Computer Simulations Advances Research Applications 8

COMPUTER SCIENCE, TECHNOLOGY AND APPLICATIONS
COMPUTER SIMULATIONS
ADVANCES IN RESEARCH
AND APPLICATIONS
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
COMPUTER SCIENCE, TECHNOLOGY
AND APPLICATIONS
Additional books in this series can be found on Nova’s website

under the Series tab.
Additional e-books in this series can be found on Nova’s website

under the eBooks tab.
COMPUTER SCIENCE, TECHNOLOGY AND APPLICATIONS
COMPUTER SIMULATIONS
ADVANCES IN RESEARCH
AND APPLICATIONS
MICHAEL D. PFEFFER
AND
ERIK BACHMEIER
EDITORS
Copyright © 2018 by Nova Science Publishers, Inc.
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or transmitted
in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical photocopying,
recording or otherwise without the written permission of the Publisher.
We have partnered with Copyright Clearance Center to make it easy for you to obtain permissions to
reuse content from this publication. Simply navigate to this publication’s page on Nova’s website and
locate the “Get Permission” button below the title description. This button is linked directly to the
title’s permission page on copyright.com. Alternatively, you can visit copyright.com and search by
title, ISBN, or ISSN.
For further questions about using the service on copyright.com, please contact:
Copyright Clearance Center
Phone: +1-(978) 750-8400 Fax: +1-(978) 750-4470 E-mail: info@copyright.com.
NOTICE TO THE READER

The Publisher has taken reasonable care in the preparation of this book, but makes no expressed or
implied warranty of any kind and assumes no responsibility for any errors or omissions. No liability is
assumed for incidental or consequential damages in connection with or arising out of information
contained in this book. The Publisher shall not be liable for any special, consequential, or exemplary
damages resulting, in whole or in part, from the readers’ use of, or reliance upon, this material. Any
parts of this book based on government reports are so indicated and copyright is claimed for those parts
to the extent applicable to compilations of such works.
Independent verification should be sought for any data, advice or recommendations contained in this
book. In addition, no responsibility is assumed by the publisher for any injury and/or damage to
persons or property arising from any methods, products, instructions, ideas or otherwise contained in
this publication.
This publication is designed to provide accurate and authoritative information with regard to the subject
matter covered herein. It is sold with the clear understanding that the Publisher is not engaged in
rendering legal or any other professional services. If legal or any other expert assistance is required, the
services of a competent person should be sought. FROM A DECLARATION OF PARTICIPANTS
JOINTLY ADOPTED BY A COMMITTEE OF THE AMERICAN BAR ASSOCIATION AND A
COMMITTEE OF PUBLISHERS.
Additional color graphics may be available in the e-book version of this book.
Library of Congress Cataloging-in-Publication Data
ISBN: HERRN
Published by Nova Science Publishers, Inc. † New York

CONTENTS
Preface vii
Chapter 1 Computer Simulations of Passenger Car Radial
Tires Using the Finite Element Method 1
Mir Hamid Reza Ghoreishy
Chapter 2 Computer Simulations as a Solution Method
for Differential Games 63
Guennady A. Ougolnitsky and Anatoly B. Usov
Chapter 3 The Computer Simulation of the Talbot Effect and
Carpet via the Iterative Fresnel Integrals Method 107
Kazi Monowar Abedin, Aamna Al-Saedi
and S. M. Mujibur Rahman
Chapter 4 Modeling of the Light Scattering by a Yarn 129
Zeguang Pei
Chapter 5 The Unsteady Analysis of a Solar Chimney
Power Plant 149
Ahmed Ayadi, Zied Driss, Haythem Nasraoui,
Abdallah Bouabidi and Mohamed Salah Abid
vi Contents
Chapter 6 Computer Simulation of Hydrogen

Thermal Desorption Spectra and
Model Parameters Identification 167
Yury V. Zaika and Ekaterina K. Kostikova
Chapter 7 Fast Hydrogen Permeability of Structural
Materials: Modelling and Parameters Estimation 195
Yury V. Zaika and Natalia I. Rodchenkova
Index 215
PREFACE
Computer Simulations: Advances in Research and Applications begins

with a concise overview and background of the topic during the past
decade. The finite element method and the commercial available codes are
introduced with an emphasis on Abaqus as the mostly used software in this
field. The next part deals with the tire geometry and creation of a
consistent and computationally effective finite element mesh from 2D and
3D drawings. The material models used for the description of the
mechanical behavior of tire constituents are also examined. The following
paper is dedicated to the computer simulation modeling as a method of
solution of differential games where analytical investigation is
problematic. The main idea is that a very small number of scenarios
including both control variables and parameters can provide a sufficiently
good qualitative representation of dynamics of the modeled system. The
Tablot Effect, which was first experimentally observed by Henry Talbot in
1836, is discussed. It is the repeated self-imaging of a diffraction grating at
regular distances in the near-field behind the grating. If the observed
diffraction images are laid out as a function of distance, a beautiful and
repetitive pattern is observed; this is known as the Talbot carpet. The
methodology of how the IFIM method was applied for the simulation of
Talbot effect is described explicitly followed by a systematic synthesis of
the Talbot carpet from the generated data. After this, Beckmann’s
viii Michael D. Pfeffer and Erik Bachmeier
scattering model is adopted to simulate the light scattering from the yarn
surface and to analyze the relation between the scattering pattern and the
surface twist angle. The solution of the scattering intensity distribution for
the yarn surface profile is derived and numerical results indicate that the
highest light intensity on the backward light scattering pattern lies along
the direction perpendicular to the surface fibers regardless of the fiber or
yarn parameters. The result of this chapter can serve as the theoretical basis
for the measurement of yarn surface twist angle based on backward light
scattering by the yarn. The authors go on to present a numerical study of
the unsteady airflow characteristics inside a solar chimney power plant.
Ansys Fluent 17.0 is used to simulate the air flow within the solar setup.
To validate the authors’ study, a solar prototype is built in the National
School of Engineers of Sfax, University of Sfax, Tunisia in the North
Africa. Subsequently, the authors also present a distributed boundary value
problem of thermal desorption with nonlinear dynamical boundary
conditions. The problem is reduced to the nonlinear functional differential
equation of neutral type for surface concentration. A numerical method is
developed for TDS spectrum simulation, where only integration of a
nonlinear system of low order (compared with, e.g., the method of lines)
ordinary differential equations (ODE) is required. Lastly, the book presents
an aggregation procedure for hydrogen permeability experiments
precluding depressurization and (or) change of the samples of investigated
material. Additionally, the corresponding mathematical software for
correct processing of measurements along with appropriate assembly of the
experimental unit is described.
Chapter 1 - This monograph is intended to briefly describe the methods
and techniques developed so far for the computer simulation of steel-belted
radial tires under different mechanical and thermal loads. The chapter
begins with a concise overview and background of the topic during the past
decade. Then, the finite element method and the commercial available
codes are introduced with an emphasis on Abaqus as the mostly used
software in this field. The next part deals with the tire geometry and
creation of a consistent and computationally effective finite element mesh
from 2D and 3D drawings. Following that, the material models used for the
Preface ix
description of the mechanical behavior of tire constituents are examined.

These include hyperelastic and linear elastic models for rubber and
reinforcing cords, respectively. The hyper-viscoelastic models and
complex frictional behavior between tire tread surface and road are also
taken into consideration for the analyses in which the non-isothermal and
dissipations effects become prominent. The latter case is mainly associated
with prediction of rolling resistance and fuel economy. Different methods
for tire analysis are then explained. These can be categorized into static
loadings (rim mounting, inflation and footprint), steady state rolling (force
and moment, slip and camber angles, material history for prediction of
rolling resistance and wear) and finally transient loading (rolling over road
disturbance and curb, accelerating and braking). The results of the analyses
of some radial tires under the mentioned loading conditions are presented
and discussed. These results, wherever possible, are also compared with
experimental data to verify the applicability of the developed models and
simulation methods.
Chapter 2 - This paper is dedicated to computer simulations as a
solution method for differential games that are difficult for analytical
treatment. The main idea is that a very small number of scenarios,
including both control variables and parameters, can provide a fairly good
qualitative representation of modeled system dynamics. This idea is
applicable to the modeling of organizational, environmental, and
socioeconomic systems in which qualitatively representative scenarios
have a clear interpretation. The proposed computer simulation method is
illustrated by two examples that model fishing in shallow-water
ecosystems and social partnership in a continuous education system.
Chapter 3 - The Talbot effect was first experimentally observed by
Henry Talbot in 1836. It is the repeated self-imaging of a diffraction
grating at regular distances in the near-field behind the grating. The
corresponding self-repeating distance is now known as the Talbot distance.
If the observed diffraction images are laid out as a function of the distance,
a beautiful and repetitive pattern is observed; this is known as the Talbot
carpet. Apart from a considerable theoretical interest, the Talbot effect has
found many applications in diverse areas of optics, for example, in
x Michael D. Pfeffer and Erik Bachmeier
imaging, refractive index measurements, displacement sensors, lithography

and array illumination, to name a few. In this Chapter, the authors have
applied the Iterative Fresnel Integrals Method (IFIM) to the simulation of
the Talbot effect, and consequently, to the generation of Talbot carpets.
The methodology of how the IFIM method was applied for the simulation
of the Talbot effect is described explicitly, followed by a systematic
synthesis of the Talbot carpet from the generated data. All the data were
generated without recourse to any experimental apparatus. Finally,
examples of Talbot carpets are presented at two different resolutions, and
suggestions are made as to how the whole process of Talbot carpet
synthesis can be automated to generate carpets of higher resolutions.
Chapter 4 - Yarn twist is an important parameter which determines the
characteristics and physical properties of a spun yarn. The yarn twist T can
tan 
be calculated by the formula: T  , where β is the surface twist angle
D
formed between the surface fibers and the yarn axis, and D is the yarn
diameter. Therefore, β and D need to be measured to obtain the twist.
Difficulty usually lies in measuring the surface twist angle due to the tiny
diameters of the surface fibers. In this chapter, the feasibility of
determining the surface twist angle based on the light scattering by the
yarn is investigated as a foundation for the online measurement of yarn
twist which is at present of great challenge to the industry. Beckmann’s
surface twist angle. The solution of the scattering intensity distribution for
the yarn surface profile is derived and numerical results indicate that the
highest light intensity on the backward light scattering pattern lies along
the direction perpendicular to the surface fibers regardless of the fiber or
yarn parameters. The result of this chapter can serve as the theoretical basis
for the measurement of yarn surface twist angle based on backward light
scattering by the yarn.
Chapter 5 - The solar chimney power plants (SCPP) could generate
electricity using solar as source. The study of the air flow characteristics is
required since the construction cost of these devices is expensive. This
Preface xi
chapter focus on numerical studies of the unsteady airflow characteristics

within a solar chimney power plant. Ansys Fluent 17.0 is used to simulate
the air flow inside the solar setup. A prototype is built in the National
School of Engineers of Sfax, University of Sfax, Tunisia in the North
Africa to validate the authors’ numerical study. The prototype is composed
by a collector, an absorber, a chimney and a turbine. The collector is
characterized by a diameter equal to 2750mm, a roof height equal to
50mm. However, the chimney is characterized by a height equal to
3000mm. The turbine diameter is equal to 150mm. The air characteristics
is presented and analyzed. In fact, the distribution of the air temperature,
the air velocity and the total pressure. Findings confirm that the ambient air
temperature and the solar radiation are important parameters for the
enhancement of the solar chimney efficiency.
Chapter 6 - One of the technological challenges for hydrogen materials
science is the currently active search for structural materials with important
applications (including nuclear power industry and the ITER project) that
will have predetermined limits of hydrogen permeability. One of the
experimental methods is thermal desorption spectrometry (TDS). A
hydrogen saturated sample is degassed under vacuum and monotone
heating. The desorption flux is measured by mass spectrometer to
determine the character of interactions of hydrogen isotopes with the solid.
The chapter presents a distributed boundary value problem of thermal
desorption with nonlinear dynamical boundary conditions. The problem is
reduced to the nonlinear functional differential equation of neutral type for
surface concentration. A numerical method is developed for low order
(compared with, e.g., the method of lines) ordinary differential equations
(ODE) is required. It is usually assumed that TDS peaks appear due to
hydrogen release from traps with different binding energies. Here, it was
demonstrated using the diffusion model (for homogeneous material) that if
surface processes are taken into account, two-peak spectra can be obtained
even for very thin experimental samples. The main final output of the
chapter is a geometrically transparent method for solving the inverse
problem of surface parameter identification where desorption and diffusion
in the bulk are dynamically interrelated (without the artificial division of
xii Michael D. Pfeffer and Erik Bachmeier
studies into the diffusion limited regime (DLR) and the surface limited
regime (SLR)).
Chapter 7 - High-purity hydrogen is required for clean energy and a
variety of chemical technology processes. The various alloys potentially
well-suited for use in gas-separation plants for the membrane technology
of high- purity hydrogen production were investigated by measuring
specific hydrogen permeability. For structural materials one had to
estimate the parameters of diffusion and sorption to numerically model the
different scenarios and experimental conditions of the material usage
(including extreme ones), and identify the limiting factors. The proposed
mathematical model matches the experimental method of hydrogen
permeability and takes into account only the main limiting factors for the
applied membrane filtering problem and the informative capabilities of the
experiment modifications. The penetration method allows determining the
diffusion coefficient by so-called lag time. The accuracy of the estimation
depends on the degree of proximity to the DLR (diffusion limited regime)
mode. The method of ‘communicating vessels’ is more sensitive to surface
processes. ‘Separate’ application of these methods leads to a situation
where the materials studied are in fact somewhat different (for example,
due to different impacts on the surface), and there appear significant
differences in hydrogen permeability parameter estimates. The main idea
of this chapter is to develop an aggregation procedure for hydrogen
permeability experiments precluding depressurization and (or) change of
the samples of the investigated material. The authors also present the
corresponding mathematical software for correct processing of
measurements along with appropriate assembly of the experimental unit.
This chapter describes the proposed cascade experiment technique and the
mathematical software: model, computer simulation and parametric
identification algorithms. This allows increasing the informative capacity
of the experimental studies and the accuracy of the estimation of hydrogen
permeability parameters (diffusion, absorption, desorption).
In: Computer Simulations ISBN: 978-1-53613-095-9
Editors: Michael D. Pfeffer et al. © 2018 Nova Science Publishers, Inc.
Chapter 1
COMPUTER SIMULATIONS OF PASSENGER

CAR RADIAL TIRES USING THE FINITE
ELEMENT METHOD
Mir Hamid Reza Ghoreishy

Iran Polymer and Petrochemical Institute, Tehran, Iran
ABSTRACT
This monograph is intended to briefly describe the methods and

techniques developed so far for the computer simulation of steel-belted
radial tires under different mechanical and thermal loads. The chapter
begins with a concise overview and background of the topic during the
past decade. Then, the finite element method and the commercial
available codes are introduced with an emphasis on Abaqus as the mostly
used software in this field. The next part deals with the tire geometry and
creation of a consistent and computationally effective finite element mesh
from 2D and 3D drawings. Following that, the material models used for
the description of the mechanical behavior of tire constituents are
examined. These include hyperelastic and linear elastic models for rubber
and reinforcing cords, respectively. The hyper-viscoelastic models and

Corresponding Author Email: M.H.R.Ghoreishy@ippi.ac.ir.
2 Mir Hamid Reza Ghoreishy
complex frictional behavior between tire tread surface and road are also
taken into consideration for the analyses in which the non-isothermal and
dissipations effects become prominent. The latter case is mainly
associated with prediction of rolling resistance and fuel economy.
Different methods for tire analysis are then explained. These can be
categorized into static loadings (rim mounting, inflation and footprint),
steady state rolling (force and moment, slip and camber angles, material
history for prediction of rolling resistance and wear) and finally transient
loading (rolling over road disturbance and curb, accelerating and
braking). The results of the analyses of some radial tires under the
mentioned loading conditions are presented and discussed. These results,
wherever possible, are also compared with experimental data to verify the
applicability of the developed models and simulation methods.
Keywords: radial tire, simulation, finite element method, steady state

rolling, hyperelastic
1. INTRODUCTION
It is generally known that the motions of a ground vehicle such as a

passenger car are highly affected by the forces and moments generated and
applied via its tires. Therefore, in order to properly evaluate the
performance of a car, a deep understanding of the tire characteristics is
essential. Tires are normally designed to be either pneumatic or solid.
Except for very specific vehicles and applications, they are commonly
manufactured to be pneumatic. A pneumatic tire is a complex and semi-
flexible engineered product made of fiber reinforced composite of rubber,
organic (such as rayon, nylon, polyester) and inorganic (mainly steel)
cords. It is fully mounted and fixed on a metallic rim, filled by compressed
air and installed to the car suspension system. Based on the arrangement of
the cords in the body plies, there are three main types of the tire
categorized into bias, bias-belted and radial tires. Owing to its superior
performance and operational properties, the use of radial tires has become
widespread so that the market share for this type reaches nearly 100% in
fully developed countries. Figure 1 shows a commercial radial tire
mounted on a rim. Moreover, a typical construction of a radial tire is
Computer Simulations of Passenger Car Radial Tires … 3
shown in Figure 2. Design of a modern pneumatic tire is mainly based on

the achievement of the following requirements:
1. Lowering the rolling resistance which is directly related to the fuel

consumption and trip economy.
2. Optimized vertical, longitudinal and lateral stiffness in both static
and rolling conditions to provide a cushion and smooth ride and
also high handling performance in conjunction with a good load
carrying and shock absorbing properties.
3. High sliding friction in both wet and dry conditions to guarantee a
safe driving.
4. Enhancing the resistance to wear and damage such as cutting,
puncturing, tearing, abrasion as well as deterioration caused by
ozone, heat oxidation and the other environmental factors.
Figure 1. A typical radial tire mounted on the rim.
Most of the above mentioned requirements are contradictory. For

example, lowering the rolling resistance using conventional compounds
also decreases the wet grip which is obviously an undesirable effect.
Therefore, tire designers must choose appropriate combinations of
materials, geometry and construction. Design of a modern pneumatic tire

that meets the mentioned requirements is not an easy task. It needs highly
sophisticated analytical tools such as advanced mathematical models and
laboratory methods to be employed. One of the best tools that can help a
tire designer to achieve an optimum design is the stress analysis of the tire
and predicts its real behavior via a virtual model. This is usually carried out
by the use of a mathematical model formulated into a computational
technique e.g., finite element method which is a very famous numerical
technique with much commercial and open access available software.
Figure 2. A 3D cross-sectional view of a typical radial tire with terminology [1].
Computer simulation of passenger car radial tires by the finite element

method plays an important role in addressing the issue of tire design. This
is because that it provides a valuable insight into the real behavior of tires
with a minimum need of carrying out highly expensive and time
consuming experiments. There are many published textbooks and sources
devoted to its theoretical bases. It is not, therefore, the aim of this chapter
to cover up the theory and mathematical formulation of the finite element
method. The reader can consult the appropriate references (see for example
[2] for a detailed formulation of the nonlinear finite element in solid

mechanics used for tire analysis).
This chapter is divided into six sections. After introduction, the second
section deals with a brief background of the application of finite element
method for tire analysis. The third section presents the details of
procedures for the development of finite element models of a typical radial
tire under different loads and service conditions. The key modeling steps
including generation of finite element mesh from initial tire drawings,
carrying out experiments for the determination of the parameters of the
selected material models and also defining of appropriate loads are
highlighted in this part. The application of these models and the obtained
results are considered in the fourth part. The fifth section gives the
concluding remarks and finally the references are given in the last part. It is
assumed that the reader should have a basic knowledge of tire construction
and mechanics. However, references [3, 4] are very comprehensive sources
for those who may need further information about tire technology and
engineering.
2. BACKGROUND
Finite element modeling of tires has been an object of intensive

researches since 1970s. There have been a considerable amount of
publications devoted to this topic. Many of these works are accompanied
with both in-door and out-door test data to experimentally verify the
applicability of developed models and techniques and also the accuracy of
the results. Reference [5] gives a thorough review on this subject up to last
decade. Since then, however, a large number of publications have been
emerged on different and new aspects of this topic. One of these topics is
the study of the interaction between tire and ground as a deformable body
such as snow and soil. The latter is an important issue for agricultural
applications. Mohsenimanesh et al. [6] have analyzed an off-road tractor
tire using ANSYS code. A non-linear multi-laminated model was
developed. They have computed the contact pressure distribution as a
function of inflation pressure and static normal load. Increased levels of

tire–road interface stress were found near the center of the tire model when
inflation pressure was increased and near the edge when load was
increased. Although they have assumed a rigid surface for ground, it was
tried to find relevant correlation with data given in other literature for soft
and deformable surfaces. Lee [7] has developed a three-dimensional finite
element model for the simulation of tire–snow interaction for low-strength
snow using Abaqus/Explicit. The pneumatic tire was modeled using
elastic, viscoelastic and hyperelastic material models while the snow was
modeled using the modified Drucker–Prager Cap material model. The
traction, motion resistance, drawbar pull, tire sinkage, tire deflection, snow
density, contact pressure and contact shear stresses were obtained as a
function of longitudinal slip and lateral slip. In a similar work, Xia [8] has
used Abaqus/Explicit code to develop a transient finite element model for
the motion of a typical pneumatic tire on soil. The Drucker–Prager Cap
model implemented in Abaqus is used to model the soil compaction. The
effects of inflation pressure, angular velocity of the tire and coefficient of
friction between tire and terrain on tire performance were numerically
studied. Xia and Yang [9] extended the previous work to simulate the
interaction between a four-tire vehicle and the pavement using similar
finite element model. Recently, Zhao and Zhang [10, 11] have developed
an in-house computer code for the simulation of the motion of a tire on
sand using combined finite element and discrete element (FEM/DEM)
methods. The tire is simulated using the FEM, and the sand road is
modeled through the DEM. The discrete element sand particles contact
each other and also have contact with the finite element of tire during the
simulation process. Another major topic in finite element modeling of tires
is the prediction of the rolling resistance force. This is due to the
motivation associated with reduction of fuel consumption. The model must
take the viscoelastic properties of the rubber and cords into consideration.
Modern finite element codes generally have capabilities of defining both
linear and nonlinear viscoelastic material models for this purpose.
Ghoreishy [12, 13] has developed a steady state finite element model for
the prediction of the rolling resistance in a passenger car radial tire based
on the steady state rolling analysis using hyper-viscoelastic material

models. Comparison of the predicted results showed that there is good
agreement between computed and experimentally measured data. Also, in
these works, a concise review was made on the other published works in
this area. Another major issue in this field that has been received
considerable attention is the development of advanced material models
especially for rubbers. Owing to the hyperelastic mechanical behavior as
well as viscoelastic nature of elastomeric materials, more complex models
are required to increase the accuracy of the simulation results. Moreover,
for a highly filled rubber compound (as normally used in tires) under
quasi-static cyclic loading, a phenomenon occurs which is referred to in the
literature as stress softening or Mullins effect [14-17]. The combination of
these behaviors into a single model is the subject of some recent
researches, especially for the prediction of tire rolling loss. Ghoreishy [18]
has recently investigated the effects of using different hyperelastic model
on the accuracy of the predicted radial, lateral, longitudinal and torsional
stiffness of a statically loaded tire. It was shown that each constitutive
model give rise to a certain degree of accuracy especially for a vertically
loaded tire under shear loads. In another recent comprehensive study,
Rafei, Ghoreishy and Naderi [19] have developed a non-isothermal
thermo-mechanical finite element model using Abaqus. They have used
this model for the simulation of the rolling of a radial tire in a flat-trac test
machine. They have compared the results including force, moment and
temperature profile with experimentally measured data which confirms the
applicability of the model. The study of using different approaches for the
composite modeling techniques was carried out by Ghoreishy et al. [20].
They have developed finite element models for a steel-cord/rubber
composite strip under tension and shear loads. The effects of the different
modeling techniques including the representative volume element (RVE),
equivalent elastic properties and rebar elements as well as cord orientation
angle on the accuracy of the results via comparison with experimental data.
The results showed that the accuracy and convergence of the
computational methods are highly dependent on the selected numerical
approach, the angle between cord and direction of applied load and also the
hyperelastic or hyper-viscoelastic model used to describe the mechanical

behavior of the rubbery part. There are also other subjects in this area that
finite element modeling could play important role in their study. For
example, noise and vibration, new tire regulation and legislations,
development of intelligent tire and tire cure process are part of this topic.
The latter topic is fully studied by the author (see for example [21-23]).
3. FINITE ELEMENT MODEL DEVELOPMENT
Due to the large deformations arise in tire and also nonlinear behavior
of rubberlike materials as well as the flexible contact area between tire and
road, the use of nonlinear finite element modeling is inevitable.
Consequently, nonlinear finite element code such as Abaqus [24] must be
used to develop a proper model for tires. As it was pointed out in the
introduction, a tire undergoes very widespread types of loads during its
service life. Therefore, various finite element models have been developed
so far for the prediction of tire response to individual or combined
loadings. These generally include 1) axisymmetric two-dimensional
models for rim mounting and applying of inflation pressure 2) three-
dimensional models under static conditions such as footprint (normal) and
shear loads 3) steady state rolling analysis for straight line motion and
cornering and 4) transient analysis for the modeling of tire under time-
dependent conditions. All the models and procedures given in the
following sections are based on the use of Abaqus code [24]. This is a very
well-known computer program which is widely used for the analysis of
tires by many investigators. Like any other commercial codes, the use of
this program needs carrying out three consecutive stages, known as pre-
processing, processing and post processing. The purpose of the pre-
processing step is to develop finite element mesh including elements and
nodes, selection of proper material models and assigning them to elements,
applying boundary and initial conditions as well as necessary loads and
finally providing supplementary information such as defining time period,
time step and function, increment size and the extra geometrical data like
thickness. This step is very time consuming and requires spending much
time by model developer. Abaqus/CAE is the pre-processing section of the
Abaqus software. Processing stage involves the solution of the equations
and obtaining the results. For tire analysis, depending on the type of model
(2D vs. 3D or steady state vs. transient) and computer hardware, it may
take a few seconds up to several hours to complete a job on the system.
Abaqus/Standard and Abaqus/explicit are the main processing modules
which are used for the tire analysis. The post-processing stage deals with
the representation of the results in such way that they can be best used for
interpretation of the outcomes and optimize of the initial designs. This step
in Abaqus is performed by Abaqus/Viewer. The details of each of these
steps for different tire model are given in next sections.
3.1. Two-Dimensional (2D) Model
3.1.1. Mesh Generation

The geometry of a tire can be virtually considered as a torus which is
produced by the revolving of a planar surface (cross-section) about a
rotational axis. Therefore, development of a finite element model for a tire
generally starts with a two-dimensional axisymmetric model of the tire
cross-section. Cross-sectional geometry of a radial tire is not a simple
shape and consists of several complex curves including circular arcs,
splines and multiple connected small curves. In addition, tires are normally
comprised of different compounds and composites in which each
component has its own specific properties. Figure 3 shows the lay-out of a
typical radial tire that is comprised of 10 sections. Hence, creating an
appropriate finite element mesh with controlled distortion and proper
aspect ratio requires careful consideration of several factors. The most
important one is to preserve the consistency between generated meshes in
different regions. For instance the number elements and nodes on the
common boundaries between tread and under tread or belts and body plies
should exactly match each other. Consequently, the unstructured or free
meshing techniques cannot be used for tire cross-section or lay-out. It is
rather preferred to use the structured meshing technique as the pattern of

the generated mesh is predictable. However, the applicability of this
method is limited to simple geometries. Thus subdividing of a complex
region (e.g., tread) into simple quadrilateral shapes should be first taken
into account.
Another significant issue that must be taken into consideration is the
non-axisymmetric topology of the tire which is mainly due to complex
tread pattern. Most of the critical performances of a radial tire are
dependent on the pattern of the tread so that tire designers normally use
highly complicated geometries and shapes. It goes without saying that in a
two-dimensional modeling approach it is not possible to fully consider the
tread pattern. Therefore, simple ribs are usually adopted to overcome this
problem. The curves of the tread region shown in Figure 3, contains four
ribs as a rational replacement of the tread pattern of the tire.
Figure 3. Lay-out of a typical passenger car radial tire.

Prior to mesh generation the lay-out of the tire must be prepared in an

appropriate standard format such DXF (Drawing Exchange Format) which
is the most common format used by almost any CAD/CAE software. After
creating the initial drawing of the tire, it is imported in the GUI (Graphical
User Interface) of the selected pre-processing software (such as
Abaqus/CAE). Following the initial opening of the drawing file, it should
be worked out so that it becomes suitable for mesh generation. For rubbery
parts, two-dimensional axisymmetric hybrid solid elements with twist and
pressure are used. In these elements the incompressibility of rubbery
materials and nonzero circumferential component of displacement tensor
are taken into consideration. The pressure in these elements is described by
either a constant values for linear elements such as CGAX4H and
CGAX3H or linear pressure for quadratic elements such as CGAX6H and
CGAX8H for quadrilateral and triangular elements, respectively. These
elements are defined in the Abaqus element library [24]. They have three
degrees of freedom in (𝑟, 𝑧, 𝜃) coordinate system as 1) 𝑈𝑟 or displacement
in radial direction 2) 𝑈𝑧 or displacement in lateral direction and 3) 𝑅𝑧 or
rotation about 𝑧 axis. An additional degree of freedom also exists to allow
for pressure inside the element.
Reinforcement materials including steel cord for belts, polyester for
body plies and Nylon for cap-plies are modeled with rebar in surface
elements, SFMGAX1 and SFMGAX3. These elements are 2-nodded linear
and 3-nodded quadratic axisymmetric surface elements with twist which
have the same three degrees of freedom(𝑈𝑟 , 𝑈𝑧 , 𝑅𝑧 ). The cord orientation
(the angle that cord makes with meridional direction) and spacing between
individual cords must also be specified. Surface and solid elements share
common nodes at identical coordinates and thus the number nodes in the
model do not increase with addition of the reinforcement to the model.
Figure 4 shows a CGAX4H element with constant pressure. Any of two
nodes in this element can be used for the definition of a SFMGAX1
surface element.
Figure 4. A quadrilateral axisymmetric hybrid element with linear interpolation for

displacement and constant pressure at the center [25, 26].
Figure 5. Finite element mesh of a 205/60R15 tire with total number of elements and
nodes equal to 1168 and 940, respectively.
Figure 6. Finite element mesh of a 185/65R14 tire with total number of elements and
nodes equal to 850 and 687, respectively.
Figures 5 and 6 show the finite element meshes of two 205/60R15 and
185/65R14 radial tires with axisymmetric solid and surface elements. As it
can be seen, due to the complex shape of the tire cross section, both
quadrilateral and triangular elements are used to create a convergent and
fully verified mesh. It is worth mentioning that although tire lay-out in 2D
view represents symmetry, due to the requiring of a full model for rolling
analysis it is preferred not to exploit symmetry to construct the mesh.
3.1.2. Material Model

The physical and mechanical behavior of each element or element set
in a FE model must be defined using a proper material model. Therefore, a
powerful finite element code (like Abaqus) is usually equipped with a
strong library consisting of wide variety of constitutive models to correctly
describe the real behavior of the materials. As noted earlier, tires are made
of rubbers and high stiffness cords which are generally modeled by
hyperelastic and linear elastic relationships, respectively. However, if the
time dependent behavior such as rolling loss computation is intended to be
considered, hyper-viscoelastic equations should be included in the model
[27, 28]. Details of these models are as follows.
3.1.2.1. Hyperelastic Model

Due to the highly nonlinear behavior and also incompressibility of the
rubbery materials, the mathematical description of their mechanical
behavior is accomplished by hyperelastic models which are based on strain
energy density (SED). It is assumed in these models that SED stored
during the application of the deformation is correlated to the invariants
(𝐼1 , 𝐼2 , 𝐼3 ) of the left Cauchy-Green deformation tensor, 𝐁 or principal
stretch ratios (𝜆1 , 𝜆2 , 𝜆3 ). The left Cauchy-Green tensor is given as:
𝐁 = 𝐅. 𝐅 𝐓 (1)
where 𝐅 is the deformation gradient expressed by:

𝜕𝑥
𝐹𝑖𝑗 = 𝜕𝑋𝑖 (2)
𝑗
In this equation, 𝑥𝑖 and 𝑋𝑗 are the coordinates of a material point in

current (deformed) and reference (initial or undeformed) configurations,
respectively. Consequently, the invariants of the tensor 𝐁 are [29]:
𝐼1 = 𝜆12 + 𝜆22 + 𝜆23 (3)

𝐼2 = 𝜆12 𝜆22 + 𝜆12 𝜆23 + 𝜆22 𝜆23
𝐼3 = 𝜆12 𝜆22 𝜆23
For a fully incompressible material, (𝐼3 = 1). However, for highly

filled rubbers such as compounds used tires, there exists a small amount of
compressibility which is normally expressed by means of the ratio of initial
bulk to shear moduli, i.e.:
𝑘
3( 0 )−2
𝜇0
𝜈= 𝑘 (4)
6( 0 )+2
𝜇0
where 𝑘0 , 𝜇0 and 𝜈 are the initial bulk modulus, initial shear modulus and
Poisson's ratio, respectively. In a fully incompressible material 𝑘0 → ∞
and thus 𝜈 = 0.5. Increasing the compressibility gives rise to lowering 𝑘0 ,
hence Poison's ratio becomes lower than 0.5. These types of materials are
called as nearly incompressible. For better description of the mechanical
behavior of these materials, the deviatoric part of the deformation is
separated from volumetric changes that may take due to compressibility or
temperature changes. In order to develop the corresponding mathematical
relationship, a new deformation gradient tensor is defined as:
1
𝐅̅ = 𝐽−3 𝐅 (5)
where 𝐽 is the total volume ratio and calculated by 𝐽 = det(𝐅). In a full

incompressible state, 𝐽 = 1 and thus equations (2) and (5) are identical.
However, for a nearly incompressible material, 𝐼3 = 𝐽2 and thus the new
form of the invariants and principal stretch ratios are given as:
2
𝐼1̅ = 𝐽−3 𝐼1 = 𝜆̅12 + 𝜆̅22 + 𝜆̅23
4
𝐼2̅ = 𝐽−3 𝐼2 = 𝜆̅12 𝜆2̅2 + 𝜆̅12 𝜆2̅3 + 𝜆̅22 𝜆2̅3 (6)
𝐼3̅ = 1
1
𝜆̅𝑖 = 𝐽−3 𝜆𝑖 (7)
The new form of the left Cauchy-Green tensor is also given by:
2
̅ = 𝐅̅. 𝐅̅ 𝐓 = 𝐽−3 𝐁
𝐁 (8)
Now, the strain energy density function can be written by the

following form:
𝑊 = 𝑊𝑑𝑒𝑣 (𝐼̅1 , 𝐼2̅ ) + 𝑊𝑣𝑜𝑙 (𝐽𝑒𝑙 ) (9)
where 𝑊𝑑𝑒𝑣 and 𝑊𝑣𝑜𝑙 are the deviatoric and volumetric components,
respectively. 𝐽𝑒𝑙 is the elastic volume ratio defined as:
𝐽
𝐽𝑒𝑙 = (10)
𝐽𝑡ℎ
𝐽𝑡ℎ = (1 + 𝜖 𝑡ℎ ) (11)
where 𝜖 𝑡ℎ is the linear thermal expansion strain. For an isothermal

deformation process, 𝐽𝑒𝑙 = 𝐽 that means that the volume change is only due
to the elastic deformation in the material.
There are several hyperelastic models introduced in literatures with
different philosophy on the derivation and physical description of their
parameters [27, 28, 30]. These models are either phenomenological in
which an empirical relationship is proposed for 𝑊𝑑𝑒𝑣 or mechanistic where
molecular structure of materials is used in conjunction with a theory to find
a suitable form for the first part of the equation (9). There are also a few
models which are called hybrid of the mentioned models i.e.,
phenomenological and mechanistic relationships. Here we give some more

important models as follows:
Neo-Hookean
This is the first and simplest mechanistic model based on the statistical
mechanics and thermodynamics given as [31]:
1 2
𝑊 = 𝐶10 (𝐼1 − 3) +
𝐷1
(𝐽𝑒𝑙 − 1) (12)
where 𝐶10 and 𝐷1 are the material parameters which can be generally
dependent on the temperature. This model can precisely predict the real
behavior of rubber up to 40% and 90% in tension and shear modes,
respectively. The initial shear and bulk moduli are also expressed by:
𝜇0 = 2𝐶10
2 (13)
𝑘0 = 𝐷
1
Mooney-Rivlin
This model is based on the nonlinear elastic theory and has been first
proposed by Mooney as:
1 2
𝑊 = 𝐶10 (𝐼1̅ − 3) + 𝐶01 (𝐼2̅ − 3) + 𝐷 (𝐽𝑒𝑙 − 1) (14)
1
Although this model can predict the tension behavior of rubbers up to

100% extension, it is unable to accurately describe its behavior in
compression. Moreover, it is not able to predict the stiffness increase in
high strains.
Polynomial Models
The neo-Hookean and Mooney-Rivlin models are special forms of a
general equation expressed by:
𝑖 1 2𝑖
𝑊 = ∑𝑛𝑖=0 ∑𝑚 ̅ ̅ 𝑗 𝑛 𝑒𝑙
𝑗=0 𝐶𝑖𝑗 ( 𝐼1 − 3) (𝐼2 − 3) + ∑𝑖=1 𝐷 (𝐽 − 1) (15)
𝑖
It is a phenomenological model in which it is assumed that 𝐶00 = 0. In

a simplest case if 𝑛 = 1 and 𝑚 = 0 then the above equation reduces to the
neo-Hookean model (equation 12); and if = 1, 𝑚 = 1 and the term 𝐶11 is
ignored then the Mooney-Rivlin equation (14) is obtained.
Reduced Polynomials
These models are like polynomial models with the exception that the
deviatoric part of the model is only dependent to first invariant of the left
Cauchy-Green tensor (𝐁 ̅ ), i.e.:
1 2𝑖
𝑊 = ∑𝑛𝑖=1 𝐶𝑖0 (𝐼1̅ − 3)𝑖 + ∑𝑛𝑖=1 𝐷 (𝐽𝑒𝑙 − 1) (16)
𝑖
The most important form of these models is introduced by Yeoh as

[32, 33]:
1 2
𝑊 = 𝐶10 (𝐼1̅ − 3) + 𝐶20 (𝐼1̅ − 3)2 + 𝐶30 (𝐼1̅ − 3)3 + (𝐽𝑒𝑙 − 3) +
𝐷1
1 4 1 6
𝐷2
(𝐽𝑒𝑙 − 3) +
𝐷3
(𝐽𝑒𝑙 − 3) (17)
It is shown by Yeoh [32-34] if limited test data are available for

material model calibration (e.g., uniaxial data), then the above model gives
highly accurately results in multi-deformational modes. It should be worth
to note that the Yeoh and neo-Hookean equations can be viewed as special
cases of the reduced polynomial model.
Arruda-Boyce Model
In this model the molecular structure of an elastomeric material is
considered to be described by a non-Gaussian behavior [35]. It is a
mechanistic model expressed by the following equation:
2
𝐶𝑖 𝑖̅ 1 (𝐽𝑒𝑙 ) −1
𝑊 = 𝜇 ∑5𝑖=1 𝑖
2𝑖−2 (𝐼1 − 3 ) + 𝐷 ( 2
− ln(𝐽𝑒𝑙 )) (18)
𝜆𝑚
where
1 1 11 19 519
𝐶1 = , 𝐶2 = , 𝐶3 = , 𝐶4 = , 𝐶5 =
2 20 1050 7000 673750
In the above equation, 𝜇 is a material parameter and 𝜆𝑚 is the locking

stretch in which the slope of stress-strain curve increases remarkably.
Ogden Model
It is also a phenomenological model which directly relates the
deviatoric part of the hyperelastic model to the principle extension ratios as
[36]:
2𝜇 𝛼 𝛼 𝛼 1 2𝑖
𝑊 = ∑𝑛𝑖=1 𝛼 2𝑖 (𝜆̅1 𝑖 + 𝜆̅2 𝑖 + 𝜆̅3 𝑖 − 3) + ∑𝑛𝑖=1 𝐷 (𝐽𝑒𝑙 − 1) (19)
𝑖 𝑖
Van der Waals Model

This model is given as [37]:
3 2
2 𝐼̃−3 2 1 𝐽𝑒𝑙 −1
𝑊= 𝜇 {−(𝜆2𝑚 − 3)[ln(1 − 𝜂) + 𝜂] − 𝑎( ) }+ ( −
3 2 𝐷 2
ln 𝐽𝑒𝑙 ) (20)
where
𝐼̃ − 3
𝐼̃ = (1 − 𝛽)𝐼1̅ + 𝛽𝐼2̅ , 𝜂 = √ 2
𝜆𝑚 − 3
In this equation, 𝜇 is the initial shear modulus, 𝜆𝑚 is the locking

stretch; 𝑎 is the global interaction parameter and 𝛽 is an invariant mixture
parameter.
Marlow Model
The Marlow model does not contain any explicit relation between
strain energy density and invariants or stretch ratios [38]. It is assumed that
the strain energy density is only a function of the first invariant of the
strain tensor, i.e.,
𝑊 = 𝑊𝑑𝑒𝑣 (𝐼1̅ ) + 𝑊𝑣𝑜𝑙 (𝐽𝑒𝑙 ) (21)
As 𝐼3̅ = 1, thus 𝐼1̅ can be written as:
1
𝐼1̅ = 𝜆̅12 + 𝜆̅22 + ̅2 ̅2 (22)
𝜆1 +𝜆 2
It can be shown that the minimum and maximum values of the 𝐼1̅ are
between 3 and +, respectively. The uniaxial form of 𝐼1̅ (𝜆̅2 = 1⁄ ) can
√𝜆̅1
be written as:
2
𝐼1̅ = 𝜆̅2 + ̅ (23)
𝜆
It is worth noting that the variation of 𝐼1̅ in equation (23) is also

between 3 and +. Therefore, it can be concluded that the uniaxial tension
test data can be used for the determination of the corresponding strain
energy density. To achieve this goal, 𝐼1̅ in a general 3D problem is first
calculated and then the associated value of the ̅ is determined from
equation (23). Having computed ̅ , the stress from the experimental data of
a uniaxial tension test is determined and the strain energy density is then
calculated by:
̅
𝜆 (𝐼)−1̅
𝑊𝑑𝑒𝑣 (𝐼 )̅ = ∫0 𝑇 𝜎(𝜖)𝑑𝜖 (24)
where  and  are the nominal strain and stress in uniaxial tension test,
respectively, that are directly substituted in equation (24).
3.1.2.2. Determination of Hyperelastic Model Parameters

Having selected a proper hyperelastic model for the description of the
mechanical behavior of the rubber, the parameters of the associated
equation must be carefully determined. To accomplish this task,
mechanical tests are carried out and the obtained results are fitted to the
selected model. On the other hand, highly filled rubber compounds show
very complicated behavior so that a single mechanical test such as a
uniaxial experiment is inadequate to accurately and precisely determine the
parameters of the model. Consequently, it is highly recommended that
more than one set of experimental data acquired from mechanical tests
performed in different modes of stress are used for curve fitting. A
complete set of experiments includes uniaxial, equi-biaxial, planar (pure
shear) and simple shear tests for the determination of the deviatoric part
and volumetric test for the parameters of the volumetric part of the
equation, respectively. However, carrying out these entire tests is time
consuming and expensive and limited test data are usually available for
calibration. In such case, the Arruda-Boyce, Van der Waals, Yeoh, or
reduced polynomial forms generally provide practical results. On the
contrary, when only one set of test data (uniaxial, equi-biaxial, or planar
test data) is available, the Marlow model is recommended. The more
accurate results is obtained with the Ogden, Van der Waals and polynomial
forms in conjunction with multiple experimental tests. All these tests are
carried out using special testing machines with specific grips, fixtures and
accessories which will be described as follows:
Uniaxial Test
Uniaxial experiment in tension mode is generally run using a universal
machine on dumbbell test specimen according to ASTM D-412. The
corresponding uniaxial compression test is performed on a cylindrical
specimen using the same machine according to ISO-7743 standard. Figures
7 and 8 show two fixtures used for tensile and compression tests installed
on a universal testing machine with their corresponding samples,
respectively.
Figure 7. Rubber sample with fixtures and extensometer used for uniaxial tensile test.
Figure 8. Rubber sample with supporting and forcing plates used for uniaxial
compression test.
Equibiaxial Test
In an equibiaxial test the rubber sample undergoes simultaneous
tensions in two perpendicular directions. To carry out such an experiment,
the testing machine needs to have two axes in which each axis has its own
independent moving mechanism. Therefore, several mechanisms have
been developed to do such test. One of the simplest methods is to use a
universal testing machine in conjunction with a special fixture that
emulates the biaxial tension in two directions. For example, the fixture
shown in Figure 9 generates tension in two directions.
Figure 9. A universal testing machine with equibiaxial fixture [39].
Planar Test
In a planar tension or pure shear test the plane strain condition is
enforced on the test specimen by preventing the edges of the specimen
from contracting. This is achieved through the use of test samples with
high aspect ratios (large width to length). Figure 10 displays a typical
rubber sample fixed between the grips of a specially designed fixture to
carry out this using any standard universal testing machine.
Figure 10. Rubber sample with fixture for planar tensile (pure shear) test.
Simple Shear Test

As most of the deformation that takes place in rubberlike materials is
in shear mode, a simple shear test can substantially help to accurately
determine the parameters of a hyperelastic model. This test is commonly
performed using either a double-sandwich construction or a quadruple
shear test piece [40]. Figure 10 shows a quadruple shear test samples fixed
on an appropriate fixture which can be attached to a universal testing
machine. The recorded force and displacements are converted to stress and
strain by [41]:
𝑑
𝜖 = 2𝑡 (25)
𝐹
𝜎 = 2𝐿𝑊 (26)
where 𝐿, 𝑊, and 𝑡 are the length, width and thickness of each sample,
respectively. It is worth noting that some software such as Abaqus does not
accept shear test data. In this case the previously mentioned test procedures
(uniaxial, equibiaxial and planar) can only be used.
Figure 11. Rubber samples with fixtures for quadruple shear test.
Volumetric Test
The last experiment from which the parameters of the volumetric part
of a hyperelastic model are computed is the volumetric test. The
experiment is carried out to determine the deviation of the Poisson’s ratio

from 0.5. In this experiment a small rubber cylinder is first prepared with
height and diameter of 19 mm and 30 mm, respectively. The specimen is
put and confined in a hollow metallic cavity with the same size and then
axially pressed under compressive force using a pin. Figure 12 shows the
metallic cavity with rubber sample inside it and the forcing pin which are
mounted on a universal testing machine.
Figure 12. A volumetric test fixture (left) mounted in a universal test machine (right).
The applied force and the displacement of the pin are recorded. These
values are then converted to the pressure and volume ratio using the
following equations:
𝐹
𝑃𝑟𝑒𝑠𝑠𝑢𝑟𝑒 = 𝐴 (27)
𝑉𝑜𝑙𝑢𝑚𝑒 𝑟𝑎𝑡𝑖𝑜 = 1 − 𝜖 (28)
where 𝐹 is the force, 𝐴 is the area of the cross section of the test sample
and 𝜖 is the nominal strain.
The minimum experiment required to calibrate a hyperelastic model is
a uniaxial test on a rubber dumbbell sample. However, the accuracy of the
predicted values of the model parameters and goodness of fitting depend
on smoothing of the measured data, stability and types of the selected tests.
Smoothing is the selection of a mathematical procedure for removing of
the noise in experimental data so that the quality of the fitting is enhanced.
Stability is taken by Drucker stability condition [42] in which the
following inequality should be satisfied:
𝑑𝜎: 𝑑𝜖 > 0 (29)
where 𝑑𝜎 and 𝑑𝜖 are infinitesimal changes in stress and logarithmic strain,

respectively. Replacing 𝑑𝜎 in the above formula by: 𝑑𝜖, the inequality (29)
becomes:
𝑑𝜖: 𝐃: 𝑑𝜖 > 0 (30)
where 𝐃 is the tangent material stiffness and thus must be positive-definite.

As it is mentioned earlier, the experimental data obtained from
different deformational modes should be used to obtain accurate and high
quality material parameters. In order to assess the applicability of different
test methods, the parameter of two phenomenological and mechanistic
hyperelastic models for a typical tread compound were determined. These
are the Arruda-Boyce (equation 18) and Polynomial with N=2 (equation
15) models. The uniaxial, planar and volumetric tests were carried. For
each model the parameters were determined adopting two approaches.
First, the uniaxial and volumetric tests were only used to predict the
parameters of the mentioned models and in the second approach all test
results including uniaxial, planar and volumetric were employed for the
determination of the parameters. The nonlinear material evaluation module
in Abaqus code was selected for these calculations. Figures 13 and 14
compare the experimentally measured nominal stress versus nominal strain
for uniaxial and planar data using the Arruda-Boyce model. As shown in
Figure 13, using only uniaxial data, a high goodness of fitting is achieved
between experimental and computed data for uniaxial test. However, when
full data sets are employed, a relatively large deviation is obtained after
nominal strain greater than 0.5. Contrary, as it can be seen in Figure 14, for
planar test, the deviation from experimental data for parameters calculated
from uniaxial results are much higher than when full data sets are utilized.
This clearly indicates that increasing the accuracy of the model parameters
in different deformational modes requires more experimental data to be
used during curve fitting. The same approaches were adopted for the
polynomial model with N = 2 as shown in Figures 15 and 16. The results
are similar to those reported for the Arruda-Boyce model. As can be in
Figure 15, for uniaxial mode, the data from both uniaxial only and full set
experiments gives a very good fitting. However, the deviation from
experimental data for planar test using only uniaxial test data, as shown in
Figure 16, is much greater than the similar case when the Arruda-Boyce
model is employed (Figure 14). This is because that the Arruda-Boyce
equation is a mechanistic model in which using only the uniaxial test data
can give rise to relatively good prediction in all deformational modes.
Nevertheless, the polynomial model is a phenomenological model which
requires data from multiple experimental tests becomes available to give a
high accurate curve fitting.
Figure 13. Experimental and computed stress-strain curve using Arruda-Boyce model
in uniaxial mode using uniaxial only and full set data.
Figure 14. Experimental and computed stress-strain curve using Arruda-Boyce model
in planar mode using uniaxial only and full set data.
Figure 15. Experimental and computed stress-strain curve using polynomial (N = 2) in

uniaxial mode using uniaxial only and full set data.
Figure 16. Experimental and computed stress-strain curve using polynomial (N = 2) in

planar mode using uniaxial only and full set data.
3.1.2.3. Hyper-Viscoelastic Models

As it is mentioned earlier, the mechanical behavior of rubber
compounds especially in applications where time scale is prominent,
should be described by a viscoelastic model in conjunction with a
hyperelastic model for their time-independent behavior. For example,
during the modeling of the rolling resistance in a tire, the dissipated energy
and heat built-up can only be predicted by the use of a proper viscoelastic
model. This is normally accomplished by combining a hyperelastic
equation as given above with a viscoelastic model in which model
parameters are described as a function of time. The time-function
equations used for viscoelastic behavior can be categorized into linear and
nonlinear models. The linear model is normally based on the Prony series
expansion [27, 28, 43] in which the dimensionless relaxation modulus is
given by:
𝑡
− 𝐺
𝑔𝑅 (𝑡) = 1 − ∑𝑁 𝑃
𝑖=1 𝑔̅𝑖 (1 − 𝑒 𝜏
𝑖 ) (31)
where 𝑔𝑅 (𝑡) = 𝐺𝑅 (𝑡)/𝐺𝑜 is the reduced shear modulus, 𝐺0 is the

instantaneous shear modulus, 𝑔̅𝑖𝑃 , 𝜏𝑖𝐺 and 𝑁 are material constants. For
hyperelastic material models, the relaxation coefficients in above equation
are applied either to the constants that define the energy function or
directly to the energy function such as Marlow model in equation (24). For
example, in polynomial functions given in equations (15) and (16), the
constants defining the deviatoric parts can be written as:
𝑡
𝑝 −
𝐶𝑖𝑗𝑅 (𝑡) = 𝐶𝑖𝑗0 (1 − ∑𝑁
𝑘=1 𝑔̅𝑘 (1 − 𝑒
𝜏𝑘
)) (32)
where 𝐶𝑖𝑗𝑅 (𝑡) and 𝐶𝑖𝑗0 are the reduced and instantaneous material constants,
respectively. Similar relations can be written for parameters in other
material models.
Figure 17. A parallel representation of the PRF model [44].
Rubber compounds used in tires show nonlinear viscoelastic behavior

at high carbon black contents. It is shown that their behavior cannot be
accurately modeled using linear viscoelastic models especially when
subjected to finite strain or large deformations. Consequently, nonlinear
viscoelastic models have been developed to tackle this problem. A very
common form for nonlinear description of the viscoelasticity in polymeric
solids such rubbery materials is the parallel rheological framework (PRF)
model. This model consists of an arbitrary number (say, 𝑁) of viscoelastic

networks connected in parallel in conjunction with an extra elastic part
which takes the equilibrium of the network into consideration as shown in
Figure 17 [44].
The elastic part in this figure as shown by a zero id, is related to the
one of the hyperelastic models used for the description of the strain energy
density such as Neo-Hookean, Mooney-Rivlin, Yeoh, etc. The other parts
(id 1 to 𝑁) corresponds to a nonlinear model in which the equivalent creep
strain rate (𝜖̅̇𝑐𝑟 ) is related to equivalent creep strain (𝜖̅𝑐𝑟 ) and equivalent
deviatoric Kirchhoff stress (𝑞) via a phenomenological or mechanistic
model. Some types of these equations are as follows:
- Strain hardening power-law
1
𝜖̅̇𝑐𝑟 = (𝐴𝑞 𝑛 [(𝑚 + 1)𝜖̅𝑐𝑟 ]𝑚 )𝑚+1 (33)
- Hyperbolic-sine law
𝜖̅̇𝑐𝑟 = (𝑠𝑖𝑛ℎ𝐵𝑞)𝑛 (34)
- Bergstrom-Boyce (BB) law
𝜖̅̇𝑐𝑟 = 𝐴(𝜆𝑐𝑟 − 1 + 𝐸)𝐶 𝑞𝑚 , 𝜆𝑐𝑟 = √𝐼1𝑐𝑟 /3 (35)
where 𝐴, 𝑛, 𝑚, 𝐵 and 𝐸 are material parameters which must be determined

using an appropriate calibration method. References [27, 28, 43] gives
details of some numerical techniques that can be used for the determination
of the material parameters for hyper-viscoelastic models.
3.1.2.4. Linear Elastic Model

Linear isotropic elastic model is the simplest form that can be used for
the description of the mechanical behavior of a material. For tire analysis,
the long fiber is assumed to be perfectly elastic and thus linear elastic
material model is used. Long fibers that are used in tires are steel cords for
belts, polyester or rayon for main body plies and nylon for cap plies. Two
material constants including elastic or Young’s modulus (𝐸) and Poisson’s
ratio (𝜈) are required to be defined. Young’s modulus is generally
determined from a uniaxial test on the preconditioned cord and for the
Poisson’s ratio a value of 0.3 is normally assumed. It should be noted that
due to the very stiff materials used for tire beads, the linear elastic model is
also used for this part of the tire.
3.1.3. Boundary Conditions and Applied Load

Once the mesh has been created and material models were assigned for
different parts in a tire, the boundary conditions and applied load must be
defined. A fixed bead is generally defined if the tire rim width (i.e.,
distance between left and right heels of the tire) is assumed to be
unchanged. This means that all degrees of freedom on outer part of tire rim
are set to zero. Following this imposition, the external load or inflation
pressure is applied on the internal surface of the tire. On the other hand, if
the tire rim width is changed during mounting of the tire on the rim, a
horizontal displacement in lateral direction is first applied on the tire beads
and then the inflation pressure is imposed. Figure 18 shows a finite
element mesh of the tire (as depicted previously in Figure 5) with applied
pressure and boundary condition marks.
Figure 18. Finite element mesh of a 205/60R15 tire with applied pressure and
boundary conditions.
3.1.4. Section and Miscellaneous Data

In addition to above mentioned data, some extra information should be
assigned to complete the model development process. These are orientation
angle, spacing and cross sectional area of the cords in body plies, belts and
cap plies, increment size and time period for defining load as function of
time.
3.2. Three-Dimensional (3D) Models
Realistic tire models require a full three-dimensional finite element

mesh to be created. These include footprint analysis under vertical load,
static analysis of a loaded tire under lateral, longitudinal and torsional
loads, steady state rolling for free rolling analysis, application of slip and
camber angles, wear analysis and transient analysis using dynamic
simulation. There are generally two methods that are used to develop a 3D
finite element mesh for a tire. In the first method, the model is generated
by revolving of the 2D elements (as shown in Figures 5 and 6) developed
from the tire cross-section about symmetry axis. In this case, the
corresponding 3D element sets are created from their 2D counterparts. For
example, in Abaqus code C3D8H and SFM3D4R elements are generated
from 2D elements CGAX4H and SFMGAX1, respectively [25, 26]. All
material and section definitions as well as the other parameters assigned in
2D model are automatically transferred to 3D state so that the generated
model is almost ready to be analyzed. Additionally, it should be note that
in all 3D analyses, tire must contact with a surface that resembles the road.
Consequently, a contact surface with corresponding boundary condition
and frictional data should also be defined. Figure 19 shows a full 3D finite
element mesh created for a 205/60R15 which its 2D mesh was previously
shown in Figure 5. As it can be seen, the finite element mesh was more
refined in the contact zone to accurately compute the variable of this
region. Total number of elements for this model is 51392. A rigid contact
surface is also defined in order to represent the road. The main drawback
associated with this method is its incapacity in taking the details of the
tread pattern into consideration. In fact, like 2D meshes, the tread patterns
are modeled using simple ribs.
Figure 19. 3D Finite element mesh of a 205/60R15 with contact surface.
Figure 20. Finite element mesh of Figure 21. Finite element mesh of
the repeating unit (symmetric part the repeating unit (tread pattern)
without tread pattern) [45]. [45].
In order to take the complex shape of the patterns including tread

blocks, ribs and sipe into account, a quite different algorithm should be
utilized. In this case, a 2D axisymmetric model of the tire without tread
part is first generated. This section was then revolved about tire symmetry
axis only for a small angle which can be the rotational angle of a repeating
unit of the tire tread blocks. The geometry of the repeating section was
then created by a 3D modeling software and meshed using 3D brick
elements. Either a uniform or variable pitch for the geometry of the tread
blocks can be assumed. Figures 20 and 21 show the finite element meshes
of the symmetric part (tire without tread) and tread blocks for a 175/70R14
steel-belted radial tire, respectively. These two meshes are then assembled
and a unique mesh is created as shown in Figure 22.
Figure 22. Finite element mesh of the repeating unit (symmetric part + tread pattern)
[45].
Figure 23. Finite element mesh of the full 3D model with tread blocks.
This single 3D sector of the model (Figure 22) is then revolved

periodically about tire symmetry axis to create a full 3D finite element
mesh of the tire with detailed tread blocks located on its surface (Figure
23). The required number of periodical repeating unit is derived from
division of a full rotation angle (360o) by angle of initial sector (as shown
in Figure 22). The key issue in generating such 3D finite element model is
that the mesh configurations at interface between the tread and symmetric
part of the tire do not usually match correspondingly, and thus numerical
difficulties in finding solutions at interfacial nodes are inevitable. A tying
algorithm which is implemented in Abaqus is generally used. In this
scheme, the displacements of each node on one surface (tire body) are
constrained such that they have the same values of the corresponding point
on the other surface (tread elements).
In addition to above-mentioned model, the global/local analysis is also
used [46]. In this technique which is also known as sub-modeling, a global
method is first adopted in which a full 3D model is developed and the
finite element analysis of the structure with a relatively coarse mesh is
performed. Then a part of the structure around the contact area is separated
and meshed with pattern of tread blocks (local model). The interpolated
displacements computed from the global method are applied to the local
model as the prescribed boundary conditions. This technique ignores the
interactions between tread blocks and carcass (especially at belt packages)
and thus cannot predict tire behavior correctly [47].
4. RESULTS AND DISCUSSION
As described in the previous section, depending on the type of

developed model and required analysis, different results are obtained. In
this section, some of these results will be presented.
4.1. Two-Dimensional
The main objectives of a two-dimensional axisymmetric (2D-axi)

model are tire sizing and prediction of the initial state of stress and strain in
a rim-mounted and inflated tire. In addition, as explained earlier, 3D model
of a tire is developed based on the 2D-axi model so that development and
running of a 2D analysis is compulsory. The results of tire sizing for some
radial tires are given in Table 1 and compared with experimentally
measured data. As can be seen, there are good agreements between
predicted and actual data which confirms the applicability of the models.
Figures 24 to 26 show the distribution of the logarithmic out-of-plane shear
strain (LE13) in the belt-edge region for a 205/60R15 tire with three
selected belt cord angles of 18o, 20o, and 22o, respectively. Here, 1, 2 and 3
refer to radial, lateral and out-of-plane directions (similar to x, y and z),
respectively. Logarithmic strain is the true strain which is expressed by:
𝐿 𝑑𝐿 𝐿
𝜖𝑡 = ∫𝐿 = 𝐿𝑛 𝐿 (36)
0 𝐿 0
Table 1. Predicted vs. measured dimensions of some tires
Tire Size Pressure Predicted Measured Predicted Measured

(MPa) OD† OD MSW†† MSW
(mm) (mm) (mm) (mm)
185/65R15 0.18 623.6 622.52±0.12 188.5 188.7±0.19
0.25 624.1 623.25±0.17 188.6 189.3±0.2
205/60R15 0.35 627.8 625±0.88 203 203.35±0.15
205/60R14 0.21 601.3 599.7±0.22 202.5 202.6±0.12
0.18 601.1 599.2±0.27 202.3 202.16±0.13
†
Outer diameter.
††
Maximum section width.
Figure 24. Distribution of the belt-edge out-of-plane true shear strain (LE13) for belt
angle = 18o.
angle = 20o.
angle = 22o.
As it can be seen in these figures, decreasing the belt angle will also
decrease the maximum shear strain which is due to the decrease in tire
shape growth. This may have direct influence on the tire durability. To
confirm this, the stored strain energy in tires have also be computed which
are 111.6, 114.12 and 117.7 Joule for tires with belt angles of 18o, 20o and
22o, respectively. A positive correlation is found between increasing the
out-of-plane shear strain in belt-edge and stored strain energy in tires.
However, it should be noted that varying the belt angle would also results
in changing of the other variables especially in contact area which must be
considered separately.
4.2. Three-Dimensional Static Analysis
Three-dimensional (3D) models are used for wide varieties of

analyses. The first model after performing of a 2D analysis is a 3D one for
modeling of the tire under static vertical load or footprint analysis. Having
generated the 3D finite element mesh and defining contact surface, the
results obtained in the first analysis (inflation) are used as the initial
estimates and the tire is analyzed under both inflation pressure and vertical
load. Using the results of a 2D model almost guarantees obtaining of a
converged solution. The tire rim is normally assumed to be fixed and the
contact surface is moved toward the tread surface to simulate the contact of
the tire with road. The friction between tread surface and contact zone can
be taken by any frictional equation ranging from simple Coulomb’s law to
arbitrary and complicated models defining via a user subroutine. In order
to avoid potential convergence difficulties the vertical load is generally
applied in two successive steps. In the first stage, a small vertical
displacement is prescribed to generate an initial contact between tire and
road and then immediately in the second step the remaining load is applied
to achieve the final deformed configuration of the tire. Details of different
3D modeling strategies of tires and the obtained results can be found in a
considerable amount of literature (see for example, [5, 18, 48]). Here we
present some results. Figure 27 shows the load-deflection data for a
185/65R15 tire under vertical condition which is obtained by experiment

and finite element calculation. As can been, the agreement between these
data are very good. The slope of this curve in the vicinity of the working
load of the tire in conjunction with damping properties is a good measure
of tire ride.
The corresponding footprint shape and contact pressure distribution for
the mentioned tire are shown in Figures 28 and 29. The predicted footprint
length and width are 125 mm and 143 mm, respectively. The
experimentally measured dimensions are 125 mm and 142 mm,
respectively which confirm the accuracy of the FE calculations. However,
the maximum calculated contact pressure as shown in Figure 29 is 136.8
N/cm2 while the experimentally recorded value is about 100 N/cm2. This
discrepancy may be explained by the type of the equation selected for the
friction between tread surface and road which was Coulomb’s law with
cf = 0.7. The actual frictional behavior of the tire in contact with road is
very complicated so that highly nonlinear models should be used [19].
Consequently, to accurately predict the pressure at contact zone a
sophisticated friction model in which the dependency of the friction factor
to different parameters such as the texture of the road, temperature, slip
rate, pressure and tread pattern must be taken into account.
The effect of tread pattern on the contact zone parameters for both
static and rolling cases of several radial tires was thoroughly investigated
and reported by Ghoreishy [45, 48-50]. Figure 30 shows the load-
deflection curve for a 175/70R14 tire (as shown in Figure 23) with and
without considering the details of the tread pattern. It can be seen that
ignoring the tread pattern has a very minor effect on the load-deflection
behavior of the tire compared to the case in which the details of the tread
design is taken into consideration.
However, for contact area variables significant differences are
obtained. Figures 31 and 32 show and compare the contact pressure
predicted of the mentioned tire by two finite element models without and
with taking the tread pattern into account. As presented in these figures,
there are important differences in calculated results.
Figure 27. Load-deflection data for a 185/65R15 tire. Experimentally measured vs.
FEM prediction.
Figure 28. Footprint area of the tire (185/65R14) predicted at 70% of the rated load
(L = 3840N).
Figure 29. Contact pressure of the tire (185/65R14) predicted at 70% of the rated load
(L = 3840N).
In both cases, the maximum contact pressure is located at edge (near

shoulder) zones. This phenomenon was already observed (both
theoretically and experimentally). On the other hand, due to the reduction
of the net contact area in patterned tire the maximum contact pressure in
patterned tire model is higher than the other model. In these cases the
analyses were performed under static loads without taking friction into
account. The effect of friction on the distribution of the contact pressure in
tread block will be considered in the next sections.
Figure 30. Predicted load-deflection curve for a 175/70R14 tire with and w/o tread
pattern detail.
Figure 31. Predicted contact pressure distribution for a 175/70R14 tire w/o tread
pattern detail.
Figure 32. Predicted contact pressure distribution for a 175/70R14 tire with tread
pattern detail.
Figure 33. Direction of applied loads in radial, longitudinal, lateral and torsional modes
on tire [18].
Once the footprint analysis has been completed, the tire can further be
analyzed and studied by applying longitudinal, lateral and torsional loads
to the contact surface. Figure 33 shows a typical radial tire presenting the
directions of the above-mentioned loads. Having analyzed the tire under
these loads, the corresponding force and deformation in tangential and
lateral directions and also moment and angular rotation for torsional mode
are determined. The tangential, lateral and torsional stiffness of the tire are
then calculated as the initial gradient or slope of the load and deflection
(moment and angular rotation for torsional stiffness) [18]. These
parameters are primarily utilized for the evaluation of the ride and handling
of tires under low speed conditions especially in urban applications.
Longitudinal stiffness is mainly used for the study of the enveloping
property of the tire encountering the road disturbances. The lateral stiffness
of a non-rolling tire is used in conjunction with cornering stiffness for the
evaluation of the relaxation property of a rolling tire and finally torsional
stiffness is useful for the assessment of steering property of the tire at
parking position and nearly zero speed condition [51].
The predicted stiffness for a 185/65R15 radial tire using above
described methods are given in Table 2 and compared with experimental
data. As can be seen, the finite element analysis were carried out using
three different hyperelastic models to compare the ability of the chosen
constitutive equations on the accuracy of the results.
Table 2. Predicted vs. experimentally measured stiffness of the tire

using different hyperelastic models
Experiment Ogden Yeoh Marlow

Radial (N/mm) 221 220.4 200.1 209.65
Longitudinal (N/mm) 329.4 272.5 222.8 251.7
Lateral (N/mm) 133.2 135.8 113.1 125
Torsional (Nm/deg) 69.45 72.2 55.9 70.6
It should be noted that as explained earlier, the radial stiffness was

determined as the gradient or slope of the load vs. deflection data at tire
operating point which was 70% of the maximum rated load. In this case the
maximum load was 5488 N. It is apparent from this table that the
mathematical form of the hyperelastic model has great influence on the
predicted stiffness especially in longitudinal or tangential direction.
Moreover, it is indicated that the accuracy of the predicted results are
significantly dependent on the mode of deformation. The current modeling
methodology used for the tire in which rebar layer technique is used to
describe the reinforcing parts cannot accurately describe the shear behavior
of the cord-rubber composite. Consequently, the hyperelastic model
selected for the description of the rubber is the dominant part for the
mechanical behavior of the cord-rubber composite in shear mode.
Therefore, the use of Ogden, Yeoh and Marlow (or any other hyperelastic
equations) determines the accuracy of the predicted stiffness in shear
modes [18].
4.3. Steady State Rolling Analysis
4.3.1. Free Rolling State

So far this section has focused on the static analysis of the tires. The
following parts will thus present and discuss the results of the tire rolling.
A very common approach that is widely used for rolling tires is the steady
state rolling analysis. This step restarts from the results of the footprint
analysis in an ALE (Arbitrary Eulerian/Lagrangian) framework
implemented in the Abaqus code. This capability uses a reference frame
that is attached to the axle of the rotating tire. In this frame the tire is not
moving but the material of which the tire is made is passing through those
points that were used to create the finite element mesh. This enables to
tackle the problem of modeling the rolling tire using the traditional
Lagrangian formulation since the frame of reference in which motion is
described is attached to the material. This technique can be viewed as a
mixed Eulerian/Lagrangian method, where rigid body rotation is described
in a spatial or Eulerian manner, and deformation, which is now measured
relative to the rotating rigid body, is described in a material or Lagrangian
manner. In other words, the finite element mesh of the tire is assumed to be
fixed (Eulerian) while the material moves through this mesh (Lagrangian).
Therefore, an identical mesh configuration can be used for both static
(footprint) and rolling analyses [25, 26, 45]. The implemented algorithm
for steady state rolling analysis in Abaqus/Standard code needs the ground
velocity and spinning speed to be specified separately. Therefore, the
ground velocity (𝑉) is assumed to be constant and the tire spinning speed
(angular velocity) is changed from minimum (𝜔𝑚𝑖𝑛 ) to maximum (𝜔𝑚𝑎𝑥 )
values calculated by:
𝑉
𝜔𝑚𝑖𝑛 = 𝑅 (37)
𝑚𝑎𝑥
𝑉
𝜔𝑚𝑎𝑥 = (38)
𝑅𝑚𝑖𝑛
where 𝑅𝑚𝑖𝑛 and 𝑅𝑚𝑎𝑥 are the minimum and maximum radii of the loaded
tire located at the center of the contact zone and the opposite side,
respectively. Figure 34 shows the reaction force parallel to the ground on
tire axle (applied at the center of the tire) at different spinning velocities
for a 185/65R15 tire. The ground velocity is assumed to be 80 km/h and
the spinning velocities are changed from 70 to 80 rad/s. As it can be seen,
the free rolling for this tire (RF = 0) occurs at spinning velocity of 73.637
rad/s. Partial braking occurs at spinning velocities smaller than this value
where the contact points between tire and road are slipping in backward
direction.
Figure 34. Longitudinal force vs. spinning velocity at constant ground velocity.
On the other hand, for those spinning velocities which are greater than
free rolling speed (i.e., >73.637 rad/s), partial traction occurs which means
that the mentioned contact points are slipping in forward direction. For a
very small or large spinning velocity, full braking or traction occurs in

which all contact points are completely slipping in backward or forward
direction, respectively.
The dynamic loaded radius of the tire can be computed from the
following relation:
𝑉
𝑅𝑑 = (39)
𝜔𝑓
where 𝑅𝑑 is the dynamic loaded radius or effective rolling radius

corresponding to the applied load in the footprint stage, 𝑉 is the ground
velocity and 𝜔𝑓 is the free rolling speed. Using the selected values for
ground velocity (80 km/h) and computed value of spinning velocity
(73.637 rad/s), the dynamic loaded radius can be computed as 301.78 mm.
The number of revolutions per km (𝑅𝑘 ) for a tire is then computed by:
106
𝑅𝑘 = 2𝜋𝑅 (40)
𝑑
The computed value of 𝑅𝑘 for this tire is about 527.4. If it is compared

with its experimental counterpart, 531, it can be seen that the difference is
quite small so that the model is capable of predicting the steady state
conditions of the rolling tire. This value is a very good parameter for
evaluating the effects of tire on vehicle fuel consumption.
4.3.2. Slip Angle Analysis

Steady state rolling analysis can be employed for the modeling of the
cornering behavior of tires. In this case, the slip angle must be defined in
the model in order to resemble the cornering of a rolling tire. Slip angle is
defined as the angle between the direction of travel and plane normal to the
axis of the rotation as shown in Figure 35. To accomplish this task, the
components of the ground velocity vector (𝑉 ⃗ ) are defined as:
𝑉𝑥 = 𝑉𝑐𝑜𝑠(𝛼) (41)
𝑉𝑦 = 𝑉𝑠𝑖𝑛(𝛼) (42)
where 𝛼 is the slip angle and 𝑉𝑥 and 𝑉𝑦 are the components of velocity
vector in 𝑥 and 𝑦 directions, respectively. The 𝑥, 𝑦 and 𝑧 coordinates refer
to longitudinal (traveling), lateral and vertical directions, respectively.
Figure 35. Slip angle and traveling direction.
A steady state rolling analysis for inclusion of the slip angle

(cornering) generally restarts from free rolling state. If a loaded tire is
steered, the slip angle is generated, and, hence the associated force and
moments are also created. The most important components of the force and
moments are lateral force (𝐹𝑦 ) and (self) aligning moment (𝑀𝑧 ) [52].
Figure 36 shows the variation of the lateral force (cornering force) vs. slip
angle at different vertical loads for a 185/65R15 tire. At zero slip angle
there is always a residual lateral force which is due to the plysteer and
conicity [52]. By increasing the slip angle, the lateral force also increases
with nonlinear behavior at higher slip angles. In addition, the rate of
increase in lateral force also increases by the increasing of the vertical load.
Figure 36. Lateral force vs. slip angle at different vertical loads.
Figure 37. Cornering stiffness vs. vertical load.
However, reducing the vertical load especially below the rated load
makes that the nonlinearity to be occurred at lower slip angles. To further
study, the cornering stiffness (CS) is defined as the absolute value of the
slope of the lateral force curve at the zero slip angle. The variation of the
cornering stiffness with vertical load is shown in Figure 37. As seen, it
initially increases with vertical load and then tends to reach a maximum
level and finally slumps to lower values with further increase of load. It
was previously shown that [52] for a better handling it is desirable to shift
the peak value in CS vs. load curve to higher values of the vertical load.
4.3.3. Camber Analysis

Camber angle,𝛾, is defined as the angle measured between the wheel
plane and the vertical direction and taken as positive if the top of the wheel
leans outwards [53]. Camber is usually imposed on tires to provide
required stability in vehicle during cornering and maneuverings. Figure 38
schematically shows camber tire vs. its normal position.
Similar to slip condition, the aligning of the tire with camber angle
generates the corresponding lateral force (𝐹𝑦 ) in rolling tires. The analysis
is restarted from free rolling state and the contact surface is rotated about
longitudinal (𝑥) axis to simulate the camber effect. Figure 39 shows the
camber force (lateral force or camber thrust) against camber angle for a
205/60R15 tire. It can be seen that, as expected, the lateral force increases
with increasing of the camber angle.
Figure 38. Schematic representation of normal and camber tires.

Figure 39. Predicted camber force vs. camber angle for a 205/60R15 tire.
Figure 40. Longitudinal force vs. spinning velocity for the steady rolling of the tire
with and without viscoelasticity [13].
4.3.4. Rolling Resistance Prediction

A very useful application of the steady state rolling analysis of the tire
is to predict the rolling resistance force which is a direct measure of the
energy loss and fuel consumption. In order to calculate the rolling
resistance by finite element method, two analyses should be performed.

First, the analysis is carried out without taking the viscoelasticity into
effect. In other words, only a hyperelastic model is used for the description
of the mechanical behavior of rubbery parts. The second analysis,
however, consists of the modeling of the tire with a hyper-viscoelastic
material model [12, 13]. Figure 40 shows the variation of the longitudinal
force versus spinning velocity for a 185/65R15 tire. The free rolling
spinning velocity is determined at the point where the reaction force is
equal to zero. As it can be seen, considering the viscoelasticity causes the
spinning velocity at free rolling state to be calculated at a higher value than
the case in which the material history is not taken into account. This is
because that by taking the viscoelasticity into consideration the tire is
forced to rotate more quickly to compensate the dissipated energy in order
to keep the linear velocity (80 km/h) to be constant. It can be assumed that
the difference between calculated reaction forces may be attributed to the
rolling resistance of the tire. This value is 54 N as depicted in Figure 40.
The actual rolling resistance of this tire was experimentally measured
which was 52.6 N. As can be seen, the calculated rolling resistance is very
close to its corresponding actual value.
4.3.5. Wear Analysis

A very useful feature of Abaqus code is the adaptive meshing
technique. This technique is generally used for re-meshing purpose such as
mesh distortion control in pure Lagrangian or ALE (arbitrary
Lagrangian/Eulerian) analyses. The combination of this meshing technique
with steady state rolling analysis provides an applied environment for wear
analysis. In this approach the wear of the tread surface in a tire is described
by the Archard [54] equation given as:
𝑘
𝑞̇ = 𝐻 𝑃𝐴𝛾̇ (43)
where 𝑞̇ is the volumetric rate of material loss due to wear of the surface, 𝑘
is a constant known as wear coefficient, 𝐻 is the hardness or elastic
modulus of the surface material, 𝑃 is the contact pressure, 𝐴 is the contact

area and 𝛾̇ is the interface slip rate. This equation describes that the rate of
worn material is proportional to the product of force (i.e., 𝑃𝐴) and slip rate
(i.e., between tread surface and road). The discretized form of the above
equation can be implemented in Abaqus code via a user subroutine in
conjunction with appropriate adaptive meshing keywords to predict the
worn shape of the tire during rolling. The details of this methodology is
described in [55]. Figure 41 shows tread elements for a 185/65R14 tire at
the start of the travelling and after 40,000 km mileage. The analysis was
carried out for a full braking (𝜔 ≪ 𝜔𝑓 ) condition to exaggerate the wearing
in the tire.
The corresponding contact pressure fields at the start of the travelling,
after 16,000, 32,000 and 40,000 km mileages are shown in Figures 42 to
45, respectively. As it can be seen, due to the changes in the tread shape
that takes place during rolling, the distributions of the pressure also vary
significantly. This means that each zone on the tread surface underwent
different history and thus in order to have a regular wear, the position of
the tires should be changed to ensure an even wear.
𝑘
Figure 41. Element set of the tire tread before and after wearing ( = 1𝑥10−14 ).
𝐻
Figure 42. Contact pressure distribution at the start of the travelling.
Figure 43. Contact pressure distribution after 16,000 km mileage.

4.4. Transient (Dynamic) Analysis
In many situations the modeling of the real behavior of a rolling tire

needs a transient or time dependent analysis to be performed. Accelerating,
braking, moving over road bump are some examples of such cases. In each
of them, a dynamic analysis should be carried out to predict the behavior of
the tire during movement. This is generally accomplished either by an
implicit or explicit dynamic analysis. In the implicit time integration
technique, the time effect is directly incorporated into the finite element
model equations. The main disadvantage of this method is that it requires
large amount of computer resource and thus it is generally limited to
simple models with low degree of freedom. On the other hand, the explicit
time integration is based on the solution of a lumped system of equations.
If the selected time increment to be small enough, there is likely to obtain
convergent and oscillation-free results. Abaqus/Explicit is a very advanced
module that is devoted to the solution of the nonlinear dynamic problem
using explicit technique. Time increment option is generally set to be
automatically determined. The code has been substantially used for the
modeling of the tires. Here, we present sample results of the rolling of a
205/60R15 tire moving over a curb or step on the road using
Abaqus/Explicit. Development of an explicit model is usually based on the

import of a model which has previously been created and run for
Abaqus/Standard i.e., steady state rolling analysis with some minor
differences. The most important one is that the discretization of the
geometry in circumferential direction must be uniform unlike mesh
developed for steady state model in which more refined elements in the
contact zone is employed. Figures 46 to 49 show the movement of the
mentioned tire over the road curb at different times. The linear velocity of
the tire for this analysis was 20 km/h.
Figure 46. Tire at the starting state.
Figure 47. Tire shape 0.06 s after moving.

Figure 50. Variations of the longitudinal force vs. time for tire moving over curb.
An important aspect of the dynamic analysis of rolling structures such

as tires is the oscillation in the computed results even for very small time
increments. Figure 50 shows the variation of the longitudinal force with
time for the moving of the mentioned tire.
As can be seen, the variation of the computed force with time is
associated with oscillations. The source of these oscillations can be
attributed to 1) discretization of a round shape into a polygonal geometry
which creates a non-smooth movement 2) short period of time from the
initial state to the point where it meets the curb and 3) lack of adequate
damping in the model that gives rise to generation of oscillations. It can
also be observed that after 5.06 s when the tire starts to touch the curb, the
longitudinal force begins to rise which is obviously due to the resistance
against tire movement. As the ground velocity is assumed to be constant, a
positive force is required to tackle the resistance enforced by the curb.
Once the tire passed the curb, as it can be seen, the force should be
returned to its preceding state. However, since this is a transient analysis it
takes a few milliseconds to reach its previous value.
CONCLUSION
This chapter presents the different aspect of the finite element

modeling of the radial tires under different loading and service conditions.
It is shown here that due to the complex behavior of the constituent
materials and also highly nonlinear behavior of the tire structure under
loads, very sophisticated computational techniques in conjunction with a
profound computer code must be employed. The results of various
modeling techniques including two-dimensional axisymmetric, three-
dimensional static and steady state rolling as well transient models are
given. These results were discussed and compared with experimental data
wherever possible to assess the applicability and accuracy of the models.
The limitations of the presented methods were also considered. Although
this topic has extensively been studied by many researchers, owing to the
complexities associated with the real behavior of radial tires in service,
further works need to be carried to narrow the gap between simulation

results and reality.
REFERENCES
[1] Mark, J. E.; Erman B; Eirich F. R. The Science and Technology of

Rubber; Third Edition; Academic Press: Burlington, MA, 2005.
[2] Bathe, K. J. Finite Element Procedures; Second Edition; Prentice
Hall: Watertown, MA, 2014.
[3] Gent, A. N.; Walter J. D. The Pneumatic Tire; U.S. Department of
Transportation, National Highway Traffic Safety Administration
(NHTSA): Washington DC, 2006.
[4] Clark, S. K. Mechanics of Pneumatic Tires; U.S. Department of
(NHTSA): Washington DC, 1981.
[5] Ghoreishy, M. H. R. Iran Polym. J. 2008, 17, 571-597.
[6] Mohsenimanesh, A.; Ward, S.; Gilchrist, M. Mater. Des. 2009, 30,
1124-1132.
[7] Lee, J. H. J. Terramech. 2011, 48, 171-197.
[8] Xia, K. J. Terramech. 2011, 48, 113-123.
[9] Xia, K.; Yang Y. Int. J. Numer. Anal. Meth. Geomech. 2012, 36, 498-
516.
[10] Zhao, C.; Zang M. J. Terramech. 2014, 55, 29-37.
[11] Zhao, C.; Zang M. J. Terramech. 2017, 72, 27-38.
[12] Ghoreishy, M. H. R. Tire Technology International Annual Review
2009, 54-57.
2013, 84-86.
[14] Bueche, F. J. Appl. Polym. Sci. 1961, 5, 271-281.
[15] Mullins, L. Rubber Chem. Technol. 1969, 42, 339-362.
[16] Ogden, R. W.; Roxburgh, D. G. Proc. R. Soc. Lond. A., 1988, 455,
2861-2877.
[17] Dorfmann, A.; Ogden, R. W. Int. J. Solids Struct. 2004, 41, 1855-
1878.
2014, 52-56.
[19] Rafei, M.; Ghoreishy, M. H. R.; Naderi, G. Math. Comput. Simulat.
2017 (in press).
[20] Ghoreishy, M. H. R.; Naderi, G.; Rafei, M. Iran. J. Polym. Sci.
Technol. (Persian) 2016, 29, 139-155.
[21] Ghoreishy, M. H. R. In Computational Materials; Oster, W. U.; Ed.;
Nova Science Publisher: New York, 2009; pp 445-478.
[22] Ghoreishy, M. H. R. Iran Polym. J. 2016, 25, 89-109.
[23] Ghoreishy, M. H. R.; Naderi, G. Iran Polym. J. 2005, 14, 735-743.
[24] Abaqus Analysis User's Guide. 2016, Simulia Inc. p. Abaqus
Documentation.
[25] Ghoreishy, M. H. R. P. I. Mech. Eng. D-J Aut. 2006, 220, 713-721.
[26] Ghoreishy, M. H. R. Plast. Rubber Compos. 2006, 35, 83-90.
[27] Ghoreishy, M. H. R.; Firouzbakht, M.; Naderi, G. Mater. Des. 2014,
53, 457-465.
[28] Ghoreishy, M. H. R. Mater. Des. 2012, 35, 791-797.
[29] Muhr, A. Rubber Chem. Technol. 2005, 78, 391-425.
[30] Ghoreishy, M. H. R.; Naderi, G.; Roohandeh, B. Iran Polym. J. 2015,
24, 1015-1024.
[31] Gent, A. N. In Engineering with Rubber; Gent A. N.; Ed.; Third
Edition; Carl Hanser Verlag: Munich, 2011; pp 37-88.
[32] Yeoh, O. H., In Engineering with Rubber; Gent A. N.; Ed.; Third
Edition; Carl Hanser Verlag: Munich, 2011; pp 345-364.
[33] Yeoh, O. H. Rubber Chem. Technol., 1993, 66, 754-771.
[34] Yeoh, O. H. Rubber Chem. Technol., 1990, 63, 792-805.
[35] Arruda, E. M.; Boyce, M. C. J. Mech. Phys. Solids 1993, 41, 389-
412.
[36] Ogden, R. W. Proc. R. Soc. Lond. A. 1972, 326, 565-584.
[37] Seibert, D. J.; Schöche, N. Rubber Chem. Technol. 2000, 73, 366-
384.
[38] Marlow, R. S. In Constitutive Models for Rubber III; Busfield, J.;

Muhr, A. H.; Ed.; Swets & Zeitlinger: Lisse, The Netherland, 2003;
pp 157-160.
[39] Shahzad, M.; Kamran, A.; Siddiqui, M. Z.; Farhan, M. Mater. Res.
2015, 18, 918-924.
[40] Sommer, J. G.; Yeoh, O.H. In Engineering with Rubber; Gent A. N.;
Ed.; Third Edition; Carl Hanser Verlag: Munich, 2011; pp 365-421.
[41] Gent, A. N. Engineering with Rubber; Third Edition; Carl Hanser
Verlag: Munich, 2011.
[42] Drucker, D. C., On the Postulate of Stability of Material in the
Mechanics of Continua; Brown Univ Providence RI., 1963.
[43] Ghoreishy, M. H. R.; Alimardani, M. R.; Zafar Mehrabian, R.;
Taghvaei Gangali, S. J. Appl. Polym. Sci., 2013, 128, 1725-1731.
[44] Hurtado, J.A.; Lapczyk, I.; Govindarajan, S. M. In Constitutive
Models for Rubber VIII; Alonso, A.; Gil-Negrete, N.; CRC Press:
London, Croydon, 2013; pp 95-100.
[45] Ghoreishy, M. H. R. Iran. Polym. J. 2009, 18, 641-650.
[46] Ghoreishy, M. H. R.; Soltani, S.; Etemadi, A. H. Iran. J. Polym. Sci.
Technol. (Persian) 2010, 23, 247-254.
[47] Gall, R.; Tabaddor, F.; Robbins, D.; Majors, P.; Sheperd, W.;
Johnson, S. Tire Sci. Technol. 1995, 23, 175-188.
[48] Ghoreishy, M. H. R. Iran. Polym. J. 2006, 15, 667-674.
[49] Ghoreishy, M. H. R.; Malekzadeh, M.; Rahimi, H. Iran. Polym. J.
2007, 16, 539-548.
[50] Ghoreishy, M. H. R.; Malekzadeh, M.; Rahimi, H. Tire Technology
International Annual Review 2007, 66-69.
[51] van Eldik Thieme, H.; Dijks, A. J.; Bobo, S. In Mechanics of
Pneumatic Tires; Clarck, S. K.; Ed.; U.S. Department of
NHTSA: Washington D.C., 1981; pp 541-720.
[52] Pottinger, M. G. In The Pneumatic Tire; Gent, A. N.; Walter, J. D.;
Ed.; National Highway Traffic Safety, Administration (NHTSA):
Washington D.C., 2006; pp 286-363.
[53] Blundell, M.; Harty, D. The Multibody Systems Approach to Vehicle

Dynamics; Elsevier Butterworth-Heinemann: Oxford, Burlington,
Mass., 2004.
[54] Archard, J. J. Appl Phys. 1953, 24, 981-988.
[55] Ghoreishy, M. H. R., Tire Technology International Annual Review
2017, 32-34.
Chapter 2
COMPUTER SIMULATIONS AS A SOLUTION

METHOD FOR DIFFERENTIAL GAMES
Guennady A. Ougolnitsky† and Anatoly B. Usov

J. I. Vorovich Institute of Mathematics,
Mechanics and Computer Sciences
Southern Federal University, Rostov-on-Don, Russian Federation
ABSTRACT
This paper is dedicated to computer simulations as a solution method

for differential games that are difficult for analytical treatment. The main
idea is that a very small number of scenarios, including both control
variables and parameters, can provide a fairly good qualitative
representation of modeled system dynamics. This idea is applicable to the
modeling of organizational, environmental, and socioeconomic systems
in which qualitatively representative scenarios have a clear interpretation.
The proposed computer simulation method is illustrated by two examples
that model fishing in shallow-water ecosystems and social partnership in
a continuous education system.

This work was supported by the Russian Science Foundation, project no. 17-19-01038.
†
Corresponding Author Email: gaugolnickiy@sfedu.ru.
64 Guennady A. Ougolnitsky and Anatoly B. Usov
1. INTRODUCTION
It is rather clear that most of the applied models describing complex

real-world systems using non-linear dynamics are analytically intractable.
This applies particularly to difficult differential games. Two principal
approaches to obtain their approximate solutions are numerical methods
and simulation modeling.
The numerical solution of optimal control problems was discussed in
many papers (Krasovskii and Subbotin, 1988; Judd, 1998; Stockey et al.,
1989; Pytlak, 1999; Powell, 2007). The first numerical method based on
Pontryagin’s maximum principle was the method of successive
approximations developed by Krylov and Chernousko (Chernousko and
Lyubushin, 1982). For a specific class of optimal control problems without
phase constraints, efficient algorithms satisfying the necessary conditions
of optimality were built. Most of them are based on gradient optimization
methods and Pontryagin’s maximum principle and have a simple software
implementation. Under phase constraints, the method of penalty functions
and modified Lagrange functions are used in simple cases. This approach
requires a multiple solution of a given optimization problem for different
values of the input parameters, and the resulting solution may have a rather
small accuracy. Besides, indirect methods can be used for solving optimal
control problems with phase constraints. In this case, for a system of
differential equations reflecting the necessary conditions of optimality, the
boundary-value problems are formulated and then solved numerically.
Optimal solution construction based on Pontryagin’s maximum principle is
reduced to the integration of a system of differential equations that model
the real system dynamics and the conjugate system for auxiliary functions.
The systems of differential equations are integrated numerically. Transition
to a discrete analog of Pontryagin’s maximum principle is performed using
the methods of value iterations or policy iterations (Guibas et al., 1983;
Judd, 1998; Jank and Kun, 2002; Stockey et al., 1989; Pytlak, 1999;
Powell, 2007).
Numerical solution methods for hierarchical optimal control problems
are less developed. First, iterative aggregation can be mentioned here
Computer Simulations as a Solution Method … 65
(Krasovskii and Subbotin, 1988). Second, the optimal strategies in

Stackelberg games can be designed as the piecewise continuous control
functions maximizing the Leader’s functional along the system trajectories,
with the end point of the trajectory corresponding to the maximal value of
an antagonistic game (Krasovskii and Subbotin, 1988). Finally, for
hierarchical systems an optimal control problem in general form is often
posed as a static optimization problem. In this case, the system under
consideration is assumed to operate on large time intervals so that the
system dynamics can be neglected. In this context, it is necessary to
discriminate between iterative and non-iterative algorithms. In the former
case (the Dantzig–Wolf algorithms, the Kornai–Liptak algorithms, the
Lagrange method of multipliers, and others), an optimal solution is yielded
by an iterative information exchange between Leader and Followers; at
each step of the iterative process, the locally optimal problems are solved
for the players. In the latter case, a one-step problem is solved and a Pareto
set is constructed (Aliev and Liberzon, 1987). Wilson (1979) proposed a
method to decompose a general control problem into subproblems.
The main idea developed in (Kononenko, 1977; Gorelov and
Kononenko, 2015) is an implementation of a two-step approach as follows.
First, hierarchical differential games are treated as minimax calculation
problems with bound variables in complex functional spaces. Second,
these problems are reduced to minimax calculation in original spaces of
smaller complexity.
Simulation modeling (Law and Kelton, 2000; Nelson, 2016) can be
viewed as another generalized numerical method. According to this
approach, a set of scenarios is constructed, and each scenario contains
strategies of all players. Note that an implementation of feedback strategies
in the simulation mode is a non-trivial problem. Under known strategies, it
is possible to calculate the corresponding trajectories of the controlled
dynamical system and the players’ payoffs, comparing them for different
scenarios. Thus, we find an approximate solution of the game on the set of
scenarios. Similar methods were developed in agent-oriented modeling
(Szilagyi, 2007).
The main problem of simulation modeling is that brute force methods

(an exhaustive search of all possible scenarios) are not implementable in
practice. There exist many techniques to reduce complete enumeration and
build incomplete factor plans (Kleijnen, 2007; Law and Kelton, 2000;
Nance and Sargent, 2002; Sargent, 2013).
This paper is dedicated to an original method of qualitatively
representative scenarios developed by the authors, which proceeds from
the following idea. For a majority of applications-relevant dynamical
models of complex real-world socioeconomic systems (in differential
games as well), it is possible to choose a very small number of scenarios
that give a fairly good picture of qualitatively different ways of system
development. For example, in models of corruption, the “kickbacks” of 10,
15, and 20% are different in principle; at the same time, there is no
essential difference between 10 and 10.5%. In the economics of public
goods, for an agent a matter of principle is whether to allocate 10, 20 or
30% percent of his budget to the production of public good. However, the
difference between 20 and 21% or 28 and 30% is crucial. We give a
mathematical formalization of this idea and also validate it numerically.
The remainder of the paper is organized as follows. Section 2 is
dedicated to the method of qualitatively representative scenarios and its
theoretical validation. In Section 3, we consider two specific applied
models, namely, a model of fishery in shallow-water ecosystems (Section
3.1), and a model of social partnership in a continuous education system
(Section 3.2). In both cases, the method of qualitatively representative
scenarios is applied and checked numerically. Section 4 gives some
concluding remarks.
2. METHOD OF QUALITATIVELY REPRESENTATIVE

SCENARIOS AND ITS VALIDATION
Let us illustrate the method of qualitatively representative scenarios

(QRSs) using an example of a hierarchically organized control system
within the game-theoretic approach (Basar and Olsder, 1999). The control
system includes one agent at the upper control level and several agents at
the middle and lower levels. The relations between the agents have
hierarchical organization. The upper-level agent exerts an influence on the
middle-level agents while the latter, in turn, have an impact on the lower-
level agents. The lower-level agents exert a direct influence on the state of
a controlled dynamical system. As a rule, the upper-level agent has no
direct impact on the lower-level agents and on the state of the controlled
system.
Such control systems are studied using the “principal - supervisor -
agent - object” hierarchical chain in different modifications within the
framework of optimal control theory and the theory of differential games.
At the beginning, the principal chooses his strategy (moves first) and
reports it to the supervisor and the agents. He maximizes his payoff
functional taking into consideration their best responses on the set of
strategies that support some conditions of sustainable development
(Ougolnitsky, 2011, 2014, 2015). The supervisor chooses her strategy
when the principal’s choice is already known but the agents’ strategies are
not fixed. She seeks to maximize her payoff functional with no regard to
the conditions of sustainable development. Any agent chooses his strategy
when the strategies at the upper levels are fixed, seeking to maximize his
payoff functional only.
In the special case of an indifferent principal (with a constant payoff
functional), we consider a simplified two-level hierarchical control system
containing one supervisor and several agents.
Suppose there are n agents. Denote by vi (t ), ui (t ) the controls of the
supervisor and agents, respectively (i  1,2,..., n) ; v(t )  (v1(t ), v2 (t )..., vn (t )) ,
u(t )  (u1(t ), u2 (t ),..., un (t )) are the control variable vectors of the players.
The payoff functionals have the following form:
- for the supervisor,

 e  T S0 ( x(T ))  max ;
T
J 0 (v(), u ()) 

0
e  t G0 (v(t ), u (t ), x(t ))dt (1)
- for the agents (i  1,2,..., n) ,
 e  T Si ( x(T ))  max .
J i (v(), u ()) 

0
e  t Gi (v(t ), ui (t ), x(t ))dt (2)
Here the notations are as follows: T as a period of consideration; t as

a current instant of time;  as a discount factor; Gi (v(t ), u(t ), x(t ))
(i  0,1,..., n) as the current payoff functions of the players; Si ( x(T )) as
their scrap value functions; finally, x(t ) as the state variable of the system
(possibly, a vector). The constraints imposed on the controls have the form
ui Ui ; vi Vi ; i  1,2,..., n , (3)
where Ui ,Vi ( i  1, 2, ..., n) are fixed sets of feasible controls.

The state dynamics with an initial condition are given by
dx
= f ( x(t ), u (t )), x( 0 ) = x0 . (4)
dt
Here f is a continuous vector-function.

If the system is analyzed from the agents’ viewpoint, then problem
(2)–(4) is considered (i  1,2,..., n) . In this case, all players are equal and
we obtain a non-antagonistic game of n persons with Nash equilibrium as
the solution concept. The functions v1(t ), v2 (t )..., vn (t ) are given.
If the system is analyzed from the supervisor’s viewpoint, then
problem (1)–(4) is considered in in a game-theoretic setup. In this case, the
solution depends on a chosen informational structure. Namely, Stackelberg

games and inverse Stackelberg games in open-loop and Markovian
strategies arise naturally. Compulsion and impulsion as hierarchical control
methods are considered and formalized, see (Ougolnitsky, 2011, 2014).
In the dynamical setup, both problems (Nash or Stackelberg
equilibrium design) are analytically intractable in the general case. Their
investigation is based on Pontryagin’s maximum principle or the
Hamilton–Jacobi–Bellman equations but the analytical solution can be
constructed only for specific model functions. The numerical construction
of the set of Nash equilibria essentially increases the computational
complexity of the algorithms for a large number of players. A numerical
investigation of differential games in hierarchical setups by Pontryagin’s
maximum principle entails initial-value problems with the supervisor’s
controls as the function-type parameters. Their solution is quite difficult.
Therefore, simulation modeling seems to be a most promising approach.
The complete enumeration of all feasible strategies is impossible, and
search reduction to an observable set composed of relatively many
scenarios becomes topical.
According to the aforesaid, we suggest a method of qualitatively
representative scenarios (QRSs). The idea of this method consists in the
following. In applied models of complex real-world socioeconomic
systems, it is possible to choose a very small number of scenarios among
all potential ones that represent all qualitatively different trajectories of the
system dynamics. These scenarios differ in principle while all the other
scenarios yield no essentially new results.
First, describe this method for a hierarchical setup of problem (1)–(4)
with Stackelberg equilibrium design.
Denote by   V1  ...  Vn  U1  ...  U n the set of game outcomes
and by J i :   R the payoff functional of player i  N  0,1,2,..., n .

The method proceeds from the assumption Ui  Ui , Vi  Vi QRS
QRS
i  1,2,..., n , where sets U iQRS and ViQRS contain the qualitatively

representative controls (strategies) of agent i and supervisor with respect
to agent i , respectively. In addition, suppose that the cardinality of the sets

ViQRS and UiQRS (i  1,2, .., n) is small enough so that they contain K
elements for each set, i.e., | U i
QRS
|| ViQRS | K , where K = 3 or 4. Then
V1QRS  ... VnQRS U1QRS  ... U nQRS  QRS is the QRS-set of the
n
game. This set includes m | QRS |  Vi U iQRS  K 2n elements.
QRS
i 1
Each representative scenario of the game (v, u)

(k )
 QRS , k  1, 2,.., m ,
has the form
(v, u)(k )  (v1(k ) , v2(k ) ,..., vn(k ) , u1(k ) , u2(k ) ,..., un(k ) ); vi(k ) ViQRS; ui(k ) UiQRS ; i  1,2,.., n.
To be more precise, we give a rigorous definition.


Definition. A set QRS  (v, u)(1) , (v, u)( 2) ,..., (v, u)( m) is called the 
QRS-set of the hierarchical game with a precision  if the following two
conditions hold:
(a) for any two elements of the set (v, u)(i ) , (v, u)( j )  QRS ,
| J 0(i )  J 0( j ) | / J 0max   ;
(b) for any other element (v, u )(l )  QRS , there exists an element
(v, u )( j )  QRS such that | J 0

(l )
 J 0( j ) | / J 0max   .
Here J 0(i ) , J 0( j ) , J 0(l ) are the corresponding payoffs of the supervisor

(in the sense of (1)); J 0  J 0 (v1 , v2 ,..., vn , u1 , u2 ,..., un ), s  i, j, l;
(s) (s) ( s) ( s) ( s) ( s) ( s)
J 0max is the global maximum of the supervisor’s payoff functional

(1) on the QRS-set of the game over the 2n functions
vi  Vi QRS , ui  U iQRS ; i  1,2,.., n ;   0 is a constant that specifies

precision. Thus, QRSs lead to an essential difference between the
supervisor’s payoffs; meanwhile, the difference between one of the QRSs
and any other scenario is negligible in this sense. For simplicity, we will
omit “with precision  ” in this definition.
For a non-antagonistic game of n agents, Nash equilibria are found. In
this case, the QRS method is also applicable.
Denote by 1  U1  ...  U n the set of game outcomes and by
Ji : 1  R the payoff functional of player i  1,2,..., n. Again, assume

Ui  UiQRS i  1,2,..., n , where the sets U iQRS contain the QRSs of
agent i . Their cardinality is | UiQRS | K , K = 3 or 4. Then
U1QRS ...  U nQRS  QRS gives a QRS-set of the game, which includes
n
m | QRS |  U iQRS  K n elements. In this case, each representative
i 1
scenario of the game u ( k )  QRS , k  1, 2,.., m, has the form

u ( k )  ( u1( k ) , u2( k ) ,..., un( k ) ), ui( k ) UiQRS , i  1, 2,.., n.
Below we employ the QRS method for solving two problems, namely,
Nash equilibrium design in pure strategies for a differential game and
Stackelberg equilibrium design for a hierarchical differential game of a
supervisor and several agents.
2.1. Nash Equilibrium Design by the QRS Method
First, let us analyze system (1)–(4) for equal independent agents

without a supervisor. In this case, a differential non-antagonistic n-person
game of form (2)–(4) arises naturally, with the Nash equilibrium as the
solution concept. The functions v1(t ), v2 (t )..., vn (t ) are supposed to be
given, vi  vi(0) (t ) ViQRS (i  1,2,.., n) .

For each agent, let a QRS-set contain K elements, i.e.,
 
UiQRS  ui(1) , ui(2) ,..., ui( K ) ; | U iQRS | K ; i  1, 2, .., n .
n
For a given supervisor’s scenario, there are m | QRS |  U iQRS  K n
i 1
qualitatively different scenarios in the game.
We propose the following algorithm to design the set of Nash
equilibria in pure strategies on the QRS -set (denoted by NE (QRS ) ).
All input functions and parameters of model (2)–(4) are given. Set
i 1.
For each agent i , construct the set of his individually non-improvable
strategies Nei by
Nei (v(0) )  (v1(0) , v2(0) ,..., vn(0) , u1 , u2 ,..., ui 1, ui(*) , ui 1,..., un ) :
v(j0) V jQRS ; u j U QRS

j ; j  i; j  1,2,..., n; ui(*) U iQRS :
(v1(0) , v2(0) ,..., vn(0) , u1 , u2 ,..., ui 1, ui(*), ui 1,..., un ) 

arg max J i (v1(0) , v2(0) ,..., vn(0) , u1 , u2 ,..., ui 1 , ui(l ) , ui 1 ,..., un ).
1l  K
Calculate the set of Nash equilibria on the QRS-set of the game as

n
NE (QRS )   Nei (v(0) ) .
i 1
The sets Nei (i  1,2,..., n) are obtained by the complete enumeration

of the QRS -set of the game for a fixed strategy of player i. Note that
n
 Ui
QRS
 K n 1 operations are required for checking of all QRSs of
i 1
j i
the game under a fixed strategy of player i . If the number of players is not
too large (for example, n  10 , K  3 or 4 ), then the calculations are
technically feasible, which gives a base for the proposed method. If there
exists a Nash equilibrium in pure strategies in model (2)-(4) but is not
yielded by the proposed algorithm (i.e., NE (QRS )   ), then it is
necessary to update the QRS-set of the game by adding new scenarios for
each player and return to Step 2 of the algorithm for i  1 . For a successful
choice of the QRS-set of the game, the Nash equilibrium will be calculated
by a finite number of iterations.
2.2. Stackelberg Equilibrium Design by the QRS Method
Now, let us study system (1)–(4) from the supervisor’s viewpoint. The
informational structure corresponds to a Stackelberg game, and the
supervisor uses impulsion (Ougolnitsky, 2011, 2014). In this case, the
supervisor’s controls are unknown and have to be found.
Again, for each agent and for the supervisor with respect to each agent,
let a QRS set contain K elements, i.e.,
   
U iQRS  ui(1) , ui(2) ,..., ui( K ) ; ViQRS  vi(1) , vi(2) ,..., vi( K ) ;
| UiQRS || ViQRS | K ; i  1, 2, .., n .
n n
Then there are m | QRS |  ViQRS  U iQRS  K 2n different
i 1 i 1
scenarios (sets of control variables) of the game. The supervisor
n
has  Vi
QRS
 K n different strategies of the form
i 1
v( k )  (v1( k ) , v2( k ) ,..., vn( k ) ), vi( k ) Vi QRS , i  1, 2,..., n, k  1, 2,..., K N ; each
agent has U iQRS  K strategies of the form ui(1) , ui(2) ,..., ui( K )
( i  1, 2, .., n ).
We propose the following algorithm to design the set of Stackelberg
equilibria on a QRS-set.
1. All input functions and parameters of model (1)–(4) are given. Set
k 1.
2. A current ( k th) supervisor’s strategy is fixed, i.e.,
v(k )  (v1(k ) , v2(k ) ,..., vn(k ) ); vi(k ) ViQRS; i  1,2,..., n .
3. For the given supervisor’s strategy, calculate the Nash equilibria
n
NE (QRS )   Nei (v( k ) ) by the complete enumeration according
i 1
to the algorithm described in section 2.1.
Find the value max J 0 (v ( k ) , u ) and compare it with the current

n
(k )
 Nei (v )
i 1
maximum of the supervisor’s payoff functional (1). Save the greater value
of (1) and the corresponding set of controls.
If there exist unexamined strategies of the supervisor, then go to the
next strategy (k : k  1 ) and return to Step 2.
n
When all feasible strategies of the supervisor ( K strategies) are
examined, the strategies of the supervisor and agents that maximize (1) or
form the set arg max max J 0 (v ( k ) , u ) , constitute the
v (k )
;vi( k ) ViQRS n Ne ( v ( k ) )
 i
i 1
Stackelberg equilibrium (v (*) , u (*) (v (*) )) in model (1)-(4) on the QRS-set

of the game.).
The screening of all feasible supervisor’s strategies at Steps 3–5 of the

algorithm requires an exhaustive search of the whole QRS-set and a
2n
comparison of K different scenarios of the game. If the number of
players is not very large ( n  5,6,...,10 , K  3 or 4 ), then the
procedure is technically feasible.
Now, consider the usage of the proposed algorithms for solving two
applied models.
3. APPLICATIONS
In this section, two applied models are considered, namely, a model of

fishery in shallow-water ecosystems (subsection 3.1), and a model of social
partnership in a continuous education system (subsection 3.2). In both
cases, the QRS method is used and validated numerically.
3.1. Modeling of Shallow-water Ecosystems
Let us study a problem of fishing in a shallow waterbody (Chistyakov

et al., 2015; Sukhinov et al., 2017). For this model, we construct a
hierarchical differential game of two players–the supervisor, or principal (a
governmental regulation body of fishing operations) and the agent, or
follower (a fishing company). Both control agents seek to maximize their
payoff functionals but the main objective of the supervisor is to maintain
the controlled ecosystem in a sustainable state (Ougolnitsky, 2015). The
informational structure corresponds to a Stackelberg game with
compulsion as a method of hierarchical control.
The ecosystem dynamics is described by the logistic Verhulst–Pearl
model with consideration of fishing. It has the form
dP P
 dP(1  )  w(t ) P; P(0)  p0 . (5)
dt R
Here d , R  const  0; R is the ecosystem’s carrying capacity for fish;

the constant d characterizes fish reproduction; t denotes a current instant
of time; P  P(t ) describes the fish biomass at the instant t ; p0 gives an
initial fish biomass; finally, w(t ) is the share of fish catch (the agent’s
control variable).
The fish biomass is calculated according to (5) while it takes positive
values. If the fish biomass becomes equal to zero, then calculations stop.
The payoff functionals are defined as follows:
J L (q(), w()) 

0
e  t [ H ( P(t )  P 0 (t )) 2  C (q(t ), w(t ))]dt  min ; (6)
- for the agent,
J F (q(), w()) 

0
e  t [a(1  s(t ))w(t )  0.5bw2 (t )]P(t )dt  max . (7)
The constraints imposed on the control variables are
0  q(t )  1; 0  t  T ; (8)
- for the agent,
0  w(t )  q(t ); 0  t  T . (9)

Here we use the following notations:  as a discount factor; T as a

period of consideration (game duration); q(t ) as a catching quota (the
compulsion control variable of the supervisor); C (q(t ), w(t )) as a convex
cost function of quota control; s(t ) as the penalty coefficient for
overfishing (a given parameter in the case of compulsion); a as the unit
price of fish biomass; b as the fishing cost coefficient; P* (t ) as an
optimal value of fish biomass in terms of sustainable development; H as
the penalty coefficient applied to the supervisor if the current biomass
deviates from the optimal value.
Thus, in this model compulsion is to assign fishing quotas, which
incurs costs. The condition of sustainable development of this ecosystem
has the form P(t )  P* (t ) for any t or, in a weaker form,
P(t )  P* (t )   ,   1, for any t . If the supervisor violates this
condition, the penalty is applied with the coefficient H  1 .
Thus, model (5)–(9) represents a differential Stackelberg game.
In any real-world system, the agents’ controls remain invariable for a
period of time due to natural inertia. Without loss of generality, assume the
strategies of all agents are constant on the same time intervals (Ugol'nitskii
and Usov, 2013), i.e.:
 b1 , if 0  t  t1 ,
 b , if t1  t  t 2 ,

b(t )   2 (10)
 ...
 bM , if t M 1  t  T ,
where bi are constant controls of the supervisor or agents for t [ti 1, ti ) ;
ti  i t; t  T / M ; M is the number of time intervals with constant
controls. Thus, the control of each agent becomes a grid function, for
example, w(t )  wi iM1 , where wi  const is the agent’s control for
t [ti 1, ti ) . Then, using (6), (7) and (10), we transform the payoff
functionals into the payoff functions as follows:
M tk
J L (wi iM1, qi iM1)   (  e t H ( P(t )  P0 (t ))2dt 
k 1 t k 1
1
C (qk , wk ) (e t k 1  e t k )) ; (11)

- for the agent,
tk

M
J F (wi i 1 , qi i 1 ) 

(a(1  sk ) wk  0.5bwk2 ) e  t P(t )dt .
M M
(12)
k 1 tk 1
The constraints imposed on the control variables are written as
0  qk  1; k  1,2,..., M ; (13)
- for the agent,
0  wk  qk ; k  1,2,..., M . (14)
Here we denote f k  f (kt ) . Then problem (11)–(14), (5) can be
solved and the grid functions w  wi iM1 and q  qi iM1 can be
calculated according to the Stackelberg informational structure.
The following algorithm of Stackelberg equilibrium design in problem

(11)-(14), (5) was proposed in (Ougolnitsky and Usov, 2016).
1. Solve the agent’s problem (12), (14), (5) as a parametrical non-

linear optimization problem. As a result, find the agent’s control as

a function of the supervisor’s control, i.e., wi* (qi ) i 1 . 
M
 
M
2. Substitute the agent’s controls wi* (qi ) i 1 into (5), (11). Solve

problem (11), (13), (5) and denote the resulting solution by qi* i 1
M
3. Construct the Stackelberg equilibrium as  q  , w (q )  .

* M
i i 1
*
i
* M
i i 1
 
Let us calculate the Stackelberg equilibrium using the QRS method. It

is necessary to find 2M numbers wi and qi , i  1,..., M , that determine the
grid functions w  wi iM1 and q  qi iM1 . Introduce the sets
Qi  qi : 0  qi  1 , Wi  wi : 0  wi  qi  and the QRS-set of the game

QRS  QRSL  QRSF  Q1QRS  Q2QRS  ...  QM
QRS
 W1QRS  W2QRS  ...  WMQRS ,
where Qi
QRS
 Qi , Wi QRS  Wi , i  1, 2,..., M .
Suppose that, both for the supervisor q  qi iM1 and for each agent
w  wi iM1 , i  1,2,..., M , the QRS-set contains K+1 element at any

fixed instant. The agent’s controls depend on the chosen supervisor’s
control, i.e.:
 1 2 K 1 
QiQRS  0; ; ; ...; ;1  and
 K K K 
 q 2q K 1 
Wi QRS  0; i ; i ;...; qi ; qi  , i  1, 2,..., M . (15)
 K K K 
In this case, for each player there are ( K  1) M possible control
scenarios, i.e., | QRSL || QRSF | ( K  1)M . The game involves two
players and hence the QRS-set of the game includes ( K  1)2 M elements,
or | QRS | ( K  1)2M .
In real-world systems, the numbers K and M are not large (not
exceeding 3 or 4), which allows a numerical implementation of the
approach.
We propose the following numerical algorithm of Stackelberg
equilibrium design in model (11)–(14), (5) on the QRS-set is proposed.
1. Fix a current representative strategy q  QRS L of the supervisor.

2. Perform an exhaustive search of all representative strategies
w QRSF of the agent to find his best response
w0  w0 (q)  QRSF to the supervisor’s strategy q .
3. Check whether the current pair of strategies (q, w0 (q)) is the best
for the supervisor on his QRS-set in the sense of (11). If so, save it.
If some feasible strategies of the supervisor are still unscreened,
return to Step 1; otherwise, go to Step 4.
4. The pair (q*, w0 (q*)) saved at Step 3, which is the best for the
supervisor on the QRS-set 2in the sense of (11), forms the
Stackelberg solution in model (11)–(14), (5) on this QRS-set.
Below we present the results of some calculations with this algorithm

for typical test data.
For each scenario of the game, equation (5) was solved numerically by
the finite differences method with an implicit difference scheme that has
the first order of approximation in time. The calculations were performed
qw
for a model penalty function C (q, w)  C1 , C1  const,   0.001,
( w   )(q   )
that possesses all the required properties; here C1 denotes a scaling

coefficient while the number  is introduced to satisfy the condition

C (0,0)  0 .
First, we constructed a QRS-set in the case M  K  2 :
qi  0 , then WiQRS  0 ; if qi 
 1  1
Q1QRS  Q2QRS  0; ; 1 . If , then
 2  2
 1 1  1 
WiQRS  0; ;  ; finally, if qi  1 , then WiQRS  0; ;1  ( i  1,2 ).
 4 2  2 
The subscript i denotes an instant of time when the player chooses a
corresponding strategy (for M  2 , i  1,2 ).
Considering all possible combinations of elements from the sets
Q1QRS and W1QRS and W2QRS and Q2QRS as well, we found 7 scenarios for
the first instant of time when the players chose their strategies, i.e., for
i  1,
 
Q1QRS  W1QRS  (q1 , w1 ) k k 1  (0,0); ( ,0); ( , ); ( , ); (1,0); (1, ); (1,1) ,
7 1 1 1 1 1 1
 2 2 4 2 2 2 
and 7 variants for the second instant of time, i.e., for i  2 ,
 
Q2QRS  W2QRS  (q2 , w2 ) k k 1  (0,0); ( ,0); ( , ); ( , ); (1,0); (1, ); (1,1) .
7 1 1 1 1 1 1
 2 2 4 2 2 2 
In the final analysis, all possible combinations of elements from the

sets Q1
QRS
 W1QRS and Q2QRS  W2QRS gave 49 different scenarios of the
game, that is,
QRS  Q1QRS  Q2QRS  W1QRS  W2QRS ; | QRS | 49 .
For these 49 scenarios of the game, we checked the first condition in

the definition of a QRS-set.
Example 3.1.1. The numerical calculations in the case

M K 2 were performed with the following input data:
a  2 000 conditional units (c.u.); b  20 000 c.u.; C1  0.1 c.u.;
P 0 (t )  0.09 mg/cm 3 ; H  600 (thousands of c.u. / cm ) / mg ;
2 2
  0.01;
T  365 days ; p0  0.05 mg/cm ; s(t )  0.5; R  0.1; d  50; and   0.001 .
3
Here conditional units measure cost or profit.

Table 3.1.1 illustrates part of the numerical calculations with the input
data of Example 3.1.1.
Table 3.1.1.
min | J 0  J 0 | / J 0
(k ) ( j) min
Strategies (q, w)( j ) J L( j ) (q0 , q1 , w0 , w1 )
1 k  49
(c.u.) k j
q0 q1 w0 w1
0 0 0 0 22.0 0.1
0.5 0.5 0.5 0.5 17.7 0.03
0.5 0.5 0.25 0.25 19.8 0.1
0.5 0.5 0 0 31 0.32
1 1 0 0 58 1
1 1 0.5 0.5 17.87 0.05
1 1 1 1 14 0.3
0 1 0 1 21.8 0.1
0 1 0 0.5 21.9 0.1
0 1 0 0 40.1 0
1 0 1 0 21.7 0.1
1 0 0 0
1 0 0.5 0 21.6 0.1
0.5 1 0 0 44.7 0
0.5 1 0.5 0.5 17.83 0.05
0.5 1 0.5 1 17.78 0.05
1 0.5 1 0.5 17.73 0.05
1 0.5 0.5 0.5 17.67 0.06
1 0.5 0 0.5 40.2 0.1
1 0.5 0.5 0 26.5 0.05
1 0.5 1 0 26.4 0.1
1 1 0.5 0 40.3 0.3
1 1 0 0.5 40 0.3
1 1 0.5 1 17.62 0.05
1 1 1 0.5 17.65 0.03
Here J 0min is the global minimum of the payoff functional (1) (payoff
function (11)), as the minimization problem was solved. The minimum was
searched in the control variables of the supervisor and agent
simultaneously (for (1) in the two functions q(t ) and w(t ) ; for (11), in the
two grid functions w  wi iM1 and q  qi iM1 ).
Table 3.1.2 illustrates part of the numerical calculations to check the
second condition from the definition of a QRS-set with precision  for the
hierarchical game with the input data of Example 3.1.1 and M  K  2 .
According to the results in Tables 3.1.1 and 3.1.2, the precision is
  0.03 . Thus, the resulting QRS-set satisfies the definition in the case
M  K  2 . Recall that the value K determines the number of QRSs for
each player at a fixed instant of time.
Table 3.1.2.
Strategies J L( j ) “Close” strategy J L(k ) | J 0(k )  J 0( j ) |
(q, w) ( j)
 QRS (c. u.) (q, w) (k )
 QRS (c.u.) / J 0min
q0( j ) ( j)
q1( j ) w0 w1( j )
(k )
q0( k ) q1 w0( k ) w1( k )
0.6 0.6 0.3 0.1 26 0.5 0.5 0.25 0 25.8 0.014
0.6 0.6 0.6 0.1 20.3 0.5 0.5 0.5 0 20.2 0.007
0.6 0.6 0.1 0.6 20.4 0.5 0.5 0 0.5 20.45 0.004
0.6 0.6 0 0 31.3 0.5 0.5 0 0 31 0.021
0.6 0.6 0.3 0.3 19.84 0.5 0.5 0.25 0.25 19.8 0.003
0.6 0.6 0.6 0.6 17.95 1 1 0.5 0.5 17.87 0.006
0.6 0.6 0.4 0.4 17.65 1 1 1 0.5 17.65 0
0.3 0.3 0.3 0.3 19.7 0.5 0.5 0.25 0.25 19.8 0.007
0.3 0.3 0.1 0.1 20.1 0.5 0.5 0.5 0 20.2 0.007
0.3 0.3 0.3 0.1 20.1 0.5 0.5 0.5 0 20.2 0.007
0.3 0.3 0.1 0.3 20.3 0.5 0.5 0 0.5 20.5 0.014
0.3 0.1 0.3 0.1 22.7 0.5 0 0.5 0 23 0.021
0.3 0.1 0.1 0.1 43.8 0.5 0 0 0 44 0.014
0.3 0.1 0 0 43.8 0.5 0 0 0 44 0.014
0.8 0.8 0.8 0.8 13.9 1 1 1 1 14 0.007
0.8 0.8 0.8 0.1 19.0 1 1 1 0 19.3 0.021
0.8 0.8 0.8 0.3 17.64 1 1 1 0.5 17.65 0.007
0.8 0.8 0.8 0.6 17.66 1 1 1 0.5 17.65 0.007
0.8 0.8 0.6 0.8 17.63 1 1 0.5 1 17.62 0.007
0.8 0.8 0.3 0.8 17.62 1 1 0.5 1 17.62 0
In Examples 3.1.2–3.1.7, we examined the changes in the supervisor’s

payoff in the Stackelberg equilibrium that are caused by varying the
number of QRSs for each player at any instant of time (i.e., by varying the
parameter K in (15)). More specifically, we increased the value K from
2 to 32 by doubling as shown below.
Example 3.1.2. The calculation results for the input data of Example
3.1.1 are given in Table 3.1.3. The value J L* describes the supervisor’s
payoff in the Stackelberg equilibrium for the corresponding value of the
parameter K . The value  was calculated by the formula   J L*  J L(0)
100%
,
J L(0)
where J L(0) is the value of the payoff function (11) for K  32 . The value
 shows the difference (in percentage terms) between the supervisor’s
payoffs in the Stackelberg equilibrium for K = 32 and for a current value
K.
Table 3.1.3.
K
J L* (c.u.)  (%)
2 21.9 2.8
4 21.5 0.9
8 21.5 0.9
16 21.4 0.46
32 21.3 0
Table 3.1.4.
K
J L* (c.u.)  (%)
2 788 0,63
4 785 0.25
8 785 0.25
16 783 0
32 783 0
Example 3.1.3. For the input data of Example 3.1.1 and

P (t )  0.04 mg/cm 3 , the results of calculations are given in Table 3.1.4.
0
Example 3.1.4. For the input data of Example 3.1.1 and

H  100(thousandsc.u.×cm2 )/ mg 2 , the results of calculations are given
in Table 3.1.5.
Table 3.1.5.
K
J L* (c.u.)  (%)
2 3.65 2.2
4 3.59 0,56
8 3.59 0.56
16 3.57 0
32 3.57 0
Example 3.1.5. For the input data of Example 3.1.1 and s(t )  0.1 , the
results of calculations are given in Table 3.1.6.
Table 3.1.6.
K
J L* (c.u.)  (%)
2 21.9 5,2
4 21.0 1,0
8 20.8 0
16 20.8 0
32 20.8 0
Example 3.1.6. For the input data of Example 3.1.1 and R  1 , the
Table 3.1.7.
K
J L* (c.u.)  (%)
2 181 312 0.55
4 180 815 0.27
8 180 567 0.14
16 180 315 0
32 180 315 0
Example 3.1.7. For the input data of Example 3.1.1 and d  5 , the
Table 3.1.8.
K
J L* (c.u.)  (%)
2 22 49.6
4 17.4 21
8 17 15.6
16 16.9 15.5
32 14.7 0
A direct analysis of the values  indicates that there is no essential

difference between two (K=2) and thirty two (K=32) representative
scenarios for the supervisor (in the sense of his payoff functional (1)) in the
Stackelberg equilibrium for a rather wide class of the input functions and
parameters. Therefore, for this model the QRS method has a quite good
performance of Stackelberg equilibrium design, and the values K=2 or
K=4 are sufficient.
3.2. Modeling of Social Partnership in a Continuous

Education System
Consider as the second extended example the following model of

social partnership in a continuous education system with three control
agents, namely, a university professor (В), an employer (Р), and a student
(С) (Dyachenko et al., 2015; Tarasenko et al., 2016). Assume all the agents
(players) are equal and make their decisions simultaneously and
independently. All the players seek to maximize their payoff functionals of
the form
T
J i   e t [ gi (ri  ui (t ))  si (t )c( x(t ))]dt  e T si (T )c( x(T ))  max, (16)
0
0  ui (t )  ri , i  B, P, C. (17)
Each agent i from the set N  {B, P, C} allocates his budget ri

between two directions as follows. A part ui (t ) (the agent’s control) is
assigned to improve the level of professional development of the students
while the residual ri  ui (t ) is used for a financial support of a private
activity. A current payoff of each player is the sum of two terms, namely,
an income from his private activity and a share in the social utility
determined by the students’ professional development. We adopt the
following notations in (16)–(17):  as a discount factor; t  [0, T ] as a
current instant of time; gi (z ) as a concave increasing function of the
variable z that reflects the income from the private activity of player i;
x(t ) as the state variable that describes the level of professional
development of the students; c(x) as a concave increasing function of the
variable x that gives a financial expression for the social utility determined
by the students’ professional development; finally, si (t ) as a share of
player i in this value.
The state dynamics are described by the equation
x  h( x(t ))  f (uB (t ), uP (t ), uC (t )), x(0)  x0 . (18)
Here h is a decreasing function of its argument, which means that the

level of professional development goes down without investments; f is
an increasing function of the players’ investments in the professional
development of students; x0 gives an initial value of the state variable.
For model (16)–(18), consider a simplified linear modification of the
form

(19)
J i (u ())  e  t [ki (ri  ui (t )) pi  si (t )cx (t )]dt  e  T si (T )cx (T )  max,
0
0  ui (t )  ri , i  N , (20)
x  ax(t )   biui (t ), x(0)  x0 . (21)

iN
In comparison with (16)–(18), model (19)–(21) involves the linear

functions c( x(t ))  cx(t ) ( c  const  0 ) and h( x(t ))  ax(t ) (a  const  0) ,
which makes it linear in the state variable (Dockner et al., 2000). For the
sake of simplicity, the linear function f (uВ (t ), uР (t ), uС (t ))   biui (t )
iN
is used in combination with the power function of income from the private
activity, i.e., gi (ri  ui (t ))  ki (ri  ui (t )) pi , ki  0, 0  pi  1, i  N ,
where bi  0 shows the contribution of player i ’s investments to the
improvement of the professional development of students. Note that such
an approach can be applied without an essential loss of generality.
By analogy with the previous subsection, assume the control strategies
of all agents are constant on the same time intervals (10). Then the strategy
of each player is his grid function ui (t )  uik k 1; i  B, P, C .   M
In this case, expressions (19), (20), and (10) can be transformed as

follows:
- the payoff functional is reduced to the payoff functions:
M T
J i (uik kM1)   (ki (rik  uik ) pi (e tk 1  e tk ) /    e t si (t )cx(t )dt 
k 1 0
e T si (T )cx(T )  max; iN ; (22)

- the constraints imposed on the control variables become
0  uik  rik , i  N ; k  1,2,..., M ; (23)
As before, fik  fi (kt ), i  N .

Model (21)–(23) is a differential game with the Nash equilibrium as a
solution concept. More specifically, it is necessary to find a triple of grid

  , u  , u 
functions  uВk
Ne M
k 1
Ne M
Рk k 1
Ne M 
Сk k 1 

such that none of the players
Ne
benefits by a unilateral deviation from the strategy uik   M
k 1 (i  N ) .
We will calculate Nash equilibria in game (21)–(23) using the QRS
method. Introduce a QRS-set of player i at the instant of time t  kt ,
assuming that it contains L+1 elements:
QRS 
 r 2r ( L  1)rik 

U ik  0; ik ; ik ; ... ; ; rik ; i  B, P, C; k  1,2,..., M .

 L L L 

Then the problem is to find 3M numbers

uik U QRS
ik ; i  B, P, C; k  1, 2,..., M .
In this case, for each player i there exist ( K  1) M scenarios:
QRSi  UiQRS  ...  UiM , | QRSi | ( K  1) , i  B, P, C.

QRS M
1
The game involves three players and hence the QRS-set of the game
contains ( K  1)3M scenarios, i.e., QRS  QRSB  QRSP  QRSS ; | QRS | ( K  1)3M .
exceeding 3 or 4), which allows a numerical implementation of the
approach.
We propose the following numerical algorithm to design the set of
Nash equilibria on the QRS-setin game (21)–(23).
1. All input functions and parameters of the model are given. Set
j  1.
2. Fix scenario j of the game, i.e.,
  , u  , u 
(uВ( j ) , uР( j ) , uС( j ) )   uВk

( j) M
k 1
( j) M
Рk k 1
( j) M 
Сk k 1 

.
3. By an exhaustive search of all strategies of the first player, check

( j)
whether it is beneficial for him to deviate from his strategy u B if
( j) ( j)
the other players still use their strategies u P and uC .
4. By an exhaustive search of all strategies of the second player,
check whether it is beneficial for him to deviate from his strategy
u P( j ) if the other players still use their strategies u B( j ) and uC( j ) .
5. By an exhaustive search of all strategies of the third player, check
( j)
whether it is beneficial for him to deviate from his strategy uC if
( j) ( j)
the other players still use their strategies u B and u P .
6. If at Steps 3–5 none of the players benefits by a unilateral
deviation from his strategy, then the Nash equilibrium on the QRS-
set is found, i.e.,
(uВ( j ) , uР( j ) , uС( j ) )  NE (QRS ) .
7. If some qualitatively representative scenarios of the game are still

screened, then return to Step 2 of the algorithm ( j  j  1 );
otherwise, go to Step 8.
8. The set of Nash equilibria NE (QRS ) of game (21)–(23) on the
QRS -set is constructed.
If NE (QRS )   , then it is required to change the QRS-set, e.g., by

doubling the step of search, with further consideration of only odd points,
i.e., L : 2 L :
 r 3r 5r 
QikQRS  0; ik ; ik ; ik ;...; rik ; i  B, P, C; k  1,2,..., M .
 L L L 
As soon as the QRS-set is updated, it is necessary to return to

Step 2 ( j  1 ).
During an exhaustive search of all strategies of the players at Steps 3–5
of the algorithm, the state equation (21) is solved numerically for each
qualitatively representative strategy, e.g., using the finite difference
method with explicit or implicit schemes.
For some input data, the results of numerical calculations using the
proposed algorithm are given below.
Example 3.2.1. Let M  2 ;   0.1; T  2 years ;
rB  100; rP  50; rC  30;
c  20c.u.; kB  8c.u.; kP  5c.u. ; kC  10c.u.;
PB  PP  PC  0.1; x0  420; bB  0.0025;
bP  0.006; bC  0.001; and sB  sP  sC  1/ 3 .
In this case, the algorithm yielded the Nash equilibrium
u NE
B u NE
P u NE
C  (0, 0) on the QRS-set for L  2 .
Example 3.2.2.For the input data of Example 3.2.1, let
kB  0.5c.u., kP  0.1c.u., kC  0.05c.u., and c  500 c.u. . In this case,
the algorithm yielded the Nash equilibrium uBNE  (rB , rB ),
uPNE  (rP , rP ), uCNE  (rC , rC ) on the QRS-set for L  2 .
Example 3.2.3. For the input data of Example 3.2.1, let
kB  0.001c.u., kP  0.002c.u., and kC  0.004c.u. In this case, the
algorithm yielded the same Nash equilibrium uBNE  (rB , rB ),
uPNE  (rP , rP ), uCNE  (rC , rC ) on the QRS-set but for L  4 .
Example 3.2.4. For the input data of Example 3.2.2, let kC  0.3 c.u.
In this case, the algorithm yielded the Nash equilibrium uBNE  (rB , rB ),
rC rC
uPNE  (rP , rP ), uCNE  ( , ) on the QRS-set for L  4 . That is, the
2 2
strategy of the third player (C) was reduced by half in comparison with
Examples 3.2.2 and 3.2.3.
Example 3.2.5. For the input data of Example 3.2.4, let k В  1.3 c.u.
rВ rВ
In this case, the algorithm yielded the Nash equilibrium uBNE  ( , ),
2 2
rC rC
uPNE  (rP , rP ), uCNE  ( , ) on the QRS-set for L  4 . That is, the
2 2
strategy of the first player (B) was reduced by half.
Example 3.2.6. For the input data of Example 3.2.5, let k Р  0.9 c.u.
rВ rВ
In this case, the algorithm yielded the Nash equilibrium uBNE  ( , ),
2 2
rР rР r r
uPNE  ( , ), uCNE  ( C , C ) on the QRS-set for L  8 . That is, the
2 2 2 2
strategy of the second player (P) was reduced by half.
According to the results of numerical calculations for a wide class of
the input functions and parameters, it is possible to find a Nash equilibrium
in pure strategies on small QRS-sets that contain 2–4 elements. Further
extension of this set doesn't affect the results.
Now, consider the following hierarchical setup of the social
partnership control problem for the continuous education system. Suppose
the players from the set N  {B, P, C} (agents) with the payoff functionals
(16) and constraints (17) are located at the lower level while the upper
level is represented by a governmental body (supervisor) that maximizes
the total payoff functional (social welfare):
T
J 0 ( s(), u ())    e t [ gi (ri  ui (t ))  si (t )c( x(t ))]dt  e  T si (T )c( x(T ))  max.
(24)
iN 0
The supervisor’s control variables are the shares of agents in the social
utility determined by the students’ professional development. Her control
constraints have the form
s()  S  {s(t ) : si (t )  0,  si (t )  1, t  [0, T ]} . (25)

iN
Then the supervisor and agents play a differential Stackelberg game

with the following informational structure.
1. The supervisor chooses an open-loop strategy (25) and reports it to

the agents N  {B, P, C} .
2. For the given supervisor’s strategy, the agents play the differential
game (16)-(18), (24), (25), with the parametrical Nash equilibrium
NE (s()) as the solution concept.
3. In fact, the supervisor chooses her strategy (25) in order to
maximize her payoff (24) on the set NE ( s()) .
As all the strategies are constant on the same time intervals (10), the
supervisor’s strategy is a grid function  s Вk

 k 1

     
, s Рk k 1 , s Сk k 1  ,

M M M 
and her payoff functional (24) can be written as the payoff function
M T 
J 0 (uik kM1 , sik kM1 , )     (k i (rik  uik ) pi (e  tk 1  e  tk ) /    e  t si (t )cx (t )dt  
iN iN iN  k 1 0 
iN
e  T cx (T )  max . (26)
The constraints imposed on the supervisor’s controls take the form

sik (t )  0,
 iN
sik (t )  1; k  1,2,..., M ; i  N . (27)
We will construct a Stackelberg equilibrium in game (21)–(23), (26)–

(27) using the QRS method.
Similarly to the Nash equilibrium design, introduce a QRS-set of agent
i at an instant of time t  kt . Assume it contains L+1 elements, i.e.,

 r 2r ( L  1)rik 

QRS
U ik  0; ik ; ik ; ...; ; rik ; i  B, P, C; k  1,2,..., M .


L L L 

In the same fashion, introduce a QRS-set of the supervisor’s controls

with respect to each agent at an instant of time t  kt , which includes at
most L+1 elements:
 i 1 i 1 i 1 
 1   sik 2(1   sik ) ( L  1)(1   sik )
i 1 
 j 1 j 1 j 1 
S ikQRS  0; ; ; ... ; ; (1   sik ) ; i  B, P, C; k  1,2,..., M .
 L L L j 1 
 
In this case, each agent i has ( K  1) M scenarios of the form
QRSi  U iQRS
1  ...  U iM
QRS
; | QRSi | ( K  1) M i  B, P, C,
while the supervisor has not greater than ( K  1)3M scenarios of the form
QRS 0  S ВQRS
1  ...  S ВM  S Р1  ...  S РM  S С1  ...  S СM
QRS QRS QRS QRS QRS
;
| QRS 0 | ( K  1) 3M
.
The QRS-set of the game contains ( K  1) 6 M scenarios, i.e.,
QRS  QRS B  QRS P  QRS S  QRS 0 ; | QRS | ( K  1) 6M .

exceeding 3 or 4), which allows a numerical implementation of this
approach. The numerical algorithm of Stackelberg equilibrium design for
game (21)–(23), (26)–(27) is the same as in subsection 3.1. The
Stackelberg equilibrium is calculated using the QRS method.
Similarly to subsection 3.1, we constructed the QRS-sets in the case
M  K  2 . There are three strategies for each agent at all instants of time
and at most three strategies for the supervisor with respect to each agent,
i.e.:
 r 
U ikQRS  0; ik ; rik ;
 2 
 i 1 
 1   sik 
 j 1 i 1 
S ikQRS  0; ; (1   sik ) ; i  B, P, C; k  1,2,..., M .
 2 j 1 
 
Then there are nine QRSs for each agent ( i  B, P, C ) of the form
QRSi  U iQRS
1  U iQRS
2 
 (ui1 , ui 2 ) 6k 1  
 ri 2 ri1 ri1 ri 2 ri1 ri 2 
(0,0); (0, ); (0, ri 2 ); ( ,0); ( , ); ( , ri 2 ); (ri1 ,0); (ri1 , ); (ri1 , ri 2 )  .
 2 2 2 2 2 2 
For the supervisor, there are six QRSs at each instant of time ( k  1,2 )
as follows:
S kQRS  S Вk
QRS
 S Рk
QRS
 SСk
QRS

 (s Вk , s Рk , sСk ) 6j 1 
 1 1 1 1 1 1 
(0,0,1); (0,1,0); (1,0,0); (0, , ); ( ,0, ); ( , ,0)  .
 2 2 2 2 2 2 
Thus, there are 26244 different scenarios of the game, i.e.:
QRS  S1QRS  S 2QRS  QRS В  QRS Р  QRSС ;
| QRS | 6 293  26244 .
For all 26244 scenarios of the game, we numerically checked the first
condition in the definition of a QRS-set. These calculations took about 320
seconds on a PC with Pentium (R) Dual Core E5400 CPU. First, we found
the global maximum J 0max of the payoff functional (24) (payoff function
(26)). The maximum J 0max was searched in the control variables of the
supervisor and all three agents (for (24), in the six functions
s В (t ), s Р (t ), sC (t ) and u В (t ), u Р (t ), uC (t ) ; for (26), in the six
uВ  uВk k 1 , uР  uРk k 1 , uС  uСk k 1
M M M
grid functions and
sВ  sВk k 1 , sР  sРk k 1 , sС  sСk k 1 ). For the input data of Example

M M M
3.2.1, the QRS algorithm yielded J 0

max
 49895 c.u.
Example 3.2.7. Table 3.2.1 presents part of the numerical results for
the input data of Example 3.2.1.
Table 3.2.2 illustrates part of the numerical calculations to check the
second condition from the definition of a QRS-set for the input data of
Example 3.2.1 and M  K  2 .
Table 3.2.1.
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , J 0( j ) (( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , min | J 0(k )  J 0( j ) |

(c.u.) 1 k  26244
Strategies k j
u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) ) / J 0max
(0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0.0012
(0,1,0,0,0,1,0,0,0,0,0,0) 49 804 0.0015
(1,0,0,0,0,1,0,0,0,0,0,0) 49 705 0.0017
(1,0,0,0,0,1,50,50,30,50,25;30) 37 096 0.02
1 1 41 860 0.023
( , ,0,0,0,1,50,50,15,50,25;30)
2 2
1 1 1 1 42 076 0.025
( , ,0,0, , ,50,50,15,0,25;30)
2 2 2 2
1 1 26 282 0.003
( , ,0,0,0,1,100,50,30,100,50;30)
2 2
1 1 42 203 0.016
( , ,0,0,0,1,50,50,30,50,25;0)
2 2
1 1 37 588 0.04
( , ,0,0,0,1,100,0,30,50,50;15)
2 2
Table 3.2.1. (Continued)
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , J 0( j ) (( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , (c.u.) min | J 0(k )  J 0( j ) |

Strategies 1 k  26244
k j
u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) )
/ J 0max
1 1 42 268 0.009
( ,0, ,0,0,1,50,50,30,50,0;0)
2 2
1 1 37 713 0.011
( ,0, ,0,0,1,100,50,15,50,25;30)
2 2
1 1 34 681 0.005
( ,0, ,0,0,1,50,50,30,50,50;30)
2 2
1 1 37 433 0.012
( ,0, ,0,0,1,0,50,30,0,25;30)
2 2
1 1 44 435 0
( ,0, ,0,0,1,50,0,30,50,25;0)
2 2
1 1 1 1 34 906 0.004
( ,0, , ,0, ,50,50,30,0,50;30)
2 2 2 2
1 1 1 1 42 017 0.016
( ,0, , ,0, ,50,0,0,50,50;30)
2 2 2 2
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , J 0( j ) (( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , sС 2 , min | J 0(k )  J 0( j ) |
(c.u.) 1 k  26244
Strategies k j
u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) u В1 , u Р1 , uC1 , u В 2 , u Р 2 , uC 2 ) ( j ) ) / J 0max
1 1 1 1 30 526 0.05
( ,0, , ,0, ,100,50,30,50,50;30)
2 2 2 2
1 1 39 456 0.007
( ,0, ,0,0,1,50,50,30,50,50;15)
2 2
1 1 1 1 41 923 0.03
( ,0, , ,0, ,50,50,15,50,25;30)
2 2 2 2
1 1 1 1 39 515 0.008
( ,0, , ,0, ,50,50,15,50,50;30)
2 2 2 2
1 1 41 803 0.024
( ,0, ,0,0,1,50,50,15,50,0;30)
2 2
1 1 47 030 0.012
( ,0, ,0,0,1,50,50,15,50,0;0)
2 2
1 1 44 177 0.0087
( ,0, ,0,0,1,0,0,15,50,25;30)
2 2
1 1 42 397 0.0078
( ,0, ,0,0,1,50,50,0,0,25;30)
2 2
1 1 47 297 0.009
( ,0, ,0,0,1,0,50,15,0,25;0)
2 2
Table 3.2.2.
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 , J 0( j ) (c.u.) ( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,

J 0( k ) (c.u.) | J 0(k )  J 0( j ) |
Strategies s “Close” strategy
С 2 , u В1 , u Р1 , uC1 , u В 2 ,
sС 2 , u В1 , u Р1 , uC1 , u В 2 ,
/ J 0min
u Р 2 , uC 2 ) ( j )  QRS u Р 2 , uC 2 ) (k )
 QRS
(0.2,0.2,0.6,0.2,0.2,0.6, 49 762 (0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0
0,0,0,0,0,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 472 (0,0,1,0,0,1,50,25,15,0,0,15) 49 466 0.0001
30,30,10,10,10,10)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 565 (0,0,1,0,0,1,0,0,15,0,0,0) 49 553 0.0002
0,0,10,10,0,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 590 (0,0,1,0,0,1,0,0,0,0,0,15) 49 615 0.0005
0,0,0,0,10,10)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 618 (0,0,1,0,0,1,0,0,0,0,0,15) 49 615 0.00006
70,40,0,0,0,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 477 (0,0,1,0,0,1,50,25,15,0,0,15) 49 466 0.0002
0,0,10,40,0,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 716 (1,0,0,0,0,1,0,0,0,0,0,0) 49 705 0.0002
0,40,0,40,0,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 421 (0,0,1,0,0,1,0,0,15,0,0,15) 49 402 0.0004
70,0,10,0,10,0)
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( j ) (c.u.) ( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( k ) (c.u.) | J 0(k )  J 0( j ) |
sС 2 , u В1 , u Р1 , uC1 , u В 2 , sС 2 , u В1 , u Р1 , uC1 , u В 2 ,
/ J 0min
u Р 2 , uC 2 ) ( j )  QRS “Close” strategy
u Р 2 , uC 2 ) (k )
 QRS
Strategies
(0.2,0.2,0.6,0.2,0.2,0.6, 49 646 (0,0,1,0,0,1,0,25,0,0,0,15) 49 652 0.0001
0,20,0,20,0,10)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 658 (0,0,1,0,0,1,0,25,0,0,0,15) 49 652 0.0001
0,10,10,0,20,0)
(0.3,0.3,0.4,0.2,0.2,0.6, 49 762 (0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0
0,0,0,0,0,0)
(0.2,0.2,0.6,0.3,0.3,0.4, 49 762 (0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0
0,0,0,0,0,0)
(0.2,0.2,0.6,0.3,0.3,0.4, 49 684 (1,0,0,0,0,1,0,0,0,0,0,0) 49 705 0.0004
70,20,0,0,0,0)
(0.3,0.3,0.4,0.2,0.2,0.6, 49 498 (0,0,1,0,0,1,50,25,15,0,0,0) 49 517 0.0004
70,20,10,0,10,0)
(0.2,0.2,0.6,0.3,0.3,0.4, 44 161 (0.5,0,0.5,0,0,1,0,0,15,50, 44 177 0.0003
0,0,30,40,10,20) 25;30)
(0.2,0.2,0.6,0.2,0.2,0.6, 44 553 (0,0,1,0,0,1,50,50,30,0,0;0) 44 556 0.00006
60,40,30,20,10,0)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 230 (0,0,1,0,0,1,50,0,15,50,0;0) 49 234 0.00008
40,10,20,30,0,0)
Table 3.2.2. (Continued)
( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( j ) (c.u.) ( s В1 , s Р1 , sС1 , s В 2 , s Р 2 ,
J 0( k ) (c.u.) | J 0(k )  J 0( j ) |
Strategies “Close” strategy
sС 2 , u В1 , u Р1 , uC1 , u В 2 , sС 2 , u В1 , u Р1 , uC1 , u В 2 ,
/ J 0min
u Р 2 , uC 2 ) ( j)
 QRS u Р 2 , uC 2 ) (k )
 QRS
(0.2,0.2,0.6,0.2,0.2,0.6, 49 238 (0,0,1,0,0,1,50,0,15,50,0;0) 49 234 0.00008
0,20,0,40,20,20)
(0.2,0.2,0.6,0.2,0.2,0.6, 49 453 (0,0,1,0,0,1,50,25,15,0,0,15) 49 466 0.0002
10,10,10,10,10,10)
(0.2,0.2,0.6,0.2,0.2,0.6, 48 807 (0,0,1,0,0,1,0,30,15,0,30,15) 48 810 0.00006
20,20,20,20,20,20)
(0.2,0.2,0.6,0.2,0.2,0.6, 39 531 (0,0,1,0,0,1,50,25,30,50, 39 532 0.00002
30,30,30,30,30,30) 25,30)
(0.3,0.3,0.4,0.2,0.2,0.6, 49 453 (0,0,1,0,0,1,0,0,15,0,0,15) 49 402 0.00008
10,10,10,10,10,10)
(0.2,0.2,0.6,0.3,0.3,0.4, 48 807 (0,0,1,0,0,1,0,30,15,0,30,15) 48 810 0.00006
20,20,20,20,20,20)
(0.3,0.3,0.4,0.3,0.3,0.4, 39 531 (0,0,1,0,0,1,50,25,30,50, 39 532 0.00002
30,30,30,30,30,30) 25,30)
(0.4,0.4,0.2,0.2,0.2,0.6, 49 762 (0,0,1,0,0,1,0,0,0,0,0,0) 49 762 0
0,0,0,0,0,0)
A direct analysis of Tables 3.2.1 and 3.2.2 implies that the precision is
  0.0012 . Thus, the resulting set of scenarios can be taken as a QRS-set
for the input data of Example 3.2.1 in the case M  K  2 .
CONCLUSION
It is clear that the QRS method may hardly be considered as a

universal solution method for dynamical control problems using computer
simulations. Nevertheless, this method yields satisfactory results for many
models of real-world systems where it is natural to separate out a few
qualitatively representative scenarios. The corresponding payoffs are
different in principle while, in comparison with the other feasible
scenarios, the difference becomes negligible.
In this paper, two problems have been investigated, namely, the design
of Nash and Stackelberg equilibria in differential games that model fishing
in shallow-water ecosystems and social partnership in a continuous
education system. The numerical examples show that the QRS method has
a good performance in the cases of 2–4 scenarios.
The QRS method can be used for solving a wider class of complex
dynamical control problems (e.g., optimal control problems). In addition, a
scenario may contain not only control variables but also system
parameters, which leads to the solution of identification problems as well.
REFERENCES
[1] Aliev R.A., Liberzon M.I. Noniterative coordination algorithms in

two-level systems//Soviet Journal of Computers and
SystemsSciences, 1987, 25(1): 114–118.
[2] Basar T., Olsder G.J. Dynamic Noncooperative Game Theory. –
SIAM: Philadelphia, 1999.
[3] Chernousko F. L., Lyubushin A. A. Method of successive

approximations for optimal control problems //Optimal Control
Applications and Methods, 1982, 3: 101-114.
[4] Chistyakov A.E., Nikitina A.V., Ougolnitsky G.A. et al., A
differential game model of preventing fish kills in shallow
waterbodies //Game Theory and Applications. Vol. 17: Game
Theoretic Models in Mathematical Ecology/Ed. V. Mazalov, D.
Novikov, G. Ougolnitsky and L. Petrosjan. – N.Y.: Nova Science
Publishers, 2015. P. 37–48.
[5] Dockner E., Jorgensen S., Long N.V., Sorger G. Differential Games
in Economics and Management Science. – Cambridge University
Press, 2000.
[6] Dyachenko V.K., Ougolnitsky G.A., Tarasenko L.V. Computer
investigation of a game theoretic model of social partnership in the
system of continuing professional education //Advances in Systems
Science and Applications, 2015, 15(4): 320–328.
[7] Gorelov M.A., Kononenko A.F. Dynamic models of conflicts. III.
Hierarchical games//Automation and Remote Control, 2015, 76(2):
264–277.
[8] Guibas L. J., Ramshaw L. Stolfi J. A Kinetic framework for
computational geometry//Proc. 24th Ann. IEEE Symp. Found.
Computer Science, 1983. P. 100–111.
[9] Jank G., Kun G. Optimal control of disturbed linear-quadratic
differential games//European Journal of Control, 2002, 8(2): 152–
162.
[10] Judd K.L. Numerical Methods in Economics. – Cambridge, MIT
Press, 1998.
[11] Kleijnen J.P.C. Design and Analysis of Simulation Experiments. –
Springer, 2007.
[12] Kononenko A.F. Game-theory analysis of a two-level hierarchical
control system//USSR Computational Mathematics and
Mathematical Physics, 1974, 14(5): 72–81.
[13] Krasovskii N.N., Subbotin A.I. Game-Theoretical Control Problems.
– Springer, 1988.
[14] Law A.M., Kelton W.D. Simulation Modeling and Analysis. –

McGraw-Hill, 2000.
[15] Nance R.E., Sargent R.G. Perspectives on the evolution of
simulation//Operations Research, 2002, 50(1): 161–172.
[16] Nelson B.L. “Some tactical problems in digital simulation” for the
next 10 years//Journal of Simulation, 2016, 10: 2–11.
[17] Ougolnitsky G. Sustainable Management. – N.Y.: Nova Science
Publishers, 2011.
[18] Ougolnitsky G. Game theoretic formalization of the concept of
sustainable development in the hierarchical control systems//Annals
of Operational Research, 2014, 220(1): 69–86.
[19] Ougolnitsky G.A. Sustainable management as a key to sustainable
development//Sustainable Development: Processes, Challenges and
Prospects. Ed. D. Reyes. – N.Y.: Nova Science Publishers, 2015.
P.87–128.
[20] Ougolnitsky G.A., Usov A.B. Solution algorithms for differential
models of hierarchical control systems//Automation and
RemoteControl, 2016, 77(5): 872–880.
[21] Powell W.B. Approximate Dynamic Programming. – Wiley, 2007.
[22] Pytlak R. Numerical Methods for Optimal Control Problems with
State Constraints. –Springer-Verlag: Berlin, Heidelberg, 1999.
[23] Sargent R.G. Verification and validation of simulation
models//Journal of Simulation, 2013, 7(1): 12–24.
[24] Stokey L.N., Lucas R.E., Prescott E.C. Recursive Methods in
Economics Dynamics. – Cambridge (MA): Harvard University Press,
1989.
[25] Sukhinov A.I., Chistyakov A.E., Ugol'nitskii G.A. et al., Game
theoretic regulations for control mechanisms of sustainable
development for shallow water ecosystems//Automation and
RemoteControl, 2017, 78(6): 1059–1071.
[26] Szilagyi M.N. Agent-based simulation of the N-person chicken
game//Advances in Dynamic Game Theory: Numerical Methods,
Algorithms, and Applications to Ecology and Economics. Eds. S.
Jorgensen, M. Quincampoix, T.L. Vincent. – Birkhauser, 2007.
[27] Tarasenko L.V., Ougolnitsky G.A., Usov A.B. et al., Concordance of

interests in dynamic models of social partnership in the system of
continuing professional education//Int. J. of Environmental and
Science Education, 2016, 11(18): 13055–13068.
[28] Ugol’nitskii G.A., Usov A.B. A study of differential models for
hierarchical control systems via their discretization//Automation and
Remote Control, 2013, 74(2): 252-263.
[29] Wilson I.D. Foundations of hierarchical control//International
Journal of Control, 1979, 29(6): 899-933.
Reviewed by:
Professor Dmitry A. Novikov
Trapeznikov Institute of Control Sciences
Russian Academy of Sciences, Russia
Chapter 3
THE COMPUTER SIMULATION OF

THE TALBOT EFFECT AND CARPET VIA
THE ITERATIVE FRESNEL
INTEGRALS METHOD
Kazi Monowar Abedin, Aamna Al-Saedi

and S. M. Mujibur Rahman
Department of Physics, College of Science,
Sultan Qaboos University, Muscat, Oman
ABSTRACT
The Talbot effect was first experimentally observed by Henry Talbot

in 1836. It is the repeated self-imaging of a diffraction grating at regular
distances in the near-field behind the grating. The corresponding self-
repeating distance is now known as the Talbot distance. If the observed
diffraction images are laid out as a function of the distance, a beautiful
and repetitive pattern is observed; this is known as the Talbot carpet.
Apart from a considerable theoretical interest, the Talbot effect has found

Corresponding Author Email: abedin@squ.edu.om.
108 Kazi M. Abedin, A. Al-Saedi and S. M. Mujibur Rahman
many applications in diverse areas of optics, for example, in imaging,

refractive index measurements, displacement sensors, lithography and
array illumination, to name a few. In this Chapter, we have applied the
Iterative Fresnel Integrals Method (IFIM) to the simulation of the Talbot
effect, and consequently, to the generation of Talbot carpets. The
methodology of how the IFIM method was applied for the simulation of
the Talbot effect is described explicitly, followed by a systematic
synthesis of the Talbot carpet from the generated data. All the data were
generated without recourse to any experimental apparatus. Finally,
examples of Talbot carpets are presented at two different resolutions, and
suggestions are made as to how the whole process of Talbot carpet
synthesis can be automated to generate carpets of higher resolutions.
Keywords: computer simulation, near-field diffraction, Iterative Fresnel

Integrals Method, diffraction gratings, Talbot effect, Talbot carpet
1. INTRODUCTION
The Talbot effect was first discovered by the English photographer and
inventor Henry Talbot in 1836. He observed that the light diffracted behind
a diffraction grating has a regular and repetitive pattern in the axial
direction [1]. A periodicity is also observed in the lateral directions due to
the periodicity of the grating. The regular distance in the axial direction is
called the Talbot distance. After its discovery, the Talbot effect was
forgotten for nearly half a century, until it was rediscovered by Lord
Rayleigh. Lord Rayleigh [2] explained this repetition of images within the
paraxial approximation in 1881, and showed that the Talbot distance was
given by D2/, where D is the slit spacing of the grating and is the
wavelength of the light used.
The Talbot effect has been extensively studied both theoretically and
experimentally in the intervening period. The theoretical investigations are
often concerned with the origin of the Talbot effect, and the associated
wave and coherence phenomena. The relevant theoretical research has
been done within the framework of Fresnel diffraction, Fresnel images,
Fourier images, and the theory of image formation [3-6]. In addition,
rigorous theoretical analyses were done using, e.g., the finite-difference
The Computer Simulation of the Talbot Effect … 109
time domain (FDTD) method [7], electromagnetic theory [8], and the
Rayleigh-Sommerfeld diffraction formula [9], for the effects of aberrations
on the self-imaging phenomena. Other authors have pointed out the
resemblance of the Talbot effect to fractals [10].
On the experimental side, many applications have been proposed for
the Talbot effect. It has been proposed, for example, for imaging [11],
sensing a distance and a displacement [12], measuring the refractive index
of a medium [13], array illumination [14], lithography [15], sub-
wavelength focusing of light [16], and laser beam combination [17].
The Talbot effect is an experimental observation in the first place. One
needs some elaborate equipment, such as light sources or lasers, diffraction
grating, translation stages, a suitable camera and other associated
equipment to observe the effect. To generate the repetitive but beautiful
pattern called the Talbot carpet is also a nontrivial task, since the image is
composed of many discrete images captured at different known axial
distances from the grating.
Previously, we used the Iterative Fresnel Integrals Method (IFIM) to
simulate the diffraction effects from rectangular apertures in the near field,
and later, from N-slits or gratings in the near-field (Fresnel) regime [18-
19]. We have used the powerful, general purpose software MATLAB,
running on an ordinary PC to perform these simulations. The method was
found to be powerful, accurate and general, being able to generate many
features of the diffraction, namely, the missing order, principal and
secondary maxima, Fresnel-Fraunhofer transitions, etc. Therefore, our idea
was to use this IFIM method to try to simulate the Talbot effect and the
Talbot carpet, entirely by computer. We were successful in our efforts.
In this Chapter, we describe how the IFIM method can be used to
simulate the Talbot effect from a periodic grating structure. In addition, we
were able to synthesize the complete Talbot carpet with a certain
resolution. The only equipment needed for this, is, of course, a PC (or a
laptop) and the well-known software MATLAB. A photo-processing
software is also required to synthesize the Talbot carpet.
The Chapter is divided into the following sections: First, we discuss
the theoretical basis of the method as it applies to the generation of near-
field images and the Talbot effect. In the next section, methods of
generating the Talbot carpet are discussed. That is followed by a section, in
which we present examples of Talbot carpet generated by the method.
Conclusions are drawn in the final section.
2. THEORETICAL BACKGROUND
2.1. Iterative Fresnel Integrals Method and the Theory of N-slits
We assume that an N-aperture system is centered at the yz coordinate

system and is illuminated by monochromatic light of wavelength  emitted
from the source S We further assume that the origin of the coordinate
system O is located at the exact center of the N-aperture. Figure 1 shows
the N-aperture system for N= 5. The selection of five slits is purely
arbitrary, being done to illustrate the principles involved. Let a be the
individual aperture width, b be the inter-aperture separation [the center-to-
center aperture separation is then D= (a+b)], and let c be the aperture
height in the z-direction.
In this geometric configuration, the two vertical edges of the central
aperture of the N-aperture system (aperture 0 in the figure) are located at
y0= -a/2 and y0’ = a/2 respectively, and the edges of the next aperture to
the right (called aperture +1 in the figure) are located at y1 = a/2+b and
y1’ = 3a/2+b. The edges of the next aperture to the right (aperture +2) are
likewise located at y2= 3a/2+2b and y2’ = 5a/2+2b. Finally, the edges of
the aperture +n are located at yn= (2n-1)a/2+nb and yn’ = (2n+1)a/
2+nb. On the left side, the edges of the first left aperture (aperture -1) will
be located at y-1’ = -a/2-b and y-1 = -3a/2-b, and, the edges of the –n
aperture will be at y-n’= -(2n-1)a/2-nb and y-n = -(2n+1)a/2-nb.
Figure 1. Geometric configuration for Fresnel diffraction for N = 5. Here, a is the

individual aperture width, b is the width of the opaque regions between apertures and
D = (a+b) is the center-to-center aperture separation. (Reprinted from Abedin et al.
Optik Vol.126. pp. 3743-3751, Copyright (2015) with permission from Elsevier).
If illuminated by a monochromatic light source at position S (as shown

in the figure), the total complex electric field E at the center P of the YZ
observation plane consists of the contributions from each of the (2n+1)
apertures. The electric field contribution from the aperture 1 is given by the
equation [18, 19],
E P1  E2u C (u)  j S (u) uu11' C (v)  j S (v) vv' . (1)
where Eu is the unobstructed electric field at P (i.e., the electric field that
would be observed if the aperture were removed). Here, C(u) and S(u) are
known as the Fresnel cosine and sine integrals, being defined by,
w w
C ( w)  cos( w' 2 / 2)dw' and S ( w)   sin ( w' 2 / 2)dw'.
0 0 (2)
Here w represents either one of the two dimensionless variables u or v.,

u and v are proportional to the Cartesian coordinates y or z, respectively,
u  y 2( p0  q0 ) /  p0 q0 ,   z 2( p0  q0 ) /  p0 q0 . (3)
The limits u1 and u1’ are the values of the dimensionless variable u
corresponding to the two edges of aperture 1, i.e., for y1 = a/2+b and
y1’ = 3a/2+b, respectively. Similarly, the limits v and v’ are the values of
the dimensionless variable v corresponding to the lower and the upper
edges of this aperture, i.e., for z = -c and z’ = +c, respectively.
The electric field contributed by the aperture n to the right is,
E Pn  E2u C (u)  jS (u) uunn ' C (v)  jS (v) vv '  Eu

2
C(un' )  C(un )  j{S (un' )  S (un )}C(v)  jS (v) vv' (4)
where un and un’ are the values of u corresponding to the two vertical edges
of aperture n, i.e., for yn= (2n-1)a/2+nb and yn’ = (2n+1)a/2+nb,
respectively. The values v and v’ are the same as in equation (1).
The electric field contributed by the central aperture 0 is,
E P 0  E2u C (u)  jS (u) u00 ' C (v)  jS (v) vv '

u
(5)
where u0 and u0’ are the values of u corresponding to the two vertical edges
of the central aperture, i.e., for y0= -a/2 and y0’ = a/2, respectively. The
values v and v’ are the same as in equation (1) or equation (4). The electric
field contribution by the -1 aperture to the left is similarly obtained from,
E P1  E2u C (u)  jS (u) uu11' C (v)  jS (v) vv ' (6)
where u-1 and u-1’ are the values of u corresponding to the two edges of
the -1 aperture, i.e., for y-1’ = -a/2-b and y-1= -3a/2-b.The electric field
contribution by the last aperture -n is given by
E Pn  E2u C (u)  jS (u) uunn ' C (v)  jS (v) vv ' (7)
where un and un’ are the values of u corresponding to the two vertical edges
of aperture -n, i.e., for y-n’= -(2n-1)a/2-nb and y-n= -(2n+1)a/2-nb,
respectively. The values v and v’ are as before.
The total complex electric field at P contributed by the (N = 2n+1)
apertures is simply the sum of all the complex amplitudes EPn, EP1, EP0,
EP-1,.EP-n,
E N  E Pn ..  E P1  E P 0  E P1  .... E P  n
{C (u )  jS (u ) un '  ...  C (u )  jS (u ) uu1'  C (u )  jS (u ) u0 '  C (u )  jS (u ) uu1'  ...  C (u )  jS (u ) u n ' }
Eu

u u u
2 n 1 0 1 n
 C (v)  jS (v) v'

v.
(8)
Separating the cosine and the sine integrals, and using the summation
notation for the Fresnel sines and the cosines, we derive the total
complex electric field at P as,
E N  E2u {i  n C (ui ' )  C (ui )  j i  n [ S (ui ' )  S (ui )]} 

i n i n
(9)
{C (v' )  C (v)  jS (v' )  S (v)} (9)
The summation on i from –n to +n for both the Fresnel cosine and sine
integrals (for the u variable only), in effect, carries out the summation of
the complex electric field contributions from aperture –n to aperture n, i.e.,
for a total of N = (2n+1) apertures. No such summation is needed for the v
variable, since only the two edges (upper and lower) of the apertures are
involved in this z-direction for all the (2n+1) apertures.
The net intensity at P is proportional to the square of the net electric
field, i.e.:
I P  40 ( EN E N ),
I *
(10)
Where I0 denotes the intensity of the unobstructed wave, i.e., I0= Eu2.
From equation (9), it is clear that the calculation of the electric field or
intensity at a point P requires, in general, the calculation of 2N Fresnel
cosine and 2N Fresnel sine integrals, corresponding to the 2N edges of the
N aperture system for the u(y) variable. In addition, two pairs of Fresnel
cosine and sine integrals are required for the v(z) variable. The cosine
integrals form the real parts of the electric field, and the sine integrals form
the imaginary parts. After calculating the complex electric field, the
intensity at P is calculated from equation (10) by multiplying the field by
its complex conjugate. All these mathematical operations can be efficiently
performed by the MATLAB software. These equations are the basis of
calculation of the complete intensity distribution of the Fresnel diffraction
pattern from an N-aperture system, as will be explained in Sec. IIIA.
Equation (9) was used to calculate first the complex electric field, and then
the intensity was calculated by taking the square of it. Equation (9) has two
factors, the first factor involves only the u (or y) coordinate, and expresses
the dependence of the electric field or intensity in the y direction. The
second factor involves only the v (or z) coordinate, and it expresses the
dependence of the electric field or intensity in the z direction.
3. SIMULATION TECHNIQUE AND SIMULATION RESULTS
3.1. Simulation Strategy and Algorithm
In the rest of this Chapter, for the sake of simplicity, we assume that
the source S is located at infinity. This can be achieved experimentally by
placing a convex lens between S and the aperture, and placing the source S
at the back focus of the lens. In this case, Eq. (3) simplifies to,
u  y 2 /  q0 ,   z 2 /  q0 . (11)
(11)
Figure 2. Flow chart of the algorithm for the N-aperture problem.

(Reprinted from Abedin etal. Optik Vol.126. pp. 3743-3751, Copyright (2015)
with permission from Elsevier).
Equations (9) and (10) describe the electric field and the intensity at
P (the center of the observation screen), respectively. In order to apply
these for an off-axis arbitrary point P', we use the following technique
described in Ref. [18] and [19]: The observation screen and the SOP line
were held fixed. Then instead of moving P, the entire aperture in the yz
plane was moved in the opposite direction, so that the relative position of
the aperture in this new position and point P remains unaltered. We then
calculate the electric field at P instead of at P'. The point P' will observe a
new set of values for the y’s and z’s (and hence for the u’s and v’s). For
example, to find the intensity at point P', 1mm to the right of P, the screen
was kept undisturbed and the aperture system was moved 1mm leftwards,
and the intensity at P in this configuration is calculated (instead of at P').
As a result, the point P will now observe a new set of values for y’ (and
therefore, for the values of u) in equations (8) and (9). The electric field
(and the corresponding intensity) at any arbitrary point P' on the image
plane (screen) can be found in this way by making appropriate virtual
movements of the aperture in the y and z directions, and in equation (8) or
equation (9), using correspondingly a new set of values for u and v.
The flow chart of the algorithm is given in Figure 2. We can realize
that for the calculation of the electric field distribution using equation (9) at
all the points (or, pixels) on the image plane, a large number of the Fresnel
cosine and sine integrals will need to be evaluated quickly, and this is
efficiently done by the special functions mfun (‘FresnelC', i:s:f) and mfun
(‘FresnelS', i:s:f)in MATLAB.
As shown in Figure 3 for N= 3 (three apertures again for simplicity),
the whole aperture system was displaced (virtually) by a distance W, and
the observed positions of the 2N edges of the displaced aperture was
determined and compared to the corresponding positions for the aperture
before the displacement, to find the required range of the u and v values.
These positions are marked in Figure 3. In a more general N-aperture
system, when the aperture is moved by W to the left, the ranges for the 2N
vertical edges of the N= (2n+1) apertures will be given by:
for yn :(2n-1)a/2+nb to W+(2n-1)a/2+nb, for yn’: (2n+1)a/2+nb

to W+(2n+1)a/2+nb
for y1: (a/2+b) to (W+a/2+b), for y1’: (3a/2+b) to (W+3a/2+b),
for y0’: (a/2) to (W+a/2), for y0: (-a/2) to (W-a/2),
for y-1’: (-a/2-b) to (W-a/2-b), for y-1: (-3a/2-b) to (W-3a/2-b),
for y-n’: -(2n-1)a/2-nb to W-(2n-1)a/2-nb, for y-n: -(2n+1)a/2-nb
to W-(2n+1)a/2-nb .
Figure 3. Apparent limits of the virtually displaced aperture seen from the observation
plane for N = 3. (Reprinted from Abedin etal. Optik Vol.126. pp. 3743-3751,
Copyright (2015) with permission from Elsevier).
Arrays of both the Fresnel cosine and sine integrals need to be

calculated corresponding to these input ranges by the mfun statements in
MATLAB, with a step size (resolution) s. By using these Fresnel array
functions, the electric field (or intensity) dependence in the y(u) direction
can be calculated (which is the first factor in equation 9). Following a
similar procedure for the entire aperture, i.e., by moving the aperture by
W upwards, the z1 and z2 ranges are determined to be: c to (W + c) for z2,
and –c to (W-c) for z1. Only four arrays of the Fresnel integrals need to be
evaluated for these two z ranges, furnishing the numerical values for the
calculation of the second factor in equation (9).
The complete MATLAB program for this problem is given in the
Appendix. The required initial parameters to the program are input through
a GUI (Graphical User Interface) generated in lines 1-2 of the program.
The GUI accepts the values of the aperture width (a), the aperture
separation (b), the aperture height (c), the width of the image area (W), the
step size or the resolution of the calculations s, the aperture-image plane

distance (q0) (all in mm), the wavelength of light (in nm), the exposure
factor (t) and the number of slits (N) as user-supplied variables. The
dimensionless quantities corresponding to the aperture dimensions a, b and
c, the step size s and the image area size W are calculated in lines (4-5) of
the program. The step size s determines the resolution of the simulated
images. A reasonably small value of the step size should be selected in the
simulations. If too small a value is selected, the simulation time will be too
long, and memory overflow may occur in the computer. On the other hand,
if too large a value is chosen, a low-resolution, blurred image will be
generated. A value of s= 0.01 mm proved to be a good starting point in the
present situation.
The summation of the complex electric field for the u variable over
the N apertures, in equation (9), is performed inside the ‘for’ loop (between
lines 7-14), where N iterations are executed. Inside this ‘for’ loop, the
required Fresnel cosine integral arrays for the u dimension are evaluated in
lines (8-9) using the relevant mfun statements, and the corresponding
Fresnel sine arrays are similarly calculated in lines (11-12). The ranges of
the input arguments in these integral arrays are controlled by the integer
parameter j, which begins from N, and goes down to (–N+2), in steps of 2
inside the ‘for’ loop. In lines 10 and 13, the  sums are evaluated inside the
loop, separately for both the cosine and the sine arrays, thus performing the
vital summation over the N apertures indicated in equation (9). In line 15,
outside the loop, the electric field in u is calculated in complex form,
corresponding to the first complex factor in equation (10). In the next four
lines, the Fresnel cosine and sine integrals are evaluated for the v
dimension. Next, in line 20, the second complex electric field in v is
evaluated, corresponding to the second factor in equation (9). In the next
lines (21-22), the matrix C, which contains the complex electric field
variations in both the u and v directions, is constructed. Finally, by
squaring it, the matrix D, which contains the intensity values for the first
quadrant of the image plane, in both the u and v directions, is calculated
(line 23). The elements of this C matrix is then normalized (line 24), and
the full E matrix for the full intensity distributions is constructed by
inverting the D matrix twice (lines 27-35). Here, we use the inversion
symmetry of the image around the u and v-axes. At the end of the program,
the generated image is displayed as a grayscale (black and white) image by
the images c statement with the appropriate scale axes (line 37).
3.2. Simulation of the Talbot Effect
We used the program to first simulate near field images for a particular
diffraction grating (with a= 0.04mm, b= 0.16mm, c= 8mm and N= 15). In
other words, the grating linewidth is D= (a+b) = 0.20 mm, and the grating
constant is 50 lines/cm. The wavelength of the light is λ = 500nm, the
image area is W= 3 mm, and the resolution of the simulation is
s= 0.01 mm. The image at an aperture-screen distance, q0= 10 mm, is
shown in the Figure 4(a). It shows the closely-spaced, characteristic pattern
of Fresnel diffraction images in the near field.
To examine the near-field images at regular intervals, and to see if they
are repeated at intervals of the Talbot lengths (self-imaging), we simulated
Fresnel images at q0= 10mm, 20mm, 80mm, 90mm, 100mm, 160mm. The
grating constants, the wavelength and all the other calculation parameters
were left unchanged. The simulated images are all shown in Figure 4.
Remembering that the Talbot distance in our case is:
ZT = (a+b)2/D2/ mm, (12)
we observe that there is a self-similarity in the simulated images between

q0 = 10mm and q0 = 90mm [(a) and (b)], between q0 = 20mm and
q0 = 100mm [(c) and (d)], between q0 = 80mm and q0 = 160mm [(e) and
(f)]. This proves that self-imaging is indeed occurring at the Talbot
distance of 80 mm, and that the essential features of the Talbot effect can
be simulated by the IFIM method. Moreover, at half the Talbot distance,
for example, q0= 40 and q0= 120, the fringe density in the images are
doubled. This is also expected for the classical Talbot effect.
Figure 4. Computer simulation for the Talbot effect for a 15 slit system with the
simulation parameters: a = 0.04 mm, b = 0.16 mm (D = 0.2 mm), c = 8 mm, w = 3 mm,
s = 0.01 mm and λ = 500 nm. q0 is increased step by step in the near field region.
Self-repeating of images at intervals of 80 mm is clearly observed.
3.3. Simulation of the Talbot Carpet
To generate the Talbot carpet, we produce a series of near-field images

for the above simulation parameters: a= 0.04mm, b= 0.16mm, c= 8mm, N=
15, W= 3 mm, and λ = 500nm. The value of q0 is increased from 1mm to
160 mm, in steps of 1 mm. About 160 images are generated in this way,
each simulated image taking about 1 min to generate.
Figure 5. Synthesized Talbot carpet at an axial resolution of 1 mm, from q0 = 1 mm to

160 mm. Simulation parameters are the same as in Figure 4. The scale shows the axial
distance in mm.
A horizontal strip across the center of each image is cut using the
software Photoshop, and the strip is laid out on a scaled map, the scale
running from 1mm to 160 mm. The scale represents the axial distance q0 of
the image behind the grating. The process is repeated for each of the
simulated images. The result is a mosaic of images, consisting of about 160
horizontal strips, each strip being placed at the appropriate distance, where
they would be observed in reality. The complete image is shown in
Figure 5.
4. DISCUSSION
Clearly, Figure 5 represents a complete Talbot carpet. The features are

repeated at a distance of 80 mm, throughout the image, starting from any
arbitrary distance. 80 mm is, of course, the calculated Talbot distance ZT.
This is the required self-imaging phenomenon of the Talbot carpet, as
mentioned in the previous section and in the introduction of this Chapter.
In addition, at half the Talbot distance, i.e., at 40 mm, the density of
the features is doubled. For example, at 80 mm, there are about 19 fringes
in the image, but at 120 mm, about 38 can be discerned. At q0= 40 mm, the
fringe density is also observed to double from that at 80 mm. At q0= 60
mm and 100 mm, the density is observed to quadruple. This observation is
valid over the entire distance scale of the generated images. We also
observe a regular, self-repeating, beautiful pattern in the whole image, and
that is essentially called a Talbot carpet. This behaviour agrees with the
main features of Talbot carpets observed in reality [20, 21].
The Talbot carpet shown has all the essential features, but it is limited
to a spatial (i.e., axial) resolution of 1 mm. This limits the detail that is
revealed in the carpet. In order to simulate a Talbot carpet at higher
resolution, we attempted to generate a Talbot carpet at about 10 times
greater resolution, i.e., at an axial resolution of 0.1 mm. All the parameters
in the previous set of simulations were kept unchanged, including the
resolution of the computations. The procedures described previously were
used to synthesize the Talbot carpet.
The complete carpet is shown in Figure 6 in six parts. The values of q0
run from 1 mm to 70 mm. The carpet is composed of approximately 700
images, and a corresponding number of simulations had to be performed.
As we can see, more intricate details within the Talbot carpet can be seen,
including the doubling and quadrupling of the fringe density at the half-
Talbot and the quarter-Talbot distances. The generation of the carpet was
not continued at larger distances, because of the extremely large amount of
computational efforts required.
Figure 6. Synthesized Talbot carpet at an axial resolution of 0.1 mm, from q0= 1 mm to
70 mm. Other parameters of the simulations were not changed.
In general, a Talbot carpet can be experimentally observed and

produced, with elaborate experimental equipment [22, 20]. It has been
observed not only with visible light, but also with x-rays [23], and with
matter waves associated with trapped atoms [24]. For light waves in the
visible region, the highest resolution carpet that can be produced is limited
by the spatial or pixel resolution of the imaging camera. But in our case,
we reproduce it completely on computer, where all the essential features
can be simulated without any experimental equipment. All the parameters,
such as the wavelength of light, the slit width, and the separation can be
arbitrary changed in the simulations, and the corresponding Talbot carpet
can be generated. The ability to generate a Talbot carpet at any wavelength
is particularly noteworthy, since it enables one to generate the Talbot
carpet for any wavelength of the illuminating light.
It takes a huge amount of computational effort and a considerable
amount of time to generate a full Talbot carpet. For example, using a
Pentium core-i7 PC with 8 MB of RAM, it took approximately 4 hours to
synthesize a Talbot carpet at 1 mm resolution, and more than 12 hrs to
synthesize it at 0.1 mm resolution to a length of 70 cm. To generate it at
higher resolutions will take such a long time as to render the process
impractical.
It may be possible to automate the whole process by devising suitable
software that can perform all the steps, i.e., generating all the simulations
and performing the synthesis of the carpet automatically. Then the whole
program should be able to automatically generate the carpet, once the
necessary input parameters, such as, the wavelength of light, the slit width,
the separation, and the axial resolution are specified. In principle, there is
no upper limit to the resolution that can be achieved; this is limited only by
the computational resources available. Viewing and printing of a such high
resolution carpet in its entirely will require a large, high-resolution
computer screen, and a correspondingly large-format printer.
CONCLUSION
In this Chapter, we have used the Iterative Fresnel Integrals Method to

simulate and observe the famous Talbot effect, and thereby, to generate
complete Talbot carpets at two different resolutions. All the expected
features of the Talbot effect and the Talbot carpet can be readily observed
in the simulated images. Our method does not require any experimental
apparatus, and moreover, it can be extended to any arbitrary axial
resolution. It also shows that the Iterative Fresnel Integrals Method is a
powerful technique to simulate the various aspects of near-field diffraction.
The only limitation of our method is the large amount of computational
effort and time involved, especially, if the Talbot carpet is to be
synthesized at a higher resolution. But even at these high resolutions, an
ordinary PC or laptop is sufficient to generate the desired Talbot carpet.
APPENDIX
MATLAB program for the near-field imaging of N-aperture Grating
1. u = inputdlg({'a mm','bmm','cmm','Wmm','s mm','q0 mm','l nm',...

2. 'exposure', 'Slit Number'},'Fresnel Diffration from N apertures',
[1,1,1,1,1,1,1,1,1]);
3. for i = 1:9; v(i) = str2num(u{i}); end
4. q0 = v(6); t = v(8); l = v(7)*1e-6;f = sqrt(2/(l*q0));
5. a = v(1)*f; b = v(2)*f; c = v(3)*f; W = v(4)*f; s = v(5)*f;
N = v(9);r = (N-1)/2;
6. CuS = 0; SuS = 0;
7. for j = N:-2:-N+2
8. Cu2 = mfun('FresnelC',j/2*a+r*b:s:W+j/2*a+r*b);
9. Cu1 = mfun('FresnelC',(j-2)/2*a+r*b:s:W+(j-2)/2*a+r*b);
10. CuS = CuS+Cu2-Cu1;
11. Su2 = mfun('FresnelS',j/2*a+r*b:s:W+j/2*a+r*b);
12. Su1 = mfun('FresnelS',(j-2)/2*a+r*b:s:W+(j-2)/2*a+r*b);
13. SuS = SuS+Su2-Su1;r = r-1;
14. end
15. A = complex(CuS,SuS);
16. Cv2 = mfun('FresnelC',c:s:W+c);
17. Cv1 = mfun('FresnelC',-c:s:W-c);
18. Sv2 = mfun('FresnelS',c:s:W+c);

19. Sv1 = mfun('FresnelS',-c:s:W-c);
20. B = complex(Cv1-Cv2,Sv1-Sv2);
21. B = B';n = size(B);B = repmat(B(:,1),1,n);
22. A = A';A = repmat(A(:,1),1,n);A = (A)';
23. C = B*A;D = C.*conj(C);
24. D = t*D/max(max(D));
25. m = 2*fix(((2*W)/s)/2);
26. E = zeros(m+1,m+1);
27. for q = 1:1:m/2+1;
28. for p = 1:1:m/2+1; E(m/2+2-p,q+m/2) = D(p,q); end
29. end
30. for q = m+1:-1:m/2+2;
31. for p = 1:1:m/2+1;E(p,-q+2+m) = E(p,q);end
32. end
33. for q = 1:1:m+1;
34. for p = 1:1:m/2;E(-p+2+m,q) = E(p,q);end
35. end
36. y = -W:s:W;ymm = y/f;
37. imagesc(ymm,ymm,E,[0,1]);colormap(gray);
REFERENCES
[1] Talbot, F. Facts relating to optical Science No IV. Philos. Mag.1836,

9, 401-407.
[2] Rayleigh, L. On copying diffraction gratings and some phenomena
connected therewith. Philos. Mag. 1881, 11, 196-205.
[3] Edgar, R. F. The Fresnel diffraction images of periodic structures. J.
Mod. Opt.1969, 16, 281–287.
[4] Winthrop J. T.; Worthington, C. R. Theory of Fresnel images. I.
Plane periodic objects in monochromatic light. J. Opt. Soc. Am. 1965,
55(4), 373–381.
[5] Kołodziejczyk, A. Realization of Fourier images without using a lens

by sampling the optical object. J. Mod. Opt.1985, 32, 74–746.
[6] Cheng, Y. S.; Chang, R.-C. Theory of image formation using the
Talbot effect. Appl. Opt.1994, 33(10), 1863–1874.
[7] Lu, Y.; Zhou, C.; Luo, H. Talbot effect of a grating with different
kinds of flaws. J. Opt. Soc. Am. A 2005, 22(12), 2662–2667.
[8] Noponen E.; Turunen, J. Electromagnetic theory of Talbot imaging.
Opt. Commun.1993, 98(1-3), 132–140.
[9] Torcal-Milla, F. J.; Sanchez-Brea, L. M.; Vargas, J. Effect of
aberrations on the self-imaging phenomenon. J. Lightwave Technol.
2011, 29(7), 1051–1057.
[10] Berry M.V.; Klein, S. Integer, fractional and fractal Talbot effects. J.
Mod. Opt.1995, 43, 2139–2164.
[11] Bryngdahl, O. Image formation using self-imaging techniques. J.
Opt. Soc. Am.1973, 63(4), 416–419.
[12] Spagnolo, G.; Ambrosini, D.; Paoletti, D. Displacement measurement
using the Talbot effect with a Ronchi grating. J. Opt. A: Pure Appl.
Opt 2002, 4(6), S376–S380.
[13] Bhattacharya, J. C. Measurement of the refractive index using the
Talbot effect and a moire technique. Appl. Opt. 1989, 28(13), 2600–
2604.
[14] Lohmann, A. W.; Thomas, J. A. Making an array illuminator based
on the Talbot effect. Appl. Opt. 1990, 29(29), 4337–4340.
[15] Isoyan, A.; Jiang, F.; Cheng, Y. C.; Cerrina, F.; Wachulak, P.;
Urbanski, L.; Rocca, J.; Menoni, C.; Marconi, M. Talbot lithography:
self-imaging of complex structures. J. Vac. Sci. Technol. B 2009,
27(6), 2931–2937.
[16] Huang, F.; Zheludev, N.; Chen, Y.; de Abajo, F. Focusing of light by
a nanohole array. Appl. Phys. Lett. 2007, 90(9), 091119.
[17] D’Amato, F. X.; Siebert E. T.; Roychoudhuri C. Coherent operation
of an array of diode lasers using a spatial filter in a Talbot cavity.
Appl. Phys. Lett. 1989, 55, 816.
[18] Abedin, K. M.; Islam M. R.; Haider, A.F.M.Y. Computer Simulation
of Fresnel Diffraction from Rectangular Apertures and Obstacles by
Using the Fresnel Integrals Approach. Opt. Laser Technol. 2007,

39(2), 237-246.
[19] Abedin, K.M.; Rahman, S.M.M. Fresnel Diffraction from N-
apertures: Computer Simulation by Iterative Fresnel Integrals
Method. Optik, 2015, 126(23), 3743-3751.
[20] Case, W.B; Tomandl, M.; Deachapunya, S.; Arndt, M. Realization of
optical carpets in the Talbot and Talbot-Lau Configurations, Opt.
Express, 2009, 17(23), 20966-20974.
[21] http://interactive.quantumnano.at/advanced/quantum-experiments/
talbot-effect//.
[22] Kim, M. S.; Scharf, T.; Menzel, C.; Rockstuhl, C.; Herzig, H. P.
Talbot images of wavelength-scale amplitude gratings. Opt. Express,
2012, 20(5), 4903-4920.
[23] Kim, J. M. et.al. Observation of the Talbot effect using broadband
hard x-ray beam. Opt. Express, 2010, 18(24), 24975-24982/.
[24] Mark, M.J. et.al. Demonstration of the temporal matter-wave Talbot
effect for trapped matter waves. New J. Phys. 2011, 13, 085008.
Chapter 4
MODELING OF THE LIGHT SCATTERING

BY A YARN
Zeguang Pei
College of Mechanical Engineering,
Donghua University, Shanghai, China
ABSTRACT
Yarn twist is an important parameter which determines the

characteristics and physical properties of a spun yarn. The yarn twist T
tan 
can be calculated by the formula: T  , where β is the surface twist
D
angle formed between the surface fibers and the yarn axis, and D is the
yarn diameter. Therefore, β and D need to be measured to obtain the
twist. Difficulty usually lies in measuring the surface twist angle due to
the tiny diameters of the surface fibers. In this chapter, the feasibility of
determining the surface twist angle based on the light scattering by the
yarn is investigated as a foundation for the online measurement of yarn
twist which is at present of great challenge to the industry. Beckmann’s

Corresponding Author E-mail: zgpei@dhu.edu.cn.
130 Zeguang Pei
surface twist angle. The solution of the scattering intensity distribution
for the yarn surface profile is derived and numerical results indicate that
the highest light intensity on the backward light scattering pattern lies
along the direction perpendicular to the surface fibers regardless of the
fiber or yarn parameters. The result of this chapter can serve as the
theoretical basis for the measurement of yarn surface twist angle based on
backward light scattering by the yarn.
Keywords: Beckmann’s model, light scattering, yarn, twist angle
1. INTRODUCTION
A yarn is a linear assembly of short-staple fibers or filaments held

together by insertion of twist to form a continuous strand [1], as shown in
Figure 1. The property of a yarn is, to a great extent, dependent on its
structure. Of the various yarn structural parameters, the twist level (or
degree of twist), or for short, the twist, is an important one since all spun
yarns possess a twisted characteristic. The twist level of a yarn is defined
as the number of turns of twist per unit length. The insertion of twist into a
yarn causes the constituent fibers (or filaments) to adopt a helical form and
the yarn twist can be calculated by the following equation:
tan 
T (1)
πD
where T is the yarn twist in turns per unit length, β is the surface twist
angle, and D is the yarn diameter. In order to control and evaluate the
quality of a yarn, measurement of the yarn twist is a critical and
indispensable process. Using Eq. (1) to determine the yarn twist requires
the measurement of the surface twist angle β and yarn diameter D. For the
past decades, various methods and devices have been proposed for the
measurement of the twist angle of a yarn. In addition to mechanical
methods [2], optical methods have also played an important role in these
measurement methods. Plöckl [3] and Hensel et al., [4] developed
Modeling of the Light Scattering by a Yarn 131
apparatus which determined the surface fiber orientation of the yarn by

analyzing the angular distribution of the intensity of the light reflected by
the surface fibers. The intensity of the light detected would reach its
maximum when the direction of detection was aligned with the surface
fibers. The surface twist angle of a yarn can also be measured using digital
image processing method or computer vision technique based on the
analysis on the microscopic image of the yarn body [5-7]. Surface twist
angle was extracted from the bright surface contours indicating the fiber
orientation using specifically developed algorithms. Paramonov et al.,
proposed a method for measuring the twist angle of a yarn by its small-
angle light scattering pattern. For the yarn at a right angle to the incident
laser beam, a light scattering pattern in a form of ‘✱’ was observed on the
screen arranged in the forward side. The angles between the central line
and each of the two lines forming the cross were taken as being equal to
the inclination angles of the fibers on both sides of the yarn to the yarn axis
[8]. However, the two lines forming the cross were not scattered by the
surface fibers but actually by the yarn body with fluffy edges. Yarn surface
twist angles were also reported to be determined based on the frequency
domain analysis of the microscopic image of yarn surface by Ozkaya et al.,
[7] and Kofnov et al., [9]. Fibers on the yarn surface can be viewed as
aligning in the similar direction and their alignment forms a periodic
texture. The dominant direction in the frequency domain lies in the
perpendicular direction to the periodically aligned surface fibers.
Therefore, the surface twist angle can be determined as complementary to
the angle formed between the direction having the maximum magnitude in
the spectrum and the yarn axis.
As can be seen from literature, although the measurement of yarn twist
angle has been a longstanding research topic, attention has still been paid
to its measurement using optical methods, especially based on the light
scattering pattern by the yarn. In view of this, there is a great need to
clarify the characteristic of light scattering by a yarn and the principle
behind it. This chapter presents a theoretical investigation on light
scattering by the yarn surface based on the Beckmann’s scattering model.
The light scattering process of an incident ray by a yarn is simulated and
132 Zeguang Pei
the solution of the scattering intensity distribution for the yarn surface
profile is derived. The feasibility of determining yarn twist angle based on
the backward light scattering by the yarn surface is determined by
analyzing the relation between the scattering pattern and the surface twist
angle. Numerical results are finally confirmed by comparing it with the
experimental results.
Figure 1. Expression of the yarn twist.
2. GENERALIZED BECKMANN’S MODEL FOR LIGHT

SCATTERING FROM YARN SURFACE
The Beckmann’s model [10] is a Kirchhoff scalar approach which has

been extensively used for modeling light scattering from rough surfaces.
Sanchez-Brea et al., adopted the Beckmann’s model to determine the far-
field diffraction pattern of a metallic cylinder of infinite length which is
illuminated in oblique incidence [11-12]. Fan et al., investigated the light
scattering from periodic rough cylindrical surfaces by applying
Beckmann’s model and superimposing the individual intensities to form
the scattering pattern [13]. Ragheb et al., used a modified Beckmann
scattering model to estimate surface roughness parameters for dielectric
surfaces using pixel brightness measurements [14]. Perez-Quintián et al.,
[15] extended the theory of Beckmann for reflection to model the diffusion
of light transmitted from rough surfaces. In this section, the Beckmann’s
scattering model is adopted to investigate the light scattering from the yarn
surface which has a rough and periodic texture. Here, the yarn is assumed
to be of a cylindrical shape and lies in the horizontal direction. At a point
on the yarn surface, the yarn surface is approximated by a tangent plane S
to the cylinder. The normal of S at this point inclines an angle α relative to
the Z axis pointing vertically upward, as shown in Figure 2. A local
Cartesian coordinate system, x0y0z0 which obeys the right-hand rule, is set
on the yarn with y0 parallel to the yarn axis, z0 superposed with the normal
of S, and x0y0 lie in S. The scattering model is applied to this tangent plane
S as an incident wave lying in the x0z0 plane and propagating
perpendicularly to both the y0 axis and Z axis strikes the yarn surface with
π
an angle of incidence θ1   α . A screen is placed at a distance DS from
2
the yarn to receive the scattered waves. A screen coordinate system XSYSZS
is set on the screen. Let k1 and k2 denote the propagation vectors of the
incident wave and scattered wave, respectively, as shown in Figure 3, we
have:
k 1  k  sin θ1x0  cosθ1z 0  (2)
k 2  k  sin θ2 cosθ3x0  sin θ2sinθ3y 0  cosθ2 z 0  (3)
2π
where k  is the incident wave number, λ is the incident wavelength,
λ
θ2 and θ3 are the scattering angles, x0, y0, z0 are unit vectors for the local
Cartesian coordinate system x0y0z0.
Let P be the point of observation in the Fraunhofer zone of diffraction,
i.e., on the screen. The scattered field E2 at point P can be expressed by the
Helmholtz-Kirchhoff integral:
1   E 
E2  P     E n   n  dS (4)
4 S
134 Zeguang Pei
e
i k2 R0 k 2 r 
with   , where R0 is the distance from point P to the origin of
R0
the coordinate system x0y0z0, r is the vector pointing from the origin to the
considered point on the yarn surface and is expressed as:
r  xx0  yy 0    x, y  z 0 (5)
where   x, y  is the function expressing the height of the surface profile

the yarn.
Figure 2. Side view of the schematic diagram of the light scattering by a yarn.
The field on S is subject to the following boundary conditions:
 E 
 E S  1  R  E1 ,    1  R  E1k 1  n (6)
 n S
ik 1 r
where E1  e is the incident plane wave of unit amplitude, R is the
Fresnel reflection coefficient, n is the normal to the surface at the
considered point.
Figure 3. Perspective view of the schematic diagram of the light scattering by the yarn
surface.
For a surface extending from x0 = -X0 to X0, and y0 = -Y0 to Y0 with its
area A = 4X0Y0 ≫ λ2 and a constant R, a scattering coefficient representing
the normalized scattered field E2 can be introduced as:
F3
  θ1 ,θ 2 ,θ3    e
ivr
dxdy (7)
A A
where
v  k 1  k 2  k  sin θ1  sin θ2 cosθ3  x0  sin θ2sinθ3y 0   cosθ1  cosθ2  z 0 

(8)
136 Zeguang Pei
R 1  cos θ1 cos θ 2  sin θ1 sin θ 2 cos θ3 

F3  (9)
cos θ1  cos θ1  cos θ 2 
As mentioned, the alignment of fibers on the yarn surface forms a

periodic texture, that is,   x, y  is periodic in both x0 and y0 directions
d d
with a period of  x  in the x0 direction and  y  in the y0
cos  sin 
direction, where d is the fiber diameter. The directions in which the
maxima of the scattered light will be obtained are determined by the
following grating equations:
λ
sin θ 2 cosθ3  sin θ1  m  m  0, 1, 2,... (10)
x
λ
sin θ 2 sin θ3  n  n  0, 1, 2,... (11)
y
π π
in which 0  θ1  , 0  θ2  .
2 2
The surface profile function   x, y  of the yarn is determined by the
alignment characteristic of the fibers on the yarn surface and the shape of
the fiber cross section. Circular cross sections are adopted by some man-
made fibers such as polyester and nylon, while some fibers used in the
textile industry may also have non-circular cross sections, such as cotton
and ramie fibers which are both natural fibers. Here for the simplification
of modeling, the yarn is supposed to have a simple helix structure and the
fibers in the yarn have circular cross sections. For calculation convenience,
the origin of the local coordinate system x0y0z0 is set at a point on the axis
of a surface fiber. Another local coordinate system x1y1z1 is also applied to
the tangential plane S. This local coordinate system shares the same origin
with x0y0z0 with x1 along the surface fiber axis and z1 superposed with z0, as
shown in Figure 4. According to geometrical relation,
y0  rC cos  (12)
x0  rC sin  (13)
  
y1  rC cos         rC sin       rC  sin  cos   cos  sin  
 2 
 y0 sin   x0 cos 
(14)
 ( x0 , y0 )   d / 2  y12   d / 2   y0 sin   x0 cos  

2 2 2
(15)
namely,
r  xx0  yy 0   d / 2   y sin   x cos  

2 2
z0 (16)
Substituting Eq. (8) and (16) into (7) yields
F3 X0 Y2 
i vx x  v y y  vz  

2X y  
 X 0 Y1
e dxdy
F3

exp ik  sin θ1  sin θ 2 cos θ3  x   sin θ 2 sin θ3  y (17)
X0 Y2

2X y  
 X 0 Y1
  cosθ1  cosθ 2   d / 2    y sin   x cos  

2 2 


2 x cos   d 2 x cos   d
where Y1  , Y2  .
2sin  2sin 
138 Zeguang Pei
Figure 4. Schematic diagram of the geometrical relations for expressing yarn surface
profile: (a) x0y0 plane; (b) y1z1 plane (A-A section).
The Fresnel reflection coefficient R depends on the local angle of

incidence, the electrical properties of the fibers in the yarn, and the
polarization of the incident light. Suppose the incoming light is randomly
polarized, then [16-17]
1  tan   r  sin   r  
2 2
R
2

1 2
R  R
2
   
2  tan 2   r  sin 2   r  
(18)
where R∥ is the reflection coefficient when light is polarized in the incident

plane, R is the reflection coefficient for the light polarized perpendicular
to the incident plane,  is the angle of incidence relative to the fiber
surface, r is the angle of refraction.
According to Snell’s law,
sin 
nf  (19)
sin r
where nf represents the refractive index of the fiber.

According to geometrical relations, the local angle of incidence  is
determined by
 π
cos   sin σ-atan  cotαcos  sin α  cos σ-atan  cotαcos  cos α cos  ,    0, 
 2
(20)
where σ   0, π  is the angle between the normal to the fiber surface and
the y0 axis.
The average value of R is calculated by taking its integral over the
illuminated part of the fiber surface:
1 π
π 0
R  Rd (21)
The spherical coordinates (θ2, θ3) is then converted to the screen

coordinates (yS, zS). As shown in Figure 3, according to geometrical
relations and for Ds ≫ D, for the forward scattering,
140 Zeguang Pei
D 
yS   sin   DS  sec  tan θ3 DS sec  tan θ3 (22)
2 
D  D
zS   sin   DS  cot δ2  cos  DS cot δ2 (23)
2  2
where δ2 is determined by
tan δ1  sin δ2 tanθ3 sec  (24)
cos δ1 cos  δ2  α   cosθ2 (25)
While for the backward scattering, the following geometrical relations

can be obtained:
 D  π 
yS   DS  sin   csc δ2 cos     δ2   tan θ3 DS csc δ2 sin   δ2  tan θ3
 2  2 
(26)
 D  D
zS   DS  sin   cot δ2  cos  DS cot δ2 (27)
 2  2
where δ 2 is determined by:
π 
tan δ1  tanθ3 cos     δ2    tanθ3 sin   δ2  (28)
2 
cos δ1 cos  α  δ2   cosθ2 (29)

3. NUMERICAL IMPLEMENTATION
Figure 5. Light scattering by a sinusoidal surface.
The light scattering model is solved by a code implemented in the

Matlab® software. In order to validate our code, the light scattering of a
2
two-dimensional sinusoidal surface   x   h cos kx , where k  ,

  10 and kh  3 for an angle of incidence 1  45 is solved and
then compared with results adapted from Brekhovskikh [18] and
Beckmann et al., [10], as shown in Figure 5. Perfect match can be found
for our results with those from literature. This code is then used to study
the relation between the scattering intensity distribution of the yarn and
various yarn parameters and its dependence on the yarn twist. The intensity
of the light scattering is calculated by Eq. (17) and the coordinates for the
corresponding scattering pattern on the screen can be determined by Eq.
(22)-(29). The modeling parameters for predicting the scattering pattern by
the yarn are shown in Table 1.
142 Zeguang Pei
4. RESULTS AND DISCUSSION
Table 1. Modeling parameters for calculating the scattering

pattern by the yarn
Parameters Case 1 Case 2 Case 3 Case 4 Case 5

Incident wavelength, λ 632.8 632.8 632.8 632.8 532 nm
nm nm nm nm
Yarn twist angle, β 25° 20° 25° 30° 35°
Fiber diameter, d 12 μm 12 μm 14 μm 16 μm 18 μm
Yarn diameter, D 200 μm 200 μm 180 μm 240 μm 220 μm
Illumination position, α 30° 90° 90° 90° 90°
Twist direction Z twist Z twist S twist Z twist S twist
Fiber refractive index, nf 1.547 1.526 1.547 1.6 1.546
R (Calculated by
0.3509 0.3348 0.3401 0.3529 0.3398
Eq.(21))
Figure 6 (a-b) show the numerical solution of the three-dimensional

light scattering process of the yarn in Case 1 (the general case). It can be
seen that the directions in which waves are scattered are symmetric with
respect to the x0z0 plane. The mode with m = 0, n = 0 corresponds to the
specular reflection, which can be deduced from Eq. (10) and (11). The
modes with same m lie on the surface of a cone with corresponding modes
of ± n located on either side of the plane of symmetry. All the scattered
waves form a series of concentric half-cones around the x0 axis with modes
m >0 lying on the forward side of the mode m = 0 and modes m <0 on the
backward side. With the increase of |m|, the scattered waves propagate
more closely to the tangent plane x0y0 toward both the forward and
backward directions until the limit of condition (10) is reached. While with
the increase of |n|, the scattered waves propagate more closely to the
tangent plane x0y0 laterally until the limit of condition (11) is reached.
Figure 6. Modeled light scattering process of an incident ray by a yarn (α = 30°): (a)
Modes of m <0; (b) Modes of m >0. Note that 1 mode in every 8 modes is shown in
(a).
As can be seen from the scattering process, a small α indicates the

portion illuminated is closer to the yarn edge. The forward scattered waves
will be more approaching to the small-angle region in the forward
direction. Therefore, the small-angle region for the forward light scattering
contains more information of the yarn body profile, i.e., the low frequency
information. Comparatively, for a larger α which indicates the incoming
wave is more perpendicular to the yarn surface, the backward scattered
waves propagate more towards the small-angle region in the backward
direction. As a result, the small-angle region for the backward light
scattering contains more information of the surface fibers of the yarn, i.e.,
the high frequency information. Therefore, the backward scattering
diagram generated by a plane wave incident normally onto the yarn surface
(α = 90°) is adopted for determining the surface twist angle of the yarn, as
shown in Figure 7. Cases 2-5 listed in Table 1 are solved and results are
shown in Figure 8(a-d), respectively. The backward scattering patterns are
144 Zeguang Pei
observed along the positive X direction, as being received by an observer

or a camera. As can be seen, an important feature that can be found on the
far-field backward scattering pattern of the yarn is that the highest light
intensity lies in the direction perpendicular to the direction of alignment of
the yarn surface fibers regardless of the light source, fiber and yarn
parameters. This result can serve as the theoretical basis for the
measurement of yarn surface twist angle.
Figure 7. Backward light scattering by the yarn.
Figure 8. Calculated scattering patterns of (a) case 2 (b) case 3 (c) case 4 (d) case 5.
In order to verify the simulation results, the image of the backward

scattering pattern of a yarn is captured by a CCD camera, as shown in
Figure 9. It can be found that the scattering pattern is generally in a shape
of an inclined ellipse whose major axis orientation indicates the direction
in which the highest light intensity distributes. The orientation of the major
axis of the elliptical pattern is determined and the angle between the major
axis and the vertical direction, β1, is measured. β1 is then compared with
the surface twist angle β2 measured on the microscopic image of the yarn
sample.
Figure 9. The backward scattering image of a Ne 32 cotton yarn with Z twist captured
by the CCD camera.
Figure 10. Comparison of twist angles measured based on the backward light
scattering and microscopic images of various yarn samples.
146 Zeguang Pei
Figure 10 shows the comparison of β1 and β2 for four groups of yarns.

Discrepancies between the measurement results of the two methods can be
observed. The differences between β1 and β2 are considered to be relatively
small and in an acceptable range. The result proves that the highest light
intensity of the backward scattering pattern lies in the direction
perpendicular to the direction of alignment of the yarn surface fibers and
validates the prediction result based on the light scattering model. The
results also show that the light scattering method is less sensitive to the
cross-sectional shape of the fibers in the yarn since the alignment direction
of the periodic structure of the yarn surface is more critical for the
direction of the scattering pattern which is used to determine the twist
angle. While the measurement results of the surface twist angles using the
two methods are comparable, the light scattering method is less labor-
intensive since the computer algorithm is straightforward for detecting the
orientation of the scattering pattern, while measuring yarn twist angle
based on the yarn microscopic image involves more sophisticated
computer algorithm and also a high-magnification objective.
CONCLUSION
In this chapter, the Beckmann’s model is adopted to investigate the

light scattering from the surface of a yarn which is approximated by a
tangent plane to the cylindrical yarn. The yarn surface is modeled as a
rough and periodic profile according to the alignment characteristic of the
surface fibers. The light scattering model is solved and both the light
scattering process of the yarn and the light scattering pattern are studied. It
is found that the highest light intensity on the far-field backward light
scattering pattern of the yarn lies in the direction perpendicular to the
direction of alignment of the yarn surface fibers. The model’s prediction is
found in agreement with experimental results. The result of this chapter
can serve as the theoretical basis for the measurement of yarn surface twist
angle.
ACKNOWLEDGMENTS
This research is financially supported by the Hong Kong Scholar

Program (No. XJ2013019 and GYZ49) and partly by the National Natural
Science Foundation of China (No. 11302048).
REFERENCES
[1] Lawrence, C. A. Fundamentals of Spun Yarn Technology. Boca

Raton: CRC Press LLC, 2003.
[2] Farah, B. D., Woo, J. L. Continuous measurement of yarn diameter
and twist. US Patent 4648054, 1987.
[3] Plöckl, J. Method of measuring the pitch angle of yarns provided
with a twist. US Patent 3985451, 1976.
[4] Hensel, R., Wampfler, H., Seitz, P. Method and apparatus for
determining the structure of yarns in the region of their surface. US
Patent 5521395, 1996.
[5] Vas, L. M., Halász, G., Takács, M., Eördögh, I., Szász K.
Measurement of yarn diameter and twist angle with image processing
system. Periodica Polytechnica Ser. Mechanical Engineering, 1994,
38, 277-296.
[6] Guo, B. P. Mechanisms of yarn twist blockage caused by frictional
contact. PhD Thesis. The Hong Kong Polytechnic University, Hong
Kong, 1999.
[7] Ozkaya, Y. A., Acar, M., Jackson M. R. Yarn twist measurement
using digital imaging. Journal of the Textile Institute, 2010, 101, 91-
100.
[8] Paramonov, A. V., Kornyukhina T. A., et al. Assay method of
determining the twist of a yarn. Tekstilnaia Promyshlennost, 1978, 3,
72-74.
148 Zeguang Pei
[9] Kofnov, O. V., Shlyakhtenko, P. G. Analysis of computed diffraction

pattern diagram for measuring yarn twist angle. Textiles and Light
Industrial Science and Technology, 2014, 3, 1-5.
[10] Beckmann, P., Spizzichino, A. The Scattering of Electromagnetic
Waves from Rough Surfaces. London: Pergamon Press Ltd., 1963.
[11] Sanchez-Brea, L. M., Salgado-Remacha, F. J. Three-dimensional
diffraction of a thin metallic cylinder illuminated in conical
incidence: application to diameter estimation. Applied Optics, 2008,
47, 4804-4811.
[12] Sanchez-Brea, L. M., Bernabeu, E. Diffraction by cylinders
illuminated in oblique, off-axis incidence. International Journal for
Light and Electron Optics, 2001, 112, 169-174.
[13] Fan, Y. Y., Huynh, V. M. Light scattering from periodic rough
cylindrical surfaces Applied Optics, 1993, 32, 3452-3458.
[14] Ragheb, H., Hancock, E. R. The modified Beckmann-Kirchhoff
scattering theory for rough surface analysis. Pattern Recognition,
2007, 40, 2004-2020.
[15] Perez-Quintián, F., Rebollo, M. A., Gaggioli, N. G. Diffusion of light
transmitted from rough surfaces. Journal of Modern Optics, 1997,
44, 447-460.
[16] Hecht, E. Optics, Fourth Edition. San Francisco: Pearson Education,
Inc., 2002.
[17] Noginov, M. A. Solid-state random lasers. New York: Springer
Science+Business Media Inc., 2005.
[18] Brekhovskikh, L. M. The diffraction of waves by a rough surface,
part II. Zh. Eksper. I Teor. Fiz. 1952, 23, 289-304.
Chapter 5
THE UNSTEADY ANALYSIS OF A SOLAR

CHIMNEY POWER PLANT
Ahmed Ayadi*, Zied Driss, Haythem Nasraoui,

Abdallah Bouabidi and Mohamed Salah Abid
Laboratory of Electro-Mechanic Systems (LASEM),
National Engineering School of Sfax (ENIS),
University of Sfax (US), Sfax, Tunisia
ABSTRACT
The solar chimney power plants (SCPP) could generate electricity

using solar as source. The study of the air flow characteristics is required
since the construction cost of these devices is expensive. This chapter
focus on numerical studies of the unsteady airflow characteristics within a
solar chimney power plant. Ansys Fluent 17.0 is used to simulate the air
flow inside the solar setup. A prototype is built in the National School of
Engineers of Sfax, University of Sfax, Tunisia in the North Africa to
validate our numerical study. The prototype is composed by a collector,
an absorber, a chimney and a turbine. The collector is characterized by a
diameter equal to 2750mm, a roof height equal to 50mm. However, the
*
Corresponding Author Email: ahmed.ayadi.gem@gmail.com.
150 Ahmed Ayadi, Zied Driss, Haythem Nasraoui et al.
chimney is characterized by a height equal to 3000mm. The turbine

diameter is equal to 150mm. The air characteristics is presented and
analyzed. In fact, the distribution of the air temperature, the air velocity
and the total pressure. Findings confirm that the ambient air temperature
and the solar radiation are important parameters for the enhancement of
the solar chimney efficiency.
Keywords: solar chimney power plant, CFD, unsteady state, turbulent

flow, temperature, solar radiation
NOMENCLATURE
d Chimney diameter (mm)

dtur Turbine diameter (mm)
D Collector diameter (mm)
e Thickness (mm)
g Gravitational acceleration (m.s-2)
G Global radiation (W.m-2)
h Collector height (mm)
H Chimney height (mm)
p Pressure (Pa)
r Radial coordinate (mm)
T Temperature (K)
u Velocity in radial direction (m.s-1)
V Air velocity (m.s-1)
w Velocity in tangential direction (m.s-1)
z Axial coordinate
µ Dynamic viscosity (m2.s-1)
ρ Density of the air (kg.m-3)
1. INTRODUCTION
The solar chimney power plants (SCPP) are devices which are
designed to generate electrical energy using the solar radiation. Generally,
The Unsteady Analysis of a Solar Chimney Power Plant 151
a solar chimney power plant is equipped by one or many turbines to extract

the kinetic energy of the wind flowing upward due to the natural
convection. The generator, which is coupled with the turbine, converts the
wind kinetic energy into electrical power. Several papers noted that the use
of the solar chimney power plant is encouraging to produce electricity
(Ayadi et al. 2018; 2017a; 2017c). The first prototype of the solar chimney
power plant was constructed in Manzanares. The prototype is characterized
by a height equal to H = 194.6 m and a collector radius equal to R = 122 m.
Haaf et al. (1982;1984) published the results of preliminary tests of the
solar chimney power plant. They presented the energy audits, the collector
efficiency values, the pressure losses due to friction and losses in the
turbine section. Ayadi et al. (2017b) established a numerical model to
assess the performance of the SCPPs while varying the collector roof
angle. The authors reported that the efficiency of the solar chimney power
plant increases with a negative angle of the collector roof. Xu et al. (2011)
analyzed numerically the influence of the solar radiation and the pressure
drop across the turbine on the power output of the SCPP. Recently, Ming
et al. (2016) carried out a numerical simulation for a solar chimney power
plant coupled with a 5-blade turbine. The considered geometry is
characterized by the chimney height equal to 400m, the chimney radius
equal to 30m and the collector radius equal to 1500m. The authors
presented the impact of the rotational speed of the turbine on the average
velocity of the chimney outlet, the average temperature of the chimney
outlet, the pressure drop across the turbine and the mass flow rate of the
system. The maximum generated power and the turbine efficiency are
about 10 MW and 50%, respectively. Guo et al. (2015) presented a 3D
numerical simulation of a solar chimney coupling with a turbine. In their
work, they have varied the rotational speed of the turbine to study the
optimal operating condition of the turbine. In their study, they have taken
account the influences of the incident angle of the sunlight. Results
revealed that the hourly variation of the zenith angle of the sun is an
important parameter to predict the annual performance of the SCPP.
Asnaghi and Ladjevardi (2012) presented a numerical model to simulate
the SCPP performance. Their work confirms that the power generation
depends on the sunshine duration, solar irradiation and the ambient

temperature. Indeed, they have noted that the power output is proportional
to chimney mass flow rate. Al-Kayiem and Aja (2016) noted that the
collector efficiency depends on the collector material such as the soil type,
the cover material, the slope of the collector and the availability of the
solar radiation. Kalash et al. (2013) experimentally analyzed a solar updraft
power plant prototype. Their study confirmed that the solar radiation and
the ambient temperature affect the chimney inlet temperature. They noted
that the chimney inlet temperature increases as the radiation increases.
However, it was noticed that the influence of the ambient temperature is
minimal during the noon period when the solar radiation starts to decrease.
In this paper, numerical studies were reported to highlight the effect of the
environmental conditions on the flow characteristics inside a solar chimney
power plant.
Although much work has been done to date, it has been noted that the
most published works presented the profiles of the air velocity and the air
temperature in a specific position within the solar setup. However, these
profiles could sometimes be misinterpreted. In fact, some researchers did
not focus on the distribution of the air temperature, the air velocity and the
pressure. These parameters are important to optimize the solar system.
Indeed, the most published works deal with a larger prototype such as the
Manzanares prototype. This chapter is identified as important to engineers
in supplying them technical solutions to enhance the performance of a
solar chimney power plant.
2. MATERIAL
The solar chimney prototype was built in the national school of

engineers of Sfax at University of Sfax located in the South East of Tunisia
of North Africa. The latitude of the setup is equal to φ=34.726˚N and the
longitude is equal to L=10.717˚ E. Tunisia has been identified as a case
study to quantify the amount of the solar energy available locally and its
distribution across the country (Ayadi et al. 2016b).
Figure 1. The SCPP configuration.
The collector diameter is equal to D = 2750mm, the collector roof

height is equal to h = 50mm, the chimney height is equal to H = 3000mm
and the turbine diameter is equal to dtur = 150mm. The collector is a
transparent membrane which covers the ground. It is made by a strong
transparent polymer. This plastic cover has a high thermal characteristic
and a resistance to breakage. However, the chimney is a polyvinyl chloride
pipe. The wood is chosen to construct the external surface of the absorber,
due to commercial availability and economic reasons. In order to analyze
the temperature field within the SCPP, ten sensors have been
used.Temperature measurement devices are thermocouple type-J. They
have been equally distributed along the collector radius. Indeed, an
electronic card is used to periodically record the air temperature values
every one hour. However, for the measurement of the air velocity in the
whole volume of the system, it was necessary to have a measuring
instrument with the best precision possible. A hot-wire anemometer is used
to measure the air velocity along the collector. This sensor is simple and
offers the greatest flexibility of use.
3. NUMERICAL METHOD
3.1. Mathematical Formulation
The CFD code, Ansys Fluent 17.0, is used to simulate the air flow
within the solar chimney power plant. The air flow is governed by the
continuity and the motion equations which are presented in the cylindrical
coordinate. The continuity equation is written as follows:
 1  (rr ) 1  ( )  (z )

   0
t r r r  z (1)
 r 
 
V    
 
 z (2)
where ρ is the density of the air, r is the radial coordinate

The motion equations are written as follows:
 r r  r 

2
r 
  r  -  z 
 t r r  z
 r

p   1 rr 

2
1  r 2 
2
 r 
-     -    g r
 r  r r 
2 2 2 2
r r  r  z  (3)
      r   

  r    z 
 t r r  r z 
1 p    1 r  2
1   2 r
2
  
        g 
r   r  r r  2

2 2
 z
2
 r r  (4)
 z z  z z 

  r   z 
 t r r  z 
p 1  z 
2
1  z
2
 z 
   r     g z
 r r  r 
2 2 2
z r  z  (5)
where µ is the dynamic viscosity and p is the pressure.
3.2. Boundary Condition
Computational results are critically affected by the boundary

conditions of the system. A 3D model was proposed for all considered
simulations. A standard atmospheric pressure and temperature conditions
are imposed. The collector inlet is specified as the pressure inlet and the
chimney outlet is specified as the pressure outlet. In fact, the pressure inlet
and the pressure outlet are equal to the atmospheric pressure defined by
Patm = 101325 Pa. Indeed, the collector roof is considered as a semi-
transparent wall.
3.3. Meshing
The establishment of efficient computational calculations requires the

generation of the different types of meshing. Otherwise, poor quality grid
will cause inaccurate solutions and slow convergence. In fact, numerical
results vary from one configuration to another (Ayadi et al. 2016a). A
meshing grid is generated using Ansys-Meshing preprocessor. The
appropriate choice of grid type depends on the geometric system and the
number of nodes. In this context, many different meshing configurations
are tested and a grid independence test was done to evaluate the optimum
node sizes and elements. In order to reduce the amount of time spending in
the meshing generation, 840,000 nodes were used for all numerical
simulations since this meshing gives a good agreement compared with the
experimental data.
4. UNSTEADY STATE OF THE SOLAR TURBINE
Four instances are chosen to highlight the effect of the environmental

conditions on the performance of the solar chimney power plant. For each
configuration, the distribution of the magnitude velocity, the air
temperature and the pressure were presented and discussed in two different
planes. The first plane is defined by z = 0 m and the second plane is in the
median of the turbine and is defined by θ = 0°.
4.1. Static Temperature
The distribution of the air temperature in the plane θ = 0° is presented

and discussed for different considered hours equal to t = 10:00a.m.,
t = 12:00a.m., t = 02:00p.m. and t = 04:00p.m. From figure 2, it is clear
that the distribution of the temperature is similar for the different cases. For
all considered instances, the minimum values are located at the zone
surrounding the inlet of the collector. However, maximum values were
recorded near the absorber and at the chimney bottom. Although, a
difference in values has been observed from one case to another. In fact,
the higher values were recorded within the noon time particularly at t =
12:00a.m. and t = 02:00p.m. This fact is explained by the maximum values
of the ambient air temperature and the solar radiation at this period. The
comparison between these results confirms that the air temperature in the
whole volume of the SCPP varies from one hour to another. In fact, at t =
10:00a.m., t = 12:00a.m., t = 02:00p.m. and t = 04:00p.m., the maximum
value of the air temperature is equal respectively to T = 318K, T = 329K, T
= 328K and T = 313K.
Figure 2. Distribution of the static temperature in the plane θ=0°.
4.2. Magnitude Velocity
Figure 3 presents the distribution of the magnitude velocity in the

plane θ=0° for the different considered hours. From these results, it is clear
that the magnitude velocity presents the same distribution over the time.
Indeed, the maximum value of the magnitude velocity is concentrated in

the chimney inlet. Although, it has been observed a difference in the
magnitude velocity values from one case to another. At t = 10:00a.m., t =
12:00a.m., t = 02:00p.m. and t = 04:00p.m., the maximum values of the air
velocity are equal respectively to V = 1.42 m.s-1, V = 1.87 m.s-1, V = 2.01
m.s-1 and V = 1.57 m.s-1. The comparison between these findings confirms
that the distribution of the magnitude velocity varies over time. This fact is
explained by the variation of the solar radiation intensity and the ambient
air temperature for the different instances.
Figure 3. Distribution of the magnitude velocity in the plane defined by z = 0 m.

Figure 4. Distribution of the static pressure in the plane θ = 0°.
4.3. Static Pressure
Figure 5 depicts the distribution of the static pressure in the plane

z = 0 over the considered hours equal to t = 10:00a.m., t = 12:00a.m.,
t = 02:00p.m. and t = 04:00p.m. Initially, the glass of the collector will heat
up the air inside the SCPP by the greenhouse. The pressure difference will
form a stream of air that head the air towards the chimney center.
From figure 6, it has been noted a difference in values of the static pressure
from one case to another. Over the considered instances, the values of the
static pressure vary. For example, at t = 10:00a.m., t = 12:00a.m., t =
02:00p.m. and t = 04:00p.m., the maximum values of the static pressure are
equal respectively to p = 101326.41 Pa, p = 101325.64 Pa, p = 101325.78
Pa and p = 101325.45 Pa.
Figure 5. Distribution of the static pressure in the plane defined by z.=.0 m.
4.4. Turbulent Kinetic Energy
Figure 6 shows the distribution of the turbulent kinetic energy in the

plane θ = 0° for the different considered hours. According to these results,
it has been observed that the wake characteristic of the maximum values of
the turbulent kinetic energy appears at the bottom of the chimney.
Elsewhere, the turbulent kinetic energy reduces to a lower value and

eventually to zero at the collector for all considered hours. From one hour
to another, the highest values were recorded when the values of the solar
radiation and the ambient air temperature are maximum. In fact, the
highest value appears at t = 02:00p.m. and it is equal to k = 0.212 m2.s-2.
However, the lowest value appears at t = 10:00a.m. and it is equal to k =
0.12 m2.s-2. The comparison between these results confirms that the
turbulent kinetic energy varies over the time.
(a) t=10:00a.m. (b) t=12:00a.m.
(c) t=02:00p.m. (d) t=04:00p.m.
Figure 6. Distribution of the turbulent kinetic energy in the plane θ=0°.

4.5. Turbulent Viscosity

Figures 7 and 8 show respectively the distribution of the turbulent
viscosity in the plane θ = 0° and z = 0 m. According to these results, it is
clear that the turbulent viscosity is found to be very weak in the collector
but it is maximum at the chimney. This fact can be explained by the
collision between the particles of air in the chimney volume. Besides, there
is a difference between the values of the turbulent viscosity for each
considered hour.
(a) t=10:00a.m. (b) t=12:00a.m.
(c) t=02:00p.m. (d) t=04:00p.m.
Figure 7. Distribution of the turbulent viscosity in the plane θ=0°.

(a) t=10:00a.m. (b) t=12:00a.m.
(c) t=02:00p.m. (d) t=04:00p.m.
Figure 8. Distribution of the turbulent viscosity in the plane z=0 m.
CONCLUSION
In this chapter, numerical investigations were focus on the unsteady

behavior of the air flow within a solar chimney power plant (SCPP).
Experimental and numerical simulations indicated that the increase of both
the solar radiation and the ambient air temperature increase the air velocity
inside the SCPP. The generated power has an obvious hourly and daily
variation. In fact, the local characteristics vary with the variation of the
environmental conditions. In the future, we are going to propose a new
model of a solar chimney power plant with a heat storage system.
REFERENCES
Asnaghi, A., Ladjevardi, S. M. Solar chimney power plant performance in

Iran. Renewable and Sustainable Energy Reviews, 2012;16:3383–
3390.
Al-Kayiem, H. H., Aja, O. C. Historic and recent progress in solar chimney
power plant enhancing technologies. Renewable and Sustainable
Energy Reviews, 2016; 58:1269–1292.
Ayadi, A., Bouabidi, A., Driss, Z., Abid, M. S. Experimental and
numerical analysis of the collector roof height effect on the solar
chimney performance. Renewable Energy, 2018 115C:649-662.
Ayadi, A., Bouabidi, A., Driss, Z., Abid, M. S. Study of the meshing effect
on the flow characteristics inside a SCPP. Handbook on Navier-Stokes
Equations: Theory and Applied Analysis, Chapter 10, Nova Science
Publishers, 2016a, 143-158. ISBN: 978-1-53610-308-3.
Ayadi, A., Driss, Z. Estimation of Solar Radiation on Horizontal Surfaces
in Gabès located in the South east of Tunisia. International Journal of
Mechanics and Energy, 2016b; 4:24-29.
Ayadi, A., Driss, Z., Bouabidi, A., Abid, M. S. Effect of the number of
turbine blades on the air flow within a solar chimney power plant.
Accepted in Proceedings of the Institution of Mechanical Engineers,
Part A: Journal of Power and Energy, 2017a.
Ayadi, A., Driss, Z., Bouabidi, A., Abid, M. S. Experimental and
numerical study of the impact of the collector roof inclination on the
performance of a solar chimney power plant. Energy and Buildings,
2017b; 139:263–276.
Ayadi, A., Driss, Z., Bouabidi, A., Nasraoui, H., Bsisa, M., Abid, M. S. A
computational and an experimental study on the effect of the chimney
height on the thermal characteristics of a solar chimney power plant,

Proceedings of the Institution of Mechanical Engineers-Part E: Journal
of Process Mechanical Engineering, 2017c, doi: 10.1177/
0954408917719776.
Guo, P. H., Li, J. Y., Wang, Y., Wang, Y. Numerical study on the
performance of a solar chimney power plant. Energy Conversion and
Management, 2015; 105:197–205.
Haaf, W., Friedrich, K., Mayr, G., Schlaich, J. Solar chimneys. Part1:
principle and construction of the pilot plant in Manzanares.
International Journal of Solar Energy, 1983; 2(1):3-20.
Haaf, W. Solar towers, Part II: Preliminary Test Results from the
Manzanares Pilot Plant. Solar Energy, 1984; 2:141-161.
Kalash, S., Naimeh, W., Ajib, S. Experimental investigation of the solar
collector temperature field of a sloped solar updraft power plant
prototype. Solar Energy, 2013; 98:70–77.
Ming, T., Xu, G., Pan, Y., Meng, F., Zhou, C. Fluid flow and heat transfer
of solar chimney power plant. Solar Chimney Power Plant Generating
Technology, 2016 95–125.
Xu, G., Ming, T., Pan, Y., Meng, F., Zhou, C. Numerical analysis on the
performance of solar chimney power plant system. Energy Conversion
and Management, 2011; 52(2):876-883.
Editors: Michael D. Pfeffer et al. c 2018 Nova Science Publishers, Inc.
Chapter 6
C OMPUTER S IMULATION OF H YDROGEN

T HERMAL D ESORPTION S PECTRA AND
M ODEL PARAMETERS I DENTIFICATION
Yury V. Zaika∗, DSc and Ekaterina K. Kostikova, PhD
Institute of Applied Mathematical Research
Karelian Research Centre, Russian Academy of Sciences
Petrozavodsk, Karelia, Russia
Abstract
One of the technological challenges for hydrogen materials science is
the currently active search for structural materials with important appli-
cations (including nuclear power industry and the ITER project) that will
have predetermined limits of hydrogen permeability. One of the experi-
mental methods is thermal desorption spectrometry (TDS). A hydrogen-
saturated sample is degassed under vacuum and monotone heating. The
desorption flux is measured by mass spectrometer to determine the char-
acter of interactions of hydrogen isotopes with the solid.
The chapter presents a distributed boundary value problem of ther-
mal desorption with nonlinear dynamical boundary conditions. The prob-
lem is reduced to the nonlinear functional differential equation of neu-
tral type for surface concentration. A numerical method is developed for
∗
Corresponding Author Email: zaika@krc.karelia.ru
168 Yury V. Zaika and Ekaterina K. Kostikova
TDS spectrum simulation, where only integration of a nonlinear system of

low order (compared with, e.g., the method of lines) ordinary differential
equations (ODE) is required. It is usually assumed that TDS peaks appear
due to hydrogen release from traps with different binding energies. Here,
it was demonstrated using the diffusion model (for homogeneous mate-
rial) that if surface processes are taken into account, two-peak spectra can
be obtained even for very thin experimental samples. The main final out-
put of the chapter is a geometrically transparent method for solving the
inverse problem of surface parameter identification where desorption and
diffusion in the bulk are dynamically interrelated (without the artificial di-
vision of studies into the diffusion limited regime (DLR) and the surface
limited regime (SLR)).
Keywords: hydrogen interaction with solids, surface processes, computer sim-

ulation of thermal desorption, dynamical boundary value problems, parameter
identification
PACS: 67.63.-r, 66.30.-h, 68.43.Nr
AMS Subject Classification: 35R30, 76R50, 80A30
1. Introduction
The interaction of hydrogen isotopes with solids has many applications [1–11].
It is sufficient to mention outlooks for the energy industry, protection of metals
from hydrogen corrosion, design of chemical reactors and rocket engineering.
Some problems of hydrogen materials science related to this paper were inves-
tigated in [12–20]. This work is a continuation of [21–24]. Practice has shown
that the limitations are not only diffusion processes but also physical-chemical
effects on the surface [1, 2]. Transfer parameters depend on the process charac-
teristics of producing the material batch, and one needs effective algorithms for
modeling instead of focusing on “tabular data”. We shall focus on the thermal
desorption method (TDS) and take into account only the limiting factors and the
information capacities of TDS experiments. Papers [16–18, 24] were dedicated
to interpretation of TDS peaks. Analysis of the causes of various TDS surges
is crucial for the selection of structural materials contacting with hydrogen iso-
topes. A sufficiently detailed review is presented in [17, 18]. Where modeling
does not take dynamic surface processes into account, TDS peaks are inevitably
Computer Simulation of Hydrogen Thermal Desorption Spectra . . . 169
interpreted as a result of capture of diffusing atomic hydrogen by the structural

defects (traps) of the material with different binding energies and(or) as a ma-
nifestation of multichannel diffusion [25]. The above listing of causes is not
exhaustive. Paper [24] shows that the appearance of a desorption peak can be
caused by some combinations of the rates of surface and diffusion processes.
This complicates the problem of TDS spectrum interpretation even more.
The main result of the paper is the method of parametric identification from
experimental data. The difficulties of inverse problems solving in mathematical
physics are well known. There is extensive mathematical literature and a num-
ber of specialized journals (inverse problems, ill-posed problems, . . . ). In exper-
imental practice, the inverse problem of multi-parameter estimation is reduced
to the one-factor-at-a-time method for DLR (diffusion limited regime) and SLR
(surface limited regime). It is done to prevent the computational ill-posedness
of the multiparametric inverse problem. In real life, however, a material is used
in the presence of a dynamic “surface–bulk” interplay. Thoroughly elaborated
techniques are available for estimation of the diffusion coefficient. The deter-
mination of desorption and dissolution parameters is far more complex (unless
the temperature is artificially lowered to “turn off” bulk processes). The paper
presents an algorithm allowing to estimate desorption and dissolution coeffi-
cients where diffusion and surface processes interact intensively. The authors
have attempted to use mathematical tools as little as possible to enable imple-
mentation of the presented algorithm using just a shareware math package.
2. Mathematical Model of Transfer

2.1. Diffusion Equation
Let us consider hydrogen transfer through a test metal or alloy sample (plate
thickness ` ). The material is well nigh uniform to neglect the interplay of H
atoms with captures. The heating is slow. Let us assume a standard model for
diffusion in the bulk
ct(t, x) = D(T )cxx (t, x), (t, x) ∈ Qt∗ , (1)

where Qt∗ = (0, t∗ ) × (0, `), c(t, x) is the diffusing H concentration. The
diffusion coefficient D is a function of the temperature T = T (t) given by the

Arrhenius equation: D(T ) = D0 exp −ED /[RT ] . We can take into account
different channels of diffusion, but the information content of the TDS experi-
ment is limited. Therefore, the coefficient D in Equation (1) is assumed to be
an integral effective index.
Seeking a write-up of the TDS peaks set it is handier to use the following
model
m
∂c ∂ 2c X h − i
aν 1 − Zν c(t, x) − a+

=D 2 − ν zν (t, x) ,
∂t ∂x ν=1
∂zν +
= a−

ν (T ) 1 − Zν c(t, x) − aν (T )zν (t, x), (2)
∂t
where zν (t, x) is the concentration of hydrogen atoms captured by defects of
different types; a∓ν are the coefficients of H capture and release by traps; Zν ≡
ν ν
zν (t, x)/zmax is the defects saturation degree zmax = max zν . Capture is
taken into account at its integral level for practical purposes. A more precise
definition of the defects’ geometry and distribution would add complexity to
the model. If, for instance, the defect is not a micro-cavity but hydride phase
inclusions, then at the degassing stage the corresponding coefficient a− j (T ) is
+
identically zero and aj (T ) value is positive only if the critical temperature is
reached:T (t) ≥ Tcrit.
It is easy to simulate the required number of TDS-peaks using different
binding energies (coefficients Ea). Algorithms based on difference schemes
and modeling results were presented in [21], [23]. In this paper we use only
Equation (1) (D = Deff ). For thin membranes of quickly permeable material
used in experiments this approximation is usually sufficient.
2.2. Dynamical Boundary Conditions

Let the plate be in contact with gaseous H2 . The surface is a potential energy
barrier (see [2, p. 177–206; Gabis, Kompaniets, Kurdymov]). Then, considering
(de)sorption processes, the boundary conditions are modeled as follows:
c(0, x) = c̄(x), x ∈ [0, `], t ∈ [0, t∗ ], (3)

c0 (t) = g(T )q0 (t), c` (t) = g(T )q` (t), (4)
q̇0 (t) = µ(T )s(T )p0 (t) − b(T )q02 (t) + Dcx(t, 0), (5)
q̇` (t) = µ(T )s(T )p` (t) − b(T )q`2 (t) − Dcx (t, `), (6)
b(T ) = b0 exp −Eb [RT ]−1 , s(T ) = s0 exp −Es [R T ]−1 .

Here, c0 (t) ≡ c(t, 0), c` (t) ≡ c(t, `) (identity sign is used in terms of equal-
ity by definition); q0,` (t) are the surface concentrations (x = 0, `, q̇ ≡ dq/dt);
g(T ) is the parameter of local equilibrium between concentrations on the surface
and in the subsurface bulk (coefficient of quick dissolution); µ is the gas-kinetic
coefficient; p0,` (t) are the gaseous (H2 ) pressures; b(T ) is the desorption coef-
ficient. The well-posedness of these boundary conditions was proved in [26].
Hydrogen occurs in the model both as molecules and in monoatomic form.
For consistency, the count is based on atoms: [c] = 1/cm3 , [q] = 1/cm2 ,
[Dcx] = [bq 2 ] = 1/(cm2 s). The factor µp determines the number of parti-
cles (H2 molecules) encountered per unit time and area. If the coefficient s
is changed to 2s, then the value of s denotes the share of adsorbed H atoms. √
Let us fix [p] = Torr. Then µ(T ) = (2πmkT)−1/2 ≈ 2.474 × 1022 / T ,
[µ] = 1H2 /(Torr cm2 s), [T ] = K (temperature numerical value under the rad-
ical), k is the Boltzmann constant, m is the hydrogen √ molecular mass. The
kinetic constant’s dependence on temperature ( µ ∝ 1/ T ) is usually neglected
on the scale of the exponential s(T ).
2.3. “Surface–Bulk” Fluxes Balance

The model (4) of quick dissolution (local equilibrium) on the surface is derived
from the more general ratios
+
k− 1 − c0,` (t)c−1 −1

max q0,` (t) − k 1 − q0,` (t)qmax c0,` (t) = ∓ Dcx 0,` . (7)
Coefficients k− (T ), k+ (T ) are the descriptors of the rate of dissolution in the

bulk and transfer to the surface. When concentrations are nowhere near maxi-
mum and Dcx ≈ 0 on the relative scale we obtain (4), where g = k− /k+ . If the
surface is isotropic (in terms of Ek− ≈ Ek+ ), then the parameter g(T ) is little
dependent on temperature. The density of the H atoms adsorption flux can be
modeled by the term µsp[1 − θ]2 for balance Equations (5), (6). For the ranges
of weak concentrations and sufficiently high working temperatures the degree
of surface coverage satisfies θ(t) = q(t)/qmax 1. These simplifications are
in agreement with the limited information capacities of the TDS experiments.
Experimental data are more easily approximated for a large number of param-
eters. But the uniqueness of the estimations is then failed. And thus essential
errors may occur at extrapolation of the results “from thin plate to wall”. Such
extrapolation is supposed by default.
2.4. TDS Experiment

Let us specify the TDS experiment under consideration. Gaseous hydrogen
under a reasonably high constant pressure is fed into a chamber with a strip (thin
plate) of the metal or alloy studied. The strip is heated to increase the sorption
rate. After the sample had absorbed a sufficient amount of hydrogen (to the
equilibrium saturated state), it is quenched (the heating current is turned off).
The chamber is then vacuumized, and the sample is simultaneously reheated.
The heatingT (t) is usually linearT (t) = T 0 + vt. The heating rate v is not too
high (< K/s). When the temperature maximum is reached (if degassing is not
yet completed) then heating is stopped: T(t) = T max . The molecular hydrogen
pressure p(t) in the chamber is measured by mass spectrometer. The desorption
flux fromthe surface controls the pressure the desorption flux density is J(t) =
b(t)q 2 (t) . We shall hereinafter use contracted notation for simplicity
b(t) ≡ b(T (t)), D(t) ≡ D(T (t)), s(t) ≡ s(T (t)).
The generalized quick dissolution coefficient g of local “surface-bulk” equilib-

rium is assumed to be constant (Ek− ≈ Ek+ ) in the given heating range. The
TDS method fulfills the symmetry conditions:
p = p0,` , q = q0,` , c0 = c` , Dcx|0 = −Dcx |` , c̄(x) = c̄ = const. (8)
The information capacities of the initial and final stages of the TDS experiment
are low. So it is sufficient to limit ourselves to t ≤ t∗ , when the flux from the
sample has decreased by an order of magnitude compared to maximum. The
experimental data are the desorption density curves J(t) or TDS spectra J(T )
(T (t) ↔ t) for different saturation conditions and heating rates. The conversion
p(·) 7→ J(·) is made more specific by taking into account the parameters of the
experimental apparatus. Modern equipment provides a means of creating deep
vacuum (10−9 − 10−7 Torr). For this reason, the component P ≡ µsp (P 1)
is the key control for the saturation stage, but resorption is neglected for the
degassing stage (P 1).
Note that only one saturation method is presented. Cryogenic technology
and the electrochemical method can also be used. There are good reasons to
bring the saturation process into agreement with the operating conditions of the
structural material. The corresponding technical adjustments are nonessential.
2.5. Equilibrium Analysis

Let us take a brief look at the equilibrium ratios at the accepted detail level of
modeling. We assume that pressure and temperature are constant. Formally,
equilibrium is characterized by all derivatives equal to zero. Keeping in mind
√
the extensively used Sieverts’ law, we shall observe proportionality to p value.
1. On the surface the following ratio is applied:
q µs
µsp[1 − θ1 ]2 − bq 2 = 0, θ1 ≡ , ν2 ≡ 2 .
qmax bqmax
Here, θ1 is the degree of surface coverage (next θi have a similar meaning). If

the factor s is changed to 2s, then the value of s denotes the share of H atoms
adsorbed from the gaseous phase. Let us formulate θ1 :
√
2 2 2 ν p √ √
pν [1 − θ1 ] = θ1 ⇒ θ1 = √ , ν p 1 ⇒ θ1 ∝ p.
1+ν p
2. In the “surface–bulk” equilibrium we have θ2 ≡ c/cmax , k1 ≡ k− qmax ,

k2 ≡ k+ cmax , γ ≡ k1 /k2 , then from Formula (7) we obtain

γθ1
k1 θ1 [1 − θ2 ] = k2 θ2 [1 − θ1 ] ⇒ θ2 = .
1 + [γ − 1]θ1
3. For definiteness, we take into account only one type of traps (m = 1).
We use Equation (2). For symmetry, we add the saturation factor [1 − θ2 ] for
z = zν . Here we have θ3 ≡ z/zmax , a ≡ a+ zmax /[a− cmax ], then we obtain

θ2
θ2 [1 − θ3 ] = θ3 [1 − θ2 ]a ⇒ θ3 = .
a + [1 − a]θ2
The “saturation-degassing” experiment provides information about a gen-
eral average concentration c̃ in the bulk V = S` (end surfaces are neglected):
V c̃ = 2θ1 qmax S + θ2 cmax V + θ3 zmax V.
Normalize c̃ by cmax and consider the dependence

2qmax c̃ zmax
Θ(p) ≡ =
θ1 + θ2 + θ3 . (9)
cmax `cmax cmax
√
Let us focus on the curves in the axes( p, Θ). Numerical results are presented
in Fig. 1 (ā = a+ /a− , cmax = 1018 , ` = 0.1cm). Quite many parameters
have an effect, but one can usually estimate the orders of magnitude to find “the
distance to linearity” (to the Sieverts’ law adequacy range).
= 3 .1015 zmax = 1016

qmax = 5 .10 15
qmax = 1015
2
g = 10 a=1
zmax = 1017 zmax = 7 .1017 = 3 .1015 = 1016

a = 10-3 a = 10-3
g = 102 g = 102
a = 10-1 a = 10-1
a = 10 a = 10 g = 104 g = 104
(a) Effect of defects (b) Effect of the surface
Figure 1. Saturation degree.
√
The curves Θ = f ( p) have the “growing wave” form. It is noticeable
in a wide pressure range only. For the given parameters, the inflection point
is the most vivid on the curves marked with pentagrams. The analysis of each
additive component in the formula (9) for the total concentration shows that the
position of the inflection point is determined by the moment when the first and
third addends have turned to the saturation mode with the following prevalent
rise of θ2 . Note that the curves for a narrower pressure range are practically
linear, in line with ranges of the Sieverts’ law. In a wide range of pressures, the
wave-like character of the graphs is observed experimentally [12].
A warning has to be made. When estimating the hydrogen permeability of
materials in experimental practices, Richardson approximation is often used for
the penetration flux density:
p p
J(t) = Φ pin (t) − pout(t) `−1 , pout < pin .

(10)
Membranes are usually very thin and their permeability is relatively high. We
can accept for the concentration gradient cx = [cout − cin ]/`. In accordance
with the Fick’s law, the formula J = −D[cout − cin ]/` is “exact”. Here, the
function J(t) is the monoatomic hydrogen permeation flux density. The ac-
tual boundary volume concentrations cannot however be measured, and thus
a quasi-equilibrium approximation is used, where the concentration c is sub-
√
stituted with the equilibrium concentration c̄ = Γ p in accordance with the
Sieverts’ law (Γ is the solubility coefficient, or solubility for short). On account
of the inequalities c̄in > cin , c̄out < cout, this substitution overrates the second
factor in the formula J = D[cin − cout]/`: c̄in − c̄out > cin − cout. Hence, the
value of D needs to be formally marked down to retain the equality. So, if the
value of Φ (permeability) is found from experimental data by fitting as indicated
in formula (10), then we obtain Φ < DΓ. The Φ and Γ values from the (typi-
cally used) formula Φ = DΓ determine the lower bound of the diffusion coeffi-
cient D. Furthermore, dissolved diffusing atomic hydrogen is mainly involved
in the steady-state (quasi-steady-state) permeability mode. In the context of the
√ p
model (4)–(6), we obtain c̄ = Γ p, Γ = g µs/b. The “saturation-degassing”
experiment yields the total concentration value c̃ > c̄ and an overstated (for the
√
permeability problem) value Γmax : c̃ = Γmax p.
Hence, we can estimate the Φ value (permeability) by fitting using for-
mula (10). This information has practical value as a convenient coefficient for
translation from pressures to flux. If, however, the D value is taken from one ex-
periment (or paper) and the Γ value is taken from another source, then, strictly
speaking, we get a ranking of three different numbers Φ < DΓ < DΓmax .
If the material has a high level of trapping by defects, then the calculated per-
meability can be an order of magnitude higher than the real permeability. The
coefficient Φ (as a parameter of (10)) has an S-form (Arrhenius-like) of the

saturation curve based on the order of pressures. It is only for relatively high
pressures (when boundary concentrations are near to Sieverts’ concentrations)
that we get Φ ≈ DΓ.
3. Functional Differential TDS Equation

Identification of TDS spectra is required not only to reveal the causes of dif-
ferent thermal desorption peaks, but also to enable numerical extrapolation and
generalization of the results received for laboratory samples (` usually is frac-
tions of mm). The model (2) gives the possibility to get any number of peaks
using traps with different parameters a± ν . The question, however, is whether
different peaks can occur when degassing an almost homogeneous material? To
answer this question let us restrict ourselves to the basic diffusion Equation (1),
but retaining symmetric dynamical boundary conditions (4)–(6), (8). The sur-
face is considered isotropic in terms of g = const over the heating range. The
resorption during vacuum building is neglected. Thus, we are limited to a mini-
mal number of parameters for the model which takes into account the dynamical
interplay of surface processes and diffusion in the bulk. In the following, let us
operate at this approximation.
The comparison of simulated and experimental TDS spectra with a focus on
parametric identification requires only the surface concentration (J = bq 2 ). It
is reasonable to try to avoid iterative solution of the boundary value problem for
interim approximations of the model parameters D0 , ED , b0 , Eb , s0 , Es , g. To
this end, we will run the transformations to reduce the problem to the integration
of a low order ODE system.
3.1. Derivation of Riccati-Type Equation

The accepted TDS degassing model is ct = D(t)cxx , c(0, x) = c̄,
c0,`(t) = gq(t), q̇(t) = −b(t)q 2 (t) + D(t)cx (t, 0), J(t) ≡ b(t)q 2 (t).
Rt
Let us replace the time t0 = 0 Ddτ (keeping the former notation t):
ct(t, x) = cxx(t, x), c(0, x) = c̄, c0,` = gq(t), (11)
cx|0 = −cx |` = q̇(t) + J(t)D −1 (t). (12)
Here q(t) is considered as the functional parameter and Equation (12) is an

additional relation for the linear problem (11). We perform a replacement to get
homogenous boundary conditions in (11):
ĉ = c(t, x) − gq(t), ĉt (t, x) = ĉxx (t, x) + f (t),

f (t) = −g q̇(t), ĉ(0, x) = ϕ̂(x) = 0, ĉ|0,` = 0.
Let us write the solution of the linear boundary value problem using the source
function (Green’s function):
Z tZ `
ĉ(t, x) = G1 (x, ξ, t − τ )f (τ ) dξ dτ,
0 0
∞ n n2 π 2 o
2X nπx nπξ
G1 (x, ξ, t) ≡ exp − 2 t sin sin .
` ` ` `
n=1
Boundary conditions contain the derivative ĉx(t, 0):
4g t n n2 π 2
Z X0 o X0 X
ĉx|0 = − q̇(τ ) exp (τ − t) dτ , ≡ .
` 0 `2 n=1,3,5...
For τ = t we have divergent series. So, term-by-term integration is implied.

For the original time t we get cx (t, 0) = ĉx (t, 0) = cx (t, `) = ĉx (t, `),
4g X0 t n n2 π 2 Z t
Z o
cx(t, 0) = − q̇(τ ) exp − 2 D(s) ds dτ .
` 0 ` τ
Finally, the dynamic boundary condition is written in the following integro-

differential form
4gD X0 t n n2 π 2 Z t
Z o
2
q̇(t) = −b(t)q (t) − q̇(τ ) exp − 2 D(s) ds dτ. (13)
` 0 ` τ
The resultant (13) with quadratic nonlinearity will be classified as a functional

Riccati equation of the neutral type. The Equation is equivalent to the origi-
nal boundary value problem in that the solution q(t) uniquely determines the
solution c(t, x). The analogy with thefunctional differential equation of the
form ẋ(t) = F t, x(t), ẋ(t), x[0,t], ẋ[0,t] of the neutral type [27] is that it is im-
possible to eliminate the derivative q̇ on the right side of the Equation through
integration by parts lest a divergent series arises. We are concerned with the
interval [t1 , t2 ], which corresponds to the TDS peak. Measurement for t ≈ 0, t∗
yields little information. There is a voluminous body of literature on Riccati-
type equations (including matrix equations for the optimal control theory).
3.2. Dimensionless Form of the Problem

For more comfortable R tmodeling we turn to dimensionless variables using sub-
stitution rules: t0 = 0 D(τ )dτ/`2, x0 = x/`, v = q/q̄ (c̄ = g q̄). Retaining the
notation t, we obtain: v(0) = 1,
X0Z t
v̇(t) = −b̃(t)v 2 (t) − 4g` v̇(τ ) exp −n2 π 2 [t − τ ] dτ,

(14)
0
b̃(t) ≡ q̄b(t)`2/D(t) is a dimensionless parameter of quadratic “desorption”.
3.3. Initial Saturation Stage

For definiteness, we will focus on the selected method of initial saturation of
the plate with dissolved hydrogen. Appropriately, we specify the factor b̃(t)
for quadratic nonlinearity. Initial saturation is conducted under relatively high
temperature T = T = const and inward pressure p̄ = const (to intensify
sorption). After the steady state of saturation is attained we get:
µ(T )s(T )p̄ = b(T )q̄ 2 , c̄ = g q̄ (g = const),

−1/2√
µsp̄ T exp Eb[2RT ]−1 ,

⇒ c̄ = gb0
b̃(t) ≡ q̄ b(t)`2D −1 = c̄ b(t)`2 [gD(t)]−1 ≡ b̃0 exp −Eb̃[RT (t)]−1 ,

Eb̃ ≡ Eb −ED , b̃0 ≡ `2 D0−1 b0 µsp̄ T exp Eb [2RT ]−1 .

p
Hereafter, let us be guided by a maximum limit of surface concentration at

around qmax ∼ 1015 − 1016 (monolayer in the context of geometric statics). If
during initial saturation the surface concentration for the given model parame-
ters is q̄ > 1014 , the degree of surface coverage must be taken into account in
numerical modeling. Then, to calculate the initial value of q̄ we take the ratio
−1 ]2 = bq̄ 2 instead of µsp̄ = bq̄ 2 .
µsp̄ [1 − q̄qmax
In the meantime, this a priori restriction (arising from the assumption that
stationary “balls” are ordered geometrically on a plane) is highly questionable.
For a dynamics model, it appears that the concentration threshold qmax may be
higher, if it is specified what meaning is attached to the term “bubbling” surface
layer (at a quite high temperature). The real q̄ value is somewhat conventional,
since it is strongly influenced by the experiment pre-treatment (building of vac-
uum before the start of heating). However, most of the hydrogen is in the bulk,
the diffusion equation (and the diffusion process itself) has a smoothing effect,
and we are interested in evident thermal desorption peaks, since the initial and
final time intervals of the experiment offer little information. For initial calcula-
tions it is therefore sufficient to correctly estimate the magnitude of the “effec-
tive” q̄ concentration. High precision requirements are non-critical here (neither
can they be realized in practice).
Let us remark that only one of the possible methods of reaching initial uni-
form saturation was discussed here. It is possible to use alternative methods.
Total hydrogen content in the sample is determined after the experiment (using
the registered desorption flux dynamic). The initial (uniform) concentration and
the corresponding q̄ value are determined in the model (using S, ` and current
g estimate). After all, data on dissolution (c̄ = c̄(p̄,T̄ )) can be used because the
material is multi-method tested.
4. Extracting Integrable Singularity

The functional differential equation of thermal desorption (14) has a singularity
hindering its numerical solution. The function
X0 X0 X
exp −n2 π 2 s ,

Θ(s) = 4 ≡ ,
n=1,3,5...
takes finite values for s > 0. The series is rapidly converging for large s.
Formally, substituting s = 0 (the integration variable τ reaches the upper limit t)
we obtain a divergent series. This can be “fixed” by term-by-term integration.
The order of terms in the series is O(n−2 ) (n = 2i − 1, i ≥ 1), wherefore
convergence is slow.
Let us formulate the problem. The equation has to be approximated by a
low order ODE system to enable application of standard software packages (for
example, the authors use Scilab freeware). Let us run some transformations
using the theory of Jacobi theta functions to explicitly extract the integrable
singularity. We consider
X∞
θ3 (t, x) = 1 + 2 exp{−n2 π 2 t} cos(2nπx), t > 0.
n=1
We have an alternative presentation for x = 0 [28, 29]:

∞ ∞ n n2 o
X 1 X
θ3 (t, 0) = 1 + 2 exp{−n2 π 2 t} = √ exp − .
πt t
n=1 −∞
The series on the left is rapidly converging for large t. But the series on the right
is rapidly converging for small t. Let us run auxiliary transformations:
∞ n n2 o ∞
1 X X X0
√ exp − = 1+2 exp{−k2 π 2 4t} + 2
πt −∞ t
k=1
∞ n n2 o
X0 1 X
=2 +√ exp − .
4πt −∞ 4t
Proceeding from here, the last series is subtracted from the first series and the
result is doubled. The following expression is obtained for the values s > 0
q ≡ − exp{−1/(4s)} :
∞
X0 1−Q X 2
exp −n2 π 2 s = √ , qn .

Θ(s) ≡ 4 Q(s) ≡ −2
πs n=1
The series Q is very rapidly converging for small s. As s → +0 we have Q → 0

√
and the integrable singularity Θ ≈ 1/ πs. The graph for Q(s) has an S-shaped
(“Arrhenius-like”) saturation curve form and S(1) ≈ 0.9996. The function
√
Q/ πs increases monotonically to a maximum (≈ 0.828 for s ≈ 0.334) and
then decreases monotonically. To represent the series Q(s) it is reasonable to
use only a small number (5–8) of series terms. As an alternative to (14), using
the above-described presentation of Θ(s) we obtain the Equation
Z t
2 1 − Q(t − τ )
v̇(t) = −b̃(t)v (t) − κ p v̇(τ ) dτ t ≥ 0, v(0) = 1 , (15)
0 π[t − τ ]
where the fraction (weak singularity under the integral) rapidly decreases from
infinity (τ = t) to near zero (τ = t − 1). For t > 1, the lower limit of
the integral can be replaced with t − 1. Thus, the neglectable background
can be easily
R t2 identified in the original physical time using the relation 1 =
0 0 2
t2 − t1 = t1 D(T (τ ))dτ /` . The compact functional differential (v̇ ≡ dv/dt)
TDS Equation (15) with initial data v(0) = 1 replaces the original nonlinear
boundary value problem (from section 3.1.) with dynamic boundary conditions
in the sense that formally only the dynamics of the surface concentration (des-
orption) is required for TDS spectrum construction.
5. Numerical Method and Computer Simulation

To be specific in the paper, we use experimental data for nickel and
steel (12Cr18Ni10Ti) [6], tungsten [5] and beryllium [30, 31]. Estimates de-
pend substantially on the experimental conditions and sample pretreatment.
So the values are perceived as a model for numerical illustrations. The
parameters common for all the materials are ` = 0.1 cm, T0 = 300 K,
[E] =J/mol. The assumed values of model parameters are the following:
(steel) b0 = 1.28 × 10−9 cm2 /s, Eb = 80 × 103 , D0 = 3.09 × 10−4 cm2 /s,
ED = 31 × 103 , g = 50 cm−1 , Eg = −5 × 103 , s0 = 0.6, Es = 110 × 103 ,
c̄ = 6 × 1017 at.H/cm3 , T = 1170 K, Ṫ = 1 K/s; (nickel) b0 = 1.53 × 10−14 ,
Eb = 43.2 × 103 , D0 = 7.5 · ×10−3 , ED = 40 × 103 , g = 100, Eg = 0,
s0 = 1.8 × 10−2 , Es = 61.4 × 103 , p̄ = 37.4 Torr, T = 770, Ṫ = 0.5; (tung-
sten) b0 = 6 · 10−4 , Eb = 69.559 × 103 , D0 = 4.1 · 10−3 , ED = 37.629 × 103 ,
g0 = 104 , Eg = 15 × 103 , s0 = 9 · 10−3 , Es = 15, p̄ = 670, T̄ = 1300, Ṫ = 5;
(beryllium) b0 = 3.08·10−9, Eb = 57.43 × 103 , D0 = 3 · 10−3 , ED = 28 × 103 ,
g0 = 200, Eg = 1.5 × 103 , s0 = 1.44 · 10−4 , Es = 1.82 × 103 , p̄ = 760,
T̄ = 1150, Ṫ = 5.
The main role in the degassing dynamics belongs to quadratic desorption.
We therefore approximate the integral term in Equation (15). The after-effect
horizon here is h < 1 (τ ∈ [t, t − h] for dimensionless time, Q(1) ≈ 0.999).
√
Let us fix h ∼ 0.3−0.4 due to the continuously differentiable function Q(s)/ s
graph (Fig.
R t 2) and the trapezoid rule for numerical integration. The replacement
of t0 = 0 Ddτ /`2 is naturally targeted at the diffusion time scale `2 /D so that
the step h corresponds to a significant segment of the experiment physical time.
Then, TDS equation (15) can be approximated step-by-step by a low-order ODE

system on segments of dimensionless time of length h.
Q(s)
Q(s)
s s
1-Q( )
d
0
N
n2
Q(s)= 2 q , q= exp{ 1/(4s)},
n=1
N > 10
s ~ 0.334 s
√
Figure 2. Functions Q(s) and Q(s)/ πs.
The greatest contribution to the integral is made by the value v̇(τ ), τ ≈ t due
to non-limited (but integrable) singularity. Thus, the quadratic approximation
v̇(τ ) ≈ v̇(t) + A[t − τ ] + B[t − τ ]2
is used due to function v̇ concavity. Let us consider the current segment of time
t ∈ [kh, (k + 1)h], k ≥ 0. The conditions
Z t
v̇(τ )|kh = v̇(kh), v̇(τ ) dτ = v(t) − v(kh)
kh
(here v(t) ≈ v(kh)+

v̇(t)[t−kh]) determine the values of A(t), B(t) constants
on τ ∈ [kh, t] :
v̇(t) − v̇(kh) v̇(t) − v̇(kh)

v̇(τ ) ≈ v̇(t) + 2 [t − τ ] − 3 [t − τ ]2
t − kh [t − kh]2

t ∈ [kh, (k + 1)h] . Represent the integral from the Formula (15) as√a sum
τ ∈ [0, kh], τ ∈ [kh, t]. For the second additive, the singularity v̇(τ )/ t − τ
is explicitly integrated by substituting (5.). The trapezoid rule√and mean value
theorem are used for the integral without singularity Q(+0)/ +0 = 0
t t
Q(t − τ ) Q(t − τ )
Z Z
p v̇(τ ) dτ = v̇(ξ)p dτ
kh π[t − τ ] kh π[t − τ ]
v̇(kh) + v̇(t) 1 √
≈ √ · · Q(t − kh) t − kh.
2 π 2
To approximate the first integral

kh X0Z kh
1 − Q(t − τ )
Z
2 2
p v̇(τ ) dτ = 4 v̇(τ )e−n π [t−τ ] dτ
0 π[t − τ ] 0
we use only several terms (for definiteness n = 1, 3) on the right-hand side

considering the factor n2 . The negative additives (v̇ < 0) thereby drop out.
Let us compensate for that by replacing kh with t, and introduce the additional
variables w1,2 (t):
Z t
wi (t) ≡ exp{−(2i − 1)2 π 2 [t − τ ]}v̇(τ ) dτ,
0
ẇi (t) = −(2i − 1)2 π 2 wi (t) + v̇(t), w1,2 (0) = 0.
As the result we obtain an ODE system of third order instead of the functional
differential Equation (15):
Q(t − kh) √

κ̂ 2
v̇(t) = −b̂(t)v 2 (t) + √ + t − kh v̇(kh)
π 15 4

− 4κ̂ w1 (t) + w2 (t) , v(0) = 1, (16)
ẇ1 (t) = . . . − [4κ̂ + π 2 ]w1 (t) − 4κ̂ w2 (t), w1 (0) = 0,
ẇ2 (t) = . . . − 4κ̂ w1 (t) − [4κ̂ + 9π 2 ]w2 (t), w2 (0) = 0,

b̃
b̂ ≡ Q(t−kh) √
,
1+ √κπ 32

15 − 4 t − kh
κ
κ̂ ≡ Q(t−kh) √
.
1+ √κπ 32

15 − 4 t − kh
Ellipsis stands for the right-hand side of the first line of the first equation
(−b̂ . . . v̇(kh)). The series for Q(s) converges very rapidly. It is sufficient to
use 5–8 terms for software implementation.
The sought function v(t) = q(t)/q̄ on the current segment of dimensionless
time (in the system (16) t = t0 , t0 ∈ [kh, (k + 1)h], k ≥ 0) is computed by
the Runge-Kutta 4th order method for integration of ODE systems (the authors
used Scilab software). If the TDS spectrum has two peaks, then w3,4 should be
additionally used to improve the modeling accuracy. Returning to the physical
time, we get the model spectrum J(T ) = b(T )q 2 (t) (t0 ↔ t ↔ T (t)).
Qualitatively, thermal desorption spectra of metal structural materials have
a typical form. Fig. 3 illustrates numerically simulated TDS spectra for the
above-listed hydrogen permeability parameter values of the structural materials
at different heating rates. Even when defects (traps) are not taken into account,
the suggested model can yield different types of two-peak graphs (where the
low temperature peak is more pronounced or where the peaks are comparable).
For reference, the change of so-called transport parameter W = `bvol c̄/D =
`bc̄/(Dg 2) is shown in the top part of the graphs. This parameter is crucial
for the study of membrane hydrogen permeability [5]. We supposed that SLR
corresponds to W < 10−2 and DLR corresponds to W > 104 . For beryllium
and steel, the low temperature TDS peak takes place in the range where diffusion
and surface processes play commensurable roles. The high temperature TDS
peak occurs in DLR. For nickel, surface processes and diffusion have similar
effect throughout the experimental temperature range, and only a peak-like step
appears in the TDS spectrum at a low temperature.
Let us focus on TDS spectra for tungsten because these spectra are qualita-
tively different from the previous ones. The aforementioned parameter values
“for tungsten” are, of course, formal. Besides micro impurities, the parameters
depend on how the sample surface was treated. This is especially true because
in practice a plate is thin, the volume is small and the effect of surface processes
is more vivid. This is one of the reasons for such a high variation of the quanti-
tative estimates of hydrogen permeability parameters. Another reason (but not

the last) is the following. Different models are used for experimental data treat-
ment (although coefficients formally have the same name). For the model data
accepted here, a narrow splash followed by a lengthy movement to the second
peak (less visible) was observed. In this model the sample has no high-capacity
traps (standard diffusion equation), but more detailed consideration is given to
the surface (see (4)–(6), the plate is thin). At first, near-surface hydrogen is
actively desorbed. Then, diffusion is slowly activated by heating. The concen-
tration gradient is substantial, and pumping to surface is growing. These effects
are usually categorically explained by the presence of traps, not by the dynamics
of “surface-bulk” interplay. An amazingly similar experimental curve (marked
with black circles) is found in [32], Fig. 1, although this paper discussed deu-
terium implantation on tungsten. The essence of our results is that various TDS
peaks may have other causes apart from the routinely blamed traps.
Fig. 4 illustrates the concentration dynamics (the parameter cannot be mea-
sured experimentally) inside the sample. Deflection of the concentration profile
(from plate center to the edge) is less visible for nickel, where a near-uniform
concentration is formed relatively quickly. In beryllium and tungsten, on the
contrary, a significant decline of the concentration near the surface happens al-
ready at the beginning of the experiment, and deflection of the concentration
profile remains during the following heating. For steel, active degassing of the
sample begins only after heating to a relatively high temperature.
Fig. 5 separately shows the dynamics of surface processes and diffusion.
High-temperature peaks correspond to the significantly enlarged desorption co-
efficient and activation of the diffusion afflux from the bulk under heating. Peaks
at a relatively low temperature take place where the diffusion towards the sur-
face is not high but near-surface concentrations are higher.
Let us briefly present the modeling algorithm.
• Set the parameters s0 , Es , b0 , Eb, D0 , ED , g. Determine the values of c̄,
b̃(t0 ), κ = g` under the saturation conditions p̄, T as described in Initial
saturation stage. If the equilibrium bulk concentration c̄ of dissolved dif-
fusing hydrogen is known, then presetting of s(T ) is not needed. Other
scenarios of initial saturation are possible (see discussion in Initial satu-
ration stage). The required correction of initial data is determined by the
specifics of the actual TDS experiment.
J(T).10 J(T) .10
-14 -14
W W
DLR
. Beryllium Nickel
.
T = 10, 5, 2.5 T = 1, 0.5, 0.25
1 2
TBe TBe TNi
T T
J(T) .10
-12
J(T) .10
-13
8.
6
9.
6
W W
DLR Tungsten 12Cr18Ni10Ti DLR
. .
T = 7, 5, 2.5 T = 2, 1, 0.5
TW1 TW2 TSt1 TSt2

T T
Figure 3. TDS-spectra.
• The low order ODE system of type (16) is numerically integrated step
by step in dimensionless time (on an h length interval). Other ODE ap-
proximations were presented in [24]. This procedure is standard for every
modern mathematical software.
• The graph J(t) = b(T (t))q 2 (t) or spectrum J(T ) is plotted in physical
time using the function v(t0 ) = q(t0 )/q̄.
6. Inverse Problem of Parametric Identification

The presented algorithm of numerical modeling allows to quickly scan different
scenarios and operating conditions of a material (including the heating law and
Beryllium Tungsten
_ _
/ /
T x/ T x/
Nickel 12Cr18Ni10Ti
_ _
/ /
T x/ T x/
Figure 4. Concentration distribution.
extrapolation of the results with ` increase). This statistical information is useful

when designing the strategy of experimental research. New materials (various
alloys) first have to be analyzed for their hydrogen permeability. In dealing
with this task we encounter inverse mathematical physics problems, which are
well-known for their difficulty. Let us suppose that the diffusion coefficient
D(T ) is known (Daines-Berrer method is usually used for DLR of permeabil-
ity). Let us formulate the problem: to estimate the surface parameters b0 , Eb ,
g using the J(t) (desorption flux) information. These conditions correspond to
the real conditions of an experiment where desorption dynamics closely cor-
JDiff . 10 / c
4

(q/q) 2
(q/q) 2 JDiff = Tungsten

Beryllium

b/b0
b/b0 12Cr18Ni10Ti
Tungsten
2
TW1 TSt1 TSt2 TW2 1
TBe TBe TW2
T T
TW1
JDiff . 10 / c
4

(q/q)2 Beryllium
JDiff =
Nickel
12Cr18Ni10Ti
2
(q/q)
Nickel

b/b0

1 b/b0 2
TBe TNi TBe TSt1 TNi TSt2
T T
(a) Surface parameters (b) Diffusion afflux to the surface
Figure 5. Dynamics of surface processes and diffusion.
relates with diffusion in the bulk. The problem of estimating the desorption
coefficient (bvol = b/g 2) for the situation where accumulation on the surface
can be neglected (q̇ ≈ 0) was presented in [22].
Since the function D(t) ≡ D(T (t)) is known, it is reasonable to move
directly to dimensionless time t0 , which is oriented at the characteristic time of
the diffusion transfer `2 /D. The old notation t is retained not to complicate the
formulas. Let us present the nonlinear term in the system (16) in the form
`
b̂(t)v 2 (t) = α(κ) · · J(t), J(t) ≡ b(t)q 2 (t).
D(t)c̄
Here, α is a function of the parameter κ. It is obtained through transformations

using notations from (14), (16). Elementary but somewhat lengthy formulas
are omitted. The function J(t) (t = t0 ) is known from measurements. After
substitution into (16) we obtain a system of three linear ODE. It is reasonable to
add the variables w3,4 to improve the calculation accuracy. The measurements
are usually noisy, but the function J(t) is comprised in the right-hand side of
the ODE, and is smoothed by integration.
Note that the right-hand sides of the equations now depend only on one es-
timated dissolution parameter κ = g`. The identification algorithm can thus
be decomposed. By setting the value of g we can eventually calculate the de-
pendence q(t) = q(t; g) in the original physical time t. Be reminded now that
Eb 103
J = bq 2 ⇒ b = Jq −2 ⇒ ln b0 − · = ln ψ, (17)
R · 103 T
where ψ(t) ≡ J/q 2 , T = T (t). Since heating is monotonic t ↔ T (t), we can
introduce the coordinates X ≡ 103 /T , Y ≡ ln ψ. The parametric curve X(t),
Y (t) is obtained on the plane (X, Y ). Judging by the ratio (17), this curve has
to be a straight line with a negative slope. Hence the criterion for the choice of
the “correct” value for g: the curve X(t), Y (t) = ln{J(t)/q 2 (t; g)} has to be
a straight line segment on the plane (X, Y ). This sub-problem is scalar, where
only g varies. Formally, we prolong the straight line segment until it crosses the
coordinate axes. The crossing with the Y axis (X = 0) yields the ln b0 value.
The crossing with the abscissa x̄ = 103 × R ln b0 /Eb determines the Eb value.
The algorithm is laborious due to iterative use of the procedure of numeri-
cal solving of the initial problem for the ODE system. It requires some effort
and familiarity with mathematical packages, but it is much easier and quicker to
use the standard built-in operation than to perform iterative solution of the orig-
inal nonlinear boundary value problem with three varying parameters (which
requires special software).
The presented linearization method is illustrated in Fig. 6. Model spectra
(for g = const) were plotted to test the algorithm. A system of five equations
(for the variables v(t), w1−4 (t)) was numerically solved to obtain more accurate
estimates for two-peaks spectra. In addition, a series of experiments was per-
formed for spectra with a random error not higher than 20% (we used standard
Scilab uniform random number generator). The identification algorithm based
on ODE system integration demonstrated the noise resistance of experimental
data treatment. The initial value of g yields the mass balance
Z t∗
−2
c̄` + 2q̄ = c̄[` + 2g ] = 2 J(τ ) dτ.
0
The experiments have demonstrated that the accuracy of such estimation de-
creases for g 1. Curves based on the “correct” numerical κ value (the re-
sultant curve is close to the straight line segment) and curves for 20% deviation
from κ are shown. The percentage error of the identification algorithm testing
is less than several percent. This error appears due to the numerical error of di-
rect and inverse problem solving. There is high sensitivity (appearing as vertical
“beak” singularity) to situations where the “true” κ value is exceeded. Math-
ematical singularity appears because, formally, the function v(t) changes sign,
taking also negative values. We did not take efforts to avoid this “non-physical”
transition because it is a vivid sign of a wrong κ value. The same numerical
experiments were done for noisy spectra. Activation energies are determined
more accurately because energy parameters more actively influence desorption
during heating. From the mathematical point of view, there is no need to strive
for visual agreement of simulated and experimental curves (because the experi-
mental error is tens of percents). ODE approximation is quite adequate.
103 103
x = 103R y Eb-1 Tungsten T y = ln b0 1.2 Beryllium T
y = lnb0 1.2 noisy data
0.8
0.8
103Ry ~
x= - 2.837
Eb
ln{J(T(t)) q-2(T(t),g)} ln{J(T(t)) q-2(T(t),g)}
103 103
12Cr18Ni10Ti T Nickel T
y = ln b0 1.2 noisy data

1.2
0.8 0.8
3
10 Ry ~
103Ry ~ x= - 6.122
x= - 2.128 Eb
Eb
ln{J(T(t)) q-2(T(t),g)} ln{J(T(t)) q-2(T(t),g)}
Figure 6. Estimation of desorption and dissolution parameters.

Conclusion
The problem of identification of the spectra of hydrogen thermal desorption
from structural materials is analyzed. This problem is of high relevance for the
nuclear power industry. Qualitatively, the identification consists in revealing the
causes of desorption peaks. It is usually assumed that TDS peaks appear due
to hydrogen release from traps with different binding energies. Here, it was
demonstrated using an elementary diffusion model (for a homogeneous mate-
rial) that if surface processes are taken into account, two-peak spectra can be
obtained even for very thin experimental samples. The tendency to resort to
the “theory of different traps” as the only explanation is understandable, but the
volume of our samples was near zero for the trapping capacity to manifest itself.
Another part of the problem is parametric identification. There are many
hydrogen permeability models, including various boundary conditions, where-
fore diffusion coefficient estimates vary even in orders of magnitude. A TDS
experiment has natural limitations in information capacity. For example, the
results have to be correctly (in a unique way) extrapolated for a higher material
thickness, considering that experimental samples are usually very thin because
the measurement time is limited. In this paper, the nonlinear boundary value
problem (standard diffusion equation with dynamical boundary conditions) is
reduced to the functional differential equation for the surface concentration,
because nothing but desorption dynamics is required to plot a TDS spectrum.
An effective algorithm oriented to the use of mathematical packages (includ-
ing freeware) is proposed. The main final output of the paper is a geometrically
transparent method for solving the inverse problem of surface parameters identi-
fication where desorption and diffusion in the bulk are dynamically interrelated.
Acknowledgments
This work was supported by the Russian Foundation for Basic Research
[Grant 15-01-00744].
References
[1] Hydrogen in metals; Alefeld, G.; Völkl, J.; Eds; Springer-Verlag: Berlin,
1978.
[2] Interactions of hydrogen with metals; Zakharov, A. P.; Ed.; Nauka:

Moscow, 1987 [in Russian].
[3] The hydrogen economy: opportunities and challenges; Ball, M.; Wi-
etschel, M.; Eds; Cambridge University Press: New York, 2009.
[4] Handbook of hydrogen storage: new materials for future energy storage;
Hirscher, M.; Ed.; Wiley-VCH: Weinheim, 2010.
[5] Pisarev, A. A.; Tsvetkov, I. V.; Marenkov, E. D.; Yarko, S. S. Hydrogen

permeability through metals; MEPhI: Moscow, 2008 [in Russian].
[6] Interaction of hydrogen isotopes with structural materials; Yukhimchuk,

A. A.; Ed.; Russian Federal Nuclear Center: Sarov, 2008 [in Russian].
[7] Varin, R. A.; Czujko, T.; Wronski, Z. S. Nanomaterials for solid state
hydrogen storage; Springer: New York, 2009.
[8] Lototskyy, M. V.; Yartys, V. A.; Pollet, B. G.; Bowman Jr, R. C. Int. J.
Hydrog. Energy. 2014, 39, 5818–51.
[9] Indeitsev, D. A.; Semenov, B. N. Acta Mech. 2008, 195, 295–304.
[10] McRae, G. A.; Coleman, C. E.; Leitch, B. W. J. Nucl. Mater. 2010, 396,
130–3.
[11] Mieza, J. I.; Vigna, G. L.; Domizzi, G. J. Nucl. Mater. 2011, 411, 150–9.
[12] Yukawa, H.; Zhang, G. X.; Watanabe, N.; Morinaga, M.; Nambu, T.; Mat-
sumoto, Y. J. Alloy Compd. 2009, 476, 102–6.
[13] Castro, F. J.; Meyer, G. J. Alloy Compd. 2002, 330–332, 59–63.
[14] Zaika, Yu. V.; Rodchenkova, N. I. Appl. Math. Model. 2009, 33, 3776–91.
[15] Evard, E. A.; Gabis, I. E.; Yartys, V. A. Int. J. Hydrog. Energy. 2010, 35,
9060–9.
[16] Schmid, K.; Rieger, V.; Manhard, A. J. Nucl. Mater. 2012, 426, 247–53.
[17] Legrand, E.; Oudriss, A.; Savall, C.; Bouhattate, J.; Feaugas, X. Int. J.
Hydrog. Energy. 2015, 40, 2871–81.
[18] Belyaev, A. K.; Polyanskiy, A. M.; Polyanskiy, V. A.; Sommitsch, Ch.;

Yakovlev, Yu. A. Int. J. Hydrog. Energy. 2016, 41, 8627–34.
[19] Denisov, E. A.; Kompaniets, M. V.; Kompaniets, T. N.; Bobkova, I. S. J.

Nucl. Mater. 2016, 472, 13–9.
[20] Zaika, Yu. V.; Bormatova, E. P. Int. J. Hydrog. Energy. 2011, 36, 1295–
305.
[21] Rodchenkova, N. I.; Zaika, Yu. V. Int. J. Hydrog. Energy. 2011, 36, 1239–
47.
[22] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2014, 39, 15819–
26.
[23] Zaika, Yu. V.; Kostikova, E. K. Adv. Mater. Sci. Appl. 2014, 3, 120–9.
[24] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2017, 42, 405–15.
[25] Andronov, D. Yu.; Arseniev, D. G.; Polyanskiy, A. M.; Polyanskiy, V. A.;

Yakovlev, Yu. A. Int. J. Hydrog. Energy. 2017, 42, 699–710.
[26] Zaika, Yu. V. Comp. Maths Math. Phys. 1996, 36, 1731–41.
[27] Hale, J. Theory of Functional Differential Equations; Springer-Verlag,

1977.
[28] Whittaker, E. T.; Watson, G. N. A course of modern analysis; Cambridge

University Press, 1996.
[29] Lang, S. Elliptic functions; Addison-Wesley publishing, 1973.
[30] Samsonov, A. V.; Korenkov, A. Yu.; Gabis, I. E.; Kurdyumov, A. A. Tech.

Phys. 1998, 43, 114–6.
[31] Tazhibaeva, I. L.; Shestakov, V. P.; Klepikov, A. Kh.; Romanenko, O. G.;

Chikhray, Y. V.; Kenzhin, E. A.; et al. National Nucl. Center Republic
Kazakhstan Bull. 2000, 1, 37–41 [in Russian].
[32] Oya, Y.; Inagaki, Y.; Suzuki, S.; et al. J. Nucl. Mater. 2009, 390–391,
622–5.
Editors: Michael D. Pfeffer et al. c 2018 Nova Science Publishers, Inc.
Chapter 7
FAST H YDROGEN P ERMEABILITY

OF S TRUCTURAL M ATERIALS :
M ODELLING AND PARAMETERS
E STIMATION
Yury V. Zaika∗, DSC and Natalia I. Rodchenkova, PhD
Institute of Applied Mathematical Research
Karelian Research Centre,
Russian Academy of Sciences
Petrozavodsk, Karelia, Russia
Abstract
High-purity hydrogen is required for clean energy and a variety of
chemical technology processes. The various alloys potentially well-suited
for use in gas-separation plants for the membrane technology of high-
purity hydrogen production were investigated by measuring specific hy-
drogen permeability. For structural materials one had to estimate the
parameters of diffusion and sorption to numerically model the different
scenarios and experimental conditions of the material usage (including
extreme ones), and identify the limiting factors. The proposed mathemat-
ical model matches the experimental method of hydrogen permeability
and takes into account only the main limiting factors for the applied mem-
brane filtering problem and the informative capabilities of the experiment
modifications.
∗
Corresponding Author Email: zaika@krc.karelia.ru.
196 Yury V. Zaika and Natalia I. Rodchenkova
The penetration method allows determining the diffusion coefficient

by so-called lag time. The accuracy of the estimation depends on the
degree of proximity to the DLR (diffusion limited regime) mode. The
method of ‘communicating vessels’ is more sensitive to surface processes.
‘Separate’ application of these methods leads to a situation where the ma-
terials studied are in fact somewhat different (for example, due to different
impacts on the surface), and there appear significant differences in hydro-
gen permeability parameter estimates.
The main idea of this chapter is to develop an aggregation procedure
for hydrogen permeability experiments precluding depressurization and
(or) change of the samples of the investigated material. We also present
the corresponding mathematical software for correct processing of mea-
surements along with appropriate assembly of the experimental unit.
This chapter describes the proposed cascade experiment technique
and the mathematical software: model, computer simulation and para-
metric identification algorithms. This allows increasing the informative
capacity of the experimental studies and the accuracy of the estimation of
hydrogen permeability parameters (diffusion, absorption, desorption).
Keywords: hydrogen interaction with solids, surface processes, computer

simulation of hydrogen permeability, parameter identification
PACS: 67.63.-r, 66.30.-h, 68.43.Nr
AMS Subject Classification: 35R30, 76R50, 80A30
1. Introduction
Studies on the interaction of hydrogen isotopes with structural materials are
mainly necessitated by problems in the energy industry, metal protection from
hydrogen corrosion and the design of chemical reactors [1–10]. Different alloys,
which may be well-suited for use in gas-separation plants for the membrane
technology of high-purity hydrogen production, were investigated by measur-
ing specific hydrogen permeability. For structural materials one had to estimate
the parameters of diffusion and sorption to numerically model the different sce-
narios and experimental conditions of the material usage (including extreme
ones), and identify the limiting factors. Some particular problems of the hy-
drogen materials science related to the topic of this study were presented and
investigated in [11–15]. Experiments show that the limiting factors are diffu-
sion processes as well as physical and chemical phenomena at the surface [1,2].
Fast Hydrogen Permeability of Structural Materials 197
The transfer parameters depend on the technological features of material batch

production. It is therefore unreasonable to target at ‘tabular data’. Instead, effec-
tive algorithms for processing experimental curves are necessary. In this study,
we consider the permeability model taking into account the main factors and the
self-descriptiveness of the experiment.
The penetration method (where a sufficiently high pressure of hydrogen gas
is built up in steps at the inlet side of the test material membrane, and the pen-
etrating flux is determined in the vacuum created at the outlet side) allows de-
termining the diffusion coefficient by so-called lag time. The accuracy of the
estimation depends on the degree of proximity to the DLR (diffusion limited
regime) mode. The method of ‘communicating vessels’ (where hydrogen from
the input volume seeps through the membrane to the isolated output vessel, and
the change of pressures is measured) is more sensitive to surface processes.
‘Separate’ application of these methods leads to a situation where the materi-
als studied are in fact somewhat different (due to different sample pre-treatment
procedures). This fact is one of the reasons for the differences in the estimates
of hydrogen permeability parameters.
The main idea of this study is to develop an aggregation procedure for hy-
drogen permeability experiments precluding depressurization and(or) alteration
of the samples of the investigated material. It also implies a corresponding
mathematical software for the correct processing of the measurements along
with appropriate assembly of the experimental unit.
Another important consideration is the uniqueness of the parameter esti-
mates of the investigated model. Mathematicians are often reproached for ‘fas-
cination with uniqueness theorems’. But after all, in justifying the choice of,
for example, structural materials for the ITER project, the results obtained on
thin laboratory samples are extrapolated to ‘walls’. Uniqueness allows for a
correct recomputation. Here, we shall not analyze the problem in general, in
particular, the adequacy of a model under specific conditions is ignored. In
the above-mentioned context the aggregation of experiments allows to make
the measurements substantially more informative for further estimation of the
parameters of bulk and surface processes in their dynamic interplay.
2. Aggregation of Experiments
Experimental practices employ various modifications of the penetration method
and thermal desorption spectrometry (TDS). The results of measurements de-
pend both on the unit design features and the procedure of preparing samples
for hydrogen permeability testing. A successive use of various methods often
causes impurities to appear on the sample surface, which significantly affects
the reproducibility of the results. These data are the input for the inverse prob-
lems of parametric identification, which are sensitive to the level of error. It is
therefore advisable to aggregate experiments to improve the accuracy and in-
formative value of the measurements. We suggest the following set-up of the
‘cascade’ experiment.
A membrane heated to a fixed temperature served as the partition in the
vacuum chamber. Degassing was performed in advance. A sufficiently high
pressure of hydrogen gas was built up in steps at the inlet side. The penetrat-
ing flux was determined by mass-spectrometry in the vacuum maintained at the
outlet side. This is a classical penetration method. Its advantage is a reliable
determination of the diffusion coefficient by the Daines - Berrer method (based
on so-called lag time). It allows distinguishing between the bulk and the surface
processes in the model, keeping in mind that surface parameters are significantly
more difficult to estimate. When the steady state level of the penetrating flux is
registered, we increase the inlet pressure and wait until a new steady state value
is established. Using (at least) three pressure jumps at the inlet side we record
the steady state flux values at the outlet side, thus determining ‘the degree of rec-
tilinearity’ of the isotherm. Then the pumping for vacuum building is stopped
and the experiment proceeds as the ‘communicating vessels’ method (this stage
is described in detail below). When pressure values become nearly equal (the
sample is almost uniformly saturated with hydrogen) it is possible to turn off
the heating, create the vacuum at both sides of the membrane and begin slowly
reheating the sample (TDS-experiment). In addition, there is no depressuriza-
tion of the diffusion cell and the sample surface remains uncontaminated with
additional impurities. We will clarify the details of the first two stages as we
describe the method of solving the inverse problem of parametric identification.
3. Hydrogen Permeability Model

Let us briefly describe the classical technique of the ‘communicating vessels’
experiment and present the corresponding mathematical model.
3.1. Distributed Transfer Model

The sample of a structural material pre-heated to a fixed temperature acts as
a vacuum vessel barrier. The sample degassing is performed in advance. At
the initial time moment pressure is built up at the inlet side by puffing of a
portion of molecular hydrogen. The declining pressure in the input chamber
and increasing hydrogen pressure in the isolated output chamber are measured.
Consider hydrogen transfer through the sample (` is the plate thickness and
S is its area). The sample temperature T is constant throughout the experiment.
The concentration of dissolved hydrogen (in monatomic state) is sufficiently
low and the diffusion flux can be considered proportional to the concentration
gradient. Some of H atoms interact with traps that can capture hydrogen. We
limit ourselves to the ‘limited sink’ concept without any additional details. The
following system is taken as the model of diffusion with limited capture within
the plate:
∂c ∂ 2c
= D(T ) 2 − f (T , z, c), (1)
∂t ∂x
∂z h z(t, x) i
= f ≡ a(T ) 1 − c − aout(T )z, (2)
∂t zmax
where c(t, x) is the concentration of diffusing (atomic) hydrogen; z(t, x) is the

concentration of the captured diffusant; D is the diffusion coefficient; a ≡ ain
and aout are the coefficients of H atoms absorption and release by traps. The
identity sign is frequently used here in the sense of equality by definition. The
magnitude zmax is considered to be quite small, so the capture is at the cor-
rection level and more detailed modelling of the capture process is not neces-
sary. Structural materials (for definiteness a metal alloy is considered) usually
have aout ain within the operating temperature range T ∈ [500, 900] K and
there is no need to further complicate the model (we suppose that hereinafter
aout = 0). Transfer parameters depend on the temperature T in an Arrhenius
way: D = D0 exp −ED /[RT ] , a = a0 exp −Ea /[RT ] . . .
Initial data for c, z (t = 0) are equal to zero.
Boundary conditions are derived from the material flux balance:

dQin ∂c
− = 2sµ p0 (t) − bc20 (t) S = −SD , (3)
dt ∂x 0
dQout ∂c
− = 2sµ p` (t) − bc2` (t) S = SD . (4)
dt ∂x `
Here, Qin (t), Qout(t) are the amounts of hydrogen atoms in the input chamber
of volume Vin and output chamber of volume Vout, c0 (t) ≡ c(t, 0), c` (t) ≡
c(t, `). In this experiment the gaseous hydrogen is in molecular form, but for
consistency (considering that atomic hydrogen diffuses through the metal) we
use atoms as the unit. According to the kinetic gas theory, the incident parti-
cle flux √density Jp is related to the pressure p by the HertzKnudsen formula:
Jp = p/ 2πmkT (k is the Boltzmann constant, m is the mass of a hydrogen
molecule). The following measurement units are chosen: [`] = cm, [p] = Torr. √
Then we numerically obtain the dependence J = µp, µ(T ) ≈ 2.474×10 22/ T
p
[µ] = 1H2 /(Torr cm2 s), [T ] = K . The processes of physical adsorption,
chemisorption, dissociation of molecules into atoms, and dissolution take place
on the surface. Only a small part of ‘incident’ H atoms will, however, be ab-
sorbed into the membrane volume. This is taken into account by the factor 2s.
One can write s (as a parameter of the model) instead of 2s, but it is more
convenient to interpret the dimensionless probability factor s as the fraction of
absorbed hydrogen atoms within the 2s notation. Thus, 2sµp is the resulting
flux of atoms through the surface into the bulk without differentiation into more
elementary stages.
Hereinafter, J0,` = bc20,` are the densities of the desorption flux from the
sample, b is the desorption coefficient. We also assume that the coefficients s
and b depend on the temperature in an Arrhenius way. Formally, the activation
energy Es in the exponent can as well be negative, being a linear combination
of the activation energies and heats of the surface processes on the way ‘from
gas to the solution’. If a constant saturation pressure of molecular hydrogen
ps = const is maintained at a constant temperature T = const on both sides
of the membrane, the equilibrium concentration c̄ of the dissolved atomic diffu-
sionally mobile hydrogen is finally established. By equating the derivatives in
the model (3), (4) to zero, we get
√ √ p
c̄ ∝ ps : c̄ = Γ ps , Γ ≡ 2sµb−1 .
Let us define the experimental conditions more precisely. The volumes
Vin,out comprise several liters, the thickness of the membrane is ` < mm, the
area S is about 1 cm2 , the inflow pressure p0 (0) is within several hundreds Torr.
The pressure range [pmin , pmax ] is narrow, so we restrict ourselves to the relation
zmax = σc̄, σ 6 0.1. This restriction will not lead to a failure of the Sieverts’
√ √
law c̄ + zmax ∝ ps while c̄ + zmax ≈ c̄ = Γ ps is within the experimental
accuracy range.
It now remains to find the magnitudes of Qin , Qout. Within the time of hy-
drogen transfer through the membrane the gas is in the thermodynamical quasi-
equilibrium with the surface, wherefore we use the formula N = pV /(kT).
Here, N is the number of gas particles occupying the volume V at the temper-
ature T and the pressure p (in the SI system [p] = Pa, [V ] = m3 , [k] = J/K).
Taking into account the relationships Torr = 133.322 Pa and Pa = J/m3 (for-
mally), we get the following expressions for the corresponding pressures and
volumes in the dynamical boundary conditions (3), (4): Q = 2N = αpV /T ,
α ≈ 1.931 × 1019. Here, p, V ,T are the numerical values of the variables in the
selected units (Torr, cm3 , K).
Within the experimental unit the membrane is situated in the tube (which is
heated to a predetermined temperature) between the inlet and the outlet cham-
bers. The tube diameter is large enough to consider the equality of pressures as
the criterion of thermodynamic quasi equilibrium between the gas in the tube
and in the chambers. The membrane temperature should be taken for the for-
mula for the kinetic constant µ(T ). The gas inside the volumes Vin,out (whose
massive walls are at room temperature) may get heated up. During the prelim-
inary experiment it is recommended to fill the chambers with a practically im-
permeable metal membrane between them with ambient gas. Then heat the tube
and record the pressure rises. Within the framework of the ideal gas approxima-
tion (equation of state) this procedure enables estimation of the increments of
the gas temperature inside the chambers. The corresponding gas temperatures
are the ones to be used in the formula given for Q (and the subsequent ones, only
excluding the value µ). The need of such a refinement arises from the character-
istics of this particular experimental unit. Such an adjustment of the values of
T should not cause difficulties in further calculations. Besides, this procedure
has relatively little effect on the final calculation of the model pressures taking
into account the measurement errors and relatively large volumes V .
3.2. Fast Hydrogen Permeability Model

It is clear from physical considerations that a quasi-stationary state is quickly
established when the membrane is thin and the material has a sufficiently high
hydrogen permeability coefficient: the traps are saturated and the diffusant con-
centration distribution is practically linear with respect to the thickness. In this
sense, the results of numerical modelling based on the ‘general’ model (the pre-
sented boundary-value problem) confirm its adequacy. Since near-to-surface
concentrations cannot be measured, the Richardson approximation is usually
used in practice to analyze the penetrating flux:
p p
J(t) = −Dcx = D`−1 c0 (t) − c` (t) ≈ JR (t) = DΓ`−1 p0 (t) − p` (t) .
Let us formulate the problem of modelling the concentrations c0,` pusing the
pressures p0,` without the quasi-equilibrium simplification c(t) = Γ p(t).
The quasi-stationary state is achieved within a time t0 , which is short com-
pared to the total experiment time (the traps become saturated and ∂x c =
−[c0 (t) − c` (t)]/`). So, the original model (1)–(4) can be simplified (taking
into account the formula Q = αpV /T ):
−1
ṗ0,` (t) = ∓β0,` c0 (t) − c` (t) , β0,` ≡ SD αVin,out` T, (5)

2sµ p0,` (t)−bc20,`(t) = ±D`−1 c0 (t)−c` (t) , t ≥ t0 > 0. (6)
Since by virtue of the ‘inlet–outlet’ balance the equalities

ṗ` (t) =
−1 −1
−Vin Vout ṗ0 (t) ⇒ p` (t) = p` (t0 ) + Vin Vout p0 (t0 ) − p0 (t) hold true, it is
sufficient to express the concentrations c0,`(t) = c0,`(p0 (t)) from the bound-
ary conditions (6) and substitute them into the first equation of (5). The sign is
chosen depending on whether the index is 0 or ` (+ for 0).
The following dimensionless variables are convenient for simulation:
X0,` (t) = 1 + 2`c0,` (t)bD −1 , a0,` (t) = 4`2 Γ2 p0,`(t)b2 D −2 − 1. (7)
In addition, the system of equations (6) is compactly written in the symmetric

form a0 + 2X` = X02 , a` + 2X0 = X`2 . For the variable X ≡ X` we obtain
the incomplete quartic equation [X 2 − a` ]2 = 4[2X + a0 ], which can be solved
in radicals (for physical considerations we are interested in the positive root).
However, the explicit expression is somewhat cumbersome and we will anyway
have to numerically integrate the first equation of (5) in the form ṗ0 = f (p0 ).
Therefore, we shall aim to derive differential equations for X0,` , since informa-
tion about the dynamics of the boundary concentrations c0,` is important.
Differentiate the equations (6) with respect to time and substitute the pres-
sure derivatives from (5). For the variables X0,` we get the ordinary differential
equations (ODE)
−1
X` − Vin Vout µST
Ẋ0 (t) = −2sM0 X0 − X` · , M0 ≡ , (8)
X0 X` − 1 αVin
X0 − Vin−1 Vout µST
Ẋ` (t) = 2sM` X0 − X` · , M` ≡ , t ≥ t0 . (9)
X0 X` − 1 αVout
Let us formulate step-by-step the algorithm of modelling the pressures
p0,` (t) (t ≥ t0 ) for the current values of D, b, s (the authors used the Scilab
freeware). We target at the ‘normal’ experimental conditions [16–20], includ-
ing the values of p, T , `, V , S.
1. We fix t = t0 : omit fast transient processes (the duration of the transient
processes is about tens of seconds on the hours-long experimental time scale).
For the variable X ≡ X` we choose the root of the biquadratic polynomial
[X 2 − a` (t0 )]2 − 4[2X + a0 (t0 )]. √
From physical considerations
p it follows that
c` (t0 ) > c̄` (t0 ) and thus X > 1 + 1 + a` = 1 + 2`Γ p` (t0 ) bD −1 .
2. The system of equations a0 +2X` = X02, a` +2X0 = X`2 (t = t0 ) yields
the missing value of X0 (t0 ). Formally, one equation is enough, but we take into
account averaging procedures including determination of the values of p0,` (t0 ).
3. We numerically integrate the ODE system (8), (9) with the obtained initial
data. The change of variables in (7) defines the concentrations c0,`(t), which are
used to calculate the model pressures p0,` (t) (t ≥ t0 ) from the equations (6).
Computational experiments show that the model curves p0,`(t) (t ≥ t0 )
almost coincide with those generated by the originally proposed model, i. e. the
nonlinear distributed initial boundary value problem.
Observe the difference from the quasi-equilibrium model (the Richard-
son approximation), where the only approximation parameter is the complex
Φ = DΓ. All the variable parameters of the original model that influence the
permeability, namely D, b, s, are important when running the above algorithm.
Thus, the fast hydrogen permeability model does not lose in informativeness
concerning transfer parameters.
There is a warning to be made. Consider the Richardson approximation in

more detail:
p p
JR (t) = Φ pin (t) − pout(t) `−1 , pout < pin . (10)
Membranes are usually very thin and have relatively high hydrogen permeabil-
ity coefficients. So, for the concentration gradient one can take cx = [cout −
cin ]/`. Then, according to the Fick’s law, the formula J = −D[cout − cin ]/`
is ‘exact’. Here, J(t) is the density of the penetrating flux of monatomic hy-
drogen. Since near-to-surface concentrations cannot be measured directly, it
is necessary to use the quasi-equilibrium approximation, substituting the equi-
√
librium concentration c̄ = Γ p for c in accordance with the Sieverts’ law. By
virtue of the inequalities c̄in > cin , c̄out < cout such a substitution overstates the
second factor in the expression J = D[cin −cout ]/` since c̄in −c̄out > cin −cout .
To maintain the equality the value of D should be formally understated.
Thus, if we fit the experimental data {p(t), J(t)} to find the value of Φ
(permeability) in accordance with the formula (10), then Φ < DΓ. Taking
the numbers Φ and Γ from the formula Φ = DΓ (which is usually used), we
get the lower estimate of the diffusion coefficient D. Thereafter, the stationary
(quasi-stationary) permeability stage mainly involves dissolved atomic diffusive
hydrogen. The ‘saturation-degassing’ experiment yields the value of the total
concentration ĉ > c̄ and an overstated (for the permeability problem) value of
√
the solubility coefficient Γmax : ĉ = Γmax p.
So, the value of Φ can be estimated by fitting using the formula (10). This
information is of practical value as a factor for converting pressures into flux.
If we take D from one experiment (published) and Γ from another source
(Γmax ), then, strictly speaking, we get a hierarchy of three different numbers:
Φ < DΓ < DΓmax . For a material with a high level of capture by traps the cal-
culated permeability can be overstated by an order of magnitude compared with
the actual one. The permeability Φ (as a parameter in (10)) has an S-shaped (Ar-
rhenius) form of the saturation curve with respect to the magnitude of pressure.
Only at relatively high pressures (when the boundary concentrations approach
the Sieverts’ ones) we have Φ ≈ DΓ.
4. Modelling of Hydrogen Permeability

4.1. Numerical Modelling of the Penetration Experiment
The proposed model is adapted to the experimental conditions and the data
range for alloys based on V group metals with high hydrogen permeability, in
particular, data for vanadium alloys which are presented in [16–23].
The sample is a membrane with the thickness ` = 0.05 cm and the area
S = 0.5 cm2 , the temperature is T = 673 K. Degassing of the membrane
was done in advance and continuous vacuum pumping was performed at the
outlet side. At the first stage of the experiment the input pressures p̄1,2,3 =
{30, 50, 70} Torr were built up in steps at the inlet, and maintained to achieve
steady state fluxes at the outlet. The gas temperature inside the inlet and outlet
chambers is assumed to be equal to 300 K. This slight difference from the room
temperature is due to the heating of the diffusion cell with the sample inside
(specified by the characteristics of the experimental equipment). The parameters
−5
of the model plot are D = 2 × 10√ cm2 /s, b = 5.7 × 10−24 cm√ 4
/s, s =
−4 20 3 15
1.2 × 10 , Γ = 2 × 10 1H /(cm Torr), Φ = 4 × 10 1H /(cm s Torr).
The values of the parameters are meaningful only within the context of the
given model (the boundary value problem). Let us therefore outline the neces-
sary specifications. The experimental conditions are such that the concentration
at the membrane outlet side is near zero and at the inlet side a stationary concen-
tration is quickly established (but it is lower than the equilibrium one): c̃ < c̄.
Within the model we determine c̄i and c̃i by the formulas
p p
c̄i = 2µsp̄i b−1 , 2µsp̄ − bc̃2 = J¯ = Dc̃`−1 ⇒ c̃i = −a + a2 + Γ2 p̄i ,
where a ≡ D[2b`]−1. For the given values of the parameters there establish
(locally) practically equilibrium concentrations:
c̃1 = 1.06 × 1021 < c̄1 = 1.09 × 1021 ,
c̃2 = 1.38 × 1021 < c̄2 = 1.41 × 1021 ,
c̃3 = 1.64 × 1021 < c̄3 = 1.67 × 1021 .
Next we express the fluxes for the boundary value problems corresponding
to the jumps of the inlet pressure.
Stage I. The boundary value problem of the penetration method is
ct = Dcxx , c(t, 0) = c̃1 , c(t, `) = 0, c(0, x) = 0.
The penetrating flux is

h ∞
X n n2 π 2 Dt oi
J1 (t) = −Dcx|` = Dc̃1 ` −1
1+2 (−1)n exp − .
`2
n=1
From the computational point of view it is convenient to introduce the di-
mensionless time t0 = Dt/`2 which is oriented at the characteristic diffusion
time `2 /D. As small t → 0 a singularity appears if we directly use the partial
sum of the expression
∞
X
f (t0 ) = 1 + 2 (−1)n exp {−n2 π 2 t0 }. (11)
n=1
Let us provide another expression for f (see also Appendix) using the properties
of the Jacobi theta function θ3 . The representation in [24, 25] is applicable:
∞ ∞ n n2 o
X
2 2 1 X
θ3 (t, 0) = 1 + 2 exp{−n π t} = √ exp − . (12)
πt −∞ t
n=1
The series on the left is rapidly converging for large t. But the series on the right
is rapidly converging for small t > 0. After some auxiliary transformations for
0 < t0 1 we get
2 X n −m2 o
f (t0 ) = √ exp 0
n ∈ N , f ≈ 1, t0 > 1. (13)
πt0 m=2n−1 4t
These representations enable a correct calculation of the flux J1 (t), t > 0. For
the considered applied problem it is sufficient to use partial sums consisting of
5–6 terms. Stage I ends with c(t∗ , x) = c̃1 (` − x)`−1 .
Stage II (t∗ → t0 = 0 is the t time zero):
ct = Dcxx, c(t, 0) = c̃2 , c(t, `) = 0, c(0, x) = c̃1 (` − x)`−1 .
The penetrating flux is
h ∞
X n n2 π 2 Dt oi

J2 (t) = D`−1 c̃1 + (c̃2 − c̃1 ) 1 + 2 (−1)n exp − .
`2
n=1
−1
Stage II ends with c(t∗ , x) = c̃2 (` − x)` .
Stage III. The formulas are similar to the cyclic interchange c̃1 → c̃2 , c̃2 → c̃3 .
The result of connecting the stages into a single curve of the penetrating flux
(conventionally, t = (t1 , t∗1 + t2 , t∗2 + t3 ), J = (J1 , J2 , J3 )) is shown in Fig. 1.
7
J3
6
J2 p3=70 Torr
5
J, 1017 cm-2 s-1
p2=50 Torr
J1
4 1
0.8
3 p1=30 Torr
0.6
f(t')
2 0.4
0.2
1
0
0 0.2 0.4 0.6 0.8 1
t'
0
0 100 200 300 400
t, s
Figure 1. The establishment of steady state fluxes.
4.2. Modelling of Fast Hydrogen Permeability

When the steady state permeability value is established during the penetration
experiment, continuous pumping at the outlet and maintenance of constant pres-
sure at the inlet are stopped. The aggregated experiment moves to the stage
of ‘communicating vessels’: inlet pressure declines and outlet pressure grows
(p0,` (t) are measured). We are so far talking about the direct problem of mod-
elling hydrogen pressures inside the volumes Vin,out. We specify the values:
√
Vin = 1500 cm3 , Vout = 2200 cm3 , p0 (0) = p̄0 = p̄3 , c̄ = Γ p̄0 .
The membrane temperature is taken in the dependences of the coefficients
D, b, s, µ on T , and the temperature of the gas inside the chambers is taken in
the expressions for Qin,out (take into account the correction to room temperature
due to the heating of the diffusion cell).
If we use the ODE system (8), (9) instead of the ‘full’ model, standard soft-
ware packages will suffice (we substitute the values of the hydrogen tempera-
ture inside the chambers Vin,out into the expressions for M0,` ). To this end one
should skip the initial time t0 within several minutes until a quasi-stationary (not
quasi-equilibrium) mode is established. Then the above algorithm is applied to
the fast hydrogen permeability model. Compared to the original boundary value
problem there is no significant additional error during the numerical simulation
(the curves visually coincide). We wait until the sample is uniformly saturated
(pressure values become nearly equal) before moving to the TDS experiment.
In this article we confine ourselves to aggregating only two types of experiments
and referencing [26–28] (TDS models for a cylindrical sample and for a powder
are presented in [14, 29, 30]).
5. Estimation of Hydrogen Permeability Parameters

5.1. Determination of the Lag Time
The initial approximation of the diffusion coefficient is determined by the
Daines - Berrer method. The flux J1 (t) asymptotically moves to the stationary
value J¯1 = Dc̃1 /`. Hence,
Z t

Q(t) = J(τ ) dτ ≈ J¯1 [t − `2 /(6D)] t ≥ t∗ ≡ t∗1 .
0
The intersection of the asymptote with the t axis gives so-called lag time τ0 =
`2 /(6D). Analytically,
Z t∗
τ0 ≈ t∗ − J1 (τ )J¯1−1 dτ, J1 ≈ J¯1 , t ≥ t∗ .
0
Note that a relative magnitude which does not require absolute values of the pen-
etrating flux (J¯1 = sup J1 (t)) is under the integral sign. In addition, the value
τ0 does not depend on c̃1 . It is usually assumed that the locally equilibrium con-
√
centration c̄1 = Γ p̄1 is quicklyp established at the inlet, so one can additionally
estimate the solubility Γ = 2sµ/b and the permeability Φ = DΓ using the
value J¯1 = Dc̄1 /`. This assumption is not used in this article. We assume
that according to the experimental conditions the stationary inlet concentration
c0 (t) ≈ c̃1 < c̄1 (t > ε, ε 1) is considered to be quickly established given
that c` (t) ≈ 0. The value of c̃1 as such is yet to be clarified. Thus, only the
estimate of the diffusion coefficient D is considered reliable at this stage.
With a new zero-time reference, integrating the expression J2 (t) we get
Z t

[J2 (τ ) − J¯1 ] dτ ≈ [J¯2 − J¯1 ] · [t − `2 (6D)−1 ] t∗ → t0 = 0 ,
0
where J¯i = Dc̃i /`, t ≥ t∗ ≡ t∗2 . Formally, we obtain the same expressions for
the lag time and the estimate of D if we change both the zero time and the flux
baseline J¯1 value overstatement. There is no additional information here (about

the target values of the surface parameters b and s), but the triple penetration
experiment allows to refine D. For the considered experimental data we have
τ01 ≈ τ02 ≈ τ03 ≈ 21 s, D = `2 /(6τ01 ) ≈ 2 × 10−5 .
5.2. Isotherm. Initial Estimates of b, s

Consider the curve of the steady state penetrating flux J¯ dependence on the inlet
pressure p̄ when vacuum pumping is performed at the outlet side. If one targets
√
at the Sieverts’ law and the (quasi)equilibrium concentration c̄ = Γ p̄ at the
¯ √p̄).
inlet side (J¯ = Dc̄/`), then it is natural to plot the dependence J¯ = J(
Let us analyze the steady state flux balance: 2sµp̄ − bc̃2 = J¯ = Dc̃`−1
p
⇒ c̃ = a − 1 + 1 + a−2 Γ2 p̄ , a ≡ D(2`b)−1 ,
p
J¯ = Dc̃`−1 = D`−1 a − 1 + 1 + a−2 Γ2 p̄ .
¯ √p̄) has a parabolic shape
Asymptotic analysis shows that the dependence J(
(J¯ ∝ p̄) at low inlet pressures p̄:
√ p
a−2 Γ2 p̄ ≡ x2 (x ∝ p̄) ⇒ c̃ = a − 1 + 1 + x2 = a[0.5 x2 + . . . ],
and a straight line form at relatively high pressures p̄:

√
a−2 Γ2 p̄ 1 ⇒ J¯ = −D 2 (2`2 b)−1 + DΓ`−1 p̄.
Using the straight-line segment of the isotherm we find DΓ/` (the slope of
the straight line) and knowing the estimate of D we determine the initial ap-
proximation of the solubility coefficient Γ. From the intersection of the straight
line with the ordinate
p axis we find b (formally setting p̄ = 0). Knowing the
values of Γ = 2sµ/b and b compute s and Φ = DΓ. A graphic illustration
(using a minimal required set of J¯1,2,3 values) is presented in Fig. 2. The initial
data are b = 5.84×10−24 , s = 1.22×10−4 , Γ = 1.98×1020 , Φ = 3.98×1015 .
5.3. The Final Stage of Estimation

We specify the values of D, b, s at the level of local variations. We solve the
ODE system (8), (9) (t0 = 2 min) with the current values of the coefficients.

8 D = 2 10-5
7 b = 5.84 10-24
s = 1.22 10-4
J, 1017 cm-2 s-1

6
1020

5 1015
0.1
4
3 0
2
-D2
J=
1 -0.1 22b
-0.1 0 0.1
0
0 1 2 3 4 5 6 8 10
pin , Torr
¯
Figure 2. Extrapolation of the isotherm of the steady state flux J.
70
60
D = 2 10-5
50
b = 5.7 10-24
p0, p, Torr
40
s = 1.2 10-4
= 2 1020
30 = 4 1015
20
10
t0=2 min
0
0 1 2 3 4 5 6
t, 103 s
Figure 3. Dynamics of pressures p0,`(t).
Fig. 3 provides a comparison of the ODE-simulated curves (dashed lines) and

the ‘full model’-simulated curves of the inlet and outlet pressures. We suc-
ceeded in achieving a high degree of agreement (visually the curves almost co-
incide) at the values D = 2 × 10−5 , b = 5.7 × 10−24 , s = 1.2 × 10−4 .
Conclusion
It is advisable to aggregate the penetration experiments (with and without vac-
uum pumping) to make the measurements substantially more informative for
further estimation of the hydrogen permeability parameters and to improve the
accuracy of the measurements. This paper suggests a cascade experiment tech-
nique and the corresponding software.
The identification algorithm uses only integral operators thus ensuring the
noise resistance of experimental data treatment. The penetration method with
vacuum pumping is characterized by a significant measurement error, and data
on the penetrating flux are required (and this, in turn, requires a more accurate
determination of the vacuum system characteristics). The model of the dis-
solved hydrogen concentration jump at the inlet side is not very precise either.
We are brought to a conclusion that the ‘communicating vessels’ stage, where
molecular hydrogen pressures are measured over a long time, is characterized
by a much higher accuracy of measurements.
The first stage of the aggregated experiment is perceived as preliminary es-
timation of the diffusion D, desorption b and absorption s coefficients. It is
essential that the solution of the inverse problem of parametric identification
is unique, since the results obtained for thin laboratory membranes are extrap-
olated (recalculated) to the dimensions of real-life structures. The results are
‘fine-tuned’ by means of local variation of the preliminary values of D, b, s in
the ODE model (8), (9).
Acknowledgment
This work was supported by the Russian Foundation for Basic Research
[Grant 15-01-00744].
Appendix
We define the function f (t0 ) as follows:
1. t0 ∈ [0, 10−4] ⇒ f (t0 ) ≡ 0; 2. t0 ∈ [10−4 , 10−1 ] ⇒ f (t0 ) : (13);
3. t0 ∈ [10−1 , 1] ⇒ f (t0 ) : (11); 4. t0 ≥ 1 ⇒ f (t0 ) = 1.
The graph for f (t0 ) has an S-shaped (‘Arrhenius-like’) saturation curve form
(see a small inset in Fig. 1).
P P
Finally, we denote 0 ≡ 1,3,5,... and present auxiliary transformations:
∞
X X0
f (t0 ) = 1 + 2 (−1)n exp {−n2 π 2 t0 } = 1 − 2 exp {−n2 π 2 t0 }
n=1
∞ ∞ n −n2 o
X
2 2 0 (12) X0 1 X
+2 exp {−4n π t } = −2 ···+ √ exp
2 πt0 −∞ 4t0
n=1
| {z }
even terms
∞ n −n2 oi
X0 1 h X X0
= −2 . . .+ √ 1+ 2 exp = −2 ...
2 πt0 4t0
n=1
n −n2 o ∞ n n2 o i
1 h X0 X
+ √ 2 exp + 1 + 2 exp − 0
2 πt0 4t0 n=1
t
| {z }
even terms
∞
(12)1 X0 n n2 o 1 h√ 0 n X oi
= √ exp − 0 + √ πt 1 + 2 exp{−n2 π 2 t0 }
πt0 4t 2 πt0 n=1
X0 1 X0 n n2 o 1
−2 exp {−n2 π 2 t0 } = √ exp − 0 + f (t0 )
πt0 4t 2
2 X0 n n2 o
⇒ f (t0 ) = √ exp − 0 .
πt0 4t
References
[1] Interactions of hydrogen with metals; Zakharov, A. P.; Ed.; Nauka:
Moscow, 1987 [in Russian].
[2] Hydrogen in metals; Alefeld, G.; Völkl, J.; Eds; Springer-Verlag: Berlin,
1978.
[3] Hydrogen isotopes. Fundamental and applied studies; Yukhimchuk,

A. A.; Ed.; Russian Federal Nuclear Center: Sarov, 2009 [in Russian].
[4] Pisarev, A. A.; Tsvetkov, I. V.; Marenkov, E. D.; Yarko, S. S. Hydrogen

permeability through metals; MEPhI: Moscow, 2008 [in Russian].
[5] Cherdantsev, Yu. P.; Chernov, I. P.; Tyurin, Yu. I. Methods of studying
metal-hydrogen systems; TPU: Tomsk, 2008 [in Russian].
[6] Gabis, I. E. Tech. Phys. 1999, 44, 90–4.
[7] Handbook of hydrogen storage: new materials for future energy storage;
Hirscher, M.; Ed.; Wiley-VCH: Weinheim, 2010.
[8] Lototskyy, M. V.; Yartys, V. A.; Pollet, B. G.; Bowman Jr, R. C. Int. J.
Hydrog. Energy. 2014, 39, 5818–51.
[9] The hydrogen economy: opportunities and challenges; Ball, M.; Wi-
etschel, M.; Eds; Cambridge University Press: New York, 2009.
[10] Varin, R. A.; Czujko, T.; Wronski, Z. S. Nanomaterials for solid state
hydrogen storage; Springer: New York, 2009.
[11] Evard, E. A.; Gabis, I. E.; Yartys, V. A. Int. J. Hydrog. Energy. 2010, 35,
9060–9.
[12] Indeitsev, D. A.; Semenov, B. N. Acta Mech. 2008, 195, 295–304.
[13] Zaika, Yu. V.; Bormatova, E. P. Int. J. Hydrog. Energy. 2011, 36, 1295–
305.
[14] Zaika, Yu. V.; Rodchenkova, N. I. Appl. Math. Model. 2009, 33, 3776–91.
[15] Zaika, Yu. V.; Rodchenkova, N. I. In Mathematical modelling; Brennan,

C. R.; Ed.; Nova Science Publishers: New York, 2013; Pp. 269–302.
[16] Dolan, M. D. J. Membr. Sci. 2010, 362, 12–28.
[17] Kojakhmetov, S.; Sidorov, N.; Piven, V.; Sipatov, I.; Gabis, I.; Arinov, B.
J. Alloy Compd. 2015, 645, S36–40.
[18] Song, G.; Dolan, M. D.; Kellam, M. E.; Liang, D.; Zambelli, S. J. Alloy
Compd. 2015, 509, 9322–28.
[19] Terrani, K. A.; Balooch, M.; Wongsawaeng, D.; Jaiyen, S.; Olander, D. R.
J. Nucl. Mater. 2010, 397, 61–8.
[20] Zhang, Y.; Maeda, R.; Komaki, M.; Nishimura, C. J. Membr. Sci. 2006,
269, 60–5.
[21] Alimov, V. N.; Busnyuk, V. N.; Notkin, M. E.; Livshits, A. I. Tech. Phys.
Lett. 2014, 40, 228–30.
[22] Zaika, Yu. V.; Rodchenkova, N. I.; Sidorov, N. I. Comput. Res. Model.
2016, 8, 121–35 [in Russian].
[23] Voyt, A.; Sidorov, N.; Sipatov, I.; Dobrotvorskii, M.; Piven, V.; Gabis, I.
Int. J. Hydrog. Energy 2017, 42, 3058–63.
[24] Whittaker, E. T.; Watson, G. N. A course of modern analysis; Cambridge

University Press, 1996.
[25] Lang, S. Elliptic functions; Addison-Wesley publishing, 1973.
[26] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2017, 42, 405–15.
[27] Zaika, Yu. V.; Kostikova, E. K. Adv. Mater. Sci. Appl. 2014, 3, 120–9.
[28] Zaika, Yu. V.; Kostikova, E. K. Int. J. Hydrog. Energy. 2014, 39, 15819–
26.
[29] Zaika, Yu. V.; Rodchenkova, N. I. Math. Model Comput. Simul. 2008, 19,
67–79 [in Russian].
[30] Rodchenkova, N. I.; Zaika, Yu. V. Int. J. Hydrog. Energy. 2011, 36, 1239–
47.
INDEX
# B
3D finite element model, 35 beryllium, 181, 184, 185

binding energies, xi, 168, 169, 170, 191
biomass, 76, 77
A
Boltzmann constant, 171, 200
boundary value problem, viii, xi, 167, 168,
Abaqus, vii, viii, 1, 6, 8, 11, 13, 23, 25, 32,
176, 177, 181, 189, 203, 205
35, 44, 51, 52, 54, 59
adaptive meshing, 51, 52
aggregation, viii, xii, 64, 196, 197 C
air temperature, xi, 150, 152, 153, 156
algorithm, 33, 35, 44, 72, 73, 74, 79, 80, 89, calibration, 17, 20, 30
90, 91, 92, 95, 96, 115, 116, 146, 169, Camber angle, ix, 2, 32, 49, 50
185, 186, 189, 190, 191, 203, 207, 211 CFD, 150, 154
ambient air, xi, 150, 156, 158, 161, 163 challenges, xi, 167, 192, 213
ambient air temperature, xi, 150, 156, 158, chemical, xii, 168, 195, 196
161, 163 chemisorption, 200
amplitude, 128, 135 clean energy, xii, 195
Arrhenius equation, 170 coherence, 108
atmospheric pressure, 155 collector temperature, 165
atoms, 124, 169, 171, 172, 173, 199, 200 commercial, vii, viii, 1, 2, 4, 8, 153
axisymmetric, 8, 9, 10, 11, 12, 13, 33, 36, complexity, 65, 69, 170
57 compounds, 3, 9, 14, 29
compression, 16, 20, 21
compulsion, 75, 77
216 Index
computer, vii, viii, ix, xii, 1, 6, 8, 54, 57, 63, distribution, viii, x, xi, 5, 36, 39, 41, 42, 53,
103, 108, 109, 118, 124, 131, 146, 168, 54, 114, 116, 130, 131, 132, 141, 150,
196 152, 156, 157, 159, 160, 162, 170, 202
computer simulation, vii, viii, ix, xii, 1, 4,
63, 103, 108, 120, 196
E
computer simulations, ix, 63, 103
configuration, 38, 44, 110, 111, 116, 153,
ecosystem, 75, 76, 77
155, 156
electric field, 111, 112, 113, 114, 115, 116,
constituent materials, 57
117, 118
construction, x, 2, 4, 5, 23, 64, 69, 149, 165,
electrical properties, 138
181
electricity, x, 149, 151
contact pressure, 5, 39, 40, 41, 42, 52, 53,
energy, 13, 15, 19, 28, 29, 30, 38, 50, 150,
54
152, 160, 161, 168, 170, 190, 192, 196,
cornering stiffness, 43, 48
200, 213
energy density, 13, 15, 19, 30
D energy parameters, 190
environmental conditions, 152, 156, 164
defects, 169, 170, 174, 175, 184 equilibrium, 30, 69, 71, 73, 74, 79, 80, 83,
deformation, 13, 14, 15, 23, 42, 43, 44 84, 86, 94, 95, 171, 172, 173, 175, 185,
derivatives, 173, 200, 203 200, 201, 202, 203, 204, 205, 207, 208,
desorption, viii, xi, xii, 167, 168, 169, 171, 209
172, 173, 176, 178, 179, 181, 184, 185, equipment, 109, 123, 173, 205
190, 191, 196, 198, 200, 211 experimental condition, xii, 181, 195, 196,
deviation, 24, 25, 89, 90, 190 200, 203, 205, 208
differential equations, 64, 203
differential games, vii, ix, 63, 64, 65, 66, 67,
F
69, 103, 104
diffraction, vii, ix, 107, 108, 109, 111, 114,
filaments, 130
119, 125, 126, 132, 133, 148
finite element, vii, viii, 1, 2, 4, 5, 8, 9, 12,
diffraction grating(s), vii, ix, 107, 108, 109,
13, 31, 32, 33, 34, 35, 38, 39, 43, 44, 51,
119, 126
54, 57
diffusion, xi, xii, 132, 168, 169, 170, 175,
finite element method, vii, viii, 1, 2, 4, 5, 51
176, 179, 181, 184, 185, 191, 195, 196,
force, ix, 2, 6, 23, 24, 42, 45, 47, 48, 49, 50,
197, 198, 199, 204, 205, 206, 207, 208,
52, 56, 57, 66
211
formula, x, 25, 84, 109, 129, 174, 175, 200,
diffusion process, 168, 169, 179
201, 202, 204
diffusion time, 181
Fresnel cosine and sine integrals, 111, 113,
discretization, 55, 57, 106
114, 116, 117, 118
displacement, x, 11, 12, 24, 31, 38, 108,
friction, 3, 6, 38, 39, 41, 151
109, 116
fuel consumption, 3, 6, 46, 50
Index 217
G L
geometry, vii, viii, 1, 4, 9, 34, 55, 57, 104, Lagrangian formulation, 44

151, 170 light, viii, x, 108, 109, 110, 111, 118, 119,
graph, 180, 181, 186, 211 123, 124, 126, 127, 129, 130, 131, 132,
gratings, 108, 109, 126, 128 134, 135, 136, 139, 141, 142, 143, 144,
145, 146, 148
light scattering, viii, x, 129, 130, 131, 132,
H
134, 135, 141, 142, 143, 144, 145, 146,
148
heating rate, 172, 173, 184
linear function, 88
height, xi, 24, 110, 117, 134, 149, 150, 151,
linear model, 28
153, 164, 165
lithography, x, 108, 109, 127
hybrid, 11, 12, 15
Lord Rayleigh, 108
hydrogen, viii, xi, xii, 167, 168, 169, 170,
171, 172, 175, 178, 179, 184, 185, 191,
192, 195, 196, 197, 198, 199, 200, 202, M
203, 204, 205, 207, 211, 212, 213
hydrogen atoms, 170, 200 magnitude, 131, 156, 157, 158, 172, 174,
hydrogen gas, 197, 198 175, 179, 191, 199, 204, 208
hyperelastic, ix, 1, 2, 6, 13, 15, 18, 20, 23, Marlow model, 19, 20, 29
24, 25, 28, 29, 30, 43, 51 mass, xi, 151, 167, 172, 189, 198, 200
hyper-viscoelastic, ix, 1, 7, 13, 30, 51 materials, xi, xii, 4, 7, 8, 11, 13, 14, 15, 23,
29, 31, 167, 168, 175, 181, 184, 191,
192, 195, 196, 197, 199, 213
I
materials science, xi, 167, 168, 196
MATLAB, 109, 114, 116, 117, 125
impurities, 184, 198
metals, 168, 191, 192, 205, 212
incidence, 132, 138, 139, 141, 148
modelling, 199, 202, 203, 213
industry, x, xi, 129, 136, 167, 168, 191, 196
modifications, xii, 67, 195, 198
integrals method, 107, 125
modulus, 14, 18, 28, 29, 31, 52
integration, viii, 54, 64, 168, 176, 177, 178,
molecular mass, 171
179, 181, 184, 189
molecular structure, 15, 17
intensity values, 118
molecules, 171, 200
interface, 6, 35, 52
monolayer, 178
invariants, 13, 14, 19
iterative Fresnel, v, x, 107, 108, 109, 110,
124, 128 N
iterative Fresnel integrals method, v, x, 108,
109, 110, 124, 128 Nash equilibrium, 68, 71, 73, 89, 90, 91, 92,
iterative solution, 176, 189 93, 94
near-field (Fresnel) regime, 109
218 Index
near-field diffraction, 108, 125

Q
nearly incompressible, 14
nickel, 181, 184, 185
qualitatively representative scenarios, ix, 63,
nodes, 8, 9, 11, 12, 35, 155
66, 69, 90, 103
nonlinear material, 25
numerical analysis, 164
R
O radial tire, viii, 1, 2, 3, 4, 5, 6, 9, 10, 13, 34,
36, 39, 42, 43, 57
operations, 73, 75, 114
radiation, xi, 150, 156, 158, 161, 163, 164
optimal control, 64, 67, 103, 104, 178
radicals, 202
optimization, 64, 65, 79
radius, 46, 151, 153
optimization method, 64
reference frame, 44
ordinary differential equations, viii, xi
refractive index, x, 108, 109, 127, 139, 142
oscillation, 54, 57
relaxation, 28, 29, 43
oxidation, 3
relaxation coefficient, 29
ozone, 3
resistance, ix, 2, 3, 6, 28, 50, 57, 153, 189,
211
P resolution, 109, 117, 118, 119, 121, 122,
123, 124, 125
parallel, 29, 45, 133 rolling resistance, ix, 2, 3, 6, 28, 50
parameter estimates, xii, 196 room temperature, 201, 207
parameter estimation, 169 rubber compounds, 20, 28
Pareto, 65 rubber(s), ix, 1, 2, 6, 7, 13, 14, 16, 20, 21,
permeability, viii, xi, xii, 167, 175, 184, 22, 24, 28, 43
185, 191, 192, 195, 196, 197, 198, 202,
203, 204, 205, 207, 208, 211, 212
S
physical properties, x, 129
plants, xii, 195, 196
saturation, 170, 173, 174, 175, 176, 178,
policy iteration, 64
179, 180, 185, 200, 204, 211
polynomial functions, 29
scattering, viii, x, 129, 130, 131, 132, 133,
polyvinyl chloride, 153
134, 135, 140, 141, 142, 143, 144, 145,
power generation, 151
146, 148
power plants, x, 149, 150
scattering intensity, viii, x, 130, 132, 141
professional development, 87, 88, 92
scattering patterns, 143, 144
project, xi, 63, 167, 197
sensors, x, 108, 153
Prony series, 28
simulation(s), v, vi, vii, ix, x, xii, 2, 4, 6, 32,
protection, 168, 196
58, 63, 64, 65, 66, 69, 104, 105, 107,
prototype, viii, xi, 149, 151, 152, 165
108, 109, 114, 118, 119, 120, 121, 122,
purity, xii, 195, 196
123, 124, 127, 128, 145, 151, 155, 156,
Index 219
163, 167, 168, 169, 171, 173, 175, 177, technology(ies), xii, 5, 173, 195, 196
179, 181, 183, 185, 187, 189, 191, 193, temperature, xi, 7, 14, 16, 39, 150, 151, 152,
196, 202, 207 153, 155, 156, 157, 158, 161, 163, 164,
slip angle, 46, 47, 48 165, 169, 170, 171, 172, 173, 178, 179,
smoothing, 25, 179 184, 185, 198, 199, 200, 201, 205, 207
software, vii, viii, xii, 1, 4, 9, 11, 23, 34, 64, tension, 7, 16, 19, 20, 21, 22
109, 114, 121, 124, 141, 179, 184, 186, test data, 5, 17, 19, 20, 23, 26, 80
189, 196, 197, 211 testing, 20, 21, 22, 23, 24, 190, 198
solar chimney power plant(s) (SCPP), viii, texture, 39, 131, 133, 136
x, 149, 150, 152, 153, 154, 156, 159, time increment, 54, 57
163, 164, 165 tires, v, ix, 1, 2, 4, 5, 8, 9, 13, 14, 29, 30, 36,
solar radiation, xi, 150, 156, 158, 161, 163 38, 43, 44, 46, 49, 52, 54, 57, 58, 60
solid state, 192, 213 transformations, 176, 180, 206, 212
solubility, 175, 204, 208, 209 tread, ix, 2, 9, 10, 25, 33, 34, 35, 38, 39, 41,
solution, vii, ix, x, 9, 38, 54, 63, 64, 65, 68, 42, 51, 52
69, 72, 79, 80, 89, 93, 103, 130, 132, treatment, ix, 63, 179, 189, 197, 211
142, 177, 179, 200, 211 turbulent flow, 150
sorption, xii, 171, 172, 178, 195, 196 twist, viii, x, 11, 129, 130, 131, 132, 141,
sorption process, 171 142, 143, 145, 146, 147, 148
steady state rolling, ix, 2, 7, 8, 32, 44, 46, twist angle, viii, x, 129, 130, 131, 142, 143,
47, 50, 51, 55, 57 145, 146, 147, 148
steel, viii, 1, 2, 7, 11, 30, 34, 181, 184, 185 twist level, 130
strain energy density, 13, 15, 19, 30
stress, 4, 6, 18, 19, 20, 23, 25, 26, 27, 28,
U
30, 36
structure, 35, 57, 69, 73, 75, 78, 93, 109,
uniaxial tension, 19
130, 136, 146, 147
uniform, 34, 55, 169, 179, 185, 189
successive approximations, 64, 104
universal test machine, 24
sustainable development, 67, 77, 105
unsteady state, 150
symmetry, 13, 32, 33, 35, 119, 142, 172,
173
V
T vacuum, xi, 167, 173, 176, 197, 198, 199,
205, 209, 211
Talbot carpet, vii, ix, 107, 108, 109, 110,
variables, vii, ix, 38, 39, 63, 65, 73, 76, 78,
120, 121, 122, 123, 124
82, 89, 92, 96, 103, 112, 118, 178, 183,
Talbot distance, ix, 107, 108, 119, 122
189, 201, 202, 203
Talbot effect, vii, ix, 107, 108, 109, 110,
vector, 46, 47, 68, 134
119, 120, 124, 127, 128
velocity, xi, 6, 44, 45, 46, 47, 50, 51, 55, 57,
Talbot, Henry, vii, ix, 107, 108
150, 151, 152, 153, 156, 157, 158, 163
techniques, viii, 1, 5, 9, 30, 57, 66, 127, 169
vessels, xii, 196, 197, 198, 199, 207, 211
220 Index
viscoelastic properties, 6
Y
viscosity, 150, 162, 163
volumetric changes, 14
yarn, v, viii, x, 129, 130, 131, 132, 133, 134,
135, 136, 138, 141, 142, 143, 144, 145,
W 146, 147, 148
water, ix, 63, 66, 75, 103, 105

water ecosystems, ix, 63, 66, 75, 103, 105
wave number, 133

Computer Simulations Advances Research Applications 8

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computer Simulations Advances Research Applications 8

Uploaded by

Copyright:

Available Formats

COMPUTER SCIENCE, TECHNOLOGY AND APPLICATIONS

Additional books in this series can be found on Nova’s website

Additional e-books in this series can be found on Nova’s website

NOTICE TO THE READER

Library of Congress Cataloging-in-Publication Data

ISBN:  HERRN

Published by Nova Science Publishers, Inc. † New York

Chapter 6 Computer Simulation of Hydrogen

Computer Simulations: Advances in Research and Applications begins

description of the mechanical behavior of tire constituents are examined.

imaging, refractive index measurements, displacement sensors, lithography

chapter focus on numerical studies of the unsteady airflow characteristics

COMPUTER SIMULATIONS OF PASSENGER

Mir Hamid Reza Ghoreishy

This monograph is intended to briefly describe the methods and

Keywords: radial tire, simulation, finite element method, steady state

It is generally known that the motions of a ground vehicle such as a

shown in Figure 2. Design of a modern pneumatic tire is mainly based on

1. Lowering the rolling resistance which is directly related to the fuel

Figure 1. A typical radial tire mounted on the rim.

Most of the above mentioned requirements are contradictory. For

materials, geometry and construction. Design of a modern pneumatic tire

Figure 2. A 3D cross-sectional view of a typical radial tire with terminology [1].

Computer simulation of passenger car radial tires by the finite element

[2] for a detailed formulation of the nonlinear finite element in solid

Finite element modeling of tires has been an object of intensive

function of inflation pressure and static normal load. Increased levels of

on the steady state rolling analysis using hyper-viscoelastic material

hyperelastic or hyper-viscoelastic model used to describe the mechanical

3. FINITE ELEMENT MODEL DEVELOPMENT

3.1. Two-Dimensional (2D) Model

3.1.1. Mesh Generation

rather preferred to use the structured meshing technique as the pattern of

Figure 3. Lay-out of a typical passenger car radial tire.

Prior to mesh generation the lay-out of the tire must be prepared in an

Figure 4. A quadrilateral axisymmetric hybrid element with linear interpolation for

3.1.2. Material Model

3.1.2.1. Hyperelastic Model

where 𝐅 is the deformation gradient expressed by:

In this equation, 𝑥𝑖 and 𝑋𝑗 are the coordinates of a material point in

𝐼1 = 𝜆12 + 𝜆22 + 𝜆23 (3)

For a fully incompressible material, (𝐼3 = 1). However, for highly

where 𝐽 is the total volume ratio and calculated by 𝐽 = det(𝐅). In a full

Now, the strain energy density function can be written by the

𝑊 = 𝑊𝑑𝑒𝑣 (𝐼̅1 , 𝐼2̅ ) + 𝑊𝑣𝑜𝑙 (𝐽𝑒𝑙 ) (9)

where 𝜖 𝑡ℎ is the linear thermal expansion strain. For an isothermal

phenomenological and mechanistic relationships. Here we give some more

Although this model can predict the tension behavior of rubbers up to

It is a phenomenological model in which it is assumed that 𝐶00 = 0. In

The most important form of these models is introduced by Yeoh as

It is shown by Yeoh [32-34] if limited test data are available for

In the above equation, 𝜇 is a material parameter and 𝜆𝑚 is the locking

Van der Waals Model

In this equation, 𝜇 is the initial shear modulus, 𝜆𝑚 is the locking

𝑊 = 𝑊𝑑𝑒𝑣 (𝐼1̅ ) + 𝑊𝑣𝑜𝑙 (𝐽𝑒𝑙 ) (21)

As 𝐼3̅ = 1, thus 𝐼1̅ can be written as:

It is worth noting that the variation of 𝐼1̅ in equation (23) is also

3.1.2.2. Determination of Hyperelastic Model Parameters

Figure 9. A universal testing machine with equibiaxial fixture [39].

Simple Shear Test

experiment is carried out to determine the deviation of the Poisson’s ratio

𝑉𝑜𝑙𝑢𝑚𝑒 𝑟𝑎𝑡𝑖𝑜 = 1 − 𝜖 (28)

𝑑𝜎: 𝑑𝜖 > 0 (29)

ISBN: HERRN