V3S : a training and decision making tool for modelling

safety interventions on SEVESO sites

Lydie Edward, Domitile Lourdeaux, Dominique Lenne, Jean-Paul Barthes,

Jean-Marie Burkhardt, Fabrice Camus, Emmanuel Plot, Martine Guerrand

To cite this version:

Lydie Edward, Domitile Lourdeaux, Dominique Lenne, Jean-Paul Barthes, Jean-MarieBurkhardt, et al.. V3S : a training and decision making tool for modelling safety inter-ventions on SEVESO sites. RICHIR, S. ; KLINGER, E.,. 9. Virtual reality InternationalConference, Apr 2007, Laval, France. Presence & Innovation Lab. Laval, pp.17-27, 2007.<ineris-00976190>

HAL Id: ineris-00976190

https://hal-ineris.ccsd.cnrs.fr/ineris-00976190

Submitted on 9 Apr 2014

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinee au depot et a la diffusion de documentsscientifiques de niveau recherche, publies ou non,emanant des etablissements d’enseignement et derecherche francais ou etrangers, des laboratoirespublics ou prives.

V3S: A Training and Décision Making Tool to Model Safety Interventionson SEVESO Sites

Edward Lydie1, Lourdeaux Domitile1, Lenne Dominique1, Barthès Jean-Paul1, Burkhardt Jean-Marie2, CamusFabrice13, Plot Emmanuel3, Guerrand Martine4

University of Technology, CNRS, Heudiasyc, 60200 Compiegne, Francehttp://www.hds.Htc.fr

2René Descartes University, LEI, 45 rue des Saints-Pères, 75270 Paris Cedex 06, Francehttp://www.psvcho.univ-paris5.fr/lei/welcome.html

3Ineris - Technologie Parc ALAT A BP 2 60550 Verneuil-en-Halatte, France"Virtual Reality, Cognitive and Interfaces Service, BP 6 - 92265 Fontenay-Aux-Roses Cedex, France

Contact author:Edward Lydie, lydie.edward@hds.utc. fr

Abstract : V3S (Virtual Reality for Safe Seveso Substractors) is an ANR/RNTL project (French nationalagency for research). In this project, we aim to design a tool allowing to scenarise hazardous working situationson SEVESO sites for risk prévention, training and décision making. The tool interprets a high level task and arelated risk model. It is meant for a manager to help him/her to make décisions. The manager plays thé scénarioof an intervention and manages a team of virtual operators (associated with autonomous agents) in thé VERP(Virtual Environment for Risk Prévention) submitted to constraints. Depending on his/her décisions, thé incurredrisks are displayed in thé virtual environment. Our architecture relies on a multi-agents platform (OMAS).

In this paper we présent some of thé features of thé tool through différent scénarios. We présent théorganisational rules of our System, how it self-adapts to thé technical competencies and thé characteristics of théoperators (human foctors). We also présent thé working environment hosting our agents.

Key-words: virtual reality, safety intervention, risk management, multi-agents system, interactive physicalsimulation

1. IntroductionThé new possibilities offered by virtual

reality in terms of scenarisation and thé émergenceof knowledge engineering tools for risk analysisgive thé opportunity of developing new tools toimprove thé training and décision making for riskprévention. Many industriai applications usingvirtual reality for risk management will appear inthé near future. However, dedicated applicationscan only represent a partial solution. Generic tools,doser to thé reality hâve to be considered. In théV3S project (Virtual Reality for Safe SEVESOSubstractors) supported by ANR, we aim atdeveloping such a tool for external maintenancecompanies that intervene on SEVESO sites. Thistool has to allow to scenarise hazardous workingsituations on SEVESO sites with avatars andautonomous virtual characters, using virtual realityand artifïcial intelligence.

The tool is meant for a manager to helphim/her to make décisions. The manager andhis/her team hâve to appear in thé VERP (VirtualEnvironment for Risk Prévention) and hâve to besubmitted to its constraints.

Our hypothesis is that thé effectiveness ofvirtual reality in terms of training and décisionmaking for safety management increase whenoperators and managers see thé impact of their

décisions on thé technical, organisational andhuman system they hâve in charge. The importanceof virtual reality is undoubtedly a priority in théconstruction of a system dedicated to thé simulationof safety interventions [1] [2]. The trainee will beable to build an adéquate mental représentation ofthé process and to learn from acting and interacting[3]. Furthermore, thé challenge of our technology isto allow building customized simulations (usualprocédures, progressive dégradation of thé workingsituations, etc.), adapted to thé needs of ourindustriai partners. Indeed, taking into account théconcerted requirements of safety is a challenge forthé industrialists throughout thé chain of décisionthat marks out thé relations between thé hazardoussites and thé subtractors companies.The tool has three main functions: (i) training; (ii )scenarisation; and (iii ) décision making.

> Training toolThe manager cornes to différent possible choiceslike to train for activity préparation or to train inconstrained situation. Thé constraints could hâvetwo origins: (i) thé environment: difficultgéographie morphology, cold, windy; and (ii) théphysical or mental characteristics of thé virtualoperators (VO): tiredness, accumulated stress,hurry, hunger, alcohol.... As a training tool forpreventing risk, thé analysis and control of thé risks

on SEVESO sites hâve to be considered. Usingvirtual reality allows us to offer thé users (operatorsand managers) thé possibilités to become familiarwith thé site while displaying thé plant underopération and finally to learn how to deteet thédefaults while playing thé scénario of anintervention. Thé traînées will be able to superviseand verify essential check-points. Depending ondifférent parameters, thé proposed pedagogicalscénarios hâve to présent thé inhérent risk in thistype of sites. The parameters are thé manager'sdécisions, thé environment évolution and thé virtualoperators' cognitive characteristics. Indeed, thévirtual characters can simulate déviant behavioursin constraint situation, risky/hazardous operativemodes or reach partial goals fixed by thé procédurein an intelligent way. Learning by action will bepossible if thé traînées can efficiently and explicitlyidentify thé technical, organisational and individualfactors simulated by thé VO.

> Scenarisation platformThe manager has to be able to rebuild accidentaicases and replay thé situation to understand théproblems encountered by thé operators, to conductpost-accidental risk analysis or to do critical eventanalysis. By immerging thé final operators, théteams, thé safety analysts and thé directors in thévirtual environments (VE) especially modelled forthé training, it will be almost possible "to live" thécritical situations, thus amplifying thé operatorsexpérience and enabling them to suitably exploitthé concepts of Human Factors and thé goodpractices of Organization.

> Help in décision making toolV3S has to offer thé possibility to model théworking environment and thé selected operativemode. The virtual reality interface wil l beconnected to a blackboard presenting théadvantages and thé disadvantages of thé simulatedsolutions. With V3S, risk analysis before or afteraccidents (HAZOP, FMEA, tree of causes, tree ofevents, preliminary risk analysis, tasks analysis,etc.) could be done doser to thé field realities, byintegrating problems related to thé Human andOrganisational factors. The VERP resulting fromV3S will improve thé communication andcoordination between thé différent actors (users andsub-contractors companies) and will thereforesupport thé setting up and thé piloting of anenterprise strategy wéll adapted to thé context ofthé interventions.

In order to support thé previous threepoints, we propose a generic tool. It wil l endow théenvironment with autonomous decisionalcapacities. We don't aim at neither reproduceperfect cognitive mechanisms nor describe théoperator's cognition but we wish to simulate thébehaviours which could occur. We will simulatethé operative mode, thé évolution of thé operators,and thé. changes in thé environment. The operators

should adapt to thé environment in an autonomousway. To give them thé necessary cognitivecapacities we propose an approach based on théconcept of agents.

In a previous related work supported by ANVA Rand VIRTHUALIS1, UTC, ECI and INERIS hâvefocused their works to demonstrate thé feasibility ofsuch a tool. A first demonstrator has allowed toscenarise two virtual operators and a real managerwho hâve to intervene on SEVESO site. A multi-agents architecture (MASVERP) has been designedto manage thé decisional process of thé virtualagents, giving them autonomous capacities andorganisational behaviours [4]. The previousarchitecture permits to show thé feasibility of sucha System by developing some of thé necessaryfimctionalities (planner, agent model and taskmodel). MASVERP endow thé virtual characterswith autonomous decisional skills. The remainingwork is to enrich, complète thé model and tocombine it with physicals and risks models. Afterthis step, supported by ANR/RNTL, thé V3Sproject has started in 2006 and wil l continue until2009. CEA-LIST, SEEMAGE and industrials of thépetrochemical area and dangerous productstransport area hâve joined thé project. In this largerproject, thé rôles of thé partners are

0 UTC for thé autonomous decisionalcapacities of thé virtual agents,

0 CEA-LIST for thé simulation of physicalphenomena, thé behaviour of virtualmanikins and thé deformable objects,

0 ECI for thé human activity analysis andthé ergonomie aspects,

0 INERIS for thé safety and risksmanagement,

0 SEEMAGE for thé software intégration,

0 Petrochemical and dangerous productstransport industrials for industrial needs.

In this article we présent thé functionalities of thétool through différent examples and scénarios. Weprésent thé architecture of our System, théorganisational rules, how it self-adapts to thétechnical specificities and thé charaaeristics of théoperators (Human Factors). We also présent anexample of a related previous work from thépreliminary step.

2. State of thé Art

2.1. Virtua l Reality and Risk Trainin g

Applications dedicated to risk prévention using

1 VIRTHUALI S : Integrated Project, in thé 6thPCRD (R&D Research Program)

virtual reality are just emerging and for thé majorityare still at thé stage of laboratory applications. TheNIOSH2 and FIOH3 first expériences published in97 were related to handling and high working[5] [6]. More recently, VTT4 worked on aprévention measure application on industrialequipment [7]. In thé domain of ergonomics,différent researches stjidy thé sensorimotor andcognitive human behaviour and thé analysis of théworking situations and tool design [8][9] . Let usmention finally thé domain of professional trainingwhere many studies and/or applications cases aredone about safety [10] [1I][12] . Thèse différentexamples highlight thé existence of an émergentneed for VERP for thé professionals. In thiscontext, INRS started a multidisciplinary project in2002, called EVICS5, in order to evaluate thécontribution of virtual reality in thé domain oftraining addressing professional risks [13] and theconception of safe system [14]. In term ofapplications for safety we can also quote SécuRéVi(Security and Virtual Reality) developed by ENIB[12]. The project helps training firemen to theoperational management and command.VIRTHUALI S is an Integrated Project, in the 6thPCRD (R&D Research Program). The main featureof the VIRTHUALI S proposal is to produce newknowledge and, as such, it is responding to theundeniable need of transforming industry towardshigh-added value organisations, i.e., moreknowledge-based ones. VIRTHUALI S proposes thedevelopment of a new user-centred technologycoupled with advanced safety methods and aspectsthat can effectively be applied in industrialapplications handling hazardous materials. TheVIRTHUALI S technology wil l make it possible:Some partners of the V3S project are participatingto the VIRTHUALI S project (INERIS, ECI) andsome links exists on safety and human factorsaspects.

2.2. Makin g décision and humans factorsapplication

This research orientation relates to thétechnical or methodological tools of decision-making in thé préparation and thé management ofsub-contractors interventions in thé field of high-risk industry, highlighting thé taking into account ofthé human factors.The expression "Human Factors" is an umbrellaterm for several areas of research that includehuman performance, technology, design, andhuman-computer interaction. The approach focuses

NIOSH - National Institute for Occupational Safety and Health(USA).3 FIOH - Finnish Instituée of Occupational Health (Finland).4 VTT : Technical Research Centre of Finland5 Virtual Environment for Safety System Conception

on how people interact with products, tools,procédures, and any processes likely to beencountered in thé modem world. It regroups ail théindividual and socio-technical factors belonging tothé working process. Human Factors practitionerscan corne front a variety of backgrounds; thoughpredominantly they are Psychologists (Cognitive,Perceptual, and Expérimental) and Engineers.However, Designers (Industrial, Interaction, andGraphie), Anthropologists, Technicalcommunication Scholars and Computer Scientistsalso contribute. It is interesting to combine ailapproaches proposed by thé researchers since theyare complementary on thé theoretical andmethodological levels [15].Among thé various approaches for safety, one candistinguish between two main groups. On thé onehand, thé reactive approaches taking place after acritical event in term of health, safety andenvironment (industrial disease, incident, accident,major accident); on thé other hand, thé pro-activeapproaches to anticipate thé occurrence ofpotentially critieal events. Thé connection betweenvarious approaches is today a major stake of théresearch in order to find ways to improve théprévention and thé reliability of thé workingsystem, and of thé environmental risks.

2.3. Behaviour modelling tools

> STEVESTEVE is a référence in thé domain of behaviourmodelling tools [16][17]. It is an autonomous,animated agent that lives in thé virtual world withstudents. Its objective is to help students learn toperforai physical procédural tasks. It candemonstrate tasks explaining his actions, as well asmonitor students performing tasks, giving helpwhen they need it. He has a cognitive mono-agentarchitecture based on SOAR [18] which allows himto know thé state of thé environment in real time, todécide what actions to undertake and how to reachhis goal. In thé MRE project [19] STEVEarchitecture is used to build an application forpeacekeeping, it allows virtual characters to cohabitwith humans.

> IRISAIRISA has developed a large number of tools forbehavioural modelling like HPTS++ or SLURGH[20][21]. HPTS++ is a behaviour modellinglanguage for autonomous agents. The agents areorganised in a hierarchy of automata. SLURGH is ascénario modelling language. It allows managingthé scénario data and also thé dialogue between thécharacters (actors). It créâtes a determinist scénarioin which thé actors hâve to share thé resourcesusing HPTS.

> GRIC-GRAALDeveloping a tool for training fïremen aims atkeeping them out of danger. To do so, thé authors

designed a virtual environment in which virtualfiremen interact, driven by a human operator. Thearchitecture of thé system is based on a multi-agents system composed of emotional and reactiveagents [22]. The agents hâve a goal to achieve; theyuse a Prolog planner to déterminé their actions. Theplan is included in a file thé agents can access.

> MASCARET (ENIB)The physical environment represents a plant wherethé exercise takes place and also includes physicalphenomena that can take place in thé plant (fire,smoke, water spreading...). The traînées play thérôle of thé différent group managers who interveneduring an incident and thé traîner participâtes to thésimulation as a troublemaker. He can createdysfunctions, help thé traînées and, as a player, playa role in the team. The MASCARET model isproposed in order to organize thé interactionsbetween agents and to give them reactive, cognitiveand social abilities to simulate thé physical andsocial environment [12].

2.4 Multi-Agent s System Approach

Proposed in thé 80' s, multi-agents Systems nowappear frequently in research activities. A multi-agents system (MAS) is composed of a number ofagents that are interacting, cooperating andbelonging to an organisation. We distinguish twokinds of agents: reactive and cognitive. Reactiveagents only react to a stimulus (intern and extern)and do not use an internai symbolic représentation.Cognitive agents are able to build their ownbehaviour and hâve a full représentation of théenvironment. But this distinction tends to disappearin hybrid agents, a mix of both species.

Multi-agents Systems are efficient to buildSystems where thé notions of coopération,organisation and autonomy are crucial.GRIC/GRAAL and MASCARET experiment thisapproach using emotional, reactive and rationalagents.

2.5 The Positioning of Our Work

Contrary to thé existing work, our approach isdifférent. Indeed, proposed approaches reliesgenerally on an architecture based on informaticsfoundation (automates, Pétri network, expertsystem). Sortie works propose to build Systemswhile iaking into account cognitive behaviourmodel [23] [24] for virtual human animation forexample [21].

In V3S, our foundation relies on cognitivemodels in thé domain of safety and humanbehaviour in risky situations [27] [28]. From thèsemodels, we propose new mechanisms to representhuman decisional process and human errors fïnallysimulate in a virtual environment. One of our added

values is to propose tools integrating artificialintelligence to (i) analyse thé human process and(ii ) generate errors.

3. Technical and Scientific Objectives

3.1. General Objective

The project aims at exploring anddemonstrating thé impact on thé operators trainingof a virtual reality tool including Human Factors inplanning and in decision-making, in thé context ofhigh risk industry.

The originality of our project dwells in thécoupling of knowledge engineering models withvirtual environment for modelling safetyinterventions on a SEVESO plant {Figure 1).Knowledge engineering allows managing théknowledge in thé différent scénarios and provideswith organisational and human generic databases.

The development of a generic tool in V3S isintended to integrate this knowledge in a riskcalculator based on a computer model and tosimulate thèse risks at a decisional level in a virtualenvironment. Another objective, at thé physicallevel, is to make thé scénarios as realistic aspossible by animating virtual characters and givingphysical behaviours to thé manipulated virtualobjects. This coupling wil l give thé opportunity tointegrate new expérience feedbacks and new plantsin a realistic simulation automatically. The tool wil lbe used by several industrials and SMEs of thépetrochemical area and dangerous productstransport area.

JftA* 'jHtfiwft:.?-jîj-wJ».. . •. . _&J

Figure 1 Coupling Knowledge Engineering andVirtual Reality

3.2. Our Architectur e

To achieve our goals, several components hâve tobe considered and developed (Figure 2).

0 A framework is necessary to model anddisplay thé scène and thé éléments needed by théscénarios like equipments, tools and characters. Thevirtual scène will be built from CAD model. Thecomplète simulation will be displayed in real-timein thé VERP, to increase thé realism for thé user. Itwil l be possible to navigate in thé scène, forexample by moving around an élément(SEEMAGE).

0 A physical layer to simulate thé physicalphenomenon implicated (objects and virtualcharacters) in thé virtual environment. The modulehas to offer a broad range of elementary behaviours(CEA-LIST).

0 A scénario catalogue grouping thé risked-situations to process; such scénarios include forexample a description of thé working environmentand thé supervised chemical process, a descriptionof thé rôles, or thé task to achieve and théprocédure to follow in a face to face operatingmode, and eventually in a degraded mode or duringparticular critical phases (start, end, etc.) (UTC).

0 A behavioural module according to théenvironment of thé virtual characters and physicalobjects. The module aims at simulating physicalphenomena such as smoke or dangerous liquidproducts propagation, virtual characters graspingtools, Connecting a flexible, etc. (CEA-LIST).

0 A decisional and diagnosis layer dealing withsafety aspects. The decisional module has to givethé virtual characters autonomous decisionalprocesses. We intend to propose a generic modelwhich would enable to easily direct, in a virtualenvironment, thé behaviours of thé operatorsintervëning on a high-risk industrial site and thévariations on thé physical environment. Thediagnosis module has to display thé pedagogicalfeedbacks. This module has to identify thé traînéebehaviours, and to associate thé causes and théappropriate feedbacks (UTC).

0 An interface module linking thé decisional andthé physical layer (CEA-LIST, UTC).

User

ScénarioCatalogue

OnlineModifications

r

/r • Virtua l EnvironmentX , Storyboarding

f i i il 1 1 1 '

f- } Use/modify C J Layer Module

Figure 2 V3S Architecture

3.2.1. Decisional Layer

This layer relies on a multi-agents system with théfollowing entries:

0 A standard model of thé activity. The model isbased on a model describing thé cognitiveactivity elaborated in thé APLG project (RIAMprqject, [25], [26]). This model results fromergonomie, cognitive and artifïcial intelligencework (ECI). Some partners (ECI, UTC) of V3Sare participating to PERF-RV26 and someconnexions exist on thèse aspects and about itsuses for virtual human scenarisation..

6 PlatEfoRme Française de Réalité Virtuelle, frenchVirtual Reality Platform

0 A risk model. The mode! will take into accountail thé risk factors defmed with thé INERISdatabases which will permit to generate data linkto thé technical, organisational and humanfactors.

The main work of UTC is to rely upon thé capacityof thé agents to take their own décisions in anautonomous way. They should be able to:

0 Generate a plan and assume différent tasks

0 Adapt and react according to thé changes in théenvironment

0 Move in a realistic way

We assume that a Multi-Agents System (MAS) is asolution to reach thèse three points in terms oforganisation, coopération and planning [4], Weconsider three types of entities in thé environment:reactive, cognitive and human. Thé reactive entitiescorrespond to thé objeçts in thé environment, thécognitive entities are our workers, and thé human isthé manager who interacts with thé VERP.The objects hâve différent behaviours but they onlyhâve to react to a spécifie action, a stimulus. Forexample: if thé scaffolding is knocked down, then itshould fall. They do not need to hâve a fullreprésentation of thé environment and thereforethey are considered reactive. The operators aremore complex as we can see on Figure 3 that showsthé architecture of our System. We should considertheir physical and mental charàcteristics accordingto thé COCOM model of Hollnagel [27] [28] .Charàcteristics could be permanent (pm) orprogressive (pg): safety (pm), prudence (pm),tiredness (pg) or stress (pg), etc. Thèsecharàcteristics are placed in thé memory of théagent and are used to determinate in real time mebehaviour (control mode) adopted by thé agent [29][30]. Such charàcteristics are taken into account toconstruct a cognitive activity model.

Hère is thé originality of our project; actually,depending on their personal characteristics they willhâve différent ways to do thé same task without orwith risk.) The way is determinate by theirdecisional motor (planner) according to severalparameters. The planner will take into account thétask model, thé risk model to provide a plan to dieoperators. It will be use also in thé diagnosismodule (Figure 4).

I tiginlivc Agent (virtual characters)

Stade basis brick

Updater Module

Decisional Module

Production

Modify

Use

Figure 3 Agent Architecture

Figure 4 Planner Working

3.2.2. BehaviouralLayers

and Physical

In thé scope of thé V3S project, CEA-LIST partneraims at developing interactive simulationcomponents. Thèse modules wil l help to simulate ina simplified way some of thé physics involved inthé scénario:

0 The motions of thé objects in thé scène0 The manipulation of pipes

0 The propagation of fluids (smokes,dangerous liquid products

0 The behaviour of virtual manikins (avatarsof thé real operators involved in thé tasks)when physically interacting with thé scènealong the scenario

Some interactive simulation components aredeveloped in PERF-RV26 and SYSTEM@TIC7

project and will be integrated in this module.

Simplified propagation of fluids will be developedaccording to international récent work[31][32][33J[34][35][36][37][38][39][40].

Uatmdiy I | MxUa

Figure 5 Behavioural and Physical LayersArchitecture

4. Organisational Rules and Rôles

Usine numérique : competitiveness clusterSYSTEM@TIC PARIS-REGION

In this project, we want to describe and simulate thécognitive mechanisms of human operators. We willtherefore implement thé physical behavioursrealistically and in particular thé cognitive décision.The operators represented by thé agents evolve inan organisation, cooperate and aim at reaching acommon goal. They are directed by a manager andtherefore submitted to his orders. We describe théstructural organisation as follows: in théorganisation, an agent can play one or more rôles,to déterminé which agents typicaUy need to ittferactwith others to exchange knowledge and coordinatetheir activities. Thèse interactions occur accordingto patterns and protocols dictated by thé nature ofthé rôle itself. The cognitive mechanisms arereflected by thé décisions taken by thé agentaccording to thé manager orders and théenvironment (other agents and resources) state(Figure 6).

Figure 6 Sructural Organisation

The virtual environment can be viewed as anorganisation to which will be connected" thé muiti-agent system. A trace of thé environment is needed.It is an open environment. The environment shouldbe accessible, dynamic, non-deterministic andcontinuous. Obviously, it is a complex system.

5. An example of a related previouswork

5.1. Context of study

Previous work in die same field had been led inUTC with INERIS, ECI (University of Paris V) anddie Polytechnic University of Milan. Partnersdeveloped a first scénario representing an opérationof pipe substitution in VTRTOOLS environment.The agents evolve in a virtual environmentrepresenting a SEVESO site modelled with 3DTools, depending on thé scénario. It consists in apure animation of succeeding scénario steps {figure9). We can represent a very realistic environmentand obtain an excellent visual rendering. Wim thédifférent virtual interfaces, our system takes anotherdimension. The user is merged into théenvironment, in a subjective way. This highlightsthé utility of virtual reality in such a project and at ahigher level thé utility of an advanced tool fordeveloping 3D environments.

5.2. Description of thé scénarioThe task model is thé heart of our system {Figure 8,a, b). It détermines which actions can be done,whjch interactions and simulations are needed.

\\P. S57PS 3:V :rt !|>e^:tft

ss,..|

1

ï^ '

It describes thé activity of an operator in idéalconditions and also in degraded conditions (missingtime, imprudent behaviours, safety behaviours,tiredness, etc.).

5.3. An example : scénario progress

We hâve two agents, a cautious (AÏ ) and animprudent one (A2). The manager orders Al tosecure thé area by installing markers and A2 touninstall thé pipe.AÏ and A2 générale a plan (tasks séquence) toachieve their goal. As A2 is imprudent he wiil notwait for AÏ beaconing to go on thé scaffolding. Hedoes not fasten himself and start to unscrew thépipe. He does not succeed, so he décides to go andlook for a grinding machine. Due to his imprudenceand his lack of knowledge, he does not verify ifthere is still residue in thé pipe and therefore apossible risk is a liquid flow on thé floor. If a fork-lif t truck driver passes by, he can slide.To prevent this danger, thé safety barrier is thé wayAÏ will beacon thé area. If thé manager orders aspécifie configuration, he has to follow it otherwise,as he is cautious he wil l beacon thé area in thésafest configuration.

Figure 7 (a) fB) Cognitive Task Model Saraple

Figure 8 (a) (b) Scénario screen-shots

6. Conclusion and Perspectives

In this paper we detailed thé goais and(jnterests ofthé V3S project We are developing" a system to

simulate thé operative mode of operators workingon a SEVESO site based on a multi-agentsarchitecture: MASVERP (Multi-Agent System forVirtual Environment for Risk Prévention). Aninteresting part of thé project is to work along anew axis: taking a model of risk, a high level modelof tasks and a cognitive activity model into account.The goal is to allow managers to visualize théscénario and to view thé effects of their décisions inthé environment. To achieve this goal we will use avirtual support helping us to give a realistic visionof thé process.

Besides, in V3S, this approach wil l allow to interestbetter more into human factors:

0 Individual (technical, organisational, humanand prescribed aspects) compétences

0 Collective compétences: understand théprocédures, thé tasks, thé working rules, thérôle of a working group, thé technical orprocédural safety barriers, work thécommunication, thé management of an action,thé building of a self reliance and a permanentcoopération.

For thé trainers, V3S will allow:

0 Quickly conceiving training programs focusingon spécifies industrial need (based on feed-backs, accidents analysis, potential risksanalysis, tasks analysis audit results),

0 Measuring thé effectiveness of théapprenticeship.

Our environment is not only a training environment(VET), it is a VERP, standing for VirtualEnvironment for Risk Prévention including trainingand décision making. At présent, thé System offersthé possibility to simulate thé operative mode andsome risk cases. According to thé operatorscharacteristics (safety, imprudent, expert,inexperienced) thé System simulate thé différentassociated behaviours.The results of this work are encouraging. Severalissues are left for further work, more precisely wewill :

> Exploit thé industrial data to builddifférent scénarios and risk cases

> Develop thé physical module for fluidsand cables objects and virtual manikins

> Coordinate thé différent basis modules ofour architecture

> Interface thé MAS with thé virtualenvironment

> Build a catalogue of scénarios according tothé industrial need (multiple scénariosengine)

> Cons&uct and coordinate thé différentbasic bricks pf ouf architecture

> Exploit thé industriaï data to builddifférent scénarios and risk cases

>• Develop thé training module

7. Références

f l ] Burkhardt J.M., Gounelle C, Cabon P., Virtualreality, Human factors and sqfety-related issues:thé HRTHUALIS project". Proceedings of VirtualConcept 2005, Biaritz, France, 8-10 Novembre

{2] Meilet-D'huart D. Michel G., Steib D.,Dutilleul B., Paugam B., Dasse & Courtois R.Usages de la réalité virtuelle en formationprofessionnelle d'adultes : Entre tradition etinnovation. In Proceedings of First InternationalVR-Learning Seminar, Laval, France, April 2005.

[3] Crison, F., Lecuyer, A., Savary, A., Mellet-d'Huart, D., Burkhardt, J.-M. & Dautin, J.-L.(2004). The use of haptic and pseudo-hapticfeedback for thé technical training of milling invirtual reality. In proceeding of EuroHaptics ().

[4] Edward L., Lourdeaux D., Lenne D., BarthesJ.P & al, Multi-Agents Approach for ModellingSafety Interventions on a SEVESO Site throughVirtual Reality. Virtual Concept 2006, Cancun,Mexico, 27 november-lrst December 2006

[5] Laitinen H., "Usability through ProductDesign: New Design Methods and IndustrialApplications". Research project, Dpt. ofOccupational Safety, FIOH, Vantaa, Fmland.

[6] Dotson B., Hsiao H., Chiou S., Dipasquale J.,"Occupational Traumatic Injury Prévention UsingVirtual Reality". National Occupational InjuryResearch Symposium, NOIRS Symposium,Morgantown, WV, USA, 15-17tb Oct. 1997.

[7J T. Màâtta, "Virtual environments in machinerysafety analysis", VTT Publications 516, Tampere,Finland, ISBN 951-38-6261-5,2003, 170 p.

[8] Burkhardt J.M., "Réalité virtuelle etergonomie, quelques apports réciproques", Letravail humain, t. 66, n° 1/2003, pp. 65-91.

{9] Chedmail P., Maille B., Ramstein E., "État del'art sur l'accessibilité et l'étude de l'ergonomie enréalité virtuelle", Industrial Ergonomics, n°3,2002, pp. 147-152.

[lOJ Duffy V. G., Parry P. W. N., Ramakrishnan A.,"Impact of a simulated accident in virtual trainingon decision-making performance", IndustrialErgonomics, 2004, vol 34-4, pp. 335-348.

[11] D. Lourdeaux, P. David, "SOFI: Virtual systemfor track opération training", WCRR'99, WorldConférence on Railway Research, Posterprésentation, Tokyo, Japon.

[12]Querrec R., "Les systèmes multi-agents pourles environnements virtuels de formation.Application à la sécurité civile ", Thèse de doctoraten Informatique, Université de BretagneOccidentale, Brest (France), 4 octobre 2002.

ll3]Gardeux F., Marsot J., Rolin A., "Unenvironnement virtuel pour la formation à laprévention des risques professionnels", l a

International VR- Learning Seminar, Laval, France,20-24 avril 2005, 8 p.

[14][Marsot J., Ciccotelli J., Gardeux F., "Virtualenvironment for Safe Design", 35* InternationalSymposium on robotics, ISR 2004, Mardi 23-26,2004, Paris, France, 5 p.

[15JDe la Garza C. D'une approche réactive à uneapproche proactive en ergonomie : apports à uneconception sûre d'équipements industriels et desystèmes de travail. HDR,, Université Paris 5, Paris,Décembre 2004.

[16]Rickel J., Lewis Johnson W., "Virtual Humansfor Team Training in Virtual Reality". In meProceedings of thé Ninth World Conférence on AIin Education, M y 1999, IOS Press.

[17]Rickel J., Lewis Johnson W. & Marcus T.,"Steve: A Virtual Human for Individual and TeamTraining", Information Sciences Etistitute &Computer Science Department, University ofSouthern California.

[I8] Laird J. E., Newell A., Rosenbloom P.S., « SOAR: An architecture for généralintelligence », Artificial Intelligence 33(1): 1- 64.

[19jRickel J., Gratch J., ffil l R., Marsella S.,Swartout W., "Steve Goes to Bosnia: Towards aNew Génération of Virtual Humans for InteractiveExpériences". In AAA I Spring Symposium onArtificia l Intelligence and InteractiveEntertainment, March 2001.

[20JDonikian, S., "HPTS: a behaviour modellinglanguage for autonomous agents". In InternationalConférence on Autonomous Agent, 2001.

[21]Donikian S., "Modélisation, contrôle etanimation d'agents virtuels autonomes évoluantdans des environnements informés et structurés",HDR, Université de Rennes I, Août 2004.

[221Pavard B., Dugdale J., Soubie J.L., "APragmatic Developpment of a Computer Simulationof an Emergency call Centre", DesigningCoopérative Systems. Frontiers in ArtificialIntelligence and Applications.

[231Chopra-Khullar, S. andN. I. Badler. Where tolook? Automating attending behaviors of virtualhuman characters. Annual Conférence onAutonomous Agents, New York: 16-23,1999

{24} N. I. Badler, D. Chi, S. Chopra, Virtual HumanAnimation based on Movement Observation andCognitive Behaviour Model, Computer AnimationConférence 1999, Geneva, Switzerland

[251 Lourdeaux D, Burkhardt J.M., Mellet D'huartD. & al. RIAM project, APLG Project: a library forLearning Virtual Environments. Poster Virtual

Realiry International Conférence. Laval: France,April , 2005.

[26]El-Kechaï N., Després C,A Plan RécognitionProcess, Based on a Task Model, for DetectingLearner's Erroneous Actions, In: IntelligentTutoring Systems ITS 2006, 26-30, Jhongli(Taïwan), p. 329-338, Edited by M. Ikeda, K.Ashley, and T.-W. Chan, © Springer-Verlag BerlinHeidelberg 2006, LNCS n° 4053, June 2006

[27] Hollnagel E., "The phenotype of erroneousactions", International Journal of Man-MachineStudies, 39,1-32,1993.

[28] Hollnagel E., "Cognitive Reliability and ErrofAnalysis Method", Oxford: Elsevier Science Ltd.,1998.

[29] Hollnagel, E. Human Reliability Analysis:Context and Control. Académie Press: London.1993

[30] Hollnagel Hollnagel, E. & Woods, D. D. Jointcognitive Systems: Foundations of cognitive Systemsengineering. Boca Raton, 2005.

[31] De la Garza C. (2004). D'une approcheréactive à une approche proactive en ergonomie :apports à une conception sûre d'équipementsindustriels et de systèmes de travail. Habilitation àdiriger des recherches, Université Paris 5, Paris, 1erdécembre 2004

[32] De la Garza C, Fadier E. (2004). Sécurité etprévention : repères juridiques et ergonomiques. InP. Falzon (s/d), Ergonomie, Paris : PUF

[33] A method for animating Viscoelastic Fluïds.T.G.Goktekin, A.W.Bargteil, J.F.O'Brien(Univesity of California, Berkeley). Siggraph 2004

[34] G Irving, J.Teran, R. Fedkiw. Invertible FiniteEléments for Robust Simulation of LargeDéformation. Eurographics. Symposium onComputer Animation (SCA) 2004.

[35] Animating gases with Hybrid meshes.B.E.Feldman, J.F.O'Brien, B.M.Klingner.(University of California, Berkeley). Siggraph 2005

[36] A semi lagrangian contouring method for fluidsimulation. A.W.Bargteil, T.G.Goktekin,J.F.O'Brien, J.A.Strain. Univesity of California,Berkeley Report. 2005

[37 j A Vortex Particle Method for Smake, Waterand Explosions. A.Selle, N. Rasmussen, R. Fedkiw.Siggraph 2005

[38] Fluids in deforming meshes. B.E.Feldman,J.F.O'Brien, B.M.Klingner, T.G.Goktekin(Univesity of California, Berkeley). Siggraph 2005

[39] Coupling Water and Smoke to ThinDeformable and Rigid Shells. E.Guendelmaii,A.Selle, F.Losasso, R.Fedkiw.Siggraph 2005'

{40] Melting and Burning Solids into Liquids andGases. F.Losasso, G.Irving, E.Guendelman,RFedkiw.. IEEE TVCG 2006

VRIC LAVA L VIRTUAL PROCEEDINGS 2007 - 27