A-TEAM:

Advanced Training System
for Emergency Management

8. TECHNICAL TRAINING IN A COMPLEX DOMAIN

A-TEAM aims at building an effective and efficient computer based training system for a complex technical domain. It builds on an operational emergency management system prototype, major components of which have been developed under the HITERM ESPRIT HPCN 22723 project under framework 4. Within an appropriate pedagogical framework, it adds didactic components into the same basic logical framework of the interactive decision support tool. The didactic components will, in part, be based on results from the Educational Multimedia Taskforce project MUTATE (MM1019).

The project is structured, over a three year period, into 13 workpackages grouped into six phases with corresponding Milestones:

  • Design phase the compiles user requirements and develops the functional specifications as well as a didactic/pedagogical framework; this provide the high-level design for the subsequent

  • Development phase, that builds the component elements both in terms of software tools and multi-media content; this is followed by an

  • Implementation phase that integrates the components into a common tools and approach, followed by a

  • Test phase with five parallel applications cases of industrial training, leading to an

  • Evaluation phase where the original specifications are compared with the results of the test cases, and a final

  • Dissemination phase the prepares the subsequent exploitation of the project results.

Advanced skill in complex technical domains are essential for industrial societies: more effective and efficient learning must therefor be a central objective in a information society.

The training for the management of technological and environmental emergencies faces the difficulty of a domain where realistic experimentation is impossible and personal experience is scarce. Learning-by-doing, the by far most effective method for practical skills in a very complex and ill-structured domain with an enormous range of possible situations is restricted to rather limited and somewhat artificial exercises, which also carry substantial costs. The possibilities of computer simulation combined with artificial intelligence and virtual reality offer an interesting new approach to this dilemma.

Three possible training modes are envisioned for the project:

  • Traditional classroom teaching, supervised by a tutor, where the interactive software provides one element in the course delivery;

  • Unsupervised learning which can also involve remote access through the Internet, thus opening the possibility of the flexibility for the student to choose the times she wants to access the system freely;

  • Learning on demand, which is a simple extension of the normal use of the system for problem solving, with the possibility to explore additional background and in-depth information in the context of a specific application.

The design of a learning environmetn must take into account the social, cultural and organisational processes as well as the technical processes of developing a training system. The Pedagogical Framework sets out the overarching principles, from a constructivist stance, that link together the technical components within the project: the case-based reasoning, the simulations, the additional explanatory materials, the student monitoring and administrative functions and the emergency management decision support system.

Learning is being promoted in an ill-structured, and very complex domain, where there is a high chance of novel scenarios that must be handled. There is a strong requirement for flexibility in the problem solving behaviour and in the knowledge used to produce a representation of the situation and make predictions and reason about the problem. So, the approach to training adopted within the Pedagogical Framework, will draw upon ideas from Spiro's research into Cognitive Flexibility Theory and Gardner's Performance View of Understanding.

The training will be embedded in realistic contexts and in individual experience gained from the simulated scenarios. An on-going process of reflective learning will be encouraged to help deal with the requirements of a changing environment. This type of learning will often occur in the context of social practice in co-operative, peer-groups.

The A-TEAM project will build upon the research and development, experiences, results and the lessons learned from the ETOILE project (29086). This transfer will occur through a partner in both consortia, the University of Lancaster. The University of Lancaster is involved in the analysis and design of the training and in evaluation of both projects. While these projects adopt different approaches to achieve their training objectives and have different target audiences for the training, there are commonalities in the technology, the theories and the methodologies used to design and evaluate these training systems. For example, both projects will deliver their instruction through virtual environments on desktop computers.

Specific examples of how A-TEAM may benefit from the ETOILE project include:

  • Evaluation instruments, such as questionnaires, which are costly to produce can be re- used/modified.

  • Building upon experience of particular evaluation methods such as observation of group/team learning situations.

  • Lessons learned from evaluation of the ETOILE demonstrators will be available in time to influence the evaluation of A-TEAM.

8.1 An Expert System framework

To provide a flexible didactic framework, the emergency management system is embedded in a hybrid forward-backward chaining real-time knowledge-based system (KBS). The KBS guides through the simulated emergency just as in a real emergency management application and co-ordinates the information resources, including the simulation models. At the same time, it monitors the trainee's responses, and can trigger additional explanatory material, tests and questionnaires, or modify the sequence of events such as returning to a previous stage to re-run a critical part.

The driving force behind this problem-solving paradigm is a real-time expert system (RTXPS) that combines forward chaining and backward chaining elements. The rule-based approach provides a reliable and rigorous logic, that is open to inspection, and can be built on the basis of the best available scientific knowledge, professional experience, and combining numerous sources.

The formalism of a rule-based knowledge based (expert) system provides the necessary rigorous and disciplined operational guidance, as well as the context sensitive flexibility essential in any emergency situation.

The real-time expert system is designed as a sequence of rule-based tasks or ACTIONS. They provide a context sensitive and dynamically evaluated sequence of instructions and information items to the operator, including requests for information, requests for external communication, instructions derived from the knowledge base and various data bases such as Material Data Safety Sheets (MSDS) for hazardous substance, and model results. ACTIONS trigger the various models or requests for external information (sensor data, imagery) autonomously and thus free the operator from the technicalities of working with a complex analytical system, so that he can concentrate on interpretation and evaluation tasks.

Since the knowledge base of the system is open for inspection, and each step of the inference procedure can be traced (WHY tracing and explain functions), laid open and criticised, the system is subject to peer review and continuous updates and improvements, based on increasing case experience. The rules driving the ACTIONS utilise the dynamically growing knowledge base for an emergency; all interactions with the system are time stamped and logged for a subsequent post-mortem analysis. The real-time nature also supports the use of timers and delays for the synchronisation with external events.

The system features the two interlinked KBS strategies which draw upon an object data base of risk objects and a GIS, as well as a set of simulation models implemented in a distributed client- server environment, that includes links to real-time data acquisition from remote sensors. The RTXPS framework maintains the dialogue with the user, e.g., an operative in an incident command centre. The real-time expert system controls communication with the various actors involved in an emergency situation, provides guidance and advice based on several data bases including Material Safety Data Sheets for hazardous substances, and triggers various simulation models for the simulation of the evolution of an emergency and the prediction of health and environmental impacts.

The expert systems compile all necessary input information for the models and perform checks for completeness, consistency, and plausibility. They then trigger, based on the available information and some simple screening and ranking methods, the most appropriate model or set of models, interpret the results, and translate that into guidance and advice for the operators.

Real-time control and logging of data availability, user inputs and decisions, model results, and communication activities provide the basis to use the system for operational management, but it also provides the possibility to monitor a learner's response and analyse performance in a post-mortem analysis.

8.2 Simulation and artificial intelligence

The A-TEAM training system is designed to support the training requirements of the Seveso II Directive, and focuses on chemical process plants as well as related transportation of hazardous material. Different classes of simulation models will be used, including release and evaporation models to define the dynamic source terms of chemical releases that can then lead to either

  • fire;
  • explosion;
  • atmospheric dispersion;
  • infiltration into soil, ground, and surface waters,
  • and any combination or sequence of the above.

    To represent the evolution of an emergency, two closely related methods will be used:

    • Numerical simulation modeling, and
    • Case based reasoning.

    8.2.1 Simulation modeling

    For the atmospheric dispersion cases, meteorological pre-processors will be used to generate the dynamic wind fields required. The models to be used will include the industry standard PHAST/SAFETI risk analysis package including: the Unified Dispersion Model (UDM) which integrates, in a seamless way, hazardous material discharge, droplet formation and evaporation, pool vaporisation, jet dispersion, heavy gas dispersion and passive dispersion, and a variety of fire, explosion and toxic effects models.

    A second set of simulation runs for conditions of complex and highly structured terrain will be using state of the art coupled modeling methods. The mesoscale to microscale coupling method consists of a three-dimensional spatial interpolation scheme, a special adjustment of values within the surface layer of the microscale model based on physical assumptions and the formulation of the lateral boundary conditions to introduce the interpolated values into the microscale model. Properties on a grid point in the microscale grid are computed by the corresponding values of the surrounding grid points in the mesoscale grid. The vertical resolution near the ground of the microscale model is much higher than that of the mesoscale grid. Consequently, the mesoscale simulation gives only poor information about the vertical structure of the boundary layer adjacent to the ground, and the resulting interpolated profiles to the microscale grid are insufficient. Henceforth, the interpolated values are adjusted using similarity theory and a prognostic mesoscale model, which allows describing the air motion over complex terrain will be used.

    In this mesoscale framework, a prognostic microscale model, which allows describing the air motion near complex building structures and industrial installations, will be embedded. The conservation equations for mass, momentum, and scalar quantities as potential temperature, turbulent kinetic energy and specific humidity are solved in terrain-influenced co-ordinates. Non-equidistant grid spacing is allowed in all directions. The numerical solution is based on second-order discretisation applied on a staggered grid. Conservative properties are fully preserved within the discrete model equations. The discrete pressure equations are solved with a fast elliptic solver in conjunction with a generalised conjugate gradient method. Advective terms are treated with the TVD scheme. Turbulent diffusion can be described with the two- equation k-e turbulence model.

    Finally, a set of common screening level models such as multi-puff Gaussian models with terrain corrections, pool, trench and BLEVE fire models, and the TNT and TNO multi-charge fuel-air blast models for explosions will be used for extensive Monte-Carlo simulations. Due to their simplicity and fast execution, they lend themselves to sensitivity analysis and uncertainty analysis in a Monte Carlo framework.

    8.2.2 Case-based reasoning

    In many emergency situations, there will be neither the data nor the time to employ a complex simulation model. Simple empirical methods will have to be used instead. These methods are based on a two-step procedure:

    • classify the situation based on the limited data available;
    • retrieve a pre-defined solution for this class of situation from some archive of such solutions.

    To improve on this basic but simple principle, we can increase the size of the archive and the dimensionality of the classification mechanism to increase the flexibility of the available set of responses.

    The case library can be extended through simulation modeling; the classification and retrieval can improved through case based reasoning.

    A particular strength of Case-based Reasoning (CBR) over most other methods is its inherent combination of problem solving with sustained learning through problem solving experience. The basic reasoning cycle of a CBR agent can be summarised by a schematic cycle (Aamodt and Plaza, 1994):

    • Retrieve the most similar case(s) to the new case.

    • Reuse or adapt the information and knowledge in that case to solve the new case. The selected best case has to be adapted when it does not match perfectly the new case.

    • Revise or evaluate the proposed solution. A CBR-agent usually requires some feedback to know what is right and what is wrong. Usually, this is performed by simulation or by asking a human expert.

    • Retain or learn the parts of this experience likely to be useful for future problem solving.

    The agent can learn both from successful solutions and from failed ones (repair). The quality of the new case(s) extracted by the CBR-agent depends upon some of the following criteria:

    • The usefulness of the case(s) extracted and selected;
    • The ease of use of this (these) case(s);
    • The validity of the reasoning process;
    • The improvement of knowledge through experience.

    8.3 From emergency management to emergency training

    To develop an emergency training system as outlined above into a training system requires a number of additional components:

    • The integration of the didactic elements into the real-time expert system; The integration of didactic multi-media material, background and explanatory information, examples, tests, and evaluation routines into the system;

    • The integration of a student administration layer.

    The organisation of didactic material for real case courses and training programs involves considerable complexity, given the diversity and hierarchical nature of the materials involved. Combining the discipline required to have homogeneous and consistent documentation with the degrees of freedom required to have innovative and solid materials is quite a challenge. SGML has been used for two decades with good results. However its inherent complexity and the lack of easy-to-use tools using it contributed for a very limited practical use. In the past two years an heir of SGML emerged in the market: XML. XML had the simplicity that SGML lacked and got almost immediately a wide acceptance. The XML approach to didactic material development has immense advantages:

    • allows for the separation of the content development task (which can be quite time consuming) and the assembling of materials in a course;

    • enables content reuse, either by "linking" to other materials in the didactic material repository or by cloning an existing content, and adapting it to a different situation;

    • separates content from presentation and format, allowing for the adaptation of the didactic materials to different graphic styles.

    The static content elements will be based on XML/CSS, and classical relational data base technology for the maintenance, storage and retrieval of the content data. The same data base technology will be used for student administration tasks such as registration, course or test case assignment, evaluation, certification, and eventual billing.


  • © Copyright 1995-2016 by:   ESS   Environmental Software and Services GmbH AUSTRIA | print page