CARESS1

Commercial Airliner Emergency Safety System

Marvin Oliver Schneider

Business School São Paulo, Universidade Anhembi Morumbi

Av. Roque Petroni Jr. 630, 9º. Andar, 04707-000 São Paulo, Brazil

João Luís Garcia Rosa

Instituto de Ciências Matemáticas e de Computação, Universidade de São Paulo

Av. Trabalhador São-Carlense, 400, 13560-970 São Carlos, Brazil

Keywords: Airliner Safety, Intelligent Safety Systems, GeneRec, Biologically Plausible Artificial Neural Networks.

Abstract: This paper presents the project CARESS1, which is a connectionist system, using an MLP as network

structure and GeneRec as learning algorithm, with the purpose to comprehend unforeseen situations in civil

aviation and treat them in order to avoid air disasters. The importance of a new safety approach is discussed,

related work is given and a system overview is presented. Shortly the first results (in a simulation

environment) shall be obtained. The approach is a new and promising means to automatic problem

treatment and might lead the way up to the final aim of a fully automatic aircraft.

1 INTRODUCTION

The issue of airplane safety is of high financial and

emotional importance. Modern passenger jets cost

frequently around 100 Million US-Dollars (see

Reuters, 2007) and thus their financial loss is of

great impact, even if insurance is involved.

Nowadays the only plausible way from one

continent to the other appears to be by airplane. In

times of globalization, it should be deemed probable

that there will be even more airlines and passengers

transported every upcoming year.

Meanwhile a feeling of lack of safety is present

to many people inside airplanes (according to

Brown, 1996, only 6% feel comfortable). This is

indirectly documented by the series of publications,

which seek to overcome this uncomfortable feeling

such as Brown (2006), Hartman and Huffaker

(1995) or Krefting and Bayaz (2000), by the great

press attention given to any of the events and by

several horror/suspense films, which include aircraft

(see the long list presented by Daniel Webster

College, 2007).

Whereas the attention given to rare fatal

accidents is large and immediate (in such a way to

not even wait for confirmation as with Air France

flight 358 to Toronto on October 2

, 2005

commented in Toronto Star, 2005), the attention

given to constantly present fears is minimal.

Yet, the well-known statistical result of an

airplane being safer than any other way of

transportation is actually not a universal truth. Weir

(1999) raises a lot of critics concerning the way

statistics are conduced and set and he identifies them

as a way to blind the ordinary passenger by

manipulating the parameters in a way to provide the

desired result.

It becomes clear that safety is one of the most

important issues on a jet plane and that every effort

should be taken to achieve it, so that this form of

transport can not only be considered statistically

more safe, but actually a reliable means of transport

- also emotionally speaking.

2 JUSTIFICATION

The following fatal characteristics may be found in

many airliner accidents (found in a series of crashes

as seen in Transportation Safety Board of Canada

(1998), Folha Online (2008) or Terra Notícias

(2005)):

223

Oliver Schneider M. and Luís Garcia Rosa J..

CARESS1 - Commercial Airliner Emergency Safety System.

DOI: 10.5220/0003568502230226

In Proceedings of the 8th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2011), pages 223-226

ISBN: 978-989-8425-74-4

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

• System malfunction (wrong path, unsuccessful

detection of the approach of danger);

• Need for rapid pilot decision (reprogramming

or evasion manoeuvres);

• High physical and emotional stress

(reprogramming in a hurry, G-forces after the

collision);

• A possibility to avoid the accident or its major

consequences, if conducted differently;

• Different behavior of the pilot is at least

difficult to impossible in the present situation,

whereas a mechanical intelligent approach

could be fast enough or accurate enough.

Out of the following reasons an automated

approach is probable to produce better results:

• It is not subject to emotional stress and stays

analytical and focused even in extreme

situations;

• It is less subject to physical “stress”

(movement, G-forces), if mounted properly;

• It may be much faster than a pilot;

• It is not subject to negligence inside its scope, it

has no corporal needs (as sleep on long distance

flights).

The aim of the present approach is to develop

CARESS1, a project, which shall lead the way to a

safety system, intended for commercial passenger

aircraft. CARESS1 shall present the following high-

level technical benefits:

• Automatic reaction on imminent extreme

dangers, overriding the pilot’s controls, if

applicable;

• Alert of dangerous situations (mechanical

failure, weather, wrong decisions etc.);

• Enhancement of automated navigation and

treatment of turbulences.

Thus the following non-technical benefits are

sought:

• Development of a feeling of high safety and

reliability;

• Help in the treatment of the fear of flying.

3 RELATED WORK

Current approaches might be divided into common

approaches (which may already be partially or fully

implemented) and novel approaches, which still

need research to reach the level of implementation.

The following shall give a short overview of some

examples.

Several collision avoidance systems were

developed and enhanced, as shown in Williams

(2004). As he stresses, trace goes back to a 1956

crash over the Grand Canyon and systems have been

improved ever since, including versions for ground

or air collisions, sounding alerts and radars. There is

yet no integrated approach in production, which

could automatically evade this type of crashes.

Waterman (2002) gives the interesting idea to be

able to override the controls of the cockpit from

outside the plane (i.e. from the tower or a mobile

station) and thus be able to lock the plane on its

course in case o hijacking. This might also be used

in case of an imminent disaster when the pilot is

unable to solve a problem alone. It, however, does

not address the issue of the speed of reaction.

The following approaches may be considered

novel and more in the scope of CARESS1. They are

currently under development, organized by the

NASA (listed in National Aeronautics and Space

Administration, 2008) and elaborated with the help

of university and industry researchers:

The Integrated Intelligent Flight Deck (IIFD)

shall offer an optimized access to controls to the

pilot as well as establish good awareness of the

aircraft condition. It shall sense internal and external

hazards and offer key information for the solution of

the problem.

The IRAC (Integrated Resilient Aircraft Control)

project is an implementation of on board systems

which shall guarantee manoeuvrability and stability

margins in case of the presence of sudden adverse

conditions (such as structural damage, control

surface failure, icing, aerodynamic problems). It

relies on integrated multidisciplinary aircraft design

tools. Math models are used to model the

interactions between control inputs, trajectory

planning and guidance and the aircraft structure and

propulsion systems.

The project Integrated Vehicle Health

Management Project (IVHM) has its focus on

automated detection, diagnosis and prognosis which

enable mitigation of adverse events during flight. It

is different from IRAC in the sense of focusing more

on the hardware and software situation of the

aircraft. Special importance is given to a proper

software analysis, for which the methodology has

yet to be developed. One of the main outputs shall

be the remaining useful life (RUL) of equipment.

Data shall be shared and mined in order to allow a

broader analysis and prevention.

CARESS1 differs from these approaches in two

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

224

main points:

CARESS1 is a connectionist approach and as

such a learning system, which is not fully pre-

modelled. It is to learn over time from its own

experiences and share them to others.

CARESS1 as a project follows a “from inside

out” approach, i.e., firstly a core is modelled with

few typical sensors and actuators (and without any

demand of completeness). Afterwards the system is

“broadened” to attend a list of the above needs.

Final objective of CARESS1 is the fully

automated aircraft.

4 OVERVIEW

The purpose of CARESS1 is to treat incidents,

which typically pose extreme difficulty on airplane

pilots and cannot either be treated by automated

means at the present point of development.

The main problems associated are:

• Air traffic: May be at crash course with the

plane and – depending on the angle or the day

light – may not be naturally seen by the pilot. In

this case, malfunctioning or deactivated

collision detection systems are an extreme

danger.

• Weather conditions: Bad weather can render the

jet uncontrollable or cause structural damage

with the respective consequences. Pilots may

not be able to cope with the weather nor the

consequences after.

• Own health status: A system of the aircraft may

suddenly fail, an engine might be lost, there may

be cracks in the structure etc. etc. A system must

be aware of such failures and promptly provide

the solution.

• Control info: Ground control may pass

important data, which is not correctly registered

by the pilot. A system must have a means to

receive information and report constantly to the

ground.

• Ground in different altitudes: A crash may be

caused even at altitude in case of an upcoming

mountain peak. This has to be fully treated and

seen by a system, at least to the point of evading

the obstacle in due time.

In its first version, CARESS1 is presented in the

form of a simulation, implemented in Java. It may be

obvious to say that the way to a commercial version,

used on board a commercial jet liner, is non-trivial,

but appears feasible. In the beginning of this project,

the focus remains on the implementation of the core

system, which shall provide a basis to judge the

feasibility of the proposed core architecture for the

tasks involved.

This means that the following elements are

currently being implemented: A measuring module

for typical aircraft sensors, a translation module for

the serialization of information to the Multilayer

Perceptron input layer, a core module with a

Recurrent Multilayer Perceptron architecture and the

use of the GeneRec learning algorithm, a translation

module for the de-serialization of information from

the MLP output layer to typical aircraft actuators

and finally an action module, which controls the

immediate action to be taken in the event.

After the definition of the core module and the

respective initial universe of sensors and actuators,

several tests in simulation shall be executed to

guarantee good function, defining sequences for

normal operation, small issues and potential hazards.

Learning shall be verified. Completing this phase,

basically sensors and actuators shall be extended

whereas the main algorithm of the system should

remain stable with only few changes.

Importantly, it shall be observed that – whereas

the actuators currently in use in an aircraft should be

almost unchanged – a series of new sensors should

also be physically implemented over time to ensure

self-awareness of the airplane. This is oriented at the

nervous structure of the human body, laying sensors

all over the plane, its fuselage (skin) and its

equipments (organs). In order to keep the wiring

low, it is suggested to establish a data-bus via fiber

optics and lead all the data to the main instance, the

server with the system (brain).

Concerning the network core architecture the

following might be said:

The Multilayer Perceptron is a standard and

easily implemented Artificial Neural Network with

Boolean inputs, at least one hidden layer of neurons

and a layer of real type outputs, which provide

values between 0 and 1 (see Haykin, 2008). A

recurrent structure (i.e., output values and next input

values are transformed into the definitive input

values) is necessary in order to work sequences and

not mere pairs of input and output.

Learning is done via GeneRec, which is a

supervised learning algorithm, considered to be

more biologically plausible. Its good function was

shown in practice in Schneider and Rosa (2009). The

algorithm generates two signals for learning: the

expectation of the network, called “minus” and the

training signal, called “plus”. Propagating these two

signals the error related to every point of the

network is found in order to have it adjusted. For

CARESS1 - Commercial Airliner Emergency Safety System

225

more details, refer to O’Reilly (1996).

Network topology and learning algorithm were

chosen in order to guarantee a fast and solid

approach, which may adjust more easily to future

advances in the field of Artificial Neural Networks.

5 PROJECT STATUS AND NEXT

STEPS

The project is currently in its implementation phase.

Shortly, results concerning the first project phase

may be given. Expectations are good as

architecturally similar problems have already been

solved using an MLP with GeneRec (see Schneider

and Rosa, 2009).

Next step shall be extensive testing of the given

approach in the simulation environment, bringing up

diverse and surprising situations and evaluating the

networks adjustment.

Having completed successfully, the project shall

be brought to the knowledge of the industry with the

aim to plan implementation in commercial aircraft.

6 CONCLUSIONS

This paper presented the objectives, importance and

overview of the project CARESS1, a novel

connectionist safety system for aircraft.

There are high expectations connected with the

research. It has the potential to help make civil

aviation much safer, to the extent to be truly

considered a safe way of transport.

As a final consequence, a fully automated and

secure aircraft may be developed, which may

revolutionize the international aviation sector.

REFERENCES

Brown, D. (1996). Flying without Fear. New Harbinger

Publications.

Daniel Webster College (2007). List of Aviation Movies.

http://faculty.dwc.edu/nicosia/apcm_list_of_aviation_

movies.htm.

Folha Online (2008). Maior acidente aéreo da história

matou 583 pessoas na Espanha, http://www1.folha.uol.

com.br/folha/mundo/ult94u435700.shtml.

Hartman, C. and Huffaker, J. S. (1995). The Fearless Flyer

– How to fly in comfort and without trepidation. The

Eighth Mountain Press.

Haykin, S. (2008). Neural Networks and Learning

Machines, 3

. Ed., Prentice Hall.

Krefting, R. and Bayaz, A. (2000). Entspannt fliegen –

Strategien gegen die Flugangst. Trias.

National Aeronautics and Space Administration (2008).

Aircraft Aging and Durability Project. http://www.

aeronautics.nasa.gov/nra_awards_aadp.htm

National Aeronautics and Space Administration (2008).

Aviation Safety Technical Conference. http://evt.grc.

nasa.gov/astc08/.

O’Reilly, R. C. (1996). Biologically plausible error-driven

learning using local activation differences: the

generalized recirculation algorithm. In: Neural

Computation, p. 895-938

Reuters (2007). Oman Air says to buy five Airbus A330-

200 aircraft. http://www.reuters.com/article/tnBasic

Industries-SP/idUSL0211920320070402.

Schneider, M. O. and Rosa, J. L. G. (2009). Application

and Development of Biologically Plausible Neural

Networks in a Multiagent Artificial Life System.

Neural Computing & Applications, 18:65–75.

Terra Notícias (2005). Acidente com Fokker 100 da TAM

completa 10 anos, http://noticias.terra.com.br/brasil/

interna/0,,OI1212450-EI306,00.html.

Toronto Star (2005). Pearson disaster was avoidable.

http://web.archive.org/web/20060718163404/http://w

ww.thestar.com/NASApp/cs/ContentServer?pagename

=thestar/Layout/Article_Type1&c=Article&cid=1128

204610920&call_pageid=1020852958068&col=10208

52955095.

Transportation Safety Board of Canada (1998). Aviation

Reports – 1998 – A98H0003, http://www.tsb.gc.ca/

eng/rapports-reports/aviation/1998/a98h0003/a98h00

03.asp.

Waterman, S. A. (2002), Airplane Security System –

Patent 6691956, http://www.freepatentsonline.com

/6691956.html.

Weir, A. (1999), The Tombstone Imperative – The Truth

About Aircraft Safety, Simon & Schuster Ltd.

Williams, E. (2004), Airborne Collision Avoidance

System. In Proc. Ninth Australian Workshop on

Safety-Related Programmable Systems (SCS 2004),

Brisbane, Australia. CRPIT, 47. Cant, T., Ed. ACS.

97-110.

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

226