A Framework for Model-Based Adaptive Training

Domain Analysis - Summary

This chapter presents distinct levels of reasoning occurring at the domain level, in the form of an expert system, within a generic training framework. The knowledge base of an industrial scheduling expert system is analysed and expanded as needed within the MOBAT framework. The Proto-MOBAT training unit attributes from the previous chapter have been extended. Further training unit design aspects are presented in terms of:

1. SOAR transfer of learning methods [van de Brug, Rosenbloom & Newell 1995] and
2. expected time duration of training units.

Both transfer of learning and training unit time estimate are attributes which can affect the specific level of generality in the domain model realisation. To avoid complex software structures, augmenting parts of the existing knowledge base is only needed if specified by training requirements. (i.e., by mapping training task features to domain knowledge within the initial framework from Proto-MOBAT). The resulting knowledge structures for training tasks in problem spaces are small and re-usable. The breakdown of existing rules into smaller sections with generic tasks and explicit problem solving methods enables a training system to reflect on what happens during problem solving. I.e., the training system can:

1. generate adequate explanations;
2. at any point recognise all possible steps a trainee might produce; and,
3. monitor transfer of learning.

The use of multiple levels of generality in domain knowledge enables a generic trainer to choose the most effective level for a trainee and dynamically switch to different levels as needed. The identified levels of domain knowledge are: procedural knowledge, associative knowledge and principled knowledge. A tool set approach is suggested for using these levels in industrial applications. The results of augmenting an existing knowledge base with distinct levels of domain knowledge provides encouragement that (for a complex industrial problem) a methodology can separate what is to be taught (the subject expertise) from how to deliver the training (the training expertise) within a generic training framework.