constraints
criterion Fn ðaÞ acceptable to an expert. These within the limits of the test tables does not lead to
constraints are repeatedly revised during solution any difficulties for an expert.
of the problem. The choice of F

n is discussed in Since we want to minimize all criteria, F

n
the following sub-section. should be the maximum values of the criteria
The design variable, functional, and criteria Fn ðaÞ, which guarantee an acceptable level of the
constraints define the feasible solution set D 
. object’s operation. If the selected values of F

n are
Let us formulate one of the basic problems of not a maximum, then many interesting solutions
multicriteria optimization. It is necessary to define may be lost, since some of the criteria are
the feasible solution set D and find a set P  D contradictory. Moreover in some cases the
such that feasible solution set may be empty.
In practice, an expert imposes the criteria
FðPÞ ¼ min FðaÞ; ð1Þ constraints in order to improve a prototype by
a2D all criteria simultaneously. If it is impossible, one
improves a prototype by the most important
where FðaÞ ¼ ðF1 ðaÞ; . . . ; Fk ðaÞÞ is the criterion criteria. In process of dialogues with computer,
vector and P is the Pareto optimal set. We mean an expert repeatedly revises criteria constraints

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
48 R. STATNIKOV ET AL.

and carries out the multicriteria analysis. The PSI necessarily stable are discussed in Lichtenstein and
method gives expert valuable information on the Slovic (2006), Wu and Azarm (2001) and Zitzler
advisability of revising various criteria constraints et al. (2003).
with the aim of improving the basic criteria. The
expert sees what price he pays for making
concessions in various criteria, i.e. what one loses 2.3. Pseudo-criteria
and what one gains. As mentioned above, often an expert cannot
Stage 3. Solvability of problem (1) via computer. determine the functional constraints correctly.
Let us fix a criterion, say Fn1 ðaÞ, and consider the For example, in practical problems ‘good’ solu-
corresponding test table (2). Let S1 be the number tions may lie beyond the limits imposed by the
of the values in the table satisfying the selected constraints.1 If informed of this, an expert may be
criterion constraint: ready to modify the constraints so that the ‘good’
solutions will belong to the feasible solution set.
Fn1 ðai1 Þp    pFn1 ðaiS1 ÞpF


n1 ð3Þ Below we present an approach to obtain such
Then criterion Fn2 is selected by analogy with Fn1 information.
and the values of Fn2 ðai1 Þ; . . . ; Fn2 ðaiS1 Þ in the test Instead of the function fl ðaÞpCl

, l ¼ 1; . . . ; t
table are considered. Let the table contain S2 pS1 with the soft constraint Cl

, we introduce an
values such that Fn2 ðaij ÞpF

n2 , where 1pjpS2.
additional criterion2 Fkþl ðaÞ ¼ fl ðaÞ, which we call
Similar procedures are carried out for each a pseudo-criterion. To find the value of the
criterion. Then if at least one point can be found constraint F

kþl one has to compile a test table
for which all criteria constraints are valid containing Fkþl ðaÞ. By using the aforementioned
simultaneously, then the set D is nonempty and algorithm, one can define F

kþl in a way that
problem (1) is solvable. Otherwise, an expert prevents the loss of interesting solutions.
should return to Stage 2 and make certain In general, when solving a problem with soft
concessions in the specification of F

n . However,
functional constraints, one has to find the set D,
if the concessions are highly undesirable, then one taking all performance criteria and pseudo-criteria
may return to Stage 1 and increase the number of into account. In other words, one must solve the
points in order to repeat Stages 2 and 3 using problem with the constraints
extended test tables (Statnikov and Matusov,
2002).
Fn ðaÞpF


n ; n ¼ 1; . . . ; k; k þ 1; . . . ; k þ t
The procedure is to be iterated until D is
nonempty. Then the Pareto optimal set is
constructed in accordance with the definition Thus, to define the feasible solution set, we
presented above. This is done by removing those consider a multicriteria problem with k þ t
feasible points that can be improved with respect criteria. Notice, however, that the pseudo-criteria
to all criteria simultaneously. are not considered when constructing the Pareto
Thus, in accordance with the PSI method, the optimal set.
criteria constraints are determined in the dialogue It is worthwhile to mention that many single-
of an expert with a computer. Then an expert criterion problems may have soft functional
should determine the Pareto set P and after constraints as well. In these cases the definition
analysing P find the most preferred solution of the feasible solution set is also very important.
FðaO Þ surpassing the prototype in all criteria, or For the definition of the feasible solution set, it is
at least the most important ones. For the class of necessary to represent functional dependencies
problems considered in this paper, experts do not (with soft constraints) as pseudo-criteria. In
encounter serious difficulties in analysing the other words, we have to consider such single-
Pareto optimal set and in choosing the most criterion problems as multicriteria ones.
preferred solution. This is because experts have a
sufficiently well-defined system of preferences for 1
If these constraints are given unreasonably rigid,
this type of problems (Statnikov and Matusov, solutions that do not satisfy them are not considered
2002) and the Pareto optimal set often contains at all. For this reason the feasible solution set can be
only a few solutions due to stringent constraints. poor or even empty.
2
More complex cases of the decision making, where Recall that we have test tables that contain performance
preferences on the Pareto optimal set are not criteria Fn ðaÞ, n ¼ 1; . . . ; k.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 49

2.4. Construction of the combined feasible and cannot say with certainty which uniformly dis-
Pareto sets tributed sequences are the most suitable. Much
Quite often the analysis of test tables points to depends on the behaviour of the criteria, the form
advisability of correcting the boundaries of the of the functional and design variable constraints,
initial parallelepiped (i.e. re-defining the con- the number of tests, and the geometry of the
straints on the design variables) and constructing feasible solution set (Statnikov and Matusov,
a new parallelepiped. Suppose that appropriate 2002). Steuer and Sun (1995) have indicated an
investigations have been performed in the new opportunity of using random number generators
parallelepiped, and that the corresponding feasible in the PSI method. Based on these recommenda-
set has been constructed. Now it is necessary to tions, we have successfully applied random num-
combine the feasible sets constructed in both these ber generators in the PSI method to solve
parallelepipeds and define the Pareto optimal multicriteria problems with very high-dimensional
solutions in the combined feasible set. The design variable vectors (dimensionality 5 1000)
procedure of the repeated correction of parallele- and to also solve the problems in the parallel
pipeds and construction of the combined feasible mode (Statnikov et al., 2005a, 2006). The experi-
and Pareto optimal sets is essential for our ments with various random number generators in
problem, as emphasized in Section 4. Constructing the PSI method are described in Statnikov et al.
the combined Pareto optimal set allows to estimate (2005b).
the contribution of each parallelepiped to this set
and the expediency of correcting the initial 2.6. Applications of the PSI method
problem (see Section 6). The PSI method is implemented in the MOVI
There is another situation where construction of (Multicriteria Optimization and Vector Identifica-
the combined feasible and Pareto sets is needed. tion) software system (Yanushkevich et al., 2004).
These are the problems where calculation of the The software package MOVI allows solution of
criteria vector for one test requires a significant problems where the number of design variables
amount of computer time (e.g. see Section 2.6). and criteria is not practically limited. Below we
Similarly, these are problems that require to summarize several problems with different char-
perform a very large number of tests. For the acteristics where the PSI method and MOVI
efficient solution of such problems, the desired software system have been successfully applied.
number of tests N can be distributed among k These examples also demonstrate that the number
computers and each computer will perform a of tests depends on constraints, dimensionality of
search for the feasible solutions for its own design variable vector, and time of calculating one
subproblem by conducting N/k tests. Next, the criteria vector for given design variable values
obtained feasible solution sets are combined and (Statnikov et al., 2006).
the Pareto optimal solution set is constructed (A) Problem of naval ship design (Anil, 2005).
(Statnikov et al., 2005a, 2006). Among the particular features of this problem are
the high dimensionality of the design variable
2.5. Number generators for systematic search in the vector (45 design variables) and the difficulties in
design variable space improving a reasonably good prototype under
To investigate design variable space, we use strong constraints on six performance criteria
uniformly distributed sequences. At present, the (propulsion power factor, electrical power factor,
so-called LPt sequences are among the best ones in volume factor, region factor, weight factor, and
terms of uniformity characteristics (Sobol’ and cost), nine pseudo-criteria, and seven functional
Statnikov, 2006; Statnikov and Matusov, 1995, dependences. Since calculation of one vector of
1996, 2002). These sequences are used to compute criteria took o1 s and the design vector is of high
N test points a1 ; . . . ; aN in the design variable space dimensionality, 200 000 tests were conducted in the
during Stage 1 of the PSI method. Other uniformly first experiment. Multicriteria analysis showed the
distributed sequences and nets (Faure, 1982; necessity of repeated correction of the constraints,
Halton, 1960; Hammersley, 1960; Hlawka and and because of this, five more experiments with
Taschner, 1991; Kuipers and Niederreiter, 1974; 200 000–500 000 tests have been performed. Each
Niederreiter, 1990; Statnikov and Matusov, 1996) subsequent experiment was carried out on the
can be also successfully used in the PSI method. basis of the previous one. In the first two
However, prior to solving a specific problem, one experiments no feasible solutions were found; in

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
50 R. STATNIKOV ET AL.

the third experiment a few feasible solutions were system. In the initial problem statement, 4096 tests
obtained; and it was only in the fourth experiment were conducted. Only seven solutions met the
where some satisfactory results were found. An criteria constraints and thus entered the feasible
analysis of these results allowed to define a region solution set. After analysing obtained results, new
of ‘good’ solutions where subsequent experiments boundaries of design variables were defined. The
were carried out. As a result, 26 Pareto optimal same number of tests (4096) was conducted. As a
solutions surpassing the prototype in all criteria result, 11 more feasible solutions were obtained.
were identified. On the basis of solution of the identification
(B) Problem of design of a car for shock problem and definition of the feasible solution set,
protection (Statnikov and Matusov, 2002). Unlike the problem of optimization by 20 performance
the previous problem where calculation of a criteria was solved next. These criteria were
criteria vector took o1 s, the criteria in this divided into the following groups: (1) comfort,
problem are based on finite element model with (2) durability, (3) load preservation, and (4) safety.
thousands of elements and nodes, and it takes Twenty parameters were varied. Optimization was
several hours to compute one vector of criteria. aimed to improve the prototype. 4096 tests were
This problem has 10 criteria: the mass of structure conducted and 21 solutions satisfied all constraints
and residual strains in the car body after impact in and entered the feasible solution set. The Pareto
the nine most dangerous points of the rear panel. set consisted of 20 solutions.
The number of design variables in this problem is
13, that at face value suggests a large
computational experiment. However, since it
3. TOOLS FOR VISUALIZATION
requires such a large amount of computational
time to compute one vector of criteria, the
Below we describe a few visualization tools that
optimization of design variables is very difficult
are particularly useful for multicriteria analysis.3 It
to implement. Therefore, the initial model was
is important to emphasize that these tools should
decomposed in two approximate models of be used together with the test tables (see Section
bumper and real panel. It was necessary to define
2.2) that allow to define feasible solutions in the
the consistent solutions between subsystems of
problems of any dimensionality. All the tools listed
bumper and real panel. For this purpose, 300 tests below are implemented in the software system
were carried out. Each test required no more than
MOVI.
10 min. The analysis of the obtained solutions has
shown that the prototype cannot be improved. As
a result, the problem has been reformulated (i.e.  Histograms of the distribution of feasible and
the designs of the bumper and the rear panel have Pareto optimal solutions. The intervals ½a
j ; a

j ,
been modified by introducing additional stiffening j ¼ 1; . . . ; r are divided into 10 identical sub-
ribs). Fifteen consistent design variable vectors intervals. Above each subinterval, the number
were defined. For these vectors we have calculated of feasible designs entering this subinterval is
all criteria using the original model. The number indicated. Analysing the histograms and graphs
of feasible solutions satisfying all constraints of the reveals how the feasible and Pareto sets are
structure was nine. The number of Pareto optimal distributed in design variable space. The
solutions was five. This approach allowed to histograms play the main role in correcting
improve the prototype in the successive design variable and other constraints.
application of the PSI method.  Graphs: ‘criterion vs. design variable.’ We con-
(C) Problem of operation development of a truck sider projections of the multidimensional points
(Statnikov and Matusov, 2002). In terms of Fn ðai Þ; n ¼ 1; . . . ; k; i ¼ 1; . . . ; N1 onto the plane
computational time, this problem falls in-between
two problems described above. The computation 3
In this paper we indicate only some basic tools. A
of a vector of criteria based on a system of detailed description of other tools, e.g. tables of
complex differential equations took 3 min. In functional and criteria failures, tables of criteria, tables
the identification phase of this problem, we of design variables, graphs of criteria vs. design
identified 16 parameters of the mathematical variables for the Pareto optimal solutions is provided
model using 65 adequacy criteria and defined to in Statnikov et al. (2005a, 2006) and Yanushkevich
what extent the model corresponds to the real et al. (2004).

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 51

Fn aj . These projections provide information feasible set is shaded in Figure 1. After analysing
about the sensitivity of criteria to the design the Pareto set PF , an expert determines the most
variables, ànd also point to localization of the preferable solution FO 2 PF . Notice that Da , Pa ,
feasible solution set. Significance of the sensi- ap , and aO are inverse images of DF , PF , Fp , and
tivity of criteria to the design variables is FO , respectively, in the design variable space.
indicated in (Sobol’, 2001). More general scenarios are discussed below.
 Graphs: ‘criterion vs. criterion.’ After N tests, N1 Next we will consider the situations when an
design variable vectors have entered the test expert should change the initial parallelepiped
1 .
table. We consider projections of the multi- Case 1b. Similarly to case 1a, we assume that
dimensional points Fn ðai Þ; n ¼ 1; . . . ; k, i ¼ 1; the values of the vector ap and vector Fp are
. . . ; N1 onto the plane Fi Fj . These projections known, see Figure 2. Let the feasible set, the
provide information about dependencies be- Pareto set PF1 (curve AB), and the most preferable
tween criteria and localization of the feasible solution FO1 be defined in the initial statement of
solution and Pareto optimal sets in criteria the problem, see Figure 2(a).
space. In many optimization problems some coordinates
of the optimal solution aO1 ¼ ðaO1 O1
1 ; . . . ; ar ) are
We note that the graphs ‘criterion vs. criterion’ located in the vicinity of the borders of the initial
by themselves are not sufficiently informative, and parallelepiped
1 . In these cases, it is logical to
it is indeed difficult or sometimes impossible to change the initial borders and to perform new
make any conclusions solely based on them. That investigations. A parallelepiped
2 with new
is why we propose here a generalized approach of boundaries is shown in Figure 2(b). As a result of
solution of the prototype improvement problem. investigating
2 , the new feasible set, the Pareto set
Specifically, test tables and histograms together PF2 (curve CD), and the most preferable solution
with graphs give a comprehensive idea about the FO2 are defined, Figure 2(a). Since the feasible sets
feasible solution set and necessity of correcting
initial problem statement. This approach allows an
expert to (1) estimate the effect of all constraints, 1 1
(2) determine significant design variables, and (3) P
P

compare the values of the criteria of a prototype 1

D P

with the results obtained by the PSI method. The D

examples of graphs and histograms are provided in
Section 6. O 1
O
P
P

4. GEOMETRIC INTERPRETATION OF P
2 2
(a) (b)
IMPROVING A PROTOTYPE PROBLEM 2

Figure 1. Improving a prototype Fp :FO is the most

Below we consider the basic scenarios for improv- preferable solution.
ing a prototype by using the PSI method and
provide guidelines to the experts.
Case 1a. The values of design variable vector ap O2

and criteria vector Fp ¼ ðFp1 ; Fp2 ) of the prototype

1
1
2
C´
are known, see Figure 1. The boundaries of the D´
initial parallelepiped
1 are defined as admissible P
B´
P A
deviations of the design variables from the 1
C
O1

B
corresponding values of the prototype ap. In
O1 P
P 1

Figure 1(b), vector ap is located in the centre of O2 P 2 A´

1
the parallelepiped
1 . Here and henceforth the D
parallelepiped
1 corresponds to the initial 2
(a) P (b) 2
statement of the optimization problem. On the 2

basis of the PSI method, the feasible set DF and Figure 2. Improving the Pareto optimal solution FO1
the Pareto set PF are defined (Statnikov and and definition of the most preferable solution
Matusov, 1995, 1996, 2002). The disconnected FO2 ; aO1 2
1 ; aO2 2
2 .

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
52 R. STATNIKOV ET AL.

were defined in
1 and
2 , an expert should 1 1

D´
construct the combined feasible and Pareto optimal P 2 2 B´
sets. In this example the combined Pareto set P 1
1

corresponds to PF2 , Figure 2(a). In other words, P A D

P
P E´
all the solutions belonging to PF2 are better than the 1
B O
E
solutions belonging to PF1 . Notice that A0 B0 , C0 D0 , O C
aO1 , aO2 , and ap are inverse images of AB, CD, FO1 , D
A´ C´
FO2 , and Fp , respectively, in the design variable
space (Figure 2b). (a) 2
P 2
(b)
2

On the basis of the PSI method, an expert can

determine the most significant design variables. If Figure 3. Construction of the combined Pareto optimal
an expert prefers to vary only the significant design set AEC; ap 2
1 , aO 2
2 .
variables, the dimensionality of the new
parallelepiped
2 can be smaller than the
dimensionality of the initial parallelepiped
1 . D
Case 1c. In contrast to the case 1b, the new 1 1
F F-1
P
combined Pareto set (curve AEC) consists of two P O
curves, AE 2 PF1 and EC 2 PF2 , see Figure 3(a). 1

Here DF is a combined feasible set. Notice that O

P
P
2
A0 B0 , C0 D0 , and A0 E0 C0 are inverse images of AB,
CD, and AEC in the design variable space P
respectively (Figure 3b). After analysing the
1
combined Pareto set, an expert determines the 2
most preferable solution FO . Notice FO 2 DC. P 2
(a) 2 (b)
From Figure 3(b) it follows that aO 2
2 . The
Figure 4. Identification of the vector ap and definition
example illustrating this case is presented in of the most preferable solution FO ; ap 2
2 , aO 2
F .
Sections 5 and 6.
Case 1d. The value of Fp is known, but many
values of the design variables of a prototype can be In general, an expert corrects the statement of
unknown, see Figure 4. This case is common in the the problem by constructing the parallelepipeds
first stage of the prototype’s operational
1 ; . . . ;
F1 ;
F , and then constructs the
development. In this stage, an expert should combined feasible and Pareto sets. These steps
identify a mathematical model and its parameters are carried out by using the PSI method. For more
using adequacy criteria but usually has only a details (see Statnikov et al., 2005a).
rough idea about the limits of many identified In the above cases, the obtained Pareto optimal
design variables. Assume that an expert cannot solution FO surpassed the prototype Fp in all
identify the vector ap in
1 and the feasible set is criteria. Situations in which an expert cannot
empty, see Figure 4(b). improve all the criteria simultaneously are shown
The parallelepiped
1 must be corrected and an in cases 2 and 3 below.
expert constructs a new parallelepiped
2 where Case 2. A sufficiently typical situation is shown
he can identify the vector ap . Then an expert states in Figure 5. After constructing the Pareto optimal
and solves the problem of improving a prototype set PF an expert revealed that Fp 2 PF . The
by the performance criteria (the second stage feasible set DF is shaded in Figure 5. The
mentioned in the Introduction). In an attempt to multicriteria analysis of the obtained Pareto
improve the prototype, an expert constructs and optimal solutions allowed to determine the most
investigates parallelepipeds
3 ; . . . ;
F1 ;
F . In preferable solution FO .
Figure 4(b) the feasible set Da and Pareto set Pa Case 3. Here an expert wishes to improve the
are defined in the parallelepiped PF. The feasible prototype, and a priori he defines the value of the
set Da is shaded. An expert determines the most criteria vector as FW ¼ ðFW W
1 ; . . . ; Fk Þ, see Figure
preferable solution FO in the Pareto optimal set 6(a). However in contrast to the former cases, here
PF , Figure 4(a). Notice that Pa , ap , and aO are all or some values FW W
1 ; . . . ; Fk can be
inverse images of PF , Fp , and FO in the design unattainable. Assume that the investigation of
variable space respectively (Figure 4b). the initial problem’s statement revealed that the

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 53

1 variable space, Figure 6(b); aW does not exist.) The

second option is not to accept a compromise
1
** solution. In this case, it is possible that the expert’s
O
wishes can be realized by creating a new object,
which is different from the prototype. Therefore,
D
P
the multicriteria analysis is helpful to answer the
P important question of how to improve the
prototype and by how much.
The identification of mathematical models of
2
** 2 different objects (vehicles, machine tools, and their
units) and their improvement were described in
Figure 5. Fp 2 PF . FO is the most preferable solution. detail in Statnikov and Matusov (2002). These
analyses correspond to cases 1d–3. Similar
problems concerning the parafoil-load delivery
P 2
P 1 system and the controllable descending system
1 1 F P F were discussed in Dobrokhodov et al. (2003) and
P F

C Yakimenko and Statnikov (2005).

C P 2 Finally, we would like to mention that
improvement of the prototype depends on (1)
W
2
constructive scheme (topology of the object), (2)
P 1
dimensionality of the design variable vector,
1 significant design variables, range of their
change, (3) materials constituting the design
2 2
(a) (b) variables (physical and chemical properties).
Figure 6. Pareto optimal set PFF cannot ‘reach’ the
solution FW . FW is unattainable, aW does not exist.
Solution FC is a compromise. 5. EXAMPLE: THE PROBLEM OF
IMPROVING A PRELIMINARY SHIP
DESIGN PROTOTYPE
feasible set is empty and an expert decides to
The mathematical model described below is based
correct the initial parallelepiped
1 . He corrects
on the references (Bales, 1980; Fung, 1991). More
the statement of the problem repeatedly, including
details are also provided in Anil (2005). Since the
the construction of the parallelepipeds
2 ; . . . ;
F ,
example is used primarily to illustrate application
as shown in Figure 6(b). The Pareto sets PF1 ; PF2 ;
of the proposed methodology, we do not empha-
. . . ; PFF were defined after the construction and
size specifics of the underlying physical problem.
investigation of the above parallelepipeds,
The problem has eight criteria that are defined
respectively, Figure 6(a). In our case, despite the
implicitly. Two of the eight are performance
corrections of the design variable and other
criteria: Resistance Performance defined using
constraints, the feasible solution set remains
Fung’s Resistance Prediction Algorithm (Bare
empty.4 Thus, the Pareto set does not contain
Hull Residuary Resistance Coefficient) and the
any feasible solutions. Therefore, the desired
Seakeeping Performance defined by Bales Formula
values of local criteria FW n , n ¼ 1; . . . ; k are (Bales Seakeeping Rank). The other six criteria are
unattainable.
pseudo-criteria. Since we have a readily available
We offer two suggestions in this situation: First,
mathematical model, we need only to conduct the
an expert can make concessions and accept a
second stage of the operational development as
compromise solution FC , taking into account
described in the ‘Introduction.’
the most important criteria, Figure 6(a).
The model is based on 14 design variables:
(Pa1 ; Pa2 ; . . . ; PaF , aC are inverse images of
PF1 ; PF2 ; . . . ; PFF , FC , respectively, in the design  a1 : length of design waterline (assumed to be
equal to length between perpendiculars), m
4
Earlier we considered cases in which feasible set D was  a2 : beam of design waterline (assumed to be
nonempty and the Pareto set P  D. equal to beam amidships), m

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
54 R. STATNIKOV ET AL.

 a3 : draft (assumed to be equal to draft amid-  f 5: waterplane area coefficient forward of

ships), m amidships
 a4 : distance from the station 0 (FP) to the cut-  f 6: waterplane area coefficient aft of amidships
up point, m  f 7: draft-to-length ratio
 a5 : waterplane area forward of amidships, m2  f 8: cut-up ratio
 a6 : waterplane area aft of amidships, m2  f 9: vertical prismatic coefficient forward of
 a7 : displaced volume forward of amidships, m3 amidships
 a8 : displaced volume aft of amidships, m3  f 10: vertical prismatic coefficient aft of amid-
 a9 : prismatic coefficient ships
 a10 : projected transom area, m2
 a11 : projected transom width, m The following rigid constraints are imposed on
 a12 : projected transom depth, m the above four functional relations:
 a13 : half entrance angle, degrees
 a14 : longitudinal centre of buoyancy from the
fore perpendicular, m  f 1p2100
 f 2p0.51
Thus, the design variable vector is  f 3X0.77
a ¼ ða1 ; . . . ; a14 Þ. Values of the prototype design  f 4p0.84
variables ap and initial design variable constraints  f 4X0.80
(i.e. parallelepiped
1 ) are provided in Table I.
We want to maximize the performance criterion The others functional constraints f 5; . . . ; f 10 are
F1 (Seakeeping Rank,) and to minimize the nonrigid, i.e. they can change in some limits,
performance criterion F2 (Residuary Resistance however it is difficult to formulate these constraints
Coefficient). The vector of performance criteria of a priori. According to the PSI method the nonrigid
prototype is Fp ¼ ð8:608; 1:968Þ, see Table II. functional relations should be interpreted as the
There are 10 functional relations: pseudo-criteria i.e. F3 ¼ f 5, F4 ¼ f 6, F5 ¼ f 7,
F6 ¼ f 8, F7 ¼ f 9, and F8 ¼ f 10, see Section 2.3.
 f 1: displacement, metric ton In this case, the constraints are defined during
 f 2: block coefficient solution of the problem (on the basis of the analysis
 f 3: maximum section area coefficient (assumed of the test table). Values F3 ; F4 ; F5 ; F6 ; F7 ; F8 for
to be equal to midship section coefficient Cm) the prototype are shown in Table II. Therefore, to
 f 4: waterplane area coefficient determine the feasible solution set we solve the

Table I. Table of design variables. Parallelepiped
1 .

Design variables Prototype ap
1 Pareto optimal solutions in initial statement

Lower bound Upper bound a26087 a75527 a81087

a1 90.700 85.700 95.700 94.749 94.485 95.592

a2 12.670 10.670 14.670 11.659 11.860 11.656
a3 3.700 3.500 3.900 3.581 3.578 3.679
a4 51.650 48.650 54.650 52.081 50.611 52.584
a5 380.10 330.100 430.100 416.938 427.530 404.837
a6 552.900 502.900 602.900 506.413 505.068 514.845
a7 991.200 891.200 1091.200 930.403 952.688 940.701
a8 1040.100 940.000 1140.000 1041.730 1066.618 1096.428
a9 0.626 0.620 0.635 0.626 0.624 0.625
a10 11.740 9.740 13.740 11.758 10.099 10.416
a11 12.120 10.120 14.120 10.555 10.982 10.622
a12 0.950 0.750 1.050 0.880 1.025 0.754
a13 13.000 12.000 14.000 12.011 13.986 12.607
a14 45.900 43.900 47.900 47.810 47.604 44.024

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 55

Table II. Table of criteria

Criteria Prototype Fp Pareto optimal solutions in initial statement Pareto optimal solutions in second statement

]26087 ]75527 ]81087 ] 74223 ]49109 ]106467

F1(max) 8.608 14.342 14.267 12.614 14.537 15.4598 15.4591

F2(min) 1.968 1.847 1.803 1.711 1.694 1.873 1.852
F3(pseudo) 0.662 0.755 0.763 0.727 0.753 0.767 0.767
F4(pseudo) 0.962 0.917 0.901 0.924 0.905 0.911 0.911
F5(pseudo) 0.041 0.038 0.038 0.038 0.038 0.039 0.038
F6(pseudo) 0.569 0.550 0.536 0.550 0.554 0.565 0.554
F7(pseudo) 0.705 0.623 0.623 0.632 0.605 0.596 0.603
F8(pseudo) 0.508 0.574 0.590 0.579 0.578 0.560 0.560
Criteria Pareto optimal solutions in final statement

]113487 ]4145 ]68410 ]39801 ]53988 ]72461 ]75110

F1(max) 15.308 15.183 15.131 15.052 14.887 14.805 14.601

F2(min) 1.677 1.674 1.657 1.645 1.644 1.629 1.622
F3(pseudo) 0.760 0.755 0.760 0.753 0.758 0.752 0.752
F4(pseudo) 0.918 0.923 0.916 0.926 0.921 0.926 0.928
F5(pseudo) 0.038 0.038 0.038 0.038 0.038 0.038 0.038
F6(pseudo) 0.538 0.547 0.558 0.534 0.551 0.538 0.554
F7(pseudo) 0.598 0.601 0.602 0.597 0.607 0.606 0.615
F8(pseudo) 0.561 0.553 0.561 0.557 0.559 0.563 0.568

problem with criteria vector F ¼ ðF1 ; F2 ; F3 ; analysing test tables. As a result, ND 5 240 vectors
F4 ; F5 ; F6 ; F7 ; F8 Þ. The vector of functional (including the prototype) entered the feasible set,
relations is (f 1, f 2, f 3, f 4). and NP 5 3 vectors are the Pareto optimal solution
In summary, we want to investigate the set (The remaining 1247 vectors did not satisfy the
problem with a 14-dimensional design variable criteria constraints). The coefficient of the effi-
space and an 8-dimensional criteria space, keeping ciency of searching the feasible solution set is gF ¼
in mind complex constraints, which we need to 240=131 072 ¼ 0:0018 (for Pareto optimal set it is
correct in order to construct feasible solutions. gP ¼ 0:00002). Very low values of coefficient gF
The analysis process summarized in the next point out the difficulties in searching the feasible
section is based on application of the PSI method solutions.
with LPt sequences. These vectors in the criteria space and design
variable space are denoted as F26087 ; F81087 ; F75527
and a26087 ; a81087 ; a75527 respectively. To simplify
6. MULTICRITERIA ANALYSIS OF THE the notation, we will denote vectors of criteria
SHIP DESIGN PROTOTYPE IMPROVEMENT corresponding to the ith test (i.e. Fi ) simply as ] i.
PROBLEM Therefore, F26087 ; F81087 ; F75527 are written as
] 26087, ] 81087, and ] 75527 (see Table II). The
6.1. Solution of the initial statement of the values of design variables and criteria of Pareto
optimization problem optimal solutions are given in Tables I and II.
We performed N 5 131 072 tests in parallelepiped Below we briefly show some visualization tools

1 and only N1 5 1487 vectors entered the test that led to improvement of the present results.
table. The 129 585 solutions did not satisfy the Graphs. ‘criterion vs. criterion.’ Figure 7 shows
functional constraints. Since we are solving dependency between the first criterion (Seakeeping
the problem of improving the prototype, the Rank) and the second criterion (Residuary
performance criteria constraints equal to the Resistance Coefficient).
values of the prototype (Fp1 ¼ F


1 ¼ 8:608 and Example of dependencies between the criterion
Fp2 ¼ F


2 ¼ 1:968) were accepted. The constraints F1 and the pseudo-criterion F3 is shown in
on pseudo-criteria were defined on the basis of Figure 8.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
56 R. STATNIKOV ET AL.

Graph. ‘criterion vs. design variable.’ Figure 9 variable (Figure 10a), it follows that the value of
shows the dependency of the criterion F2 vs. the the upper boundary can be decreased to a

10 ¼
design variable a10 . 13:0 (recall that in the initial parallelepiped
1
Histograms of feasible solutions. The we had a

10 ¼ 13:74).
distributions of feasible solutions for the range of  The other boundaries of the parallelepiped
2
the 2nd and 10th design variables in the initial were determined similarly. In the process of the
parallelepiped
1 are shown in Figure 10(a). We definition of boundaries of parallelepiped
2 ,
can see that there are large ‘gaps’ that do not the values of the prototype design variables and
contain any feasible solutions. Similar Pareto optimal solutions ] 26087, ] 75527 and
distributions with ‘gaps’ are observed for the 5th, ] 81087 were also considered.
6th, and 9th design variables.
Again, N 5 131 072 tests were conducted.
N1 5 18 270 vectors entered the test table. The
6.2. Second statement and solution of the optimiza- pseudo-criteria constraints were defined based on
tion problem. Parallelepiped
2 the analysis of the test tables, while constraints on
The analysis of the feasible solutions has pointed performance criteria were kept unchanged. This
ways for correction of the initial problem state- time ND 5 17 302 vectors (including the
ment. The reasoning for construction of the prototype) entered the feasible solution set, and
parallelepiped
2 is given below: NP 5 14 Pareto optimal solutions were identified
(] 87511, ] 21855, ] 63919, ] 49109, ] 106467,
 From the histogram of the second design ] 90819, ] 78907, ] 116871, ] 31819, ] 64407,
variable (Figure 10a), it follows that the value ] 74223, ] 80159, ] 105823, and ] 22671). It
of the upper boundary can be decreased to a

2 follows that the coefficient of the efficiency of
¼ 13:5 (recall that in the initial parallelepiped searching the feasible solutions (gF ) was increased

1 we had a

2 ¼ 14:67). more than 70 times. Furthermore, the histograms
 From the graph ‘Criterion vs. Design Variable’ in
2 have much better distributions of the feasible
(Figure 9) and the histogram of the 10th design solutions than in
1 (see Figure 10).

2.4

2.3

2.2

2.1
Criterion 2

2.0

1.9
#26087

1.8
#75527
#81087
1.7
2 3 4 5 6 7 8 9 10 11 12 13 14 15
Criterion 1

Infeasible Solutions Feasible Solutions Pareto Optimal Solutions Prototype

Figure 7. The dependency between criteria F1 and F2. Initial statement: feasible solutions ND 5 238; Pareto optimal
solutions NP 5 3.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 57

0.78
#75527
0.76
#26087
0.74

0.72 #81087
0.70
Criterion 3

0.68

0.66

0.64

0.62

0.60

0.58

0.56
2 3 4 5 6 7 8 9 10 11 12 13 14 15
Criterion 1

Infeasible Solutions Feasible Solutions Pareto Optimal Solutions Prototype

Figure 8. The dependency between criterion F1 and pseudo-criterion F3.

2.4

2.3

2.2

2.1
Criterion 2

2.0

1.9

#26087
1.8
#75527

#81087
1.7
9.70 10.20 10.70 11.20 11.70 12.20 12.70 13.20 13.70
Design Variable 10

Infeasible Solutions Feasible Solutions Pareto Optimal Solutions Prototype

Figure 9. The dependency of criterion F2 on design variable a10 .

After combining the feasible sets in both problem has entered into the combined Pareto set. In other
statements, we have constructed a combined Pareto words, all the results of optimization in the initial
set. No solution belonging to the initial statement statement have been improved. The preference of the

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
58 R. STATNIKOV ET AL.

criteria constraints now were determined on the

basis of analysis of the test tables. In particular,
more rigid criteria constraints (F

1 ¼ 14:342 and
F


2 ¼ 1:711) were formulated. The pseudo-criteria
constraints were also revised. As a results, we
obtained ND 5 847 feasible solutions and NP 5 7
Pareto optimal solutions.
The smaller number of the feasible solutions
compared to the previous problem statement can
be explained by much stronger performance
criteria constraints. The Pareto optimal solutions
(] 113487, ] 4145, ] 68410, ] 39801, ] 53988,
] 72461, ] 75110) are shown in Figure 11. The
values of criteria of these solutions are given in
(a)
Table II.
The analysis of Pareto optimal solutions
revealed that solution ] 75110 surpassed seven
solutions from the second statement (] 80159,
] 87511, ] 78907, ] 31819, ] 64407, ] 63919,
] 74223) by two criteria simultaneously. Solution
] 113487 surpasses five solutions from the second
statement (] 22671, ] 90819, ] 105823, ] 116871,
] 21855) by two criteria simultaneously. The
expert’s preference was given to solution ] 113487.
Using the feasible solutions from the second
and final statement, the combined Pareto set was
constructed. The combined Pareto set includes all
seven Pareto optimal solutions from the final
statement and only two solutions ] 106467,
(b) ] 49109 belonging to the second statement, see
Figure 10. Histograms of the distribution of feasible Figure 12. Further attempts to improve the
and Pareto optimal solutions for the 2nd and 10th obtained solutions have not yielded any new
design variables: (a) corresponds to initial parallelepiped interesting results.
P1 and (b) corresponds to parallelepiped P2. The The stability of the most interesting Pareto
percentage of designs entering the corresponding inter- optimal solutions was investigated with respect to
val is indicated on the right of each histogram. The
small variations of the parameters in the vicinity of
prototype is marked with a green diamond. The Pareto
optimal vectors are marked with red triangles. The these solutions. To this end, we constructed
‘gaps’ of the initial range for the 2nd and 10th design parallelepipeds centred in the Pareto optimal
variables are circled. solutions and performed 1024 tests in each
parallelepiped. The corresponding variations in
the criteria were small and insignificant, which
expert was given to the vector ] 74223. This vector indicated the stability of the solutions.
surpassed all solutions in the initial statement by two The overall dynamics of improving a prototype
performance criteria simultaneously, see Table II on the basis of two corrections of the problem
and Figure 12. statement is shown in Figure 12.
In summary, the problem of improving a
6.3. Final (third) statement and solution of the preliminary ship design prototype has been
optimization problem. Parallelepiped
3 solved. This process included changing the range
By analysing the feasible solutions from
2 , a new of variation of design variables and revising all
parallelepiped
3 was constructed similarly as in criteria constraints. As a result of multicriteria
Section 6.2. After N 5 131 072 tests, N1 5 34 986 analysis, the parallelepiped
2 was constructed.
vectors entered the test table. Compared to the The volume of parallelepiped
2 was considerably
initial and second statements of the problem, all decreased in comparison with the volume of

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 59

1.97
1.95
1.93
1.91
1.89
1.87
1.85
1.83
1.81
Criterion 2

1.79
1.77
1.75
1.73
1.71
1.69 #113487
1.67
#4145
1.65 #68410
#39801
1.63 #53988
1.61 #72461
#75110
1.59
8.5 9.5 10.5 11.5 12.5 13.5 14.5 15.5
Criterion 1

Infeasible Solutions Feasible Solutions Pareto Optimal Solutions Prototype

Figure 11. Dependency between criteria F1 and F2. Final statement: feasible solutions ND 5 847; Pareto optimal
solutions NP 5 7.

parallelepiped
1 . Similarly, the parallelepiped
3

was constructed with a smaller volume than of
parallelepiped
2 . This allowed a more careful and
efficient investigation of the design variable and
criteria spaces.

7. CONCLUSION

Construction of the feasible solution set is of

fundamental importance in multicriteria real-life
problems, especially when improving a prototype.
The primary contribution of the present work is
that on the basis of PSI method we provided
various visualization and analysis techniques that
facilitate construction of the feasible solution set
and help solve the problem of improving a
prototype. Furthermore, we described some basic
features of this problem, provided a geometrical
interpretation, and showed the necessity of con-
structing the combined Pareto set. Finally, we
presented a case study where we aimed at
improving a preliminary ship design prototype.

ACKNOWLEDGEMENTS

Figure 12. Pareto optimal solutions in the three state- The authors would like to thank Ralph Steuer for
ments and the combined Pareto optimal solutions. his interest in our work. We are grateful to Dan

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
60 R. STATNIKOV ET AL.

Boger, Terry Smith, Karl Pfeiffer, Fotis Papoulias Meisel WS. 1973. Tradeoff decision in multiple crite-
(Naval Postgraduate School), Vladimir Astashev ria decision making. In Multiple Criteria
Konstantin Frolov (Russian Academy of Sciences), Decision Making, Cochrane JL, Zeleny M (eds).
Alexander Zhilyaev (Russian Academy of University of South Carolina Press: Columbia,
Sciences), Kadir Sarioz (Istanbul Technical Uni- USA; 461–476.
Niederreiter H. 1990. Statistical independence properties
versity), and Noyan Kilinc (Turkish Navy) for their of pseudorandom vectors produced by matrix gen-
help and contribution. erators. Journal of Computational and Applied Mathe-
matics 31: 139–151.
Ozernoy VM. 1988. Multiple criteria decision making in
the USSR: a survey. Naval Research Logistics 35:
REFERENCES 543–566.
Sobol’ IM. 2001. Global sensitivity indices for nonlinear
Anil KA. 2005. Multi-criteria analysis in naval ship mathematical models and their Monte Carlo esti-
design. Master’s Thesis, Naval Postgraduate School, mates. Mathematics and Computers in Simulation
Monterey, California. 55(1–3): 271–280.
Bales NK. 1980. Optimizing the seakeeping performance Sobol’ IM, Statnikov RB. 2006. Selecting Optimal
of destroyer-type hulls. Proceedings of the 13th ONR Parameters in Multicriteria Problems (2nd edn).
Symposium, Tokyo, Japan. Drofa: Moscow (in Russian).
Barrans SM, Bhat N. 2003. Optimization of air bearings Stadler W, Dauer JP. 1992. Multicriteria optimization in
with the aid of finite element analysis. Proceedings of engineering: a tutorial and survey. In Structural
NAFEMS World Congress, Orlando, USA. Optimization: Status and Promise. Progress in Aero-
Cleveland WS. 1985. The Elements of Graphing Data. nautics and Astronautics, Vol. 150, Kamat MP (ed.).
Wadsworth: Belmont, CA. American Institute of Aeronautics and Astronautics:
Dobrokhodov V, Statnikov R, Statnikov A, Yanushke- Washington, DC; 209–249.
vich IV. 2003. Modeling and simulation framework Statnikov RB. 1999. Multicriteria Design. Optimization
for multiple objective identification of a controllable and Identification. Kluwer Academic Publishers:
descending system. Proceedings of the International Dordrecht, Boston, London.
Conference on Adaptive Modeling and Simulation Statnikov RB, Bordetsky A, Statnikov A. 2005a.
(ADMOS-2003), Geteborg, Sweden. Multicriteria analysis of real-life engineering optimi-
Faure H. 1982. Discrépance ade suı́tes associées a une zation problems: statement and solution. Nonlinear
systéme de numération (en dimension s). Acta Analysis 63: e 685–e 696.
Arithmetica 41: 337–351. Statnikov RB, Bordetsky A, Statnikov A. 2006. Multi-
Fung SC. 1991. Resistance and powering prediction for criteria analysis tools in real-life problems. Journal of
transom stern hull forms during early stage ship Computers and Mathematics with Applications
design. SNAME Transactions, Vol. 99, Department of 52(1–2): 1–32.
the Navy, NAVSEA, Washington, DC; 29–84. Statnikov RB, Matusov JB. 1995. Multicriteria
Gobbi M, Mastinu G, Catelani D, Guglielmetto L, Optimization and Engineering. Chapman & Hall:
Bocchi M. 2000. Multi-objective optional design of New York.
road vehicle sub-systems by means of global approx- Statnikov RB, Matusov JB. 1996. Use of Pt nets for the
imation. Proceedings of the 15th European ADAMS approximation of the Edgeworth– Pareto set in multi-
Users’ Conference, Rome, Italy. criteria optimization. Journal of Optimization Theory
Halton JH. 1960. On the efficiency of certain quasi- and Applications 91(3): 543–560.
random sequences of points in evaluating multi- Statnikov RB, Matusov JB. 2002. Multicriteria Analysis
dimensional integrals. Numerische Mathematik 2: in Engineering. Kluwer Academic Publishers: Dor-
84–90. drecht, Boston, London.
Hammersley JM. 1960. Monte Carlo methods for Statnikov RB, Statnikov A, Yanushkevich IV. 2005b.
solving multivariable problems. Annals of the New Construction and analysis of the feasible set in
York Academy of Sciences 86: 844–874. engineering optimization problem. Problemy Mashi-
Hlawka E, Taschner R. 1991. Geometric and Analytic nostroenia i Nadezhnosti Mashin (Journal of Machin-
Number Theory. Springer: Berlin. ery Manufacture and Reliability, Russian Academy of
Kuipers L, Niederreiter H. 1974. Uniform Distribution of Sciences) 4: 3–12 (in Russian).
Sequences. Wiley: New York, USA. Steuer RE, Sun M. 1995. The parameter space
Lichtenstein S, Slovic P. 2006. The Construction of investigation method of multiple objective nonlinear
Preference. Cambridge University Press: New York, programming: a computational investigation. Opera-
USA. tions Research 43(4): 641–648.
Lotov ÀV, Bushenkov VA, Kamenev GK. 2004. Wu J, Azarm S. 2001. Metrics for quality assessment of
Interactive Decision Maps. Kluwer: Boston, USA. a multiobjective design optimization solution set.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda
 PROTOTYPE IMPROVEMENT PROBLEM 61

Journal of Mechanical Design (Transactions of the Yanushkevich IV, Statnikov RB, Statnikov A, Matusov
ASME) 123(1): 18–25. JB. 2004. Software Package MOVI 1.3 for Windows:
Yakimenko OA, Statnikov RB. 2005. Multicriteria User’s Manual.
parametric identification of the parafoil-load delivery Zitzler E, Thiele L, Laumanns M, Foneseca CM, Grunert
system. Proceedings of the 18th AIAA Aerodynamic da Fonseca V. 2003. Performance assessment of multi-
Decelerator Systems Technology Conference and objective optimizers: an analysis and review. IEEE
Seminary (AIAA 2005), Germany. Transactions on Evolutionary Computation 7(2): 117–132.

Copyright r 2008 John Wiley & Sons, Ltd. J. Multi-Crit. Decis. Anal. 15: 45–61 (2008)
DOI: 10.1002/mcda