A Framework for Model-Based Adaptive Training

Training System Research - The Motivation

There are various viewpoints about what the definition of a methodology is. With respect to software systems the term methodology is often linked to project life-cycle and software development methods. The term methodology can be defined as:

  1. a system of methods and rules applicable to research or work in a given science or art; or,
  2. the evaluation of subjects taught and principles and techniques of teaching them [MOBIT 1994, W1].

A method can be defined as a way of doing anything in a regular orderly procedure. A framework in this research work is saying that the methodology must be systematic and usable in ‘real world’ situations; and also, that the methods used must be flexible for a range of industrial training solutions. The purpose and driving force for this work are industrial training needs.

The term Intelligent Tutoring Systems (ITS) is generally used to refer to software systems in the education domain (e.g., teaching programming languages, principles of electronics, mathematics). ITS’s have mainly been created by computer scientists and currently very few methodological approaches have been developed.

The term Intelligent Training Systems is used to indicate a specific emphasis on tools and techniques that are appropriate for specific job related industrial training (e.g., operator training for product assembly, process control, equipment maintenance, quality control).

This type of industrial application can demand different priorities in methods of instruction (e.g., more drill and practice training), a different focus on tools (e.g., more simulation techniques and an industrial knowledge base) and different problem specification methods (e.g., coping with changing training requirements and the lack of a formal course curriculum).

Both Intelligent Tutoring Systems and Intelligent Training Systems incorporate techniques from Artificial Intelligence or Knowledge Based Systems. In this work, the term Intelligent Training Systems is expanded to Model-based Intelligent Training Systems to indicate a principled approach for using multiple modelling dimensions. A set of model dimensions for scope, ontology, generality, perspicuity, precision, accuracy and uncertainty have been adapted from [ITSIE 1992, D7].

Creating an Intelligent Training System can be viewed as mapping from a hierarchy of training analysis levels into lower computer representation levels. This procedure should result in an application that is usable in solving a real business problem. For a technology to be commercially viable, users and managers require that the technology gets used in a verifiable, repeatable and maintainable way. For KBS technology this has led to a number of systematic life cycle methods which formalise and direct the knowledge engineering process.

A methodological approach is new for Intelligent Training Systems but is under development in other Knowledge Based System application domains (e.g., fault diagnoses, product configuration, planning and scheduling). The most widely known KBS methodology in Europe is the approach used in KADS [Tansley & Hayball 1993]. As presented in Chapter Two, the dominant theme in this work for ensuring a complete training solution methodology is based on work from the Knowledge Based System field.

Difficulties in Creating Industrial Solutions

Industrial solutions demand a coherent and consistent foundation. No one will use an application that is not solving a real business problem. The training solution methodology should be driven by training needs and should cope with multiple training objectives and multiple user profiles (note: this does not mean any type of subject for any type of trainee).

Another difficulty in creating a KBS is in adaptability. The architecture and methodology needs to be adaptable for multiple training subjects or it would only be useful for a one-shot application. It is not easy to create a KBS that is scaleable. Often a KBS is shelved shortly after the first prototype.

The training subject knowledge-base has to scale up beyond demonstrator level. Scaling up from limited subject areas to a practical system is a more important goal than trying to create a completely generalised training system (which may only be suitable for “laboratory-scale” training solutions).

The training solution methodology needs to consider that the application is to interoperate with other systems. Few problems are strictly one type or another, embedding of one technology with the other is essential. The training subject knowledge-base should be maintainable for a changing subject.

Maintenance is a crucial issue that has to be thought through before any coding. Maintenance is not just about correcting bugs – maintenance is about the mapping of changed information from the environment into defined problem spaces.

The difficulties above provide the motivation to develop a framework for model-based adaptive training solutions.
Flower Show
© | | Sitemap