Editor-in-Chief

Dr. Sergey Victorovich Ulyanov

State University “Dubna”, Russian Federation

Editorial Board Members

Federico Félix Hahn-Schlam, Mexico Luis Pérez Domínguez, Mexico

Luiz Carlos Sandoval Góes, Brazil Abderraouf MAOUDJ, Algeria
Reza Javanmard Alitappeh, Iran Ratchatin Chancharoen, Thailand
Brahim Brahmi, Canada Shih-Wen Hsiao, Taiwan
Behzad Moradi, Iran Siti Azfanizam Ahmad, Malaysia
Hassan Alhelou, Syrian Arab Republic Mahmoud Shafik, United Kingdom
Lihong Zheng, Australia Hesham Mohamed Shehata, Egypt
Nguyen-Truc-Dao Nguyen, United States Hafiz Alabi Alaka, United Kingdom
José Miguel Rubio, Chile Abdelhakim DEBOUCHA, Algeria
Fazlollah Abbasi, Iran Karthick Srinivasan, Canada
Chi-Yi Tsai, TaiWan Ozoemena Anthony Ani, Nigeria
Shuo Feng, Canada Rong-Tsu Wang, Taiwan
Mohsen Kaboli, Germany Yu Zhao, China
Dragan Milan Randjelovic, Serbia Aslam Muhammad, Pakistan
Milan Kubina, Slovakia Yong Zhong, China
Yang Sun, China Xin Zhang, China
Yongmin Zhang, China ANISH PANDEY, Bhubaneswar
Mouna Afif, Tunisia Hojat Moayedirad, Iran
Yousef Awwad Daraghmi, Palestinian Mohammed Abdo Hashem Ali, Malaysia
Ahmad Fakharian, Iran Paolo Rocchi, Italy
Kamel Guesmi, Algeria Falah Hassan Ali Al-akashi, Iraq
Yuwen Shou, Taiwan Chien-Ho Ko, Taiwan
Sung-Ja Choi, Korea Bakİ Koyuncu, Turkey
Yahia ElFahem Said, Saudi Arabia Wai Kit Wong, Malaysia
Michał Pająk, Poland Viktor Manahov, United Kingdom
Qinwei Fan, China Riadh Ayachi, Tunisia
Andrey Ivanovich Kostogryzov, Russian Federation Terje Solsvik Kristensen, Norway
Ridha Ben Salah, Tunisia Andrey G. Reshetnikov, Russian Federation
Hussein Chible Chible, Lebanon Mustafa Faisal Abdelwahed, Egypt
Tianxing Cai, United States Ali Khosravi, Finland
Mahmoud Elsisi, Egypt Chen-Wu Wu, China
Jacky Y. K. NG, China Mariam Shah Musavi, France
Li Liu, China Shing Tenqchen, Taiwan
Fushun Liu, China Konstantinos Ilias Kotis, Greece
Ebtehal Turki Alotaibi, Saudi Arabia Junfei Qiu, United Kingdom
 Volume 1 Issue 2 · October 2019 · ISSN 2661-3220 (Online)

Artificial Intelligence
Advances

Editor-in-Chief
Dr. Sergey Victorovich Ulyanov
 Volume 1 | Issue 2 | October 2019 | Page 1-55
Artificial Intelligence Advances

Contents
Article
5 Probabilistic Rationale of Actions for Artificial Intelligence Systems Operating in Uncer-
tainty Conditions
Andrey I. Kostogryzov
24 School Debit Transaction Using Fingerprint Recognition System
Wai Kit Wong, Thu Soe Min, Shi Enn Chong
38 Using the CVP Traffic Detection Model at Road-Section Applies to Traffic Information
Collection and Monitor - the Case Study
Shing Tenqchen, Yen-Jung Su, Keng-Pin Chen
44 Application of Feature Curves and Shape Blending on Yacht Designing
Shih-Wen Hsiao, Jeng-Horng Chen, Ting-An Yeh

Review
1 The Comparison of Current Development, Technology and Governments’ Attitudes of
Driverless Car at Home and Abroad
Chuyi Zhang

Copyright
Artificial Intelligence Advances is licensed under a Creative Commons-Non-Commercial 4.0 International Copyright
(CC BY- NC4.0). Readers shall have the right to copy and distribute articles in this journal in any form in any medium,
and may also modify, convert or create on the basis of articles. In sharing and using articles in this journal, the user must
indicate the author and source, and mark the changes made in articles. Copyright © BILINGUAL PUBLISHING CO. All
Rights Reserved.
 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Artificial Intelligence Advances

https://ojs.bilpublishing.com/index.php/aia

REVIEW
The Comparison of Current Development, Technology and Govern-
ments’ Attitudes of Driverless Car at Home and Abroad
Chuyi Zhang*
Beijing University of Posts and Telecommunications, Beijing, 100876, China

ARTICLE INFO ABSTRACT

Article history Driverless car, as a direction for future automobile development, greatly
Received: 5 August 2019 improves the efficiency and safety of the traffic system. It’s one of the
most popular technical fields. In recent years, driverless car has developed
Accepted: 20 August 2019 rapidly. The related development is concerned by governments, business-
Published Online: 30 October 2019 es, consumers and stakeholders widely, and most of countries have been
actively studying this technology. This paper first introduces the current
Keywords: development of driverless car at home and abroad. Besides, the basic
Driverless car technologies of driverless car are briefly analyzed. In addition, the author
compares the American government’s attitudes with Chinese government’s
Current development attitudes towards driverless car. Specifically, the article makes an analysis
Technology of contents of literature and periodicals at home and abroad and policies
Governments’ attitudes and documents which have already been published. The analysis shows that
there is no great difference between the attitudes of Chinese and American
Future development direction
governments. Both of two governments actively support the development
of driverless car. Finally, this paper expounds the development direction
of the driverless car field in future by dividing into two categories through
road conditions: automatic driving on expressways and automatic driving
in cities.

　
1. Introduction Engineers (SAE) have all graded automatic driving tech-

A
nology (see Table I), which, to a certain extent, represents
driverless car is a vehicle that is capable of the development phase of the driverless technology [2].
sensing its environment and navigating without In China, the driverless technology has developed well
human input. Driverless cars use a variety of tech- with the encouragement and support of the government.
niques to detect their surroundings, such as radar, laser Beijing, Shanghai and some other cities have promulgated
light, computer vision, GPS, IMU and so on. The poten- the detailed rules of automatic driving road test. In 2014,
tial benefits of driverless cars include reducing mobility Google fired a shot heard all the way to Detroit. Google’s
and infrastructure costs, increasing safety, increasing newest driverless car had no steering wheel and no brakes.
mobility, increasing customer satisfaction and reducing The message was clear: cars of the future will be born
crime [1]. In recent years, driverless cars have been de- fully autonomous, with no human driver needed [3]. Com-
veloping rapidly. The National Highway Traffic Safety panies, such as Uber and Waymo, are also actively testing
Administration (NHTSA) and the Society of Automotive driverless cars hoping for combining autopilot technology

*Corresponding Author:
Chuyi Zhang,
Beijing University of Posts and Telecommunications, Beijing, 100876, China;
Email: [email protected]

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1115 1

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

with car calling services. In Dubai, the first driverless taxi mixed test of urban road and high-speed traffic conditions.
in the world was put into operation by the Dubai highway In 2015, Toyota Corporation, cooperated with Standford
and Transport Administration (RTA) in February 28, 2018. and University of Michigan, allocated 1 billion to the
These show that the automotive industry is undergoing a autopilot project. In 2016, Toyota Corporation invested
revolution. It is obvious that driverless cars are likely to strategically in American Technology Corp Uber, and also
be widely applied in the future, which has great certain established the cooperative relationship in the research of
research significance. driverless technology. In 2016, Ford Fusion ushered in the
However, no one has investigated both of the technol- debut of the driverless car. Google has built an independent
ogy and governments’ attitudes of driverless cars at home automated driving company Waymo. Its actual test mile-
and abroad so far. There are few book resources about this age of driverless car has reached about 2 million miles. In
topic. But there are many Internet sources and periodical March 2017, Tesla launched the Autopilot 8.1 system on
sources that appear to be reliable. Contents of literature the basis of Autopilot 2, greatly improving the grade of the
and periodicals at home and abroad and policies and doc- driverless car. According to statistics, Tesla has been driv-
uments which have already been published are analyzed. ing over 222 million miles in the Autopilot mode [4].

Table 1. Automatic Drive Classification 3. The Development Status of Driverless Car

Automatic
in China
Drive Main Part
Classification
Name (SAE)
The development of domestic driverless vehicle is lagging
Peripheral behind compared with other developed countries.
NHT- Driving System
SAE Monitor- Support
SA Operation Scope
ing In 1992, National University of Defense Technology
0 0 No Automation
Human System developed the real driverless car in China. In 2000, the
Drivers Scope
fourth generation of driverless cars developed successful-
Human
Human ly with a speed of 76km/h. In 2005, Shanghai Jiao Tong
1 1 Driving Support Driving Human None
Drivers
System Driv- University developed the first urban driverless car. Since
Partial Automa- ers 2009, with the support of the National Natural Science
2 2
tion
Conditional
Fund, the “China smart car future challenge”, the only
3 3 Part
Automation competition aimed at driverless car in China, has been
System
4
Highly Automa-
System successfully held for eight sessions. It has played a great
tion Sys-
4 role in promoting the development of driverless cars in
Complete Auto- tem Whole
5 China. In 2010, the pilotless car developed by the Nation-
mation Domain
al University of Defense Technology and FAW has real-
2. The Development Status of Driverless Car ized autonomous navigation without human intervention,
in Foreign Countries and can drive stably on high speed roads.
After 2013, IT giants in China and some large auton-
In the early 1970s, the developed countries such as the omous vehicle companies gradually joined in the field of
United States have already begun to study driverless car technology development in the field of driverless vehicle.
technology. In 1995, Carnegie Mellon University devel- The unmanned pilot project, led by the Baidu Institute
oped the Navlab series of driverless cars, which complet- for deep learning and research, started in 2013. In 2015,
ed the pilotless experiment across the eastern and western Baidu realized the road test of the driverless vehicle. At
parts of the United States. In 2005, in the “big challenge” the same time, it announced the establishment of the auto-
competition organized by the United States Department matic driving department, and plans to complete commer-
of defense, the multi-functional driverless car launched by cialization as soon as possible. In addition, BYD and other
Stanford University realized long distance travel on the auto companies have gradually stepped into the field of
rugged road. self driving [5].
In recent years, as Google, Tesla, BMW and other fa- In 2016, Baidu Inc and Ford automobile company in-
mous companies have joined the research of driverless cars, vested in laser radar manufacturer Velodyne, hoping to
unmanned technology has made great progress. In July reduce the production cost of driverless car laser radar.
2014, BMW and Baidu Inc established long-term cooper- In addition, Baidu has started to work with NVIDIA, a
ation on driverless cars. By the end of 2015, the driverless global graphic technology and digital media processor
car jointly developed by them successfully completed the industry leader, to form an automatic driving platform. At

2 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1115

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

the same time, Jingdong announced that its unmanned ve- produces control
hicle began to enter the road test phase and planned to run commands for the throttle, the brake, the steering wheel
the test and may be able to use it on a large scale in 2017. and the transmission lever to track the planned trajectory
Didi announced that driverless car would be one of their and send out the corresponding instructions. At present,
major strategic layouts, and will soon achieve the goal of the most commonly used method is the classical intelli-
carrying a passenger car on a driverless car. gent PID algorithm.

4. The Basic Technologies of Driverless Car 5. Governments’ Attitudes Towards Driver-

less Car
The most important part of driverless car is the automatic
driving system, which mainly consists of three parts: en- 5.1 The United States
vironmental perception, vehicle location and navigation,
and motion control. These three aspects complement each The United States is the leader in the development and
other and constitute the foundation of driverless car [6]. application of driverless car in the world. The U.S. De-
partment of Transportation (DOT) published the world’s
4.1 Environmental Perception first self-driving cars policy document Federal Automated
Vehicles Policy in September 20, 2016 [7]. This policy
In the driving process, people need to observe and analyze
includes the policy objectives and Taxonomy and Defi-
the state of the vehicle itself, other vehicles, pedestrians,
nitions for Terms Related to On-Road Motor Vehicle
road, traffic lights and so on, which is called the per-
Automated Driving Systems. The previous day, President
ception of the traffic environment. This perception first
Obama introduced a new deal in Pittsburgh Post-Gazette:
obtains information through multi-senses, and then uses
“Automated vehicles could change their lives. Safer, more
information and experience to make sense information
accessible driving. Less congested, less polluted roads.
and make decisions. Like people, driverless car also needs
That’s what harnessing technology for good can look like.
to perceive the environment when driving autonomously.
But we have to get it right. That’s why my administration
Environmental perception, which is similar to the driver’s
is rolling out new rules of the road for automated vehicles
eyes, usually made up of camera devices, sensors, and
– guidance that the manufacturers developing self-driving
so on. It is used to perceive the driving environment, and
cars should follow to keep us safe. [8]”
then to produce behavioral decisions based on the useful
environmental information obtained. 5.2 China
4.2 Vehicle Location and Navigation In November 15, 2017, China’s Ministry of science and
technology announced that it would rely on Baidu to build
Vehicle location and navigation is similar to the driver’s
a new generation of automatic driving national innovation
map. Currently, the most commonly used technologies in-
platform. In December 18, 2017, the Beijing Municipal
clude magnetic navigation and visual navigation. Among
Transportation Committee, the Public Security Manage-
them, magnetic navigation is the most mature and reliable
ment Bureau, the Economic Information Committee and
scheme at present. The biggest advantage of magnetic nav-
other departments issued two guiding documents on the
igation is that it’s not affected by natural conditions such
guidance of speeding up the road test of automatic driving
as weather. However, magnetic navigation methods often
vehicles and the detailed rules for the implementation of
need to bury certain navigation devices on the road, and
the road test management for automatic driving vehicles
the implementation process is cumbersome. The advantage [9]
. It’s clearly pointed out that the independent legal entity
of visual navigation is that it can prevent cars deviate from
registered in China can apply for temporary driving on
the target lane. But with low visibility caused by natural
account of automatic driving related research and stereo-
factors, the navigation system will not work. Due to the low
type test. Beijing has taken the lead in making a complete
requirement for infrastructure, visual navigation is general-
set of laws and regulations for driverless car. To a certain
ly recognized as the most promising location method.
extent, it supports the development and popularization of
4.3 Motion Control driverless car.

Motion control system is similar to the driver’s brain, 6. The Future Development Direction of the
which is used to analyze road information. It’s based on Driverless Car Field
the driving trajectory, speed planning and the current po-
sition, posture and speed of the driverless car. And then it The development direction of driverless car can be divid-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1115 3

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

ed into two categories through road conditions: automatic nology, people would use intelligent driverless cars in the
driving on expressways and automatic driving in cities. near future. It is a kind of intelligent car, which combines
The driving environment and traffic signs on the free- detection, identification, judgment, decision, optimization,
way are relatively better, but long time driving at high optimization, execution, feedback, and control function.
speed is rather boring for drivers. The application of driv- It also can learn, summarize and improve skills. Besides,
erless car in this direction can solve this problem well, and it integrates the top scientific and technological achieve-
also can effectively improve the efficiency and safety of ments such as microcomputers, green environment power
traffic. In the future development, as long as driverless car system, new structure material and so on. The driverless
completes the mark line tracking and vehicle identification technology at home and abroad is developing continuous-
function on a well structured highway, and can avoid the ly, and its function and reliability are constantly improv-
same track as far as possible, the full automatic driving of ing. What’s more, further improvements are needed in the
the highway can be achieved successfully. aspects of security, intelligence and regulation of driver-
In cities, there are more vehicles and people on the less cars.
road. Additionally, the environment is much more com-
plex. In the autonomous driving test which took place
References
in Parma in July 2013, the vehicle has proven to be able [1] Wikipedia. Autonomous car. [Online], Available [Ac-
to drive autonomously, without any human intervention, cessed Mar. 30, 2018]:
sharing roads with other vehicles, and manage round- https://en.wikipedia.org/wiki/Autonomous_car
abouts, intersections, priority roads, stops, tunnels, cross- [2] Q.Y. Lv. Prospects for cooperation between un-
walks, traffic lights, highways, and urban roads. But one manned technology and traditional automotive indus-
of the aspects that need to be further investigated and try. Auto Time. U463.6, 2017.
developed is the driving efficiency and speed: the test was [3] H. Lipson. A Driverless World. in Driverless: intelli-
carried out considering safety as a priority, and the most gent cars and the road ahead. Cambridge, Massachu-
complicated maneuvers were carried out at a reduced setts: MIT Press, 2016: 23-45.
speed. Plus some perception problems still have to be [4] W.G. Qiao and X.J. Xu. Development status and di-
solved: the problem of merging in large and multi-lane rection of unmanned vehicle. Shanghai Car. U469,
roundabouts, where vehicles travel at high speeds, has not 2007.
been completely solved [10]. So it still needs further techni- [5] Y.L. Jia. The origin and development of unmanned
cal support and research, and puts forward higher require- vehicles. Information Recording Material. U463.6,
ments for the technology. 2018.
[6] Z. Ai. Is the pilotless time really coming. China
7. Conclusions Equipment Engineering. U463.6, 2018.
[7] U.S. DOT. Federal Automated Vehicles Policy – Ac-
This paper investigates the current development of driv- celerating the Next Revolution In Roadway Safety.
erless cars at home and abroad, analyses the basic tech- Washington, DC, 2016.
nologies of driverless cars, compares the American gov- [8] B.H. Obama. Barack Obama: Self-driving, yes, but
ernment’s attitudes with Chinese government’s attitudes also safe. Pittsburgh Post-Gazette. Pittsburgh, PA,
towards driverless cars, and expounds the future develop- 2016.
ment direction of the driverless car in two ways. [9] S.X. Guo. Policy escort “unmanned” ahead of speed.
The results indicate that the development of driverless New Production. F426.471, 2018.
cars would be the development direction of the automo- [10] B. Alberto, C. Pietro, D. Stefano, L.M. Chiara,
tive industry in the future, and plays an important role in M. Paolo; M. Daniele, P. Matteo, and P. Antonio.
promoting the technology innovation in automobile indus- PROUD-Public Road Urban Driverless-Car Test.
try. With the development of economy, science and tech- IEEE Trans. Phys. Electron. 2015, 16: 3508 - 3519.

4 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1115

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Artificial Intelligence Advances

https://ojs.bilpublishing.com/index.php/aia

ARTICLE
Probabilistic Rationale of Actions for Artificial Intelligence Systems
Operating in Uncertainty Conditions
Andrey I. Kostogryzov1,2*
1. Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences, Vavilova Str. 44,
bld.2, Moscow, 119333, Russia
2. Gubkin Russian State University of Oil and Gas (National Research University) Leninsky Av. 65, bld. 1, Moscow,
119991, Russia

ARTICLE INFO ABSTRACT

Article history The approach for probabilistic rationale of artificial intelligence systems
Received: 10 September 2019 actions is proposed. It is based on an implementation of the proposed
interconnected ideas 1-7 about system analysis and optimization focused
Accepted: 15 October 2019 on prognostic modeling. The ideas may be applied also by using another
Published Online: 30 October 2019 probabilistic models which supported by software tools and can predict
successfulness or risks on a level of probability distribution functions. The
Keywords: approach includes description of the proposed probabilistic models, optimi-
Analysis zation methods for rationale actions and incremental algorithms for solving
the problems of supporting decision-making on the base of monitored data
Artificial intelligence systems and rationale robot actions in uncertainty conditions. The approach means
Model practically a proactive commitment to excellence in uncertainty conditions.
Operation A suitability of the proposed models and methods is demonstrated by ex-
amples which cover wide applications of artificial intelligence systems.
Prediction
Probability
Rationale
Risk
System
System engineering

　
1. Introduction processes”, ISO 17359 “Condition monitoring and di-

D
agnostics of machines - General guidelines”, IEC 61508
ifferent mathematical models and methods are “Functional safety of electrical/ electronic/ programmable
applied in system analysis. System analysis is electronic safety-related systems” etc. It is recommended
required at level of the international standards of for using every time across all life-cycle to analyze per-
system engineering - for example, ISO/IEC/IEEE 15288 formance, system behaviour, feasibility, affordability, crit-
“System and software engineering – System life cycle ical quality characteristics, technical risks, sensitivity for

*Corresponding Author:
Andrey I. Kostogryzov,
Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences, Vavilova Str. 44, bld.2, Moscow,
119333, Russia; Gubkin Russian State University of Oil and Gas (National Research University) Leninsky Av. 65, bld. 1, Moscow,
119991, Russia;
Email: [email protected]

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 5

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

changes of critical parameters values etc. Artificial intel- (2) the problem 2 - to rationale a robot actions under
ligence systems (AIS) which are understood here as sys- limitations on admissible risks of “failures” (according to
tems, performing functions by logic reasoning on the base ISO Guide 73 risk is defined as effect of uncertainty on
of data processing, also needs system analysis because of objectives considering consequences. An effect is a devia-
their complexities and uncertainty conditions. tion from the expected — positive and/or negative).
Note: System is combination of interacting elements or- Note: Some relevant problems (such as the problems of
ganized to achieve one or more stated purposes (according robotics orientation, localization and mapping, informa-
to ISO/IEC/IEEE 15288). tion gathering, the perception and analysis of commands,
Considering AIS specificity there may be some scien- movement and tactile, realizations of manipulations for
tific problems devoted to: which different probabilistic methods are also applicable)
(1) system analysis of uncertainty factors, capabilities have not been covered by this work.
of operation in real time, information gathering and pro- The proposed approach for solving AIS problems are
cessing, protection from authorized access and dangerous based on theoretical and practical researches [15-45] and
influences; need to be used either in combination or in addition to ex-
(2) analysis of system requirements to acceptable con- isting methods which are used now in AIS. There, where it
ditions; is required often prognostic system analysis or where the
(3) system analysis and optimization in architectural used approaches are not effective, the proposed probabi-
design; listic approach can be used as rational basis or alternative.
(4) comparative and prognostic estimations of quality, The ideas of this approach may be applied also by using
safety, interaction “user-system” and conditions, optimi- another probabilistic models which supported by software
zation of different processes, rationale of operation in un- tools and can predict success or risks on a level of prob-
certainty, etc. ability distribution functions (PDF). The structure of this
Now there isn’t enough universal effective approach research is shown by the Figure 1.
to rationale of actions for AIS operating in uncertainty
conditions. In practice for each concrete case it is often
used subjective expert estimations, a regression analysis
of collected data, a simulation of processes [1-14]. It means,
that search of new methods for advanced rationale actions
of AIS and by AIS is today very important. The proposed
approach is focused on probabilistic rationale of actions
to operate in uncertainty conditions against existing ap-
proaches for which applied mathematical methods cover
mainly information processing in the logician if …, that
… and/or tracing situations by a man-operator. An ap-
plication scope of this paper covers AIS supporting de-
cision-making in intellectual manufacture (for example,
in dispatcher intelligence centers) and robotics systems
operating in uncertainty conditions and used to provide
Figure 1. The structure of the research
operation efficiency or/and increase reliability and safety
(including aerial, land, underground, underwater, univer- Various fields of the examples applications have been
sal and functionally focused AIS). chosen purposely to demonstrate universality and analyti-
The main efforts of this paper are not connected with cal usefulness of the probabilistic approach. The proposed
illustrating the capabilities of AIS, but they are focused models and methods are an original Russian creation, they
on demonstrating the applicability of original probabilis- have been presented at seminars, symposiums, confer-
tic models and methods to improve some of the existing ences, ISO/IEC working groups and other forums since
capabilities of AIS [15-45]. For this goal by the use of these 2000 in Russia, Australia, Canada, China, Finland, France,
probabilistic models the next specific problems are cov- Germany, Italy, Kuwait, Luxembourg, Poland, Serbia,
ered: the USA, etc. The supporting software tools were award-
(1) the problem 1 - to rationale a rational variant for de- ed by the Golden Medal of the International Innovation
cision-making on the base of monitored data about events and Investment Salon and the International Exhibition
and conditions, and “Intellectual Robots”, acknowledged on the World's fair

6 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

of information technologies CeBIT in Germany, noted frequency of failures λ - is popular today. This implies
by diplomas of the Hanover Industrial Exhibition and the the use of corresponding exponential PDF – see Figure 2.
Russian exhibitions of software. Only one connection of the frequency of failures λ with
random time variable τ between losses of system integrity
2. The essence of the approach may be interpreted as the requirement: “to provide no fail-
ures during required time with probability no less than the
The AIS behaviour corresponding to the rationale of ac-
given admissible probability Padm. this required time should
tions for AIS operating in uncertainty conditions means
be no more than treq. = 1/ λadm., here λadm.=-ln(1-Radm.)”. But
proactive commitment to excellence. Such behaviour is
for AIS element it is rough and unpromising engineering
based on an implementation of the next proposed inter-
estimations because capabilities of monitoring conditions
connected ideas 1-7.
and recovery of the lost element integrity are ignored.
Idea 1 is concerning the usual concept and properties
Such disregard deforms very essentially probabilistic es-
of probability distribution functions (PDF) [15] for a con-
timations of probabilistic risk values and can’t be useful
tinuous random variable of time. PDF for a time variable
for scientific search of effective counteraction measures
τ is nondecreasing function P(t) whose value for a given
against different threats. Deviations from more adequate
point t≥0 can be interpreted as a probability that the value
PDF estimations are very high [33,44,45]. On Figure 3 the
of the random variable τ is less or equal to the time value t,
limitations to admissible risks, fragment of exponential
i.e. P(t)=P(τ≤t). Additionally P(t)=0 for t<0, and P(t)→1
and an adequate PDF of time between losses of system
for t→∞. In general case the solutions for the problems 1
integrity with identical frequency of system integrity
and 2 are based on using concept of the probabilities of
losses are illustrated (in conditional units). It means more
“success” and/or “unsuccess” (risk of “failure”) during
adequate PDF allows more right understanding of prob-
the given prognostic time period treq.. This probability is a
abilistic AIS vision of events prediction with scientific
value for a point treq. and is defined by created PDF.
interpretation considering situations in time line.
Idea 2. The processes, connected with data processing,
Note: System integrity is defined as such system state
and used information should provide required AIS oper-
when system purposes are achieved with the required
ation quality (because AIS is a system, performing func-
quality.
tions by logic reasoning on the base of data processing).
And corresponding probabilistic methods should appro-
priate for prognostic estimations.
Idea 3. The PDF should be presented as analytical de-
pendence on input parameters. It needs to solve direct and
inverse problems to rationale of actions in a real time of
AIS operation. For example, for a simple element PDF
P(t) of time τ between losses of element integrity may be
presented by analytical exponential approximation, i.e.
Figure 2. The possible variants of correlations for ad-
P(t) = 1-exp(-λt), where λ is frequency of failures (losses
missible risks, exponential and an adequate PDF of time
of element integrity). At the same time frequency of fail- between losses of system integrity with identical frequen-
ures may be represented as a sum of frequencies of fail- cy of losses
ures because of specific reasons for each failure type – for
example, failure from “human factor” λ1, from hardware
λ2, from software λ3 and so on. For this use case PDF may
be presented as P(t) =1-exp[-(λ1+λ2+λ3+…)t]. Then if the
adequate function P(t) is built in dependence on different
parameters and if admissible level for probability is given
than inverse problem may be solved.
Note. The rationale for exponential approximation
choice in practice see, for example, in [28, 30].  
Idea 4. The PDF should be adequate, it means a de- Figure 3. All requirements to admissible risk are met for
pendence on several essential parameters which define an adequate PDF of time between losses of system integ-
AIS operation and on which “success” or “failure” of rity
AIS operation is mainly dependent. For example the way Idea 5. Because an AIS is a complex system and this
for risks prediction based on uses only one parameter -

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 7

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

AIS may be subsystem or element of comprehensive ered information is used for proper AIS specificity. The
complex system, the proposed approach should allow abstract view on a quality of used information is presented
a generation of probabilistic models for prediction of on Figure 4.
“success” or “failure” of AIS actions in uncertainty con-
ditions. In general case an input for generated models
used in real time should consider system complexity,
periodical diagnostics, monitoring between diagnostics,
recovery of the lost integrity for every system element
and also processes, connected with data processing, and
used information. As an output of such generated models Figure 4. Abstract explanation for a quality of used (real)
adequate PDF of time τ between losses of system (sub- information against required one
system, element) integrity should be produced in analyti- The proposed models for the estimation of information
cal form. systems operation quality are described in Table A.1 of
Idea 6. Input for probabilistic modeling should be Appendix.
formed mainly from gathered data and established specific The main analytical models and calculated measures
order of AIS actions. are the next:
Idea 7. To probabilistic rationale of actions for AIS (1) “The model of functions performance by a complex
operating in uncertainty conditions the problems of opti- system in conditions of unreliability of its components”;
mization should be solved. Optimization should be per- (2) “The models complex of calls processing”;
formed in real time by defined beforehand optimization (3) “The model of entering into IS current data con-
problem statement. Every time the used optimization cerning new objects of application domain”;
problem statement should be appropriated for solving (4) “The model of information gathering”;
specific problem 1or 2. For probabilistic rationale of (5) “The model of information analysis”;
actions the prognostic period should be defined so to be (6) “The models complex of dangerous influences on a
in time to do the given action or complex of actions on protected system”;
acceptable level according to optimization criterion or to (7) “The models complex of an authorized access to
perform preventive action (with which the initiation of system resources”.
performing an action or solving a problem is connected) Risk to lose integrity (R) is an addition to 1 for prob-
or/and to recover operation capabilities (which can be ability of “success” (P), i.e. R=1-P considering conse-
lost). quences.
For the approach implementation the next probabilistic These models, supported by different versions of soft-
models are proposed. ware Complex for Evaluation of Information Systems Op-
eration Quality, registered by Rospatent №2000610272 [46],
3. The Description of the Proposed Models may be applied and improved for solving problems 1 and
In general case a probabilistic space (Ω, B, P) for probabi- 2.
listic modeling is created [15], where: Ω - is a limited space
3.2 About Risks Prediction for System Formalized
of elementary events; B – a class of all subspace of Ω-space,
as “Black box”
satisfied to the properties of σ-algebra; P – is a probabil-
ity measure on a space of elementary events Ω. Because, The proposed models helps to implement ideas 1, 3, 4.
Ω={ωk} is limited, there is enough to establish a reflection In general case successful system operation (not only
ωk→pk =P(ωk) like that pk ≥ 0 and ∑ p k = 1 . AIS) is connected with system counteraction against var-
k
In order not to overload the reader with mathematical ious dangerous influences on system integrity - these may
details, the final formulas for calculations are presented in be counteractions against failures, defects events, “human
the Appendixes A and B. factors” events, etc. There are proposed the formalization
for two general technologies of providing counteraction
3.1 About AIS operation quality against threats: periodical diagnostics of system integrity
The proposed models help to implement ideas 1 and 2. (technology 1, without monitoring between diagnostics)
In general case AIS operation quality is connected with and additionally monitoring between diagnostics (technol-
requirements for reliable and timely producing complete, ogy 2). As a rule these technologies are implemented by
valid and/or, if needed, confidential information. The gath- AIS.

8 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Technology 1 is based on periodical diagnostics of makes an error but a dangerous influence occurs if the
system integrity, that is carried out to detect danger danger is activated before the next diagnostic. Otherwise
sources penetration into a system or consequences of the source will be detected and neutralized during the next
negative influences (see Figure 5). The lost system in- diagnostic.
tegrity can be detect only as a result of diagnostics, after It is supposed for technologies 1 and 2 that the used di-
which system recovery is started. Dangerous influence agnostic tools allow to provide necessary system integrity
on system is acted step-by step: at first a danger source recovery after revealing danger sources penetration into a
penetrates into a system and then after its activation system or consequences of influences.
begins to influence. System integrity can’t be lost be- The probability of correct system operation within the
fore a penetrated danger source is activated. A danger is given prognostic period (i.e. probability of “success” - P)
considered to be realized only after a danger source has may be estimated as a result of use the models presented
influenced on a system. in Appendix B. Risk to lose integrity (R) is an addition
to 1 for probability of correct system operation (P), i.e.
R=1-P considering consequences.

3.3 About a Generation of Probabilistic Models

for Complex System
The proposed method for a generation of probabilistic
models helps to implement ideas 1 and 5.
The basic ideas of correct integration of probability
metrics are based on a combination and development of
models. For a complex systems with parallel or serial
structure described there are proposed the next method
Figure 5. Some accident events for technology 1 (left –
to generate adequate probabilistic models [25,26,28-30]. This
correct operation, right – a lose of integrity during prog-
nostic period Treq. ) method uses the usual way of probability theory for
independent random variables. However, given the im-
Technology 2, unlike the previous one, implies that portance to rationale the generation of new probabilistic
system integrity is traced between diagnostics by operator models for complex system, the approach is described
(operator functions may be performed by a man or special below.
AIS component or their combination). In case of detecting Let’s consider the elementary structure from two in-
a danger source an operator recovers system integrity. The dependent parallel or series elements. Let’s PDF of time
ways of integrity recovering are analogous to the ways of between losses of i-th element integrity is Вi(t) =Р (τi≤ t),
technology 1 – see Figure 6. then:
(1) time between losses of integrity for system com-
bined from series connected independent elements is
equal to a minimum from two times τi: failure of 1st or
2nd elements (i.e. the system goes into a state of lost in-
tegrity when either 1st, or 2nd element integrity is lost).
For this case the PDF of time between losses of system
integrity is defined by expression

В(t) = Р[min (τ1,τ2)≤t]=1- Р[min (τ1,τ2)>t]=1-Р(τ1>t)Р(τ2 >

t)= 1 – [1-В1(t)] [1- В2(t)](1)

(2) time between losses of integrity for system com-

Figure 6. Some accident events for technology 2 (left –
bined from parallel connected independent elements (hot
correct operation, right – a lose of integrity during prog-
reservation) is equal to a maximum from two times τi:
nostic period Treq. )
failure of 1st and 2nd elements (i.e. the system goes into
Faultless operator’s actions provide a neutralization of a state of lost integrity when both 1st and 2nd elements
a danger source trying to penetrate into a system. A pen- have lost integrity). For this case the PDF of time be-
etration of a danger source is possible only if an operator tween losses of system integrity is defined by expression

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 9

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

В(t)=Р[max(τ1,τ2)≤t]=Р(τ1≤t)Р(τ2≤t)=В1(t)В2(t)(2) and on the current moment of time. For example, the rang-
es of possible values of conditions may be established:
Applying recurrently expressions (1) – (2), it is possi-
“Working range inside of norm”, “Out of working range,
ble to build PDF of time between losses of integrity for
but inside of norm”, “Abnormality” for each separate crit-
any complex system with parallel and/or series structure
ical parameter. If the parameter ranges of acceptable con-
and theirs combinations.
ditions are not established in explicit form than for mod-
An example of complex system integrating two serial
eling purpose the may be implead and can be expressed
complex subsystems (abstraction) is presented by Figure 7.
in the form of average time value. These time values are
For this integration the next interpretation of elementary
used as input for probabilistic modeling. For example, for
events is used: complex system integrating compound
coal mine some of many dozens heterogeneous parameters
components “Intellectual tructure 1 and 2” is in condition
are: for ventilation equipment - temperature of rotor and
“correct operation” (“success”) during given period Treq. if
engine bearings, a current on phases and voltage of stator;
during this period “AND” component “Intellectual truc-
for modular decontamination equipment - vacuum in the
ture 1” “AND” component “Intellectual tructure 2” (both
pipeline, the expense and temperature of a metano-air
are special complex subsystems including AIS subsystems
mix in the pipeline before equipment, pressure in system
and elements) are in condition “correct operation” (“suc-
of compressed air, etc. It may be interpreted similarly by
cess”).
light signals – “green”, “yellow”, “red” - see Figure 8 and
All ideas for analytical modeling complex systems are
following Example 6.3.
supported by the software tools “Mathematical modeling
of system life cycle processes” – “know how” (registered
by Rospatent №2004610858), “Complex for evaluating
quality of production processes” (registered by Rospatent
№2010614145) and others [46-51].

Figure 8. An example of universal elementary ranges for

monitoring data about events and conditions

4. Optimization Problem Statements for Ra-

tionale Actions
The proposed optimization problem statements for ratio-
nale actions helps to implement idea 7. For example the
proposed ideas 2-6 may be supported by the next typical
optimization problem statements for AIS [25,28,30]:
(1) on the stages of system concept, development, pro-
duction and support: system parameters, software, tech-
Figure 7. An example of complex system integrating two nical and control measures (Q) are the most rational for
serial complex intellectual structures which also are com- the given prognostic period if on them the minimum of
plex subsystems (abstraction) expenses (Zdev.) for creation is reached

Zdev. (Q rational) = min Z dev. (Q),

3.4 About Data Forming for Probabilistic Model-
ing Q

The proposed practical way to data forming helps to im- (1A) at limitations on probability of an admissible level
plement idea 6. of quality Pquality (Q) ≥ Padm. and expenses for operation
For each critical parameter (for which prognostic esti- Сoper. (Q) ≤ С adm. and under other development, operation
mations are needed to do actions) the ranges of acceptable or maintenance conditions; or
conditions can be established. The traced conditions of (1B) at limitations on admissible risk to lose system
monitored parameters are data about a condition before integrity R≤Radm. and expenses for operation Сoper. (Q) ≤
С adm. and under other development, operation or mainte-

10 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

nance conditions; or are based on using the models and methods above.
(1C) at limitations presented as combination 1A) and
1B); 5.1 The Algorithm for Solving Problem to Ratio-
(2) on utilization stage: nale a Rational Variant for Decision-making on
(A) system parameters, software, technical and control the Base of Monitored Data About Events and
measures (Q) are the most rational for the given period of Conditions (problem 1)
AIS operation if on them the maximum of probability of It is supposed that the terms “success” and accordingly
correct system operation is reached “unsuccess” (“failure”) are defined in terms of admissible
Pquality (Q rational) = max Pquality (Q), condition of interested system to operate for the purpose
according to required quality.
Q Note: For example for each parameter of equipment
the ranges of possible values of conditions may be esti-
(2A.a) at limitations on probability of an admissible
mated as “Working range inside of norm” and “Out of
level of quality Pquality (Q) ≥ Padm. and expenses for opera-
working range, but inside of norm” (“success”) or “Ab-
tion Сoper. (Q) ≤С adm. and under other operation or mainte-
normality” (“failure”), interpreted similarly light signals
nance conditions; or
– “green”, “yellow”, “red”. For this definition a “ailure”
(2A.b). at limitations on admissible risk to lose system
of equipment operation characterizes a threat to lose
integrity R≤Radm. and expenses for operation Сoper. (Q) ≤С
system norm integrity after danger influence (on the logic
adm. and under other operation or maintenance conditions;
level this range “Abnormality” may be interpreted analyt-
or
ically as failure, fault, losses of quality or safety etc.).
(2A.c). at limitations presented as combination 2A.a)
The proposed steps for solving problem 1 to rationale a
and 2A.b);
rational variant for decision-making on the base of moni-
tored data about events and conditions may be carried out
(B) system parameters, software, technical and control
by the next 4 steps – see Figure 9.
measures (Q) are the most rational for the given period of
system operation if on them the minimum of risk to lose
system integrity is reached

R (Q rational) = min R(Q), Figure 9. Steps for solving problem 1

Q Step 1. The complete set of variants for actions is
defined, including for each variant – a definition of com-
(2B.a) at limitations on probability of an admissible
pound components is being. Each use case may be charac-
level of quality Pquality (Q) ≥ Padm. and expenses for opera-
terized by an expected benefit in comparable conventional
tion Сoper. (Q) ≤С adm. and under other operation or mainte-
units. If the objective value of a benefit can’t be defined,
nance conditions; or
expert value of a level of “success” may be established,
(2B.b) at limitations on admissible risk to lose system
for example, on a dimensionless scale from 0 to 100 (0 –
integrity R≤Radm. and expenses for operation Сoper. (Q) ≤С
«no benefit», i.e. “failure”, 100 – «the maximal benefit »,
adm. and under other operation or maintenance conditions;
i.e. complete “success”).
or
Step 2. The measures and optimization criteria are cho-
(2B.c). at limitations presented as combination 2A.a)
sen (see sections 3 and 4). As criteria there can be accept-
and 2A.b);
ed:
These statements may be transformed into the problems
(1) maximum of benefit as a result of system operation
of expenses minimization in different limitations. There
under the given conditions and limitations on the accept-
may be combination of these formal statements in system
able risk of “failure” and/or other limitations;
life cycle.
(2) maximum probability of “success” or minimum risk
Note. Another variants of optimization problem state-
of “failure” under limitations.
ments are possible.
Step 3. The knowledge is used to refine the input for
5. The incremental algorithms for solving the modeling. Using the probabilistic models and methods for
each variant, the “success” measures are calculated for the
problems 1 and 2
given prognostic period. From a set of possible variants
The proposed algorithms for solving the problems 1 and 2 the optimal one is chosen according to the step 2 criterion.

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 11

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Note: Formal statements of optimization may be con- use data from various sources (for example, from air
nected with maximization of benefit at limitations on drones, intelligent buoys on the water or sensors under
admissible levels of quality and/or risks measures or with water, etc.). If necessary, possible damages are taken into
minimization of risks at limitations on admissible levels of account. For example, each use case may be character-
benefit and/or quality and/or risks measures and/or under ized by an expected damages in comparable conventional
other operation or maintenance conditions (see section 4). units. If the objective value of a damage can’t be defined,
Step 4. A decision for the optimal variant of actions expert value of expected level of “failure” for each variant
(defined in step 3) is made. In support of the efficiency of may be set, for example, on a dimensionless scale from 0
the functions, the achievable benefit calculated at step 3 is to 100 (0 – «no damages», i.e. “success”, 100 – «the max-
recorded. New knowledge is improved and systematized imal damage»).
by comparing it with reality (including comparisons of The index i of the first part of the selected route is set
probabilistic estimations and real events). to the initial value i=1.
Note: A solution that meets all conditions may be not Step 2. The knowledge is used to refine the input for
existing. In this case, there is no optimal variant of system prognostic modeling. Using probabilistic model, a cal-
operation on the base of monitored data about events culation of the probability of “failure” (risk of “failure”)
and conditions. is carried out for each variant. From the set of variants
(remaining route) the optimal one is chosen, for its the
5.2 The Algorithm for Solving Problem to Ratio- minimum probability of “failure” (risk of “failure”) is
nale a Robot Actions under Limitations on Ad- achieved.
missible Risks of “Failures” (problem 2) Step 3. The robot overcomes the i-th part of the select-
The approach for solving problem 2 to rationale a robot ed route. If the part can’t be overcome, the comeback to
actions under limitations on admissible risks of “failures” the initial point of the part is being. If an alternative route
is demonstrated in application to robot route optimization isn’t here, the comeback to initial point of the previous
in conditions of uncertainties. part is being. The input for modeling every part of possi-
For a robot, the concept of “failure” under uncertainty ble route for each of the variants are updated. New knowl-
is defined as the failure to achieve the goal within a given edge is improved and systematized by comparing it with
time. It is assumed that there are several possible routes to reality (including comparisons of prognostic risks and real
achieve the goal, and uncertainties may include both the events).
conditions for robot operation (including random events in Step 4. If, after overcoming the i-th part, the robot
orientation, localization and mapping). The minimum risk arrived at the intended point of route (i.e., the last part
of “failure” under the existing conditions and limitations of the route is overcome and the goal is achieved), then
is set as a criterion of optimization. the solution for optimizing the route is complete. If the
The proposed steps for solving problem 2 of robot robot hasn’t yet arrived at the intended point (i.e. the last
route optimization under limitations on admissible risks part of the route isn’t overcome), then the complete set of
of “failure” under conditions of uncertainties may be car- different route variants for achieving the goal is redefined
ried out by the next 4 steps – see Figure 10. (similar to step 1). The input for modeling every part of
possible route for each of the variants are updated, i= i+1.
Then steps 2-4 are repeated until the last part of the route
is overcome on the set of possible variants (i.e. it means
the goal is achieved and problem 2 is solved).
If the set of possible options is exhausted and the goal
is not achieved, it is concluded that the goal is unattain-
able with the risk of “failure” less than the acceptable risk
(i.e., it means an impossibility of solving problem 2 in the
defined conditions).
Figure 10. Steps for cognitive solving problem 2
Thus, to rationale a robot actions under limitations on
Step 1. The complete set of route variants to achieve admissible risks of “failures” (i.e. to a “successful” solu-
the goal within the given time, and for each variant – a set tion of problem 2) in real time, information gathering,
of components, is defined (redefined). Data characterizing probabilistic predictions for possible route variants, their
every part of route for each of the variants are gathered comparison, the choice of the best variant, the implemen-
(refined) for modeling. To do this, a specific robot can tation of further actions, the improvement, systematization

12 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

and use of knowledge are being. time between operator’s error during continuous monitor-
ing of system integrity is estimated not less than 1 year (for
6. Examples general technology 2). Initial input data for probabilistic
modeling are reflected by the Table 1, the used model is
6.1 About a Period of Successful System Opera-
described in subsection 3.2 of this paper.
tion by AIS Capabilities
The example is related partly to solving the problem 1 Table 1. Input for estimation
and concerning an estimation of successful system opera-
Variants for comparisons
tion during a long time by AIS capabilities in comparison Input for modeling
1-st (an usual system) 2-nd (an AIS)
against an usual system without or with usual sensors
(without artificial intelligence capabilities to logic reason- The given prognostic period (“in
3 years 5 years
future”)
ing). The frequency of influences for
How long time may be a period of successful system 1 day-1 1 day-1
penetrating into system
operation by AIS capabilities? And what about conditions The mean activation time 6 hours 6 hours
for this long period? The time between the end of 4 hours (by
Those threats to system operation which are known, diagnostic and the beginning of 1 month AIS capabili-
traced at diagnostics and do not cause irreversible con- the next diagnostic ties)
1 hour (by AIS
sequences at the first influence, are considered only. Be- The diagnostic time 4 hours
capabilities)
sides, it is supposed, that an integrity can be operatively The mean time between oper-
recovered after AIS recovering reaction at the earliest ator’s error during continuous 1 year 1 year
monitoring of system integrity
stages of detection of dangerous or guarding symptoms.
Moreover, at modeling the time of full integrity recover- Some probabilities of providing system integrity in
ing is artificially reduced till diagnostic time. Thus, the dependence on input, changing in diapason -50%+100%
elementary condition “acceptable integrity” means such from Table 1 data, are presented on Figures 11-139. They
system state when system purposes are achieved with cover dependences on the given prognostic period, the
the required quality, i.e. absence of danger source or time between the end of diagnostic and the beginning of
neutralization of a penetrated source at the earliest stage the next diagnostic, the mean time between operator’s er-
prior to the its danger influence after activation. It (as ror during continuous monitoring of integrity. Deviations
supposed by the model) enough for successful AIS oper- for other dependences are insignificant.
ation.
Note: The above assumptions are supposed for model-
ing. In a reality it may be not always so. These conditions
are considered for interpretation of modeling results.
To compare system operation with AIS capabilities
against an usual system (without artificial intelligence
capabilities) for the same conditions we consider AIS pos-
Figure 11. The probability of providing system integrity
sibilities to provide “acceptable integrity” by continuous in dependence on the given prognostic period
monitoring with artificial intelligence logic reasoning.
Let's the threats to system integrity are being about 1
time a day because of natural or technogenic threats and
“human factor”. Let's also after occurrence of a danger
source an average activation time is equal to 6 hours,
during which else it is possible to prevent or neutralize
negative influence.
Two variants of reaction caring of AIS integrity are
compared. 1 st variant (an usual system) considers the Figure 12. The probability of providing system integrity
address to a recovering center about 1 time a month and in dependence on the time between the end of diagnostics
and the beginning of the next diagnostics
reception of necessary recovering procedures within 4
hours after diagnostics. 2nd variant means AIS perform-
ing functions of diagnostics every 4 hours and recovering
acceptable integrity within one hour. For all variants mean

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 13

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

oil-and-gas systems), emergency services etc. [20,25,28,30-33,35-

45]
. The results are based on described modeling to provide
quality of output information producing, quality of used
information IS and security of IS operation (see Table A.1
from Appendix A).
According to this generalization for the best practice of
IS operation the acceptable requirements are the next (see
Figure 13. The probability of providing system integrity the measures from Table 1):
in dependence on the mean time between operator’s errors (1) to provide quality of output information producing:
during continuous monitoring of system integrity A. Probability of providing reliable function perfor-
Results of modeling show, that for 1st variant (for an mance during given time should be no less than 0.99;
usual system) the probability to provide “acceptable in- B. System availability should be no less than 0.9995;
tegrity” during 1 year is equal to 0.39, during 2 years – C. Probability of well-timed calls processing during the
not less than 0.16, during 3 years – only 0.07. It means required term should be no less than 0.95;
practically the inevitability of a failure during 2-3 years. D. Relative portion of well-timed processed calls of
2nd variant (for AIS) with operative recovering is more those types for which the customer requirements are met
effective. Really, it is possible to provide “acceptable in- should be no less than 95%;
tegrity” for system operation with AIS capabilities within (2) to provide quality of used information:
3-5 years with probability about 0.90-0.93 – it may be in- A. Probability that system contains information about
terpreted as successful operation 9 times from 10 possible states of all real object and coincides should be no less
five-year periods. These results of modelling should serve than 0.9;
a rationale for development counteractions against threats. B. Probability of information actuality on the moment
Conditions for five-year period of successful system op- of its use should be no less than 0.9;
eration with AIS capabilities are presented in Table 1 for C. Probability of errors absence after checking should
2nd variant. be no less than 0.97;
Note. Serrated and nonmonotonic character of depen- D. Probability of correct analysis results obtaining
dence on Figures 11, 12 (left) is explained by the periodic should be no less than 0.95;
diagnostics, monitoring presence or absence and their E. Probability of providing information confidentiality
quantitative values, and also because of parameter “N” is during objective period should be no less than 0.999;
integer part – see Appendix B. Detais see in [30]. (3) to provide security of IS operation:
Of course the concepts “acceptable integrity” and “fail- A. Probability of faultless (correct) operation under
ure” of special system should be defined in details (which dangerous influence on IS during given time should be no
produced input for modeling). However the expected less than 0.95;
modeling results against typical plausible input for this B. Probability of system protection against unautho-
this simple example has also demonstrated for readers a rized access should be no less than 0.99.
suitability of the proposed probabilistic “Black box” mod- These values characterizes some admissible limitations
els (from section 3). for probabilities of “success” (P) and risks of “unsuccess”
(R=1-P) for information systems operation quality.
6.2 Example 2 of Acceptable Requirements to The fulfillment of these requirements is a certain sci-
Solve Problem 1 for Information Systems Opera- entifically proved guarantee of the quality of information
tion used by AIS.

The example is connected with rationale a rational re- 6.3 Example of Solving Inverse Problem to Es-
quirements to information system (IS) operation for pro- timate the Mean Residual Time before the Next
viding high information quality for using in an AIS. Infor- Parameters Abnormalities for a Coal Company
mation systems are systems for which input is information
and output (as result of IS operation) also is information The example demonstrates an AIS possibility on the base
for following use according to purpose. This example of solving inverse problem by model described in subsec-
summarizes the numerous results of researches performed tion 3.2 and Appendix B to a rationale of actions in a real
for IS operating in government agencies, manufacturing time for a coal company
structures (including power generation, coal enterprises, Conditions of parameters, traced by dispatcher intelli-
gence center, are data about a condition before and on the

14 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

current moment of time, but always the future is more im- Here R(Tpenetr, t, Tbetw, Tdiag, Treq.) is risk to lose integ-
portant for all. With use of current data responsible staff rity, it is addition to 1 for probability P(Treq) of providing
(mechanics, technologists, engineers, etc.) should know system integrity (“probability of success”), for calcula-
about admissible time for work performance to maintain tions the formulas (B.1)–(B.3). Tpenetr is the mathematical
system operation. Otherwise because of ignorance of a expectation of PDF Ωpenetr (τ), it is defined by parameter
residual time resource before abnormality the necessary statistics of transition from “green” into “yellow” range
works are not carried out. I.e. because of ignorance of this (see Figure 8). The others parameters Tbetw, Tdiag in (3) are
residual time it is not undertaken measures for prevention known – see Appendix B. The main practical questions
of negative events after parameters abnormalities (failures, are: what about Treq. and what about a given admissible
accidents, damages and-or the missed benefit because of risk Radm.(Treq)? For answering we can use the properties of
equipment time out). And on the contrary, knowing resid- function R(Tpenetr, t, Tbetw, Tdiag, Treq.):
ual time before abnormality these events may be avoided, (1) if parameter t increases from 0 to ∞ for the same
or system may be maintained accordingly. For monitored another parameters, the function R(…, t, …) is monoto-
critical system the probabilistic approach to estimate the nously decreasing from 1 to 0 (for N – real, i.e. no integer
mean residual time before the next parameters abnormali- part), if the mean activation time of occurred danger (threat
ties for each element and whole system is proposed. - from the 1-st input at the “yellow” range to the 1-st input
For every valuable subsystem (element) monitored pa- in the “red” range) is bigger to lose integrity is less;
rameters are chosen, and for each parameter the ranges of (2) if parameter Treq increases from 0 to ∞ for the same
possible values of conditions are established: “In working another parameters, the function R(…,Treq) is monoto-
limits”, “Out of working range, but inside of norm”), “Ab- nously increasing from 0 to 1, i.e. for large Treq risk ap-
normality” (interpreted similarly light signals – “green”, proaches to 1.
“yellow”, “red”) – see Figures 8 and 14. The condition It means the such maximal x exists when t=x and
“Abnormality” characterizes a threat to lose system integ- Treq.=x and 0<R(Tpenetr, x, Tbetw, Tdiag, x)<1. I.e. the residual
rity. time before the next parameter abnormality (i.e. time be-
fore first next coming into “red” range) is equal to defined
x with confidence level of admissible risk R(Tpenetr, x, Tbetw,
Tdiag, x). So, if Tpenetr =100 days, for Radm.= 0.01 residual
time x≈2.96 weeks (considering decisions of recovery
problems of integrity every 8 hours). Adequate reaction of
responsible staff in real time is transparent for all interest-
ed parties. Details see [35].

6.4 Example of Solving Problem 1 by AIS Oper-

ating for Providing Safety of a Floating Oil and
Gas Platform
For estimation and rationale the possibilities of a floating
oil and gas platform operation (considered as a system)
Figure 14. Example of a prognozed residual time before the probabilistic modeling is being to answer the next
the next parameter abnormality question: “What risks to lose system integrity may be for
For avoiding the possible crossing a border of “Abnor- a year, 10 and 20 years if some subsystems are supported
mality” a prediction of residual time, which is available by special AIS on the levels which are proper to skilled
for preventive measures, according to gathered data about workers (optimistic view) and to medium-level workers
parameter condition fluctuations considering ranges is car- (realistic view)?”
ried out. The approach allow to estimate residual time be- Let for studying efficiency a system is decomposed on
fore the next parameter abnormality (i.e. time before first 9 subsystems, for example - see Figure 15. System com-
next coming into “red” range) [35]. The estimated residual ponents are: 1st - a construction of platform; 2nd - an AIS
time Tresid is the solution t0 of equation: on platform for robotics monitoring and control; 3rd - an
underwater communication modem; 4th - a remote con-
R(Tpenetr, t, Tbetw, Tdiag, Treq.) = Radm.(Treq)(3) trolled unmanned underwater robotic vehicle; 5th - a sonar
beacon; 6th - an autonomous unmanned underwater ro-
concerning of unknown parameter t, i.e. Tresid = t0.
botic vehicle; 7th - non-boarding robotic boat - a spray of

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 15

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

the sorbent; 8th - non-boarding robotic boat – a pollution system diagnostics are used once for work shift 8 hours, a
collector; 9th - an unmanned aerial vehicle. Data is moni- mean duration of the system control is about 10 minutes,
tored from different sources and processed by the models mean recovery time of the lost integrity of object equals
described above in section 3. to 1 day. The workers (they may be robotics, skilled me-
Note: Of course every subsystem also may be consid- chanics, technologists, engineers etc.) are supported by
ered as a special complex system. capabilities of an intellectual system allowing estimations
in real time the mean residual time before the next pa-
rameters abnormalities. Formally they operate as parallel
elements with hot reservation. Workers are capable to
revealing signs of a critical situation after their occur-
rence owing to the support of intellectual systems. If all
subsystems are supported by intellectual systems on the
level which is proper to skilled workers (optimistic view),
workers can commit errors on the average not more often
once a year. If all subsystems are supported by intellec-
tual system on the level which is proper to medium-level
workers (realistic view) only one difference is – medi-
um-level workers can commit errors more often in com-
parison with skilled workers, for one element it is equal to
1 time a month instead of once a year.
Further we do the steps 1-4 from Figure 9. Computed
risks to lose system integrity on Figure 15 means the risks
of “failure” for every subsystem which can be detailed
to the level of every separate critical parameter of equip-
ment.
The fragments of built PDFs on Figure 15 show:
(1) if all subsystems are supported by intellectual sys-
tem on the level which is proper to skilled workers (opti-
Figure 15. Subsystems operating for providing safety of mistic view) the risk of “failure” increases from 0.000003
a floating oil and gas platform for a year to 0.0004 for 10 years and to 0.0013 for 20
years. The MTBF equals to 283 years;
The information from monitored data and a time data (2) if all subsystems are supported by intellectual sys-
of enterprises procedures are used as input for using tem on the level which is proper to medium-level workers
models from Table A.2 and performing steps 1-4 (from (realistic view) the risk of “failure” increases from 0.0009
Figure 9) in real time. Here risks to lose system integrity for a year to 0.0844 for 10 years and 0.25 for 20 years.
during given period Tgiven means risks to be at least once The MTBF equals to 24 years. It is 11.4 times less against
in state “Abnormality” within Treq.. The functions of mod- the results for optimistic view.
eling may be performed on special servers (centralized or Such effects (MTBF = 283 years for optimistic view
mapped). If virtual risks are computed for all points Treq. and MTBF = 24 years for realistic view) are owing to
from 0 to ∞, the calculated values form a trajectory of the implemented technology of counteractions to threats.
PDF. The mathematical expectation of this PDF means These are some estimations for example assumptions.
the mean residual time to the next state “Abnormality”. It Please, compare the effects against primary frequency of
defines mean time before failures (MTBF) from this PDF. occurrence of the latent or obvious threats is equal to once
Requirements to IS operation quality should meet admis- a month, mean activation time of threats is about 1 day +
sible levels recommended in Example 2. workers errors.
To answer the question of the example let the next in-
put are formed from data monitored and the time data of 6.5 Example of Solving Problem 2 for a Robot
enterprises procedures. Route Optimization
Let for every system component a frequency of occur-
rence of the latent or obvious threats is equal to once a Applicability of the proposed probabilistic methods and
month, mean activation time of threats is about 1 day. The models to solving problem 2 (of robot actions optimiza-
tion under limitations on admissible risks of “failure”) is

16 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

demonstrated to improve some of the existing capabilities 10 minutes; given prognostic period =2 hours.
of a rescue robot, interconnected with accessory drone,
for route optimization in conditions of uncertainties. Sim- i=1.
ilar problems of specific rescue robot route optimization Step 2 (i=1). Using probabilistic model, a calculation of
from point A (Start) to point E (End) can arise in burning the probability of “failure” is carried out for each variant.
wood, in mountains, in the conditions of a city, and in From the set of variants ABCDE, ABGKLDE, ABGHL-
other situations in conditions of uncertainties. Specific DE the shorter variant ABCDE for which risk is equal to
cases of uncertainties can be connected additionally with 0.034 is chosen (for the route ABGKLDEF risk=0.051, for
complex conditions of environment and necessity of ro- route ABGHLDEF risk=0.067). The relevant data from
botics orientation, localization and mapping that influenc- the drone about the local fire conditions and the weather
es on input for the proposed probabilistic models. on the part BCDE to 8.00 a.m. are taken into account.
On this simplified hypothetical example of moving Step 3 (i=1). The robot overcomes the part AB of route.
some rented values by means of the pilotless car from For the new initial point B the input for modeling every
point A to the final point E of a route (from where the SOS part of possible route are updated in real time for routes
signals are following) we will demonstrate the proposed BCDE, BGKLDE, BGHLDE.
approach to route optimization with acceptable risk of Step 4 (i=1). The robot hasn’t yet arrived at the intend-
“failure” less than 0.1 (i.e. a probability of success should ed point E (i.e. the last part of the route isn’t overcome).
be more than 0.9) under conditions of uncertainties during
the route – see Figure 16. i= i+1=2.

Step 2 (i=2 for variants BCDE, BGKLDE, BGHLDE).

Input for modeling isn’t changed. Risks are the same.
From the route variants BCDE, BGKLDE, BGHLDE the
shorter one BCDEF (with minimal risk) is chosen.
Step 3 (i=2 for variant BCDE). The robot overcomes
the part BC. For the new initial point C the input for mod-
eling every part of possible route are updated in real time:
dense fog in forest thicket on the CD part does not allow
further movement. And additional information for robot
is: the local weather improvements in the next 2 hours are
not expected. Part CD is impassable. The comeback to the
initial point B of the part is being.
Step 2 (i=2 for two remaining variants). From variants
BGKLDE, BGHLDE the shorter one BGKLDE (with
minimal risk 0.051) is chosen.
Step 3 (i=2 for variant BGKLDE). The robot over-
Figure 16. Possible robot route from point A (Start) to comes the part BG. For the new initial point G the input
point E (End)
for modeling every part of possible route are updated in
The next steps from Figure 10 are performed. real time: according drone from 9.00 a.m. on parts GK
Step 1. The complete set of route variants to achieve and KL the imminent fire is detected. The gathered infor-
the goal within given 2 hours, and for each variant – a mation and knowledge are used to clarify the input for
set of components, is defined: ABCDE, ABGKLDE, AB- modeling, namely: the frequency threats in the part GKL
GHLDE. Let data characterizing every part of route for increases from 1.5 to 2.5 times at 10 hours. Using a prob-
each of the variants are gathered from drone-informant, abilistic model for each variant, a recalculation of the risk
processed and prepared for modeling - frequencies of the of failure is carried out. Of the variants GKLDE, GHLDE
occurrences of potential threats are: for ABCDEF = 1 time the variant GHLDE is chosen (risk is equal to 0.067, for
at 10 hours, ABGKLDEF = 1.5 times at 10 hours, ABGH- the route GKLDE risk equals 0.083).
LDE = 2 times at 10 hours (since 08.00 a.m. to 18.00 a.m. Step 4. After overcoming the part GHLDE the robot
what is connected with drone capabilities); mean activa- arrived at the intended point E of route in given time.
tion time of threats = 30 minutes; time between the end of Thus the way ABCBGHLDE is the result of optimi-
diagnostics and the beginning of the next diagnostics = 2 zation before and on the route. The robot purpose was
minutes; diagnostics time = 30 seconds; recovery time = achieved owing to preventive measures which were de-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 17

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

fined by using risk control on the way (with controlled The proposed models include models to estimate AIS
probability of “success” more than 0.9). operation quality and risks prediction for system formal-
ized as “Black box”, algorithm to build new probabilistic
6.6 What about the Possible Pragmatic Effects models for complex system. The practical way to data
from Probabilistic Rationale of Actions for AIS? forming for probabilistic modeling is described.
Author of this article took part in creation of the Com- A suitability of the approach is demonstrated by exam-
plex of supporting technogenic safety on the systems of ples about:
oil&gas transportation and distribution and have been (1) a period of successful system operation by AIS ca-
awarded for it by the Award of the Government of the pabilities;
Russian Federation in the field of a science and tech- (2) acceptable requirements to solve problem 1 for in-
nics. The AIS is a part of the created peripheral posts are formation systems operation;
equipped additionally by means of Complex to feel vibra- (3) solving inverse problem to estimate the mean re-
tion, a fire, the flooding, unauthorized access, hurricane, sidual time before the next parameters abnormalities for a
and also intellectual means of the reaction, capable to coal company;
recognize, identify and predict a development of extreme (4) solving problem 1 by AIS operating for providing
situations – see engineering decisions on Figure 17. safety of a floating oil and gas platform;
(5) solving problem 2 for a robot route optimization;
(6) the possible pragmatic effects from probabilistic
rationale of actions for AIS.
The proposed approach means practically a proactive
commitment to excellence in uncertainty conditions.

Appendix A. The Models to Estimate AIS Opera-

tion Quality
The probabilistic models for the estimation of informa-
tion systems operation quality are presented by the formu-
las (A.1) – (A.14) in Table A.1.

Table A.1 The probabilistic models for the estimation

of information systems operation quality (the proof and
details - see [20-22,24,25,28])
Figure 17. The AIS as a hard-software part to support Models. Input Evaluated measures
technogenic safety on the objects of oil&gas distribution
The model of functions perfor-
The applications of this Complex for 200 objects in mance by a complex system in
several regions of Russia during the period 2009-2014 conditions of unreliability of
have already provided economy about 8,5 Billions of its components.
Input:
Roubles. The economy is reached at the expense of effec- N(t) - is the probability
tive implementation of the functions of risks prediction distribution function (PDF)
and processes optimization [32]. of time between neighboring Probability P rel of providing reliable
failures (TMTBFnk is the mean function performance during given time.
time); W(t) – is the PDF of 
∞ ∞
 ∞

7. Conclusion P =  V (t - t ) dN (t )  dt ∫ td [ N * W (t ) ] ,
repair time (Trep. is the mean rel ∫0  ∫t  0
time); V(t) – is the PDF of (A.1)
The proposed approach for probabilistic rationale of AIS given time if this time is
actions includes description of the proposed models, opti- random value (Treq. is the mean * - is the convolution sign.
time).
mization methods for rationale actions, incremental algo- Note. The next variants are
rithms for solving: used by the software tools [35-
37]
(1) the problem 1 - to rationale a rational variant for de- :
N(t), W(t)  are exponentially
cision-making on the base of monitored data about events distributed (i.e. enough mean
and conditions, and times - TMTBFnk , Trep.), V(t) is
(2) the problem 2 - to rationale a robot actions under determined (i.e. Treq. is const).

limitations on admissible risks of “failures”.

18 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Models. Input Evaluated measures Models. Input Evaluated measures

Probability Ptim..i false of well-timed The model of information
processing of i-type calls during the gathering.
required term Input:
γ i2Т req.i Т full.i C(t) is the PDF of time between
e dt
γ i −1 − t
∫ t essential changes of object
The models complex of calls P tim.i = P ( t full .i ≤ T req.i ) = , (A.2)
0
∞
states, xi – is the mean time;
e dt
γ i −1 − t
processing for the different ∫t
0 B(t) is the PDF of time for
dispatcher technologies. information gathering and
Т full.i Probability Pact of information actuality
Input for M/G/1/∞: γi = ,
Т full.i 2 − T full.i
2 preparing, transfer and entering
λi – frequency of the i–th type on the moment of its use:
into system;
calls for processing; Relative portion of all well-timed (1) for the mode D1 when information is
Q(t) is the PDF of time interval
βi – mean processing time of processed calls – S and relative portion gathered in order “immediately after an
between information updating,
the i–th type calls (without of well-timed processed calls of those essential object state change:
q is the mean time (only for
queue). types for which the customer require- mode D2); 1∞
Note. The software tools [35- ments are met – C:
= P act ∫ B(t)[1−C(t)]dt, (A.4)
37] the mode D1 of gathering: ξi 0
allow to estimate and to information is gathered in order (2) for the mode D when information
compare effectiveness of the I
“immediately after an essential is gathered without any dependencies
2

next dispatcher technologies ∑ λ i P tim.i

S= i =1
, object state change; on changes of objects current states
for modeling by M/G/1/∞: I
the mode D2 of gathering:
- technology 1 for aprior- ∑ λi (including regulated information gather-
i =1 information is gathered without ing)
ity calls processing: in a I any dependencies on changes of
consecutive order for sin- ∑ λ i P tim.i  Ind (α1) + Ind (α 2)  1  ∞ ∞

=
objects current states (including P q ∫   [1 −
act
Q(t)][1 − ∫ C(t + τ)dB(τ)]dt, (A.5)
gle-tasking processing mode; С= i =1
I
, regulated information gather-
i
0 0 

in a time-sharing order for ∑ λi ing).

i =1
multitasking processing mode; Note. The next variants are used
- priority technologies of 0 , if α = true
consecutive calls processing
Ind ( α ) =  , by the software tools [35-37]:
1, if α = false B(t), C(t) are exponentially
2-5: technology 2 for calls distributed, Q(t) is V(t) is
processing with relative priori- a1=(there is used criterion 1 and
Tfull i ≤ Treq.i); determined or exponentially
ties in the order “first in - first distributed.
out” (FIFO); technology 3 for a2=(there is used criterion 2 and
Ptim i ≥ Preq.i). The model of information Probability Pafter of errors absence after
calls processing with absolute
Criterion 1 is if there is required analysis. checking (probability Pafter of correct analysis
priorities in the order FIFO;
Tfull i ≤ Treq.i to be i-type calls processed Input: results obtaining):
technology 4 for batch calls
in time, criterion 2 is if there is required Treq. - assigned term for analy- Variant 1. An assigned term for analysis is no
processing (with relative pri-
Ptim i=P(Tfull i ≤ Treq.i) ≥ Padm.i sis; less than the real analysis time (Treal≤Treq.) and
orities and in the order FIFO
to be i-type calls processed in time, Padm.i N(t) is the PDF of time be- the content of analyzed information is such
inside a batch) [7]; technology
- is admissible level for well-timed tween type I analysis errors, h-1 small that it is required only one continuous
5 is a combination of technol-
processing of i-type calls during the is the mean time; analyst’s work period (Treal ≤ Tcont.).
ogies 2, 3, 4 [8].
required term Preq.i M(t) is the PDF of time be- P after (1)(V , m ,n , n,T MTBF ,T cont .,T req.) =
The formulas for mean response time tween the neighboring errors = 1 − Nˆ (V / n )  ⋅
in checked information; А(t)   (A.6)
Tfull of i-type calls and for 2nd moment V /n
is the PDF of analyzed type II  dA(t )[1 − M (V / n − t )] + dA(t ) 
 ∞
Tfull.i2 - see [20-22,24,25,28] . 
∫ ∫
errors, TMTBF is the mean time;  0 V /n 
m is the relative fraction of Variant 2. An assigned term for analysis is
errors in information content no less than the real analysis time (i.e. Treal ≤
The model of entering into (destined for problems of Treq.). But the content of analyzed informa-
system current data concern- checking) or the relative frac- tion is comparatively large, i.e. Treal>Tcont..
ing new objects of application tion of information essential Рafter(2) = {Рafter(1) (Vpart(2) N
, m, n, h, ТMTBF, Тcont.,
domain. for analysis (destined for tpart(2))} , (A.7)
Input: N=V/(n Тcont.), Vpart(2)=V/N, tpart(2) = Тreq./N.
problems of analysis);
qm - the probability that m new Variant 3. An assigned term for analysis is
Treal= V/n - is the real time for
less than the real analysis time (Treal>Treq.) and
objects appear in random mo- complete information analysis; the content of analyzed information is such
ment, intervals between these V – is a content of analyzed small that it is required only one continuous
Probability Pcomp. that system contains
moments are exponentially information; analyst’s work period (Treal ≤ Tcont.).
information about states of all real
distributed with parameter l. n - is an analyzed speed; Рafter(3) = (Vpart(3)/V)× Рafter(1) (Vpart(3) , m, n, h,
object and coincides
m
Φ ( z ) = ∑ qm z -is productive Tcont. - is time of continuous ТMTBF, Тcont., Тreq.)+
m >0  ∞ 
(generating) function; P comp.
= exp −l ∫ 1 − Φ ( Β(t ) )  dt  , (A.3)
 
analyst’s work. +[(V-Vpart(3))/V]× Рwithout,(A.8)
0 Note. The next variants are where Vpart(3) =nТreq.. , Рwithout = e -m (V-V part(3) ).
В(t) – is the PDF of time for
used by the software tools [35- Variant 4. An assigned term for analysis is no
new information revealing 37]
: less than the real analysis time (i.e. Treal>Treq.),
and preparing, transfer and but the content of analyzed information is
Treq. - is an assigned term
entering into data base. comparatively large, i.e. Treal>Tcont..
(deadline) for analysis; V, ν,
Note. The next variants are  [V ( ) / V ] × P ( ) (V ( ) , m , v,h , Т ,Т ,Т ) +
Tcont. and Treq. are assigned as part 4 after 1 part 4 MTBF cont . req .

used by the software tools [35- 

 + [(V − V ( ) ) / V ] × e (
( )) − m V −V part 4
,if Т ≤ Т ;
37] deterministic values; 
Р [V / V ] × {Р (V
=
part 4 req . cont .

: Ф(z)=z; B(t) is exponen- 

after
( ) , m , v,h , Т , Т , t ( ) )} N

N(t)=1–exp(-t×η);
( ) ()
part 4 after 1 part 4.2 MTBF cont . part 4.2

tially distributed.  +[(V − V ( ) ) / V ] × e (
( )) − m V −V part 4
,if Т > Т ,
M(t) =1 –exp(-t×μ×ν);
part 4 req . cont .

(A.9)
А(t)= 1–exp(-t/ TMTBF).

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 19

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Models. Input Evaluated measures Appendix B. The Models to Predict Risks for
The models complex of an autho- “Black box”
rized access to system resources
during objective period. The proposed models allow to estimate preventive risks
Input (for estimation of confi-
dentiality): for being control in real time. The approach for modeling
М  is the conditional number of a is based on algorithmic building new probabilistic models
barriers against an unauthorized Probability Pvalue of system protection
access; against unauthorized access during – see Table B.1.
Fm(t) is the PDF of time between objective period The probabilistic models for the estimation of preven-
M
changes of the m-th barrier P value = 1 − ∏ Pover .m , (A.10)
parameters; m =1 tive risks for being control in real time is presented by the
Um(t) is the PDF of parameters where Pover m – is the risk of overcom- formulas (B.1) – (B.6) in Table B.1.
decoding time of the m-th secu- ing the m-th barrier by violator during
rity system barrier, um – the mean objective period when resources value
time of a barrier overcoming; is high, Table B.1 – The models to predict risks for “Black box”
H(t) – is the PDF of objective 1 ∞ ∞ t (the proof and details - see [24, 25, 28, 30, 44-45])
period, when= resources value is P over f 0∫ dt ∫t d F m( t )0∫ d U m( θ )[1 − H ( θ )] ,
m
high. Evaluated
Models, methods Formulas
Note. The next variants are used measures
by the software tools [35-37]: The model for technolo-
Um(t)is exponentially distributed; gy 1 (“Black box”).
Fm(t) and H(t) are determined or Note. Technology 1
exponentially distributed. (without monitoring
The models complex of danger- between diagnostics)
ous influences on a protected Probability Pinfl of faultless (correct) is based on periodical
system. operation during given time: diagnostics of system
Input: variant 1 – the assigned period Treq is less integrity, that are carried
Wpenetr(t) – is the PDF of time be- than established period between neigh-
out to detect danger
tween neighboring influences for boring diagnostics (Treq < Tbetw.+Tdiag)
penetrating a danger source, for sources penetration into
-st Pinfl.(1)(Treq) = 1 - Wpenetr* Wactiv(Treq), a system or conse- R=1-P considering consequences.
Wpenetr.(t)=1-e , s - is the frequen-
(A.11) quences of negative Variant 1 – the given prognostic
cy of influences for penetrating;
Wactiv(t) – is the PDF of activa- influences. The lost period Treq is less than established
tion time of a penetrated danger variant 2 – the assigned period Treq is system integrity can be period between neighboring diag-
source, for Wactiv(t)=1-e-t/b, b – is more than or equals to established period detect only as a result nostics
the mean activation time; between neighboring diagnostics (Treq ≥ of diagnostics, after (Treq < Tbetw.+Tdiag):
Treq – is the required period of Tbetw.+Tdiag): which system recovery P(1) (Treq) = 1 - Wpenetr* Wactiv(Treq).
permanent secure system opera- N (Т betw. + T diag . ) is started. Dangerous (B.1)
⋅ P inf l .(1) (Т betw. + T diag . ) +
N
=P inf l .(2)
tion; T req. influence on system is Risk Variant 2 – the assigned period Treq
Tbetw. – is the time between the T req. − N (Т betw. + T diag . ) acted step-by step: at to lose is more than or equals to estab-
end of diagnostic and the begin- + P inf l . (Т betw. + T diag . ) ,
T req. first a danger source system lished period between neighboring
ning of the next diagnostic, Tdiag penetrates into a system integrity diagnostics
– is the diagnostic time. (A.12)
and then after its activa- (R). (Treq ≥ Tbetw.+Tdiag):
Note. The next variants are used where N=[ Тreq./(Тbetw.+ Тdiag.)] – is the
integer part. tion begins to influence. Proba- measure a)
by the software tools [35-37]:
System integrity can’t bility of P(2) (Treq) = N((Tbetw +Tdiag)/Treq) P(1)
Wpenetr(t) and Um(t) are exponen- N
tially distributed. be lost before a pene- providing (Tbetw +Tdiag) + (Trmn/Treq) P(1)(Trmn),
The models complex of an
trated danger source is system (B.2)
authorized access to system activated. A danger is integrity where N=[Tgiven/(Тbetw.+Тdiag.)] is the
resources. considered to be real- (P) integer part, Trmn = Tgiven -N(Tbetw
Input (for estimation of confi- ized only after a danger +Tdiag);
dentiality): source has influenced measure b)
М  is the conditional number on a system. P(2) (Treq) = P(1)N(Tbetw +Tdiag) P(1)
of a barriers against an Input: (Trmn),
unauthorized access; Probability Pprot of system protection Wpenetr(t) – is the PDF of (B.3),
Fm(t) is the PDF of time between against unauthorized access: time between neigh- where the probability of success
changes of the m-th barrier M
P prot = 1 − ∏ Pover m , (A.13) boring influences for within the given time P(1)(Treq) is
parameters; m =1
penetrating a danger defined by (B.1).
Um(t) is the PDF of parame- where Pover m – is the probability of over- source;
ters decoding time of the m-th coming the m-th barrier by violator,
1 ∞
Wactiv(t) – is the PDF of
security system barrier, um – the =P over ∫ [1 − F m(t )]U m(t )dt. (A.14) activation time of a pen-
mean time of a barrier overcom- fm0
m

etrated danger source;

ing.
Note. The next variants are used
Tbetw. – is the time be-
by the software tools [35-37]: tween the end of diag-
Um(t)is exponentially distribut- nostic and the beginning
ed; of the next diagnostic,
Fm(t) is determined or exponen- Tdiag – is the diagnostic
tially distributed. time.
Note: The final clear analytical formulas are received by Lebesque-inte-
gration of (A.1). (A.3) – (A.6), (A.10), (A.14).

20 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Models, methods
Evaluated
Formulas Software, 2017, 2: 393.
measures
[8] R. Valencia, J. Andrade-Cetto, Mapping, Planning
The model for technolo-
gy 2 (“Black box”). and Exploration with Pose SLAM, Springer, 2018:
Note. Technology 2, 124.
unlike the previous one,
[9] The DARPA Robotics Challenge Finals: Humanoid
implies that operators
alternating each other Robots To The Rescue, Springer, 2018: 692.
trace system integrity [10] G. Antonelli, Underwater Robots, Springer, 2018:
between diagnostics 374.
(operator may be a
man or special device [11] Cognitive Reasoning for Compliant Robot Manipula-
or their combination). tion, Springer, 2019: 190.
In case of detecting a [12] G. Venture, J.-P. Laumond, B. Watier Biomechanics
danger source an op-
erator recovers system R=1-P considering consequences. of Anthropomorphic Systems, Springer, 2019: 304.
integrity. The ways of Variant 1 - (Treq < Tbetw.+Tdiag): [13] A. Santamaria-Navarro, J. Solà, J. Andrade-Cetto,
integrity recovering are T req T req

1 − ∫ dA(t ) ∫ d Ω penetr * Ω act (θ ). (B.4) Visual Guidance of Unmanned Aerial Manipulators,

P (1)(T req ) =
analogous to the ways Risk
Springer, 2019: 144.
0 t

of technology 1. Fault- to lose Variant 2 – (Treq ≥ Tbetw.+Tdiag):

less operator’s actions system measure a) [14] S. Tadokoro, Disaster Robotics, Springer, 2019: 532.
provide a neutralization integrity P(2) (Treq) = N((Tbetw +Tdiag)/Treq) P(1) [15] Feller W. An Introduction to Probability Theory and
N
of a danger source try- (R). (Tbetw +Tdiag) + (Trmn/Treq) P(1)(Trmn),
ing to penetrate into a Proba- (B.5) Its Applications. Vol. II, Willy, 1971.
system. When operators bility of [16] Martin J. System Analysis for Data Transmission.
alternate a complex providing measure b) V. II, IBM System Research Institute. Prentice Hall,
diagnostic is held. A system P(2) (Treq) = P(1)N(Tbetw +Tdiag) P(1)
penetration of a danger integrity (Trmn),(B.6), Inc., Englewood Cliffs, New Jersey, 1972.
source is possible only (P) [17] Gnedenko B.V. et al. Priority queueing systems,
if an operator makes an where N is the same and the prob- МSU, Moscow, 1973: 448.
error but a dangerous ability of success within the given
influence occurs if the time P(1)(Treq) is defined by (B.4) [18] Kleinrock L. Queueing systems, V.2: Computer ap-
danger is activated be- plications, John Wiley & Sons, New York, 1976.
fore the next diagnostic. [19] Matweev V.F. & Ushakov V.G. Queuing systems.
Otherwise the source
will be detected and MSU, Moscow, 1984: 242.
neutralized during the [20] Kostogryzov A.I. Conditions for Efficient Batch Job
next diagnostic. Processing of Customers in Priority-Driven Com-
Input:
Additionally to Input puting Systems Where the Queueing Time Is Con-
for technology 1: strauned, «Avtomatika i telemehanika». 1987, 12:
A(t) - is the PDF of P.158-164.
time from the last finish
of diagnostic time up to [21] Kostogryzov A.I. Study of the Efficiency of Combi-
the first operator error nations of Different Disciplines of the Priority Ser-
Note: The final clear analytical formula (B.4) is received by its Leb- vice of Calls in the Computer Systems, «Kibernetika
esque-integration i sistemny analiz». 1992, 1: 128-137.
[22] Kostogryzov, A.I., Petuhov, A.V. & Scherbina, A.M..
References Foundations of evaluation, providing and increasing
output information quality for automatized system.
[1] Alan Turing, Computing Machinery and Intelligence,
Moscow: “Armament. Policy. Conversion”, 1994.
Mind, vol. LIX, 1950, 236: 433—460.
[23] Gnedenko B.V., Korolev V. Yu., Random Summa-
[2] Experimental Robotics. Springer, 2016: 913.
tion: Limit Theorems and Applications. – Boca Ra-
[3] A. Ajoudani, Transferring Human Impedance Regu-
ton: CRC Press, 1996.
lation Skills to Robots, Springer, 2016: 180.
[24] Kostogryzov, A.I. Software Tools Complex for Eval-
[4] Geometric and Numerical Foundations of Move-
uation of Information Systems Operation Quality
ments. Springer, 2017: 417.
(CEISOQ). Proceedings of the 34-th Annual Event of
[5] Robotics Research. Springer, 2017: 646.
the Government Electronics and Information Asso-
[6] Cybernetics Approaches in Intelligent Systems. Com-
ciation (GEIA), Engineering and Technical Manage-
putational Methods in Systems and Software, vol. 1,
ment Symposium, USA, Dallas, 2000: 63-70.
Springer, 2017: 405.
[25] Kostogryzov A., Nistratov G.: Standardization, math-
[7] Applied Computational Intelligence and Mathemati-
ematical modelling, rational management and certi-
cal Methods. Computational Methods in Systems and
fication in the field of system and software engineer-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 21

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

ing. Armament.Policy.Conversion, Moscow, 2004. (ICSRS), Milan, Italy, 2017: 368-373. 

[26] Zio En.: An Introduction to the Basics of Reliabil- [36] Kostogryzov A., Stepanov P., Nistratov A., Nistratov
ity and Risk Analysis, World Scientific Publishing G., Klimov S., Grigoriev L.: The method of ratio-
Co.Pte.Ltd. , 2006. nal dispatching a sequence of heterogeneous repair
[27] Korolev V.Yu., Sokolov I.A., Mathematical Models works. Energetica, 2017, 63(4): 154-162.
of Non-Homogeneous Flows of Extremal Events. -- [37] Kostogryzov A., Stepanov P., Nistratov A., Atakish-
Moscow: TORUS-PRESS, 2008. chev O.: About Probabilistic Risks Analysis During
[28] Kostogryzov A.I., Stepanov P.V.: Innovative man- Longtime Grain Storage. In: Proceedings of the 2nd
agement of quality and risks in systems life cycle Internationale Conference on the Social Science and
(modern standards and ideas of system engineering, Teaching Research (ACSS-SSTR), Volume 18 of
mathematical models, methods, techniques and soft- Advances in Social and Behavioral Science. Edited
ware tools complexes for system analysis, including by Harry Zhang. Singapore Management and Sports
modelling through Internet, 100 examples with an Science Institute, PTE.Ltd., 2017: 3-8 .
explanation of logic of the reached results, useful [38] Kostogryzov A., Stepanov P., Grigoriev L., Atakish-
practical recommendations). Moscow, Armament. chev O., Nistratov A., Nistratov G.: Improvement of
Policy.Conversion, Moscow, 2008. Existing Risks Control Concept for Complex Sys-
[29] Kolowrocki K., Soszynska-Budny J.: Reliability and tems by the Automatic Combination and Generation
Safety of Complex Technical Systems and Processes, of Probabilistic Models and Forming the Storehouse
Springer-Verlag London Ltd., 2011. of Risks Predictions Knowledge. In: Proceedings of
[30] Kostogryzov A., Nistratov G. and Nistratov A.: Some the 2nd International Conference on Applied Mathe-
Applicable Methods to Analyze and Optimize Sys- matics, Simulation and Modelling (AMSM), Phuket,
tem Processes in Quality Management. Total Quality Thailand. DEStech Publications, Inc., 2017: 279-283.
Management and Six Sigma, InTech, 2012: 127-196. [39] Kostogryzov A., Atakishchev O., Stepanov P., Nis-
[31] Grigoriev L., Guseinov Ch., Kershenbaum V., tratov A., Nistratov G., Grigoriev L.: Probabilistic
Kostogryzov A. The methodological approach, based modelling processes of mutual monitoring operators
on the risks analysis and optimization, to research actions for transport systems. In: Proceedings of
variants for developing hydrocarbon deposits of Arc- the 4th International Conference on Transportation
tic regions. Journal of Polish Safety and Reliability Information and Safety (ICTIS), Canada, Banff,
Association. Summer Safety and Reliability Semi- 2017: 865-871.
nars, 2014, 5(1-2): 71-78. [40] Kostogryzov A., Panov V., Stepanov P., Grigoriev L.,
[32] Akimov V., Kostogryzov A., Mahutov N. at al. Se- Nistratov A., Nistratov G.: Optimization of sequence
curity of Russia. Legal, Social&Economic and Sci- of performing heterogeneous repair work for trans-
entific&Engineering Aspects. The Scientific Founda- port systems by criteria of timeliness. In: Proceedings
tions of Technogenic Safety. Under the editorship of of the 4th International Conference on Transporta-
Mahutov N.A. Znanie, Moscow, 2015. tion Information and Safety (ICTIS), Canada, Banff,
[33] Kostogryzov A., Nistratov A., Zubarev I., Stepanov P., 2017: 872-876.
Grigoriev L. About accuracy of risks prediction and [41] Kostogryzov A., Nistratov A., Nistratov G., Atakish-
importance of increasing adequacy of used adequacy chev O., Golovin S., Grigoriev L.: The probabilistic
of used probabilistic models. Journal of Polish Safe- analysis of the possibilities to keep “organism integ-
ty and Reliability Association. Summer Safety and rity” by continuous monitoring. In: Proceedings of
Reliability Seminars, 2015, 6(2): 71-80. the International Conference on Mathematics, Mod-
[34] Eid, M. and Rosato, V. Critical Infrastructure Disrup- elling, Simulation and Algorithms (MMSA), Cheng-
tion Scenarios Analyses via Simulation. Managing du, China. Atlantis Press, Advances in Intelligent
the Complexity of Critical Infrastructures. A Model- Systems Research, 2018, 159: 432-435. 
ling and Simulation Approach, SpringerOpen, 2016: [42] Kostogryzov A., Grigoriev L., Golovin S., Nistratov
43-62. A., Nistratov G., Klimov S.: Probabilistic Modeling
[35] Artemyev V., Kostogryzov A., Rudenko Ju., Kur- of Robotic and Automated Systems Operating in
patov O., Nistratov G., Nistratov A.: Probabilistic Cosmic Space. In: Proceedings of the International
methods of estimating the mean residual time before Conference on Communication, Network and Arti-
the next parameters abnormalities for monitored ficial Intelligence (CNAI), Beijing, China. DEStech
critical systems. In: Proceedings of the 2nd Interna- Publications, Inc., 2018: 298-303.
tional Conference on System Reliability and Safety [43] Kostogryzov A., Grigoriev L., Kanygin P., Golovin

22 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

S., Nistratov A., Nistratov G.: The Experience of №2000610272.

Probabilistic Modeling and Optimization of a Cen- [47] Kostogryzov A.I., Nistratov G.A., Nistratova E.N.,
tralized Heat Supply System Which is an Object for Nistratov A.A. Mathematical modeling of system life
Modernization. International Conference on Physics, cycle processes – “know-how”, registered by Ro-
Computing and Mathematical Modeling (PCMM), spatent №2004610858.
Shanghai, DEStech Publications, Inc., 2018: 93-97. [48] Kostogryzov A.I., Nistratov G.A., Nistratova E.N.,
[44] Artemyev V., Rudenko Ju., Nistratov G.: Probabi- Nistratov A.A. Complex for evaluating quality
listic modeling in system engineering. Probabilistic of production processes, registered by Rospatent
methods and technologies of risks prediction and №2010614145.
rationale of preventive measures by using “smart [49] Kostogryzov A.I., Nistratov G.A., Nistratov A.A.
systems”. Applications to coal branch for increasing Remote analytical support of informing about the
Industrial safety of enterprises. Edited by Andrey probabilistic and time measures of operating system
Kostogryzov, IntechOpen, 2018: 23-51. and its elements for risk-based approach, registered
[45] Kershenbaum V., Grigoriev L., Kanygin P., Nistratov by Rospatent №2018617949.
A.: Probabilistic modeling in system engineering. [50] Kostogryzov A.I., Nistratov G.A., Nistratov A.A.
Probabilistic modeling processes for oil and gas sys- Remote rationale of requirements to means and con-
tems. Edited by Andrey Kostogryzov, IntechOpen, ditions for providing “smart” systems operation qual-
2018: 55-79. ity, registered by Rospatent №2018618572.
[46] Kostogryzov A.I., Bezkorovainy M.M., Lvov V.M., [51] Kostogryzov A.I., Nistratov G.A., Nistratov A.A.
Nistratova E.N., Bezkorovainaya I.V. Complex for Remote probabilistic prediction of informatized
Evaluation of Information Systems Operation Quali- systems operation quality, registered by Rospatent
ty – “know-how” (CEISOQ), registered by Rospatent №2018618686.

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1195 23

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Artificial Intelligence Advances

https://ojs.bilpublishing.com/index.php/aia

ARTICLE
School Debit Transaction Using Fingerprint Recognition System
Wai Kit Wong*　Thu Soe Min Shi Enn Chong
Faculty of Engineering and Technology, Multimedia University (MMU), Jalan Ayer Keroh Lama, Melaka, 75450, Ma-
laysia

ARTICLE INFO ABSTRACT

Article history This paper proposed a fingerprint based school debit transaction system us-
Received: 11 September 2019 ing minutiae matching biometric technology. This biometric cashless trans-
action system intensely shortens the luncheon line traffic and labour force
Accepted: 22 October 2019 compared to conventional cash payment system. Furthermore, contrast with
Published Online: 30 October 2019 card cashless transaction system, fingerprint cashless transaction system
with benefit that user need not carry additional identification object and re-
Keywords: member lengthy password. The implementation of this cashless transaction
Fingerprint recognition system provides a more organize, reliable and efficient way to operate the
school debit transaction system.
Biometric Authentication
Image processing

　
1. Introduction friends, and sometimes the card can be scanned twice,

N
lead to a double charged for a single transaction. Another
owadays, parents need not to give cash directly concerning issue is if students forgot their identification
to their primary/secondary schools' children. number and PIN number of their card upon transaction, it
Some schools already practice to use their own might lead to a longer queuing time for other students, as
debit card system, whereby students or their parents just primary and secondary students normally have 15-20 min-
need to bank in/ deposit the money to the school treasury utes’ break, the heavy traffic might lead to a waste of food
department, the school treasury department will issue and time. Therefore, a higher level of security and reliable
a debit card to students for purchasing food in canteen, system should be implemented to replace the conventional
stationaries, fees etc. This debit card system in favour by debit card transaction system, in order to create a conve-
parents because they can monitor their children better due nient and safer environment for the children. Thus, bio-
to the reason no cash for students to be get lost/stolen or metric based debit transaction system is being proposed
purchasing outside drugs, tit-bit or unhealthy entertain- by researchers to overcome the above issues [1,2].
ment. Biometric authentication is a method of recognizing
However, parents still have concerns, such as debit a human being according to the physiological measure-
cards being stolen or misplaced by their children. There ments or physically features and traits [3]. The human
were a lot of wasteful resources for the system maintain- physical characteristics such as fingerprints, face, hand
er, as a new card had to be made for them who losing it. geometry, voice and iris are known as biometrics. Bio-
Besides that, students would also share their cards among metric technologies are becoming the foundation of an

*Corresponding Author:
Wai Kit Wong,
Faculty of Engineering and Technology, Multimedia University (MMU), Jalan Ayer Keroh Lama, Melaka, 75450, Malaysia;
Email: [email protected]

24 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

extensive array of highly secure identification and person- correct person and section IV reports some experimental
al verification solutions. Since biometric identifiers are results. Finally, in section V, some conclusion and envi-
associated permanently with the user, it is more reliable sion future developments is drawn.
than the conventional authentication methods such as PIN
and password. Thus, biometric based authentication can 2. Fingerprint Recognition School Debit
provide extra confidential in transactions by securing the Transaction System Design
personal information and privacy data.
This section outlines the system design for the fingerprint
In fact, there are many methods for biometric authen-
recognition school debit transaction system. It consists
tication. The most commonly used methods for biomet-
of the hardware architecture and software modules of the
ric authentication are iris scanning [4], hand scanning [5],
system as discuss in section 2.1 and section 2.2 below.
fingerprint recognition [6], face recognition [7], and voice
recognition [8]. The biometric information will never ever 2.1 Hardware Architecture
match with another individual because everyone has their
own unique biometric features. Researchers [9,10,11] had Figure 1 shows the hardware architecture for the proposed
conducted numerous analysis and comparison among dif- fingerprint recognition school debit transaction system.
ferent types of biometric recognition methods. The com- The overall system consists of fingerprint collection mod-
parison of biometric methods is mainly based on charac- ule and PC/tablets placed at each shop/stall, all linked to
teristics such as universality, uniqueness, performance, a centre hub (server) with students’ information database
permanence, and measurability. for info matching and business transaction.
Based on the researchers’ results [9,10,11], fingerprint Stall 1
Functional Push Buttons
recognition has the highest market share among the tech- LCD and LED Transaction
System
Display
nologies of biometric security system in the market. The R305 Optical
Fingerprint
Microcontroller
Arduino Uno PC/
preference of fingerprint recognition in the current market Module Board Fingerprint
Collection Module
Tablets

Server
is mostly due to its high accuracy, performance and sta-
Stall 2 TransactionSystem
bility. Besides that, fingerprint recognition has a moderate
pricing compare to iris scan which required a high capital
…

cost and level of skills to operate and maintain. Moreover, Stall n Transaction System

it also has a relatively low percentage of False Acceptance

Rate (FAR) and False Reject Rate (FRR) [12] which make Figure 1. Fingerprint Recognition School Debit Transac-
tion System
it more reliable to use either in police or industrial area.
Thus, fingerprint biometric recognition is widely accept-
able and preferable compare to the other biometric securi- 2.1.1 R305 Optical Fingerprint Module
ty systems. It is an optical biometric fingerprint sensor with a TTL
In this paper, a fingerprint recognition system will be UART interface for direct connection to the PC. The user
proposed for replacing the debit card in business trans- is allowed to store fingerprint data and verify identity with
action system. This can provide an even more secure the module. The module consists of two main function
studying environment for students, since debit card can be which are enrolment and matching of fingerprint. How-
stolen and use within the school too. With the fingerprint ever, in this project, only the enrolment of fingerprint
recognition system, the student's fingerprint itself is the function is in used. The build in fingerprint matching
debit card. Nobody can steal it. Fingerprint scanning de- function of R305 is slow when datasets achieving 1000 or
vices in hardware and recognition algorithm in software more. There is an advance algorithm, namely the Hough
will be developed to verify and identify the identity of a Transform Minutiae Pairing (HTMP) fingerprint matching
student’s debit account through fingerprint scan. This will technique is proposed in the computing side (PC/tablet) to
enable the student to do transaction by fingerprint scan in run the fingerprint matching task. Besides that, the optical
the school and does not need to go through the hassle of fingerprint module will capture the image of a finger by
showing the debit card or paying by cash. utilizing the light ray. However, the image of the finger-
The paper is organized in the following way: Section print can be affected by external environment such as dirt,
II will be briefly comments on the fingerprint recognition wet, quality of skin and humidity. It will be more easy to
school debit transaction system. Section III presents the study and trouble shoot the fingerprint mismatch in the
proposed Hough Transform Minutiae Pairing Fingerprint proposed external matching algorithm (HTMP), rather
Matching (HTMP) technique applies in identifying the than using the R305 build in matching algorithm.

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 25

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Figure 2. R305-optical fingerprint module

Figure 4. Push buttons for Enrolment, Verification and

2.1.2 Arduino UNO
Delete
The R305 fingerprint module is interfacing by Arduino
UNO board. There are several types of Arduino board 2.1.3 Functional Push Buttons
such as Uno, Mega, Nano, and etc. However, Arduino
Uno which based on ATmega328 microcontroller was se- There will be three push buttons as shown in Figure 4,
lected due to its inexpensive cost, cross platform, simple placed on the fingerprint collection module for three im-
and clear programming environment, with open source portant functions, namely: Enrolment, Verification and
and extensive software, together with large support of Delete. The three buttons are connected each with a 1kΩ
community. The proposed Arduino UNO board is supplied pull up resistor to the Arduino UNO board.
with multiple sets of digital and analog input/output pins.
2.1.4 LED and LCD Display
The operating voltage of the Arduino Uno will be 9Vby
connecting to the power supply of PC/tablet. The four dig- One red LED, one green LED and three yellow LEDs are
ital pins (TXD, RXD, VIN and GND) of the R305 finger- used as indicators on the fingerprint collection module.
print module are connected to the Arduino UNO board, as The three yellow LEDs are connected to pin number 12,
shown in Figure 3 below. The Transmit Data (TXD) pin is 10 and 4, as shown in Figure 5, indicate the mode selec-
connected to pin D3 of Arduino UNO board for data out- tion of Enrolment task, Verification task and Delete task
put, the Receive Data (RXD) pin is connected to pin D2 respectively. The red LED is connected to digital pin 13 to
of Arduino UNO board for data input. The Input Voltage indicate task failure, the green LED is connected to digital
(VIN) pin is connected to pin 5V of Arduino UNO board pin 11 to indicate task successful. The (I2C) LCD display
for a constant 5V supply to fingerprint module, and the is used to display the short messages to the shopkeepers
Ground (GND) pin is connected to pin GND of Arduino about the task execution of the fingerprint collection mod-
UNO board for ground. ule, as shown in Figure 6 below.

Figure 3. Fingerprint module connected to Arduino UNO Figure 5. Circuit Diagram of the LED Indicators and
board Arduino UNO Board

26 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

2.2.1 Activity Diagrams

Activity diagrams are used to demonstrate the processes
carried out in the system. Generally, there are five ac-
tivity diagrams used in the proposed fingerprint based
school debit transaction system: (1) Activity diagram
Figure 6. A working (16*2) LCD Display of fingerprint enrolment, (2) Activity diagram for fin-
gerprint verification, (3) Activity diagram for top-up
2.1.5 PC/Tablet process, (4) Activity diagram of transaction process (5)
Activity diagram for checking purchase history.
The PC, laptop or tablet will read in fingerprint signals trans- The fingerprint enrolment needs at least two identi-
ferred from the Arduino UNO and perform the fingerprint cal fingerprint inputs for the sake of wipe out potential
image processing for users’ identity matching. It comes with errors during the feature extraction process. The image
Graphic User Interface to registration page, selling products processor (PC/laptop/tablet) will match the fingerprints
menus, purchase histories etc. Other than that, it can display to determine whether the two inputs are from the same
the product database, and allow other operations such as user’s finger. If yes, the template will be saved. Else, if
viewing employees clock in-out times, sales report, etc. the second fingerprint input is not match with the initial
one, the system might not generate the template. The
2.1.6 Server
complete fingerprint enrolment process is shown in Fig-
The server work as a centre hub which manages the access ure 8.
of multiple stall stations to a centralized resource (users’ For fingerprint verification process, the quality of the
personal information, databases of selling products name fingerprint will be enhancing by some pre-processing
lists, price, credit balance, top-up and purchase histories steps like classifying fingerprint image into the 8 ma-
etc.) jor pattern categories, identify the essence point in the
fingerprint image and crop the core region concentrat-
2.2 Software Implementation ed in the essence point, so that the important features
like minutiaes, ridges and bifurcation points can be
The fingerprint based school debit transaction system begin
extracted for future matching. The fingerprint matching
with a graphical user interface that authorize the users to
algorithm will further discuss in Section 3. If the input
register their personal information into the database 1. After
fingerprint matches with the template stored in the da-
that, user is required to enrol their fingerprint into the sys-
tabase, the LCD screen will display the user identity
tem for future verification purpose. When the user wishes
for shopkeeper to further verify with user. However,
to proceed for credit transaction (top-up credit or purchase
if the input fingerprint does not match any of the tem-
items/services), a verification of fingerprint is needed.
plate stored in the database, the LCD will not display
There are only two possibilities for the verification results:
any user identity, and perhaps will show message “Not
Successful or Fail. Only the user with a valid fingerprint
found in database”. The whole fingerprint verification
stored in the database can proceed to further operations
process is shown in Figure 9.
such as top-up credit or purchase items/services. After the
If the input fingerprint is successfully matched with
transactions being performed, the system will update the
the template stored in the fingerprint database, the user
latest account balance into the database. The overall soft-
can carry out functions like topping up their debit ac-
ware architecture for fingerprint based school debit transac-
count, paying for their meals, groceries, stationaries,
tion system is shown in Figure 7 below.
services etc. and checking their top-up and purchase
history. The whole process for the functions above is
graphically illustrated in Figure 10, Figure 11 and Figure
12 respectively.

Figure 7. Software Architecture for Fingerprint based

school debit transaction system

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 27

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Figure 8. Activity diagram of fingerprint enrolment

Figure 10. Activity diagram for top-up process

Figure 9. Activity diagram for fingerprint verification Figure 11. Activity diagram of transaction process

28 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

(c) (d)

Figure 13. Microsoft Visual Studio 2017 (a) Blank tem-

plate of GUI layout form (b) Toolbox (c) Properties box (d)
Server Explorer
The database of the proposed fingerprint based school
debit transaction system is built by Microsoft SQL Server
Management Studio. The database of this project is to
store and retrieving data such as personal information and
the account balance. There will be total four tables of the
product. Each of the tables is used to store and manage the
collected data that are inserted from the GUI application,
and it will be retrieved when it is requested by users. The
connectivity of the database of the database is explained
Figure 12. Activity diagram for checking purchase history below. First, thee relational database management system
is connected to the ASUS server host after the SQL Server
2.2.2 GUI Design and Database Software Management Studio has been assessed. Then, a new da-
In this sub-section, the software implementation and tabase from the Object Explorer is created. The tables in
procedures for designing a user interface is discussed. the database can be used to assign and manage different
The Graphical User Interface (GUI) will be created on input data directing from the GUI application. After a new
Microsoft Visual Studio 2017 as it can be easily created table has been created, a name is assigned to each of the
by clicking and dragging the desired components such as column in that table. Figure 14 shows an example of the
label, text box, buttons, check box and data grid into the created database.
GUI layout form. Figure 13 (a) shows a blank Microsoft
Visual Studio 2017 GUI layout form and Figure 13 (b)
shows its corresponding Toolbox. The component param-
eters and properties such as appearance, data and layout
can be checked and changed by double clicking on the
components. Furthermore, the database connection can be
checked from the server explorer. This assures that all the
data is entered into the database table. Figure 13 (c) shows
the Microsoft Visual Studio 2017 Properties box and Fig-
ure 13 (d) shows the Microsoft Visual Studio 2017 Server
Explorer. The completed design of the GUI will be shown
in Section 4.

Figure 14. An example of database built by Microsoft

SQL Server Management Studio

3. Hough Transform Minutiae Pairing Fin-

gerprint Matching (HTMP) Technique
(a) (b)
The Hough Transform Minutiae Pairing (HTMP) fin-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 29

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

gerprint matching technique is proposed to the develop due to the reason that the same device is used to capture
fingerprint based school debit transaction system and it is fingerprint images for both the offline processing stage
run by the following six steps: and online authentication stage. The generalized Hough
Step 1: Scan Live Fingerprint: Retrieve user’s finger- Transform [15] is applied to determine those parameters.
print image using R305-optical fingerprint module. The entry E (l , p, q ) sum up the verification of the orien-
Step 2: Classify Fingerprint Image: Classify the fin-
gerprint image into any of the below eight categories: (i) tation transformation Gθl ,∆x p ,∆y p , where (θ l , ∆x p , ∆y p )
Plain Arch (ii) Tented Arch (iii) Ulnar Loop (iv) Radial
Loop (v) Plain Whorl (vi) Central Pocket Loop Whorl (vii)
are the quantized values of (θ , ∆x, ∆y ) correspondingly.
Double Loop Whorl and (viii) Accidental Loop Whorl, as The normalized E (l , p, q ) in the range from 0 to 1 is
depicted in Figure 15.
represented by f (θ , x, y ) and is treated as the probabil-
Step 3: Identify Region of Interest: identify the essence
point in the fingerprint image and crop the core region ity density function for θ, x, y transformation parameters.
concentrated in the essence point. (θ , x, y ) are independent from each other and can be es-
Step 4: Enroll Fingerprint Minutiae: extract two timated separately, hence f (θ , x, y ) can be re-structured
minutiae spot sets M and N from two fingerprint im-
as:
ages (database and inquiry) with undisclosed scale,
rotation and translations by the following notations:
f (θ , x, y ) = f θ (θ ) * f x ( x ) * f y ( y ) (2)
{( ) (
M = m1x , m1y , γ 1 ,..., m xP , m yP , γ P )} ,

{( ) (
N = n1x , n1y , δ 1 ,..., n xQ , n Qy , δ Q )} where P and Q are ( 1
(1) Rotation Parameter: Let Od = Od ,..., Od d be the
P
)
orientation field from the database fingerprint image and
the total number of minutiaes in set M and set N respec-
tively. (m , m
i i
) ( )
, γ i and n xi , n iy , δ i are those three
( )
Oi = Oi1 ,..., OiPi be the orientation field from the in-
x y

features (x-position, y-position, orientation) correlated quiry fingerprint image, where Pd and Pi are the lengths
with the i-th minutiae in set M and set N respectively. of the Od and Oi arrays respectively. The generalized
Hough Transform based method to estimate the rotation
parameter (θ) consists of following 2 main steps:
(a) Estimate the probability density function of the ro-
tation parameter fθ (θ )
(b) Search the correct rotation parameter of the two
transformation among the two fingerprint images based on
the results obtain in (a.)

An accumulator array E (l ) is used to gather the ver-

ification for each possible rotation by l degree, whereby
Figure 15. Fingerprint patterns [13] l = OdP − Oiq degree which map Oiq to Odp . The correct
Step 5: Determine Rotation and Translation Pa- rotation transformation among two fingerprint images is
rameters: the transformation orientation function: computed by mass center [14]:

G s ,θ ,∆x ,∆y : R 2 → R 2 is given by [14]:

m l (θ ) =
∑ l l × fθ (l ) (3)
l∈

 x   cos θ sin θ  x   ∆x  ∑ l fθ (l )
l∈
G s ,θ ,∆x ,∆y   = s   +   (1)
 y   − sin θ cos θ  y   ∆y  where λ is the densest interval among the extracted mi-
nutiaes from the collected fingerprint images
where s, θ, (∆x, ∆y ) are the scale, rotation and trans- (2) Translation Parameter: the computation of the trans-
lation parameters correspondingly. For fingerprint authen- ( )
lation parameters t x , t y can be done with the rotation
tication/identification, the scaling factor s is set to unity, parameter θ obtained above. Consider that the minutiae

30 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

set M and N are extracted from the database and inquiry

fingerprints correspondingly, utilizing the determined rota- ml ( y ) = ∑ l q × f (q ) (6)
q∈ y

tion transformation Gθ upon the minutiae in set N, a new ∑ l f (q )q∈ y

rotated version of the N point set, N r can be acquired. whereby q = m y − n y

i r, j
is the y-translation which map
The following notation are in used:
n yr , j to miy . miy is the y-coordinate for the i-th minutiae in
N = n , n ,δ
r
{( r ,1
x
r ,1
y
r ,1
),..., (n r ,Q
x ,n r ,Q
y ,δ r ,Q
)} set M and n y
r, j
is the y-coordinate for the j-th point in set

where (n r ,i
x )
, n yr ,i , δ r ,i are the three features (spatial, Nr.
Step 6: Set Threshold to Compute Matching Score:
position, orientation) related to the i-th minutiae in set
Pair up the minutiae set if the two fingerprint images’ fea-
N r . It can be calculated by [14]:
tures or components ( ml (θ ) , ml ( x ) , m l ( y ) ) are identical or
 n xr ,i   cos θ − sin θ 0  n xi   0  within a range of tolerance. A tolerance box is generated
 r ,i       throughout each minutiae feature for coping with the
 n y  =  sin θ cos θ 0  n iy  +  0  (4)
shifting in the minutiaes pairs. The minutiaes pairs are
 δ r ,i   0
   0 1  δ i  θ  gathered among the pairs that fulfil the below geometric
constraints [14]:
(a) The two minutiaes’ Euclidean distance does not ex-
where (n , n i
x
i
y )
, δ i is the i-th minutiae in set N, θ is ceed a certain value ∆d .
(b) The two minutiaes directions’ angular difference
the estimated rotation angle. Two minutiae points sets M
below a certain tolerance ∆θ
and N r are used to calculate the translation parameters (c) Supposing that in excess of one pairs situate in the
(t x , t y ) . The steps for calculating x-translation, t x : same bounding box, the two minutiaes with the minimum
Euclidean distance opt as the matched pair.
(a) Estimate the probability density function of the
Concerning to weight out the similarity among two fin-
x-translation parameter f x (x ) gerprints, a similarity level measurement method applying
(b) Calculate the x-translation, t x with f x ( x ) . the matching score ψs is adopted, with the below formula
[14]
:
An accumulator array E ( p ) is used to gather the
verification for each possible x-translation, whereby N pair
ψs = (7)
p = m −n i
x
r, j
x is the x-translation which map n
r, j
x max{P, Q}
i i
to m x . m x is the x-coordinate for the i-th minutiae in set
where N pair is the number of matched minutiae pairs,
r, j
M and n x is the x-coordinate for the j-th point in set P and Q are the total number of minutiaes extracted from
N r . The correct x-translation transformation among two the database and inquiry fingerprint images correspond-
ingly. ψs is with value ranging from 0 to 1. If ψs is tends
fingerprint images is computed by mass center [14]:
to 0 implies that the two fingerprints are non-matching,

ml (x) =
∑ l p × f ( p ) (5)
p∈ x
else if ψs is tends to 1 implies that the two fingerprints are
good match.
∑ l f ( p) p∈ x
4. Experimental Results
where λ is the densest interval among the extracted mi-
In this section, the application of the proposed fingerprint
nutiaes from the collected fingerprint images.
school debit transaction system will be illustrated. It will
The steps for calculating y-translation, t y are identi- cover the product overview, test run results and the prod-
cal with that of calculating x-translation, t x . The correct uct survey analysis. The fingerprint school debit transac-
tion system was tested in an institution of higher learning
y-translation transformation among two fingerprint images
within Malaysia, who wants to remain anonymous.
is computed by mass center [14]:

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 31

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

4.1 Product Overview Figure 18. The basic information to insert is course, phone
number, full name, gender and age. After inserting all the
The final prototype of the fingerprint school debit transac- relevant information, the “ADD” button is pressed to en-
tion system is shown in Figure 16. The circuit board and rol in the system. In this case, the system will generate an
Arduino UNO microcontroller module are stored inside an ID number for that particular student. Moreover, the sys-
external custom made casing. The function of all the but- tem can be used to update the latest information or delete
tons are clearly labelled with laminated instruction signs. the student’s previous account.
The LCD and LED will begin to operate once the USB Next, students are instructed to enrol their fingerprint
cable is connected to PC/laptop/tablet. When the finger- with the assigned identification number in the fingerprint
print verification process is successful, the green LED will collection module. The “Enrol” button must press to enrol a
light up, whereas if unsuccessful fingerprint verification fingerprint, the message “Please insert your ID” will be dis-
process detected, the red LED will be light up. played on LCD screen, as shown in Figure 19(a). Once the
ID number has been inserted, the student may place his/her
finger on the fingerprint module to scan fingerprints. The
student must place his/her finger twice in each registration
process, because the second fingerprint input is required to
compare and check with the initial template to ensure that
both fingerprint inputs are matched. The system will then
display “Print Matched, Enrol successful” to indicate a
success case (as shown in Figure 19(c). Else it will display
“Error” in the LCD screen for fail case. Students who have
successfully enrolled their fingerprint into the database sys-
tem may now start deposit /top up credit into the database.
The top-up page is shown in Figure 19d.

Figure 16. Final Prototype of the Fingerprint School Deb-

it Transaction System (Fingerprint Collection Module)
The Admin Login GUI page is shown in Figure 17.
This is the main page of the fingerprint school debit
transaction system. Only the related school shopkeepers
(Canteen, Cafeteria, Mini Mart, Printing shop, laundry
shop, computer shop etc.) have their own username and
password to login to the server to perform the business
transaction (transfer students deposited money into their
account). This is to ensure the privacy and confidentiality
of every shopkeeper.
Figure 18. Registration Page at Server

Figure 17. Admin Login GUI page

The first important step for the system to work is to
enrol the students’ (users’) fingerprint and personal infor-
mation into the system. The registration page is shown in
(a)

32 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

shows the Drinks GUI page of the canteen menu page.

The meals and drinks selected by the students are later
added to their shopping list, under Mycart GUI page, as
shown in Figure 20c. This page shows the price and quan-
tity of items ordered by students.
In order to check out the items, students must authenti-
cate their identity by evaluating their fingerprint for mak-
ing payment. Students are required to press the “Verify”
button on the fingerprint collection module, as shown in
Figure 20d. After the “Verify” button pressed, the LCD
screen will display the message “Waiting for valid fin-
gerprint”, as shown in Figure 20e, while processing the
fingerprint to seek for the correct students’ ID. The LCD
screen will display the message “Found ID #..” for suc-
(b)
cessful verification attempt, as shown in Figure 20f.
On the fingerprint collection module, the red LED will
initially illuminate to indicate that there is no finger place
on the sensor. The green LED will light up once the stu-
dents’ fingerprint had been identified. The LCD screen will
display the students’ identification number and the accuracy
of the matching fingerprint. Only student who has suc-
cessfully obtained his/her identification number can make
transaction. The canteen operator will insert the students’
identification number to complete the payment process.

(c)

(a)

(d)

Figure 19. Enrolment (a) Message shown after Enrol

Button Pressed (b) Scanning Fingerprint (c) Enrolment
successful (d) Top-Up Credit page
Students may start purchasing meals, goods, services
within the school after enrolled and deposited credit into
the database. A canteen case is chosen for further studies.
Figure 20a shows the Meal GUI page and Figure 20b (b)

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 33

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

4.2 Test Run Results

The fingerprint school debit transaction system was test
run in the canteen of the institution of higher learning.
The experiment tested between 1,000 different students
and staff. Each person must contribute 2 fingerprint in-
puts, which are thumb and index finger during the experi-
ment. Moreover, each fingerprint entry will be tested with
total number of 5 times. There will be a total of 10,000
attempts in this experiment. The main experiment objec-
tive is to test the reliability of this fingerprint recognition
(c)
system by calculating the successful authentication rate
between 10,000 attempts. The rate of accuracy will be cal-
culated with the below formula:

 Successful Attempt 
Rate of Accuracy
=   × 100% (8)
 Total Attempt 

Table 1 below shows the number of successful authen-

tication in each of the thumb and index finger attempts.
The total number of successful attempts in this authenti-
cation experiment is 9,090. Therefore, the accuracy rate
(d) for this system is 90.9% by using equation (8). Figure 21a
shows the failure rate between the thumb and index finger
and Figure 21b shows the failure rate between the (<2)
attempts and (>2) attempts.
The pie chart in Figure 21a shows clearly that index
finger has higher unsuccessful attempt rate (60.55%) com-
pare to thumb finger (39.45%). The pie chart in Figure
21b shows that most of the unsuccessful attempts took
place in the first two attempts of the experiment. It has
failure rate up to 79.23% in the first two attempts of the
authentication process. However, there is a drastic drop
after the first two attempts (20.77%).
(e) There are several reasons that could affect the accuracy
of the authentication process such as the finger pressure
exerted on the sensor and the immobility and fixed posture
of finger during verification. Besides that, finger’s skin
conditions such as dry, moisture and dirt may also affect
the authentication result. Thus, it can conclude that the
initial failure rate was solely caused by human factor.

Table 1. Results of Fingerprint Authentication Attempts

No. of Authentication Attempts

Finger Types
1 2 3 4 5 Total
Thumb 761 960 979 971 970 4,641
(f)
Index Finger 679 879 972 960 959 4,449
Figure 20. Menu Selection and Authentication for Pay- Total Successful At-
1,440 1,839 1,951 1,931 1,929 9,090
tempts
ment (a) Meals GUI page (b) Drinks GUI page (c) Mycart
Total Unsuccessful
GUI page (d) Identity Verify Button Pressing (e) Identity Attempts
560 161 49 69 71 910
Verify Process On-going (f) Results of Verification

34 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

is 30 second plus). Furthermore, most respondent do not

believe that the use of fingerprints would be an invasion
of privacy and are willing to trust the fingerprint based
debit transaction system if it is introduced in the school
business area.

(a)

(b)

Figure 21. Pie Chart on Unsuccessful Attempt (a) Be-

tween Thumb and Index Finger (b) Number of attempts

4.3 Product Survey Analysis

For the 1,000 participated students and staff in testing the
fingerprint collection module, a survey was conducted
onto them to determine the awareness and market value
of the fingerprint based school debit transaction system.
The questionnaire and the results are shown in Figure
22. Generally, majority of the respondents agreed that all
schools should have a cashless future and 40% of them
have heard about biometric identification, particularly in
fingerprint technology. In addition, 98% of them agreed
that a fingerprint based debit transaction system can elim-
inate the need for a student to carry cards and remember
their password/identification numbers. Most of them are
willing to replace the conventional payment method with
this new fingerprint based debit transaction system. Fur-
thermore, 82% of the respondents believe that the appro-
priate time taken to identify the fingerprints should be less
than 1 minute (our proposed system current achievement

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 35

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

system does have some limitation. It depends critically on

the quality of the fingerprint’s image. The resolution of
the fingerprint’s image has a vital influence on the accura-
cy of the matching system. The current optical fingerprint
module might not be able to extract the important features
from the fingerprint if the input is in extremely low quali-
ty (blur user’s fingerprint surface). Therefor it can lead to
rejection or acceptance. Furthermore, the pressure exerted
and the finger’s skin condition are also constraints to the
current system.
The current fingerprint collection module is powered
by electricity from the PC/laptop/tablet by each shop with
the school. However, if there is a power failure occurs, the
system will stop working. In this case, it will restrict the
registration and verification process, and may stop other
users from making payment with the system. An indepen-
dent power source will be developed in future to prevent
such incidents cause by power failure occurs. At this mo-
ment, the system only works in an offline mode within
school. For instance, it does not allow parents to check the
purchase history and credit balance via online. In addi-
tion, the proposed system does not support online banking
yet. Parents have to come to the finance department to
transfer or deposit funds to their children’s debit account.
Figure 22. Questionnaire and the results surveying finger- In future, the debit transaction system will be integrated
print based school debit transaction system and corporates with several banks to allow online banking
transaction.
5. Conclusion
References
A fingerprint-based school debit transaction system is
developed to promote a smooth flow and cashless transac- [1] D. Thaktar. Biometric Solution for Schools- Fin-
tion environment within the school. In this case, it makes gerprint lunch line. Bayometric, 2018. From:
all the procedures and step become more convenient and https://www.bayometric.com/biometric-solu-
efficient. Furthermore, the implementation of fingerprint tion-schools-fingerprint-lunch-line/
-based transaction system can totally avoid using cash [2] J. Trader. Why School Districts Should Implement
money as well as prevent issues like debit card get lost Cashless Fingerprint Payment System, M2SYS Tech-
or stolen. All of the transactions made within the school nology, 2016. From:
will be independent of PIN and cards. Therefore, with this h t t p : / / w w w. m 2 s y s . c o m / b l o g / e d u c a t i o n /
cashless payment system, all the money can be safely se- why-school-districts-should-implement-cashless-fin-
cured in the student’s account. Moreover, one of the high- gerprint-payment-system/
lights of this system is able to trace back all the purchase [3] C. Kalyani. Various Biometric Authentication Tech-
history. This function can help the parent to monitor and nique: A Review. Journal of Biometrics & Biostatis-
restrict their children from consuming any unhealthy diet tics, 2017, 08(05).
[4] S. Hashemi, H. Tann, F. Buttafuoco and S. Reda.
and junk foods. Thus, parent can ensure that their chil-
Approximate Computing for Biometric Security Sys-
dren’s money is being spent on the right intention. In addi-
tems: A Case Study on Iris Scanning. 2018 Design,
tion, a survey was conducted with 1,000 students and staff
Automation & Test in Europe Conference & Exhibi-
in the corresponding institution of higher learning, and
tion (DATE), 2018: 319-324.
majority of them agreed that all schools should be pursu-
[5] N. Charfi. Biometric recognition based on hand
ing a cashless future and willing to use fingerprint-based
schape and palmprint modalities. Image Processing
transaction system.
[eess.IV]. Ecole nationale supérieure Mines-Télécom
The proposed fingerprint-based school debit transaction
Atlantique, 2017.

36 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

[6] M. M. H. Ali, V. H. Mahale, P. Yannawar and A. T. tion Rate (FRR). Bayometric, 2017. From:
Gaikwad. Overview of fingerprint recognition sys- https://www.bayometric.com/-acceptance-rate-far--
tem. 2016 International Conference on Electrical, recognition-rate-frr/
Electronics, and Optimization Techniques (ICEEOT), [12] W. Yang, S. Wang, J. Hu, G. Zheng and C. Valli.
Chennai, 2016: 1334-1338. Security and Accuracy of Fingerprint-Based Biomet-
[7] R. Mansoor & B. Parisa. A Review Of Face Rec- rics: A Review. Symmetry, 2019, 11(141): 1-19.
ognition Methods. International Journal of Pattern [13] "Intergalactic Vault," [Online]. Available:
Recognition and Artificial Intelligence, 2013, 27(4): http://www.intergalacticvault.com/if-you-have-a-spi-
1356005(1 -35). ral-whorl-fingerprint-pattern-this-iswhat-it-means/
[8] M. K. Sharma & O. Kumar. Speech Recognition: A [14] J. Qi, Z. Shi, X. Zhao and Y. Wang. A Novel Finger-
Review. Special Conference Issue: National Confer- print Matching Method Based On The Hough Trans-
ence on Cloud Computing & Big Data, 2014: 62-71. form Without Quantization Of The Hough Space.
[9] J. Choudhary. Survey of Different Biometric Tech- Proceeding of the Third International Conference on
niques. International Journal of Modern Engineering Image and Graphics (ICIG’ 04) Hong Kong, China,
Research (IJMER), 2012. 2004: 262-265.
[10] M. N. Anjana Doshi. Biometric Recognition Tech- [15] N. Saroha and N. S. Gill. Hough Transform Based
niques. International Journal of Advanced Research Fingerprint Matching Using Minutiae Extraction.
in Computer Networking, Wireless and Mobile Com- International Journal of Advanced Research in Com-
munications, 2015, 2(1): 143-152. puter Science, 2013, 4(10): 117-120.
[11] D. Thaktar. Acceptance Rate (FAR) and Recogni-

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1202 37

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Artificial Intelligence Advances

https://ojs.bilpublishing.com/index.php/aia

ARTICLE
Using the CVP Traffic Detection Model at Road-Section Applies to
Traffic Information Collection and Monitor — the Case Study
Shing Tenqchen1*　Yen-Jung Su2　Keng-Pin Chen1
1. ChungHwa Telecom Telecommunication, Laboratories (CHTTL), Taiwan
2. Graduate student of Dept. of Electronic & Computer Engineering, NTUST

ARTICLE INFO ABSTRACT

Article history This paper proposes a using Cellular-Based Vehicle Probe (CVP) at
Received: 12 September 2019 road-section (RS) method to detect and setup a model for traffic flow infor-
mation (info) collection and monitor. There are multiple traffic collection
Accepted: 23 October 2019 devices including CVP, ETC-Based Vehicle Probe (EVP), Vehicle Detector
Published Online: 31 October 2019 (VD), and CCTV as traffic resources to serve as road condition info for
predicting the traffic jam problem, monitor and control. The main project
Keywords: has been applied at Tai # 2 Ghee-Jing roadway connects to Wan-Li section
Intelligent Transport Systems (ITS) as a trial field on fiscal year of 2017-2018. This paper proposes a man-flow
turning into traffic-flow with Long-Short Time Memory (LTSM) from re-
ETC-Based Vehicle Probe (EVP) current neural network (RNN) model. We also provide a model verification
Vehicle Detector (VD) and validation methodology with RNN for cross verification of system per-
Long-Short Time Memory (LTSM) formance.
Recurrent Neural Network (RNN)

　
1. Introduction into big data model for regular daytime.

A
This paper provides proposed traffic monitoring re-
dvanced traffic information collection can pro- sources from CVP, EVP, VD, and CCTV, etc. This model
vide an efficient, reliable, instant and mass traffic
combines a LTSM RNN structure to rapidly setup and
info at instant governmental road information
validate with RNN several trained into model with peak
publication. It can also be used for traffic managing strat-
and non-peak model for one year.
egies, monitor and control. This paper provides a trial
The main project combines Tai # 2 Gee-Jing Roadway
area at Tai # 2 Gee-Jing Roadway connecting to Wan-Li
section on the fiscal year of 2017-2018[1,2]. This project connecting to Wan-Li section and Mai-Kin road (RD)
provides a new increased CVP detecting road-section (RS) as RS collection area. It includes Tai #2 northern costal
from Man-flow turning into vehicle-flow Model applying RD (Emerald Bay to Gee-Jin 2nd RD), Gee-Jin 3rd RD,
to Traffic Information Collection and Monitor for 7 days Gee-Jin 2nd RD, Gee-Jin 1st RD, Mai-Jin RD, and An-Ler
per week with 5 minutes interval traffic info. Via the new RD. Figure 1 shows the flowchart of CVP road condition
improved man-flow turning to vehicle flow model, we detection procedure. This paper just includes the above
proposed a Long-Short Time Memory (LTSM) recurrent area as shown 56 RS in total at Figure 2 [1] (24 RS) and
neural network (RNN), to rapidly setup several trained Figure 2a [2] (32 RS). Thus, there are 56 RS in total at

*Corresponding Author:
Shing Tenqchen,
ChungHwa Telecom Telecommunication, Laboratories (CHTTL), Taiwan, China;
Email: [email protected]

38 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Figure 2 (a) & (b) and 5 minutes traffic info for CVP with
RS traffic travelling time, vehicle flow, and RS-turning
vector updated periodically. Figure 3 shows the detected
CVP Road Condition detection flowchart. The flowchart
is designed to have 5 procedures, which is the first step,
CVP signal command original; data collection; the second
step is related traffic info collection, like CVP, EVP, VD,
and CCTV. The third step is merged from step 1 and step
2. Then, the major work of this project is traffic flow tra-
jectory modelling setup and Analysis. The deployed at the
trial region is shown at Figure 2a & b.
(1)

CVP Signal command (3) (4) (5) Figure 3. Signal timing at selected intersection points
original data Collection.
Traffic flow Trajectory RS CVP Road Condition Model
Modelling Setup and Detection and Model Calibration

Table 1. Peak and Non-peak Traffic modes at different

Related Traffic info Analysis Setup And Verification
Collection

(2)
types of Intersection

Figure 1. CVP Road Condition Detection Flowchart

Emerald RD
Gung-west Link RD

Tai #2nd RD
Gee-Jin 3rd RD II Gee-Jin 2nd RD Gee-Jin 1st RD II Gee-Jin 1st RD I
Gee-Jin 3rd RD I
N N N N N N
Gin-Shan 010 010 010 010 010 010
Keelung
5 4 3 2 1 0 Downtown
HW #3
National
Tai #62

Mai-Jin RD

Ma-Lan RD
Gin-Mei RD Gee-Jin 3rd RD N N
030 020
0 0
N N
040 040
1 An-Ler RD sect. 1 3

N N Keelung
CVP Detecting RS Ba-Do
Si-Ze 050
Intersection 0
040 Downtown
2
(Da-Ye Tunnel)
2018 new increased 2 RS

Figure 2 (a). 2017 de facto CVP Detecting RS Demo

Project [2]

Resource: CHT Data Division Company

Jong-Jeng RD

Yi- 2nd RD
Yi- 1st RD

Jong-Jeng Public Park

Shin 2nd RD
2. CVP Road Condition Detection and Model
Jong-Shan 1st RD

Yi- 3rd RD

Yi- 4th RD
Gun-Shi St.

Hsin 2nd
Parking lot
Shin 1st RD Setup
Road condition detection is defined as to include the
Zen 1st RD

Keelung Maritime Plaza Dong-An

Parking lot
Zen 2nd RD traffic flow and travelling time detection together. Based
Jong 1st RD
Miaokou Night Market on the traffic flow volume, one can postulate the instant
signaling data and related position of actual position. We
Zen 3rd RD
Keelung
Ai 4th RD

Jong 2nd RD
Station

Jong 4th RD
Zen 4th RD can use statistics to calculate each individual timing on
the certain road-section (RS) for every customer. Then,
Jong 3rd RD

Zen 5th RD

we use customer volume and actual real traffic volume to

SIAO 4th RD

SIAO 2nd RD

Ai 1st RD
SIAO 3rd RD

Ai 3rd RD

setup a postulated model. Thus, we can probably postulate

CVP Detecting RS

this RS traffic flow volume. Simultaneously, based on one

customer’s adjacent instant signal interval difference, one
Figure 2 (b). 2018 newly increased CVP Detecting RS can postulate the traffic volume at that RS moving traffic
Demo Project [2] flow. When we calculate all of the mobile customers indi-
Figure 3 demonstrates the signal timing at selected in- vidually contribute the traffic volume by statistics, one can
tersection points during the weekday / holiday morning, induct an embedded road-user related reference value at
evening, and holidays with major investigated intersec- RS traffic volume. However, Telecom signal will have an
tion. Table 1 shows the peak and non-peak traffic modes averaged 5 minutes delay from signal reception, cleaning,
at different types of Intersection. making a pair, model analysis, and providing a operation
result.

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211 39

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

2.1 Traffic-flow Model Setup and Analysis It can memorize the time-series info of traffic volume
and serve as the reference point of next time stamp. Com-
Via the mobile signal info, we can estimate the traf- pared with RNN model, LSTM model has more gates
fic-flow volume at that RS from the historical info by such as forget gate, input gate, and output gate. They can
neural network training data module to postulate the RS be used as a more complicated parameters to be used
traffic volume. When the system receives the instant memorize and forget to reach a better postulated capabili-
traffic mobile data, one can use this module to postulate ty. We compare the experimental models of CNN, BPNN,
traffic-flow volume. Furthermore, when each accumulated and LSTM from simulation. The LSTM has a better result
period of time from the historical data, one can increase [2]
.
the accuracy of traffic-flow volume. For analyzing the
CVP road- condition to postulate the trend of traffic-flow 2.2 Traffic-flow Estimation
volume. This paper uses the supervised learning from
machine learning model. The training procedure is to tell Via the mobile signal info, this paper proposes a postu-
the true value of that RS to let them learn themselves. The lated estimated traffic –flow volume estimation from neu-
training model is used to postulate the future change of ral network training RS during a historical info module
traffic flow. This paper uses the model of long short- term over a period of time. For accumulated historical data, one
memory, (LSTM) to be an extended model, as shown in can reconstructed the model of RS to increase the accura-
Figure 4 for system architecture. It is an artificially recur- cy of traffic-flow estimation.
rent neural network (RNN) architecture. LSTM has feed- To analyze the trend change of road condition for the
back connections and not only process single data points variation of CVP to postulate the traffic-flow. This paper
(such as images), but also entire sequences of data. The adopts the supervised learning style of machine learning.
compact forms of the equation for the forward pass of an That is at the training level to input the true value of traf-
LSTM with a forget gate are: [7,8] fic-flow to let if self-adjust at the training model. One just
see the MSE error of training model enter the allowance
range to reach the accepted setup model.
We can use this model to postulate the variation of traf-
fic flow in the future. In the model, we propose the major
structure of RNN because the info of time-series model
can record the past info change from memory and use this
model to postulate the future change of traffic flow. There
are five major elements in the RNN model:
tanh
(1) The input of RNN is an X, which is the Telecom
Forget
Gate
Input Gate signal info. Those contain man-flow, date, time (5 minutes
Output
Gate
as one units), and traffic-flow, etc.
(2) The output of RNN is a postulated traffic-flow.
tanh sigmoid
sigmoid sigmoid (3) The parameters of RNN are the weightings of U, V,
W with the final recursive trained values.
(4) The hidden state is to represent the RNN’s memo-
ry’s S.
Input
Output (5) A series of continued periods of time from t-1, t,
t+1.
Figure 4. LSTM system architecture (6) The traffic-flow postulated model is based on the
Where the initial values are c0 = 0 and h0 = 0 and the time-series recurrent neural network, which is based on
operator ο denotes the Hadamard product (element-wise the info of CVP, can be regarded the input of RNN and
product). The subscripts t indexes the time step. LSTM the trend of traffic-flow is regarded as the output of RNN;
handles the cell memory to be a controlled gate state. That where the hidden state at the time instant of t-1 and the
is it can delete, increase, and output the message. ft is de- hidden state of time t. The formula is written as
fined as forget gate, which message is deleted, increased,
and outputted from cell. And, it is defined as the input =
gate (
St f [U ] X t + [W ] St −1 (6) )
which message is deleted, increased, and inputted from
cell. Ot is defined as the output gate which message is de- The Timing Sequence of RNN can be depicted as Fig-
leted, increased, and outputted from cell. ure 5.

40 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

the original score and the distance of population mean.

(4) The 4th stage:
The 4th stage is the correspondence of EVP info and
produces the training data and test data. Then, provide
the CVP man-flow turning into traffic-flow model with
cross-validation by RNN. The flowchart of construction of
model and testing procedure is given at Figure 7.

Figure 5. Timing Sequence of RNN [3,4] Data Model Training Model Testing Model

Because the traffic-flow is a continuous state; thus, it

Retrain
is quite to refer the past-timing state to change the cur- Y Query DB
historical info of
rent state to reach a more postulated state for the trend CVP/EVP

N
of variation. To reduce the effects of the irregular states, Data Pre-
processing
Extract info

one needs to pretreat the traffic info. The pretreatment Train Model

of traffic info include the info become a time-series info; Finish model setup
and store at AP server
Return Model
Parameters

which contains 4 stages, sort out time-series info, com- Message Queue
Acquire an instant
bine RS, info standardization, and correspond to CVP necessary CVP info

info, respectively. The total flowchart is depicted as fig- Data Pre- Download

ure 6. processing Model

Extract info

Model Postulation

Sort-out Time- Time is arranged in Write Postulated Return Postulated

Step 1: Grasp RS Info.
series Info. chronological order. Results into DB Results

Path and Road Section Merge Road Extreme values

Step 2: Merged Road in corresp o n d en ce Section into (outliers) is
Section. table paths deleted.

D a t a i s Z-score is
Step 3:
standardized. standardized. Figure 7. Man-flow turning into traffic-flow postulated
Connect & P r o d u c e
framework
Step 4: Correspond to
Correspond to Training data
CVP data
EVP data.
or testing data. The framework of human-flow given at Figure 8
turning into vehicle-flow contains input layer, LSTM
Figure 6. Pretreatment of Man-flow turning into traf- layer, hidden layer and output layer; where t represents
fic-flow time stamp, t-5, represents a priori 5 time stamp. X1 –
The procedure is arranged into the following stages: X4 represents different characteristics, j1 to jn represents
(1) The 1st stage: a hidden neural unit, y(t) represents the traffic-flow. At
The 1st stage is to arrange time-series data and telecom input layer, there are numerous characteristics such as
user individual time-series data. the total signal numbers, travel time, a rough estimated
(2) The 2nd stage: total number of vehicles at that RS to be a set of input
The 2nd stage is to combine RS. At first is to investigate info. Furthermore, one-time estimated info needs to be
the path road section in correspondence table and find out input 6 sets data from (t0)~(t-5). After passing numerous
the needed RS info of required path and combine them long-term memory LSTM layer (as shown in Figure 7)
and delete the outlier’s value and vacant value. and one hidden layer, it finally come out an estimated
(3) The 3rd stage: traffic-flow result. The flowchart of human-flow turning
The 3rd stage is to standardize data and proceed the into vehicle-flow postulated model is given in Figure
data to meet standard normal distribution. That is, the 8. Mainly, it can be casted with three folds. There are
mean value is 0, standard deviation error is 1, and, its data model, training model, and testing model, respec-
transfer function is tively. Their responsibilities are data processing at data
model, training model at trained model, and executing

z=
(x − µ)  (7)
vehicle-flow postulation at testing model. At first, one
σ comes into data model. They execute the judge of retrain
to decide those data whether needs to be retrained. If,
where x is needed to be standardized score, µ is the av- it is ‘yes’, it shall enter to query the historical data and
erage value, σ is the standard error and σ ≠ 0, z represents proceed to handle data preprocessing. Via the training

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211 41

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

model, one can get the extracted trained model to be the fication result.
input model of neuron. After finishing the setup model, This paper adopts the Lewis [7] as the accuracy of
we store it into Application (AP) server. Furthermore, evaluation criteria with MAPE. If, when the value of
from the data model, one can acquire the necessary MAPE is between 10%-20%, it the model has high accu-
postulated CVP data and similarly proceed info pre-pro- racy; when the value is 20%~50%, it means the predic-
cessing. Therefore, we can get a better capability from tion is reasonable.
that kind of reference record. The individual training and
parameter adjustment are important to reach an expected Table 2. MAPE Evaluation criteria [7]
result.
MAPE(%) illustration

y(t): the postulated traffic-flow at time-stamp t. <10 High accuracy prediction

10-20 Good prediction

20-50 Reasonable prediction

>50 Not accuracy prediction

x1(t): the signal volume at time t
x2(t): the CVP travelling time at time t [7]
x3(t): the calculated vehicle # at time t Resource: Lewis, C.D. (1982) .
x4(t): the time-stamp at time t

3.3 Cross Validation by Neural Network

The cross validation by RNN is given in Figure 3.1
(a). When the values of characteristics (include man-
Figure 8. the flowchart of man-flow turning into vehi- flow, date, time (5 minutes as a unit), traveling time, etc.
cle-flow model the output of o for traffic-flow), times weighting W and
add it together. Then, one can get the mapping value
3. Model Verification and Validation from activation (sigmoid) function f; that is, the postu-
lated traffic-flow volume. Then, we get final value of t
3.1 Signal Info Analysis
via eTag detecting traffic-flow volume. Finally, we use
Although collecting traffic signaling info, it can pro- the Adam [6] algorithm to optimize the weightings of W.
mote the comprehensiveness and integrity of traveling The compared adjusted postulated value of a with true
info. However, the signal info is restricted to the law of value of t is very close. This procedure is called a one-
Privacy Act. Due to different geometrical geospatial base time calibration cycle. Through various trainings, the
station coverage, the market share of telecom operator error of MSE will gradually reduce to the wanted results
and user habit are different. The possibility of accuracy as shown in Figure 3.1 (b) at Gee-Gin 1st RD and 2nd RD
of signal may produce prejudice. Therefore, we need to during holiday model.
proceed model verification and validation through other Figure 9 (a) and 9 (b) were executed at the last fiscal
transportation info to increase the accuracy of analytical year of 2017 results. The weekday’s and holiday’s verifi-
results. cation results of independent CVP model at Gee-Gin 1st
RD-2nd RD were given respectively. The CVP traffic-flow
3.2 Model Verification and Validation verification at National HW #1 (From Bar-do to Mai-Gin
The major RS have implanted electronic tag (eTag) RD) and the Manual Computing Unit (MCU) vs. CVP
detectors at the described deployed testing area. Thus, are given at Figure 10 (a) and 10 (b), respectively. The
we want to use eTag’s info to proceed the verification CVP traffic-flow postulated verified testing results at Gee-
and validation. This paper wants to use neural network to Gin RD at Table 2. Using EVP traffic-flow has the same
train the traffic-flow and use eTag detector RS to get cross result as CVP did it before. According to [7], If, MAPE is
validation and then calculate the error rate. If the error less than 20%, it is a good prediction; the value of MAPE
rate is larger enough than expected, we need to calibrate between 20~50 % is a reasonable prediction. Thus, if we
the module to reach the required accuracy. Furthermore, found that the newly increased detecting are has no the
if the RS has no eTag detector or vehicle detector (VD), resources of EVP or VD at this project of 20, we will use
this project will use the traveling time from Google Map CCTV instead as vehicle detector. Using image to transfer
to be reference point for cross validation. About the Traf- into traffic-flow and then compare/validate with the result
fic-flow, we use AI image recognition to analyze the veri- with CVP model.

42 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Table 2. The CVP traffic-flow postulated verified testing 4. Conclusion

results at Gee-Gin RD
This paper proposes a using CVP method at RS to
Independent Model Shared Model
detect and setup a model for traffic flow info collection
Road Section Week-
Holiday Weekday Holiday and monitor. Via the new improved man-flow turning to
day
MAPE MAPE MAPE
MAPE traffic-flow model, we proposed a LTSM with RNN to
Gee-Gin 1st RD-2nd validate data model for regular daytime and peak and non-
20.4% 15.5% 40.1% 32.2%
RD
Gee-Gin 2nd RD-
peak model for one year.
18.9% 16.7% 32.4% 28.6%
1stRD
Gee-Gin 2nd RD-3rd References
14.4% 14.8% 25.0% 27.3%
RD
Gee-Gin 3rd RD-2nd [1] Lo, Bin-Rong , Jiang, Jhe-Ye, Tung, Shen-Lung.
18.2% 12.2% 25.1% 24.3%
RD Northern Bin-Hai Railroad at Gee-Gin Road Section
Average 17.9% 14.8% 30.6% 28.1%
with Crowded Stable Corridor with from ITS System
Constructed Plan Technical Service Report. Gee-
Lung City Hall Transportation and Tourism Depart-
ment, 2018: 27-63.
[2] Lo, Bin-Rong , Jiang, Jhe-Ye, Tung, Shen-Lung.
Northern Bin-Hai Railroad at Gee-Gin Road Section
with Crowded Stable Corridor with from ITS System
Constructed Plan Technical Service Report. Gee-
Lung City Hall Transportation and Tourism Depart-
Figure 9 (a). the weekday’s verification result of indepen-
ment, 2019, 03: 1-30.
dent CVP model at Gee-Gin 1st RD-2nd RD
[3] Tenqchen, S., Chen, Keng-Pin, Tung, Shen-Lung,
Jiang, Jhe-Yi. Using Neural Network Technology to
Develop CVP Traffic Flow Information - the Case
Study,TL Technical Journal, 2018, 48(4): 16-25.
ISSN-1015-0730
[4] Adolf D. May. Fundamentals of Traffic Flow. Pren-
tice - Hall, Inc. Englewood Cliff New Jersey 07632,
Figure 9 (b). the holiday’s verification result of indepen-
second edition, 1990.
dent CVP model at Gee-Gin 1st RD-2nd RD
[5] Green Shields Model
http://www.webpages.uidaho.edu/niatt_labmanual/
300 chapters/trafficflowtheory/theoryandconcepts/Green-
200 shieldsModel.htm
100
0 [6] Kingma, Diederik P., Jimmy Ba. Adam: A method
for stochastic optimization. arXiv preprint arX-
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…
2019-01…

iv:1412.6980, 2014.
[7] Lewis, C. D. Industrial and business forecasting
methods: A radical guide to exponential smoothing
and curve fitting. London, Boston: Butterworth Sci-
Figure 10 (a). The CVP traffic-flow verification at Na- entific, 1982: 305.
tional HW #1 (From Bar-do to Mai-Gin RD) [8] Sak, Haşim, Andrew Senior, and Françoise Beaufays.
Long short-term memory recurrent neural network
architectures for large scale acoustic modeling. Fif-
teenth annual conference of the international speech
communication association, 2014.
[9] Yann LeCun, Leon Bottou, Yoshua Bengio, Patrick
Haffner. Gradient-Based Learning Applied to Docu-
ment Recognition, 1998.

Figure 10 (b). The Manual Computing Unit (MCU) vs. CVP

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1211 43

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Artificial Intelligence Advances

https://ojs.bilpublishing.com/index.php/aia

ARTICLE
Application of Feature Curves and Shape Blending on Yacht Designing
Shih-Wen Hsiao1*　Jeng-Horng Chen2　Ting-An Yeh1
1. Department of Industrial Design, National Cheng Kung University, Tainan, 70101, Taiwan
2. Department of Systems and Naval Mechatronic Engineering, National Cheng Kung University, Tainan, 70101, Tai-
wan

ARTICLE INFO ABSTRACT

Article history The global yacht market share and industry might expand continually since
Received: 16 September 2019 increasing people are willing to enjoy the yacht life in lower price nowa-
days while yacht activities were regards as a luxury acitivity in the past.
Accepted: 23 October 2019 Additionally, Taiwanese yacht manufacturers are well-known worldwide.
Published Online: 30 October 2019 They show excellent performances on the international annual rankings
which implies Taiwan has excellent manufacturing technologies. However,
Keywords: Taiwanese manufacturers so far do not have a mature local design team.
Artificial intelligent Therefore, this study goals to developing a systematic and objective design
method for hull designing, which facilitates designers to design innovative
Yacht industry yachts or create a series product with brand recognition. This study is divid-
Shape morphing ed into three parts: the first part is the investigation of the yachts market; the
Computer aid design second is establishing a shape blending platform; the third is 3D forming.
Finally, it is used the existing ship CAD software to design an innovative
Product design
yacht based on the blended curves and then calculates its basic hydrostatic
Aesthetics performance. This study provides a quantitative method to create a new
form and to preserve the features for a brand. The exist yacht combines
with other graphics to create a new form and maximizes the features of the
original graphics.

　
1. Introduction In the past, the yacht market was mainly distributed

T
in North America, Europe, the Mediterranean, and New
he yacht industry has had a stable demanding mar-
Zealand and Australia. With the rise of Dubai and China
ket in Europe and American. In recent years, the
in recently, the demand in these two regions has grown
market for yachts, water amusement facilities and
rapidly. According to the global luxury yacht market sur-
related industries has expanded depending on the people’s
vey report of Wise Guy Reports in 2017 [1], the market and
attention to leisure and entertainment. At present, the mar-
ket capacity of the entire yacht industry in the world has growth trend of the future yacht industry are pointed out
exceeded USD 40 billion. Because of the world economic in the article because of the growth of millionaire families
growth, the adjustment of the supply and demand scales, and the popularity of yachts as luxury sports. Thus, they
and the large number of old ships that need to be renewed, have increased the market demand for private yachts. In
the overall demand for new ships will remain above a cer- addition, the market size in the following ten years is fore-
tain amount. cast to grow. By 2022, the global market can reach USD

*Corresponding Author:
Shih-Wen Hsiao,
Department of Industrial Design, National Cheng Kung University, No. 1, University Road, Tainan, 70101, Taiwan;
Email: [email protected]

44 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

74.7 billion [1]. uct makers in the market, it is critical to stand out from
Taiwan has a land area of about 36,000 km2 where the vast competitors in the case where the market share of
contains 31 yacht manufacturers and several of which are the product is close to saturation and the product function
well-known worldwide. Our construction abilities have and manufacturing technology are highly mature. When
been favored around the world as well. Also the yacht the brand is identified as a combination of elements, noth-
manufacturers have been ranked in the global rankings ing can be ignored and the product conveys brand value
recently. For example, Horizon Yachts has been ranked through its appearance [7]. Overall, family features and the
among the top ten manufacturers in the world for many establishment of perceived connections between brands
years [2], and the Ocean Alexander, their number of yacht and consumers, with their design needs and business eco-
sales in 2016 also jumped to the fourth place in the world nomic considerations, which may be a key factor in de-
[3]
. Those show Taiwanese yacht manufacturers have good termining whether the product is sold successfully in the
reputation and are trusted by the international customers. market.
However, Hsing Hang Marine states on its official website Regarding the basic principle of aesthetics within the
that the yachts produced by Taiwanese brand have not yet past days, Plato once stated that the thing with symmetry
established their own distinctive style and brand charac- and regularity was beautiful. Most forms of beauty were
teristics. It is to say that Taiwan still hasn’t had its own possessed regular geometric shapes, conforming to the
yacht designer team. Most of the current ship plans are appearance of natural look. Birkhoff, the first scholar
designed abroad. It is more difficult to develop the specif- proposing quantitative beauty, applied the calculations of
ic characteristics of the company. beauty on forming, poetry, music and other fields in his
In recent years, the manufacturers in Asia, such as book Theory of Aesthetic Mathematic. He applied the
South Korea and China, has also been continuously im- above concepts to quantify the beauty by mathematics.
[8]
proving. Among these countries, China progress the most. Later, Staudek used Birkhoff’s aesthetic formula to
In such a competitive environment, to maintain market discuss the contribution of the measurement method re-
share and the company’s profitability, Taiwan should garding the aesthetic quantification of the appearance of
break through the present operation modes of buying de- the object and he verified it by vases (shown in Figure 1)
[9]
signs from abroad, and consider both future trends and . The several types of characteristic points were created
one of the major features of Taiwan's yacht industry. Add- along the outline: terminal point, point of inflection, point
ing features to their products highlights the values and of vertical tangent, and point of abrupt change in the di-
significances behind the brand, while the product can be rection of tangent. (Eq. (1)).
modularized, launched a series of products, and used for
innovative design to increase recognition. H +V + P +T
M= (1)
For many modern industries, the application and de- C
sign extension of the brand series are the design direction
of the new era. Yacht design can also use this concept to Where H is the number of the horizontal order, H ≤ 4;
develop specialty products. The purpose of this study is V is the number of the vertical order, V ≤ 4; P is the num-
to apply computer-aided design (CAD), systematic col- ber of the proportional order, T ≤ 4; C is the number of
laboration design, and the concept of brand feature to es- the tangent order, ; is the complexity of the characteristic
tablish a system that is suitable for yacht design: develop points and M is the aesthetic measure.
new forms with the shape blending method, and judge the
form with the principles of aesthetics, which is no longer
relying on the designer's personal aesthetic to design the
appearance of the yacht.

2. Related Works
2.1 Feature Forms and Aesthetics
In the product design, the appearance is the most basic
Figure 1. Tomáš Staudek’s aesthetic measurement of
form of a product. Regardless of the functions, the ap- vases
pearance has a tremendous influence on product sales, and
sometimes plays an important role in whether the product Hsiao & Tian added the appropriate parameters into
can survive successfully [4-6]. Therefore, for mature prod- the original aesthetic measure. The new formula could

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 45

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

be applied to the 2D curves and 3D surface. Also, they and the conceptual yacht design. Finally, it is to present
proposed the measurement of the order, complexity, and a new yacht with distinctive features and a system model
aesthetics of the form as the evaluation basis of the form that could be applied to design concept yachts.
aesthetics for designers. Finally, it assisted designers to
design new products. [10]

2.2 Shape Morphing

Shape homogenization and shape blending can gener-
ate new forms by interpolating and extrapolating the
parameters of product forms, following certain rules
to produce changes in the form, such as displacement,
shrinkage, elongation, inversion, etc. Hui and Li in their
feature-based shape blending study decomposed the form
by semantics, components and points, thereby proposing
features that could be blended and establish the correspon-
dence between object features. The normalized weights
were used in interpolation to find the corresponding re-
lationship of feature points and to blend two models. [11]
With the operations, the features of the form could be pre-
served during the blending process. In addition, the prop-
erty of the interpolation was that the new blended feature Figure 2. The process of the research
was predictable when the two patterns were blending and
tended to be similar with which pattern. After blending the 3.1 Limitation and Specification
2D curves, the final 3D model was synthesized by the 2D This study selects the yacht's feature form as the research
curve boundary outlines. Finally, 3D model was generat- object and focuses on the exterior design only although
ed. [12] numerous factors have to be considered in designing a
Hsiao et al used reverse engineering in combination yacht. The color of the appearance is also excluded so
with shape blending to create a reverse engineering meth- all the yacht samples in the study show black and white.
odology which could be applied to shape morphing. Com- Except for the forming factors affected by regulations, the
bining with the existing CAD software provided industrial other factors might make a concession to the exterior de-
designers to structure ideation more quickly. The different sign. The interior compartments will be referenced to the
part from the former study was the corresponding various planning of the original yacht. Due to the huge scale of
point methods. Ray firing method calculated the angle the yacht, this study focuses on the elements and features
between the end-to-center point line and the horizontal of the sideways. Other perspectives are used as an assis-
line through the center axis of the input shape. It was tance. Finally, it presents in 3D computer model.
determined that the two end points were on the same ray In this study, the rulers of an existing yacht are used as
diverging from the center point. When the difference be- a reference to obtain a datum model. In order to simplify
tween the two angles was the closest, it was regarded as the research scale, few factors are taken into consideration,
a group of similar corresponding points. After the points such as speed, stability, resistance, load, and buoyancy
were determined, blending operations could be performed. requirements. There is not a specific owner and shipyard,
This study proposes four blending methods and uses dif- so the budget and the manufacturing technology cannot be
ferent weights to adjust the specific gravity of the two confirmed, and this study doesn’t consider the main engine
components. [13-14] as the variable. Also, the overall cost and total weight can-
not be estimated. Finally, the calculation of manipulability,
3. Methodology
resistance to wave and sinking are excluded as well.
This study contains only determination of “design re-
quirements and major dimensions” and “ship design and 3.2 Setting of the Design Samples
hydrostatic performance”. The rest is not considered in 3.2.1 Yacht Samples Selecting
the study. The research process is mainly divided into two
parts, as shown in Figure 2, which are the forming design In this study, it was going to combine the existing

46 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

yacht's form with other features, so the eight samples, inal canoe, orchid and Taipei 101(shown in appendix
named in order from A to H, were selected as a reference A.2). The eight options for the subject to choose and to
based on the survey results. (See as appendix A.1) The add in the yacht design were ranked. According to the
first choice for yacht samples was that the manufacturer top five elements of the most wanted element and based
designed and built the length between 25 to 40 meters on the result, the highest score of the aboriginal pattern
of yacht. For its style to reach the modern trend, the ship and the second highest score of Taiwanese leopard cat
should be built within five years, and its main function are selected to add in the design.
was leisure. There was at least one closed interior space
on the deck. It was mainly driven by fuel and achieving
the buoyancy requirements and the regulations.
To concentrate on the exterior design, the samples pre-
sented to the participants should be displayed by outlines
and feature curves, and excluded materials, color. At the
same time, the influence of brand preferences on the par-
ticipants was excluded as well. At the time of the investi-
gation, the brands of the eight yachts were not disclosed: Figure 4. Preference ranking and scores of the Taiwanese
the participants only knew the length, beam and draught elements going to be blended (Score; Selection; Percent-
of the yachts. age)
3.2.2 Expert Survey
3.3 Shape Blending and Selecting the Feature
The study interviewed 11 yacht experts and conducted a Curves
survey of preferences, including engineers, designers and
According to the results, this study selected the top two
operations managers of ship researchers, shipping agents,
scores: Taiwanese aboriginal pattern and Taiwanese leop-
and who were interested in yachts and had a certain level
ard cat as the Taiwanese feature elements for the new
of understanding multiple yachting experience and under-
concept yacht design, respectively decomposing the fea-
stand the Taiwanese yacht brands, concerning about the
ture curves, and then composing new elements and shape
latest yachts often.
blending of hull.
11 experts conducted style analysis on eight ships and
The hull sample used the G yacht as a reference and the
ranked them in order of preference. The interview results
study selected its local feature curves. According to the
showed that the eleven experts disagreed on the style
side-view diagram above the waterline, the yacht curves
view, but only the G (Figure 3) had a similar view. In
could be divided into outlines, windows, structural lines,
terms of the preference ranking, the number G also was
details, the parts of which to adjusted are selected. Then,
scored the highest. Thus, it was going to use the number
they were blended with the same or different feature el-
G yacht on the following study.
ements. The datum hull was created by DELFTshipTM
free, from which the outlines of the various views were
obtained, and the side view and top view of the G yacht
were used to obtain more detailed curves, shown as Fig-
ure 5. With the lines in the front view, top view, and right
view, the curves that represent the features of the Taiwan-
ese leopard cat could be selected. The curves of Taiwan-
Figure 3. The yacht G
ese leopard cat's round head and strong musculoskeletal
In addition, the interview also questioned the ex- features were the final decision to join the new form of the
perts’ views on the innovative yacht styling and their yacht, shown as Figure 6.
acceptance of the design of the Taiwanese elements.
The acceptance was also divided into 11 stages, more
score more acceptance. The answers were all positive,
the score statistics were shown in Figure 4. At the same
(a) The outline of the main deck (side view)
time, this study provided several Taiwanese elements,
namely: butterfly, Taiwanese salmon, Taiwanese leopard
cat, Hakka pattern, aboriginal pattern, Taiwanese aborig- (b) The fence of the deck (side view)

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 47

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

The hundred pacer is an important god in the group

of Taiwanese aborigine, called Paiwan, faith. The people
(c) Rooftop
believe that the hundred pacer could bring good luck, and
only the nobles can engrave the hundred pacer pattern on
the ornament, which is a symbol of status. The eye of the
sun is the decoration on the Dawu’s canoe, another group
of Taiwanese aborigine. This pattern means the eyes of
their unique fishing boats, leading the way for seafarers
(d) The outline of the main deck (top view) and blessing safe and healthy.

Figure 5. The feature curves of the G yacht 4. Results

The study, based on the feature parameterization method,
shape morphing and other methods combining with the
existing 3D CAD software DELFTshipTM, established a
new concept yacht design process and connects them in
JAVA programming language.
(a) Outline of head (front view)
4.1 The Framework
After understanding the steps of the overall yacht design,
the preferences of the current existing yachts from the ex-
perts and the speculations of experts on the future forming
trends were obtained, as well as the features of Taiwan to
(b) Outline of a section (side view)
be blended through the data from interviews with experts.
Then the platform in JAVA could calculate the new feature
curves by shape blending method, and exported the result
image to DELFTshipTM in order to build a 3D model and
calculate the hydrostatic performance. After adjusted to
a workable vessel, it was finally calculated its aesthetics
(c) Outline of a section (top view) values.

Figure 6. The selected feature curves of Taiwanese leop- 4.2 The Interface and Method of Shape Blending
ard cat
A blending platform was written by JAVA, shown as Fig-
The function of the aboriginal patterns cloud be classi-
ure 8. After the hull and one of the sectional curves of
fied to records, legends, blessings, etc. After referring to
Taiwanese leopard cat were inserted into the platform, the
the shape, story behind the pattern, the study selected the
blending operation could be performed. The steps were as
hundred pacer (Deinagkistrodon), a kind of snake, and the
follows:
eye of the sun as the blending elements.
(1) Inserted the point data (.txt) of the yacht and the el-
ement separately.
(2) Adjusted the yacht weight W S and the element
weight WE, WS + WE = 1.
(3) Clicked the Blending button to blend.
(4) The SAVE button stored the blending result in the
image file (.jpeg).
(a) Hundred pacer (Deinagkistrodon)
(5) The Clear button cleared all point data and display
curves shown on the interface.

(b) The eye of sun

Figure 7. Taiwanese aboriginal patterns

48 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

the curve of Taiwanese leopard cat from back to hip. Ad-

ditionally, there were selecting part of two samples, name-
ly the front end of the upper deck of the main deck and
hundred pacer’s head, as shown in Figure 12.

(a) the weight of ship, WS: 0.4; the weight of element, WE:0.6

(b) the weight of ship, WS: 0.5; the weight of element, WE:0.5

Figure 8. The blending interface

After inserting the feature curve data, the point clouds (c) the weight of ship, WS: 0.8; the weight of element, WE:0.2
would be converted into eigenvalues in the platform
which would be normalized, and then the blending opera- Figure 9. The blending results with different weights
tion would be performed, converting the new eigenvalues from the waterline of the datum hull and the top view of
calculated in the platform. After the data was formed, the Taiwanese leopard cat
points were connected in order by interpolation and a new
feature curve was created. The result was accessed in the
JPEG file.
There were four different methods for the above-men-
tioned blending operations: arithmetic mean method, (a) the weight of ship, WS: 0.4; the weight of element, WE:0.6
geometric mean method, harmonic mean method and gen-
eralized weighted mean method. The four algorithms were
input into the same weighting operation and compared.
Because this study still emphasized on the feasibility of
(b) the weight of ship, WS: 0.5; the weight of element, WE:0.5
the hull, the blending pattern should be closer to the orig-
inal one. The arithmetic mean and the geometric mean
were closer to the original hull line plan. The harmonic
mean showed more features of the blended element. The
generalized weighted mean method had obvious differ- (c) the weight of ship, WS: 0.8; the weight of element, WE:0.2
ences between the original hull line plan and the charac-
teristic elements (the four methods under the same weight Figure 10. The blending results from the side outline of
value). Therefore, the arithmetic average was chosen as the datum hull and the side outline of Taiwanese leopard
the calculation method of shape blending. cat

4.3 The Results of Shape Blending

(a) The deck fence
The point clouds of the new form curves were calculated
by the arithmetic mean and the new curves were calcu-
lated by interpolation in order to find the most suitable
result with different weight values. The two curves of
(b) The blending result
the blending process must be at the same view, such as
the most upper waterline of the basic hull and Taiwanese
leopard cat in the top view and the side hull outline and
the side outline of Taiwanese leopard cat, as shown in the (c) The part of the outline from Taiwanese leopard cat
Figures 9 and 10. Some were mainly selected from yachts
and blended with part of Taiwanese leopard cat curves. Figure 11. The blending results from the deck fence and
Figure 11 shows the outline of the fence on the deck and part of the Taiwanese leopard cat’s outline

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 49

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Figure 12. The blending results from the top view of main
deck and the hundred pacer’s head

4.3.1 3D Forming and Hydrostatics Test

The basic model had been obtained by DELFTshipTM free,
and then a new feature curve was obtained by the software
calculating. The original model was adjusted according
to the new feature curves to obtain a new model of the
yacht. Finally, it was necessary to calculate the hydrostatic
performance of the new yacht. If it failed the testing, the
form of the yacht had to be adjusted. It was three revises
(a) the first result
to obtain the yacht that achieved hydrostatic performance
standard in the study.

4.3.2 Aesthetic Test

After the hydrostatic performance test, the 3D stereo mod-
el was output, and then returned to the JAVA platform for
aesthetic testing. The new yacht model was divided into
two main objects and elements. The centroid position and
plane projection position were calculated respectively. It
was five aesthetic factors, such as balance, unity, equi-
librium, proportion and cohesion to be tested. Finally,
combined the five factors to calculate the average aes-
thetic value (all weights were 1), and the values were all
between 0 and 1. The closer to 1, the more objects met the
ideal standard of aesthetics.
The final result obtained in the previous step was im-
ported into the JAVA for aesthetic test. The following (b) the second result

results obtained were 0.649 for balance, 0.620 for equilib-

rium, 0.878 for proportion, 0.770 for unity and 0.471 for
cohesion, as well as 0.678 for the average (Figure 13 (a)).
The result was greater than 0.5 for the aesthetic test, so in
this test, the cohesion project did not achieve the standard.
It was necessary to get back to the previous steps to adjust
the yacht model.
After the revise, the aesthetic test results were shown in
Figure 13 (b). It could be known that the cohesion value
in the second result had not passed the test, so the yacht
model had a fine adjustment again. The third results ob-
tained by the test were balance 0.622, equilibrium 0.569,
proportion 0.886, unity 0.770, cohesion 0.487, average
0.667 (Figure 13 (c)). Although the value of cohesion
was improved, but it was still less than 0.5. However, this
study focused on the accessibility of the yacht, so the step (c) the third result
could move on to the next step if it failed over three times.
Figure 13. The results of aesthetic test

50 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

4.3.3 The Final Result The eye of the sun is installed on the bow which follows
its original meaning: the eyes of vessel. Such pattern is
After repeatedly adjusted the yacht model through the bound to be engraved on the canoe of the Dawu people to
hydrostatic performance and aesthetic test, the final result protect the crews from the navigation. In the results of this
was shown in Figures 14 and 15 This new yacht was a study also adds such elements. However, in order to make
private yacht for leisure and entertainment, suitable for it more concise, this design slightly simplifies the original
families or entertaining guests. The design had a length of graphics, but carries the same meaning as the origin, a
33.40 m, a total length of 33.59 m, a design beam of 7.30 wish of safe sail.
m, a total beam of 7.78 m, a design draft of 1.7 m, and a
hydrostatic performance result of a relative water densi- 5. Discussion and Conclusion
ty of 1.025 and a mean shell thickness of 0.015 m. The
molded volume was 122.038 m3, the displacement was Based on the feature parameterization method, shape
127.37 tons, the longitudinal center of buoyancy is 16.301 blending and other methods combined with the existing
m, the block coefficient was 0.2998, and the waterplane 3D CAD software DELFTshipTM free and Rhino 5.0, this
coefficient was 0.4932. There were two closed cabins and study established a new concept of yacht design process,
an open platform above the deck. and uses the JAVA programming language to carry out the
calculation of point data.

5.1 Feature Curves and Blending Method

In this study, the feature forming is decomposed, and the
feature curves are defined by the quantitative method. The
necessity of the original yacht feature curves is analyzed,
which can match the appropriate product image to know
the understanding of the forming curve on the consumers’
perceptive, and then it is possible to select the appropriate
Figure 14. The final result
feature curves, blended with the feature curves of other
items to obtain a new feature curves. The feature data es-
tablished through this mode is not only fast and objective,
but accessible and possible to redevelopment in the future,
which is easy to analyze in the research. The method can
help the designers to develop new form, provide the de-
signers new inspirations, and be able to mix the features
curves of different things altogether at the same time.
Such benefits are conducive to the development of region-
al style. Also, it is easy to preserve the feature data, which
can be further developed and updated from this data, re-
newing the brand characteristics.
Figure 15. The rendering diagram of the final result The platform written by JAVA program reads and
blends the point data, and generates the image files that
The hull is mainly blended with the Taiwanese leopard can be used for DELFTshipTM free in the study. The ob-
cat, showing the body curve of the Taiwanese leopard cat. tained two data of point clouds are sorted out and then
The other parts that has more angular curves and the fence the remaining points are relocated and arranged in order.
on the deck are also blended with the Taiwanese leopard Finally, the two points data are integrated by ray firing
cat's feature curves to make it more round. Also, above the method to obtain the same number of points for following
main deck, a part of the yacht outline segments uses the calculation; in the blending process, the sorted point data
hundred pacer’s head to blend. The hundred pacer is a re- is first converted into feature values and then normalized.
spected god in the belief of the Paiwan people. It implies Blending is performed on the same scale to avoid dis-
blessings that the hundred pacer on the deck leads the tortion due to different sizes. Calculate the new feature
vessel going forward, wishing sailing safe. In detail, the value, and then go backward of the previous steps to get a
new yacht shows the features of the aboriginal patterns, new feature curves.
repeated and regularly arranged. Thus, a little decoration The comparison of the results of the four formulas of
is added to enhance the image of the aboriginal patterns.

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 51

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

the blending method shows that the results of the arith- 5.3 Conclusion
metic mean and the geometric mean method are similar
to the original graph, and the blended curves are more This study provides a systematic method to design the
predictable. The result of the harmonic mean method is concept of yacht design: the existing yachts, with the
significantly different from the original graph. However, it actual animal, graphics, etc. into the yacht, constitute
can still be seen that it is a little similar to the original, and new curves, so that the figures can be properly combined
the generalized weighted mean method has the biggest with the existing yacht, the maximum present the special
difference. Because this study focuses on the feasibility of characteristics of the original graphics, and can meet the
the vessel, the arithmetic mean and geometric mean meth- basic performance of the yacht and the legal norms. The
od are more suitable so this study uses the least difference new curve obtained from the shape blending method can
method - arithmetic mean. At the same time, it can be seen also be completely preserved and provided for the next
that the harmonic mean method is suitable for the devel- design use. The future development is to further improve
opment of brand new product, subverting the impression the shape blending software, making the selected feature
of the past. points more accurate and more representative in order to
preserve the feature. Also, improving the software can de-
5.2 Problems and Discussions crease the time of the design process. Widely application
for the brand company, it can be effectively established
The results of interviews with relevant industry workers and save its own brand features, and further add new
and experts show that the participant's perception of each elements into the old product style. If a brand company
yacht is different, only a few in common. The possible wants a cross-over with different domains, this method
reason causing the problem is that the definition of each can provide a higher level of cooperation.
adjective is not clearly defined, so the participants might
follow their thinking based on growth background, experi- Appendix
ence and understanding of the vocabularies. These results
in different imaginations, which lead to divergence of
A.1 Eight Selected Yacht Samples
results. Therefore, it is necessary to guide the participants
to understand before they fill in the questionnaire, to de-
scribe the connection between words and images, and to
present photos in a monochrome stereoscopic model.
In the discussion of style and feature curves, this study
only interviews workers and experts in related industries.
In the future, it is better to add a large number of con-
sumer opinions in this part, and further subdivide into (a) Raised pilothouse yacht A
statistical data of different regions, and even join the age
segment for analysis. It can better reflect the market ori-
entation, assist in the decision-making of the design, and
then the accurately position in the market can capture con-
sumer curiosity and inspire desire of purchase.
In this study, some blended feature curves are seg-
ment of the initial curves. This research method does not
provide a feature curve that can be more automatically
(b) Luxury leisure yacht B
selected. In contrast, it is necessary to manually select the
required curve segments. In this aspect, the computer pro-
gram of the platform can add the ability about selecting
partial curves, making the whole process more automatic
and convenient. In addition, each software and platform
are independent on each other at present. When data is
processed, it is often necessary to switch between several
software. If the platform for calculation can be rewritten
as a plug-in for a forming software or doing operations
(c) Luxury sport fisher yacht C
inside the CAD software, it will be more convenient from
blending to the forming process.

52 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

(d) Luxury leisure yacht D

(b) Taiwanese salmon

(e) Sailing support yacht E

(f) Luxury leisure yacht F (c) Taiwanese Leopard cat

(g) Luxury leisure yacht G

(d) Hakka pattern

(h) Luxury leisure yacht H

A.2 Taiwan’s feature pictures used in the survey

(a) Butterfly (e) Taiwanese aborigine pattern

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 53

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

References
[1] QYResearch Group. Global Luxury Yacht Market
Professional Survey Report 2017. [R]. Wise Guy Re-
ports., 2017. Retrieved from
h t t p s : / / w w w. w i s e g u y r e p o r t s . c o m / r e -
ports/887080-global-luxury-yacht-market-profes-
sional-survey-report-2017
[2] Raphael Montigneaux and the editors of Show Boats
International. Annual Report: 2016 Global Order
Book. [R]. 2015. Retrieved from
https://www.boatinternational.com/yacht-mar-
(f) Taiwanese aboriginal canoe ket-intelligence/luxury-yachts-on-order/annual-re-
ports/2016-global-order-book—28783
[3] Yo, Hsiao-Yen. Mysterious Taiwanese yacht facto-
ry dominates the US market: not to help American
OEM manufacturing but to find American to OEM
manufacture. [J]. Wealth Magazine, 2016, 499: 103-
105.
[4] Bloch, P. H.. Seeking the ideal form: product design
and consumer response. [J]. The Journal of Market-
ing, 1995, 59(3), 16-29.
DOI: 10.2307/1252116
[5] Moulson, T., & Sproles, G.. Styling strategy. [J].
Business Horizons, 2000, 43: 45-52.
[6] Page, C. & Herr, P. M.. An investigation of the pro-
cesses by which product design and brand strength
interact to determine initial affect and quality judg-
ments. [J]. Journal of Consumer Psychology, 2002,
12(2): 133-147
(g) Orchid [7] Mccormack, J. P., Cagan, J., & Vogel, C. M.. Speak-
ing the Buick language: capturing, understanding,
and exploring brand identity with shape grammars.
[J]. Design Studies, 2004, 25(1): 1-29.
[8] Birkhoff, George D.. Aesthetic Measure. [M]. Cam-
bridge, MA: Harvard University Press, 1993.
[9] Staudek, T.. On Birkhoff’s Aesthetic Measure of Vas-
es. [J]. FI MU Report Series, 1999.
[10] Hsiao, S. W. & Tian, Z. C.. Use aesthetic measure to
analyze the consumer preference model of product
forms. [J]. International Journal of Engineering and
Innovative Technology (IJEIT), 2018, 12(7): 5-11.
[11] Hui, K.C. & Li, Y.. A feature-based shape blending
technique for industrial design. [J]. Computer-Aided
Design, 1998, 30(10): 823-824.
[12] Wang, G.J., Wang, C.C. & Frank Chuang, S.H.. Re-
verse engineering of sculptured surfaces by four-axis
non contacting-scanning. [J]. The International Jour-
nal of Advanced Manufacturing Technology, 1999,
15(11): 800-809.
[13] Hsiao, S.W. & Chuang, J. C.. A reverse engineering
(h) Taipei 101
Resources: Google Images, www.google.com.tw/imghp based approach for product form design. [J]. Design

54 Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220

 Artificial Intelligence Advances | Volume 01 | Issue 02 | October 2019

Studies, 2003, 24(2): 155-171. curve. [J]. Advanced Engineering Informatics, 2018,
[14] Hsiao, S.W., Lee, C.H., Chen, R.Q. & Lin, Chien-Yu. 38: 14-26.
A methodology for brand feature establishment based DOI: https://doi.org/ 10.1016/j.aei.2018.05.008
on the decomposition and reconstruction of a feature

Distributed under creative commons license 4.0 DOI: https://doi.org/10.30564/aia.v1i2.1220 55

 Author Guidelines
This document provides some guidelines to authors for submission in order to work towards a seamless submission
process. While complete adherence to the following guidelines is not enforced, authors should note that following
through with the guidelines will be helpful in expediting the copyediting and proofreading processes, and allow for
improved readability during the review process.

Ⅰ. Format

● Program: Microsoft Word (preferred)

● Font: Times New Roman
● Size: 12
● Style: Normal
● Paragraph: Justified
● Required Documents

Ⅱ. Cover Letter

All articles should include a cover letter as a separate document.

The cover letter should include:
● Names and affiliation of author(s)
The corresponding author should be identified.
Eg. Department, University, Province/City/State, Postal Code, Country
● A brief description of the novelty and importance of the findings detailed in the paper
Declaration
v Conflict of Interest
Examples of conflicts of interest include (but are not limited to):
● Research grants
● Honoria
● Employment or consultation
● Project sponsors
● Author’s position on advisory boards or board of directors/management relationships
● Multiple affiliation
● Other financial relationships/support
● Informed Consent
This section confirms that written consent was obtained from all participants prior to the study.
● Ethical Approval
Eg. The paper received the ethical approval of XXX Ethics Committee.
● Trial Registration
Eg. Name of Trial Registry: Trial Registration Number
● Contributorship
The role(s) that each author undertook should be reflected in this section. This section affirms that each credited author
has had a significant contribution to the article.
1. Main Manuscript
2. Reference List
3. Supplementary Data/Information
Supplementary figures, small tables, text etc.
As supplementary data/information is not copyedited/proofread, kindly ensure that the section is free from errors, and is
presented clearly.

Ⅲ. Abstract

A general introduction to the research topic of the paper should be provided, along with a brief summary of its main
results and implications. Kindly ensure the abstract is self-contained and remains readable to a wider audience. The
abstract should also be kept to a maximum of 200 words.
Authors should also include 5-8 keywords after the abstract, separated by a semi-colon, avoiding the words already used
in the title of the article.
Abstract and keywords should be reflected as font size 14.

Ⅳ. Title

The title should not exceed 50 words. Authors are encouraged to keep their titles succinct and relevant.
Titles should be reflected as font size 26, and in bold type.

Ⅳ. Section Headings

Section headings, sub-headings, and sub-subheadings should be differentiated by font size.

Section Headings: Font size 22, bold type
Sub-Headings: Font size 16, bold type
Sub-Subheadings: Font size 14, bold type
Main Manuscript Outline

Ⅴ. Introduction

The introduction should highlight the significance of the research conducted, in particular, in relation to current state of
research in the field. A clear research objective should be conveyed within a single sentence.

Ⅵ. Methodology/Methods

In this section, the methods used to obtain the results in the paper should be clearly elucidated. This allows readers to be
able to replicate the study in the future. Authors should ensure that any references made to other research or experiments
should be clearly cited.

Ⅶ. Results

In this section, the results of experiments conducted should be detailed. The results should not be discussed at length in
this section. Alternatively, Results and Discussion can also be combined to a single section.

Ⅷ. Discussion

In this section, the results of the experiments conducted can be discussed in detail. Authors should discuss the direct and
indirect implications of their findings, and also discuss if the results obtain reflect the current state of research in the field.
Applications for the research should be discussed in this section. Suggestions for future research can also be discussed in
this section.

Ⅸ. Conclusion

This section offers closure for the paper. An effective conclusion will need to sum up the principal findings of the papers,
and its implications for further research.

Ⅹ. References

References should be included as a separate page from the main manuscript. For parts of the manuscript that have
referenced a particular source, a superscript (ie. [x]) should be included next to the referenced text.
[x] refers to the allocated number of the source under the Reference List (eg. [1], [2], [3])
In the References section, the corresponding source should be referenced as:
[x] Author(s). Article Title [Publication Type]. Journal Name, Vol. No., Issue No.: Page numbers. (DOI number)

Ⅺ. Glossary of Publication Type

J = Journal/Magazine
M = Monograph/Book
C = (Article) Collection
D = Dissertation/Thesis
P = Patent
S = Standards
N = Newspapers
R = Reports
Kindly note that the order of appearance of the referenced source should follow its order of appearance in the main manu-
script.
Graphs, Figures, Tables, and Equations
Graphs, figures and tables should be labelled closely below it and aligned to the center. Each data presentation type
should be labelled as Graph, Figure, or Table, and its sequence should be in running order, separate from each other.
Equations should be aligned to the left, and numbered with in running order with its number in parenthesis (aligned
right).

Ⅻ. Others

Conflicts of interest, acknowledgements, and publication ethics should also be declared in the final version of the manu-
script. Instructions have been provided as its counterpart under Cover Letter.
About the Publisher
Bilingual Publishing Co. (BPC) is an international publisher of online, open access and scholarly peer-reviewed
journals covering a wide range of academic disciplines including science, technology, medicine, engineering,educa-
tion and social science. Reflecting the latest research from a broad sweep of subjects, our content is accessible world-
wide – both in print and online.

BPC aims to provide an analytics as well as platform for information exchange and discussion that help organizations
and professionals in advancing society for the betterment of mankind. BPC hopes to be indexed by well-known
databases in order to expand its reach to the science community, and eventually grow to be a reputable publisher
recognized by scholars and researchers around the world.

BPC adopts the Open Journal Systems, see on http://ojs.s-p.sg

Database Inclusion

National Library, Singapore Asia & Pacific area Science China National Knowledge Creative Commons
Citation Index Infrastructure

Google Scholar Crossref J-Gate My Science Work