A Framework for Model-Based Adaptive Training

Methodology Review - Discussion

The approach in ITSIE is based on a cognitive architecture supporting both learning and adaptation based on multiple domain models. However, because ITSIE was designed from a control engineering perspective, ITSIE is still untested on a wide range of changing industrial training problems. In ITSIE, the view is that the process of training an operator for a task in an industrial environment can be done by removing the trainee from the actual environment in which they operate (the “real world”) and training them in an environment which mimics the industrial environment.

In this way the training process and the industrial process are separated. In both ITSIE and MOBIT methodologies, the domain expertise module which represents categories of expertise appropriate to the industrial domain is separated from simulation module. Simulation provides a means of replicating the behaviour of the “real world”, and, as such, provides a means of demonstrating and validating procedures, rules and principles.

As in KADS [Tansley & Hayball 1993], the issue of modelling is the basis for systematically capturing domain expertise in ITSIE. The componential expertise framework for ITSIE involves only three components: procedural knowledge, associative knowledge and principled knowledge. For each of these 3 categories a specific implementation language is available. Their utilisation is specified in the ITSIE methodology [Sime & Leitch 1993].

In emphasising executable domain models, ITSIE characterises the split between specifications and control different from the usual expert system “what & how”, “declarative & procedural” dichotomies, which often leave the underlying models implicit. A componential expertise framework can lead to consistency, tractability and completeness in the methodology of KBS design, architecture and analysis [McDermott 1990, Steels 1990]. Components of expertise promote re-usability of symbol level code chunks. In ITSIE the symbol level is created with 3 tools: Component Based Language (CBL), Rule Based Language (RBL) and Event Graph Language (EGL).

ITSIE does not separate model design at the knowledge level and symbol level. To help specify the contents and structure of a knowledge base at the knowledge level, chapter 2 introduced the concept of problem solving methods and knowledge roles [van de Brug et al. 1985]. A problem solving method that is sufficiently specific to impose various roles on domain knowledge can result in effective knowledge-base maintenance.

The roles that the domain knowledge can play in ITSIE is dictated by aspects of instructional strategies in the didactic methods, diagnostic methods, remediation methods, and explanation methods. The ITSIE approach suggests the implementation of three instructional strategies (tutoring, coaching and discovery learning) within the methods of various modules of the generic architecture.

However, no engineering methodology exists on how to sub-divide from these abstract levels into control methods for the ITSIE expert and tutor modules. The ITSIE demonstrator experiments have been based on clearly defined physical systems with a component-centred ontology. I.e., a set of component-objects represent an internal model. Key questions remaining unresolved in ITSIE are:-

  1. what are the appropriate subject areas for Model Based Training (e.g., what is the scope of an appropriate domain model) in a (changing) industrial environment;
  2. given a set of domain models, which model should be used and when should there be a change of model?
Flower Show
© | | Sitemap