A Framework for Model-Based Adaptive Training

Multiple Models - Summary

In this chapter the specification of an industrial training problem with a set of qualitative modelling dimensions has been adapted from the European Community ITSIE project [ITSIE 1992, D7]. The Qualitative Modelling (QM) community has made substantial progress in identifying model dimensions for reasoning about physical systems. The recognition of multiple modelling dimensions provide benefits:

• (a) this copes with complexity in industrial training domains; and,
• (b) to a generic trainer agent this supports different ways of providing instruction by varying and switching models. The modelling dimensions adapted and refined for the MOBAT framework are:
  1. Ontology – the source of knowledge;
  2. Scope – the part of the domain within which training takes place;
  3. Generality – how broadly the model can be applied;
  4. Perspicuity – the ease-of-use of the model representation;
  5. Precision - the exactness of the detail in the model description;
  6. Accuracy – how near the model description is to a reference model; and
  7. Uncertainty – the belief in the model description.

A demonstration system has been developed which incorporates and extends each of these seven modelling dimensions within the MOBAT framework. The training application in this chapter is to qualify manufacturing operators to a level where they can inspect printed circuit boards to acceptable workmanship standards. The results indicate that a set of explicit modelling dimensions can be used by a set of generic trainer tasks which enables training delivery to be highly adaptability to changing needs of individual trainees.

Introduction

The domain of training application in this chapter is Printed Circuit Board manufacturing workmanship. The training purpose is to qualify manufacturing operators to a level where they can inspect printed circuit boards to acceptable workmanship standards. The workmanship application experiment uses the training system architecture as shown in Figure 1-1 [Slater, Brown, van de Brug & Brown 1994]. The aspect of research relevant for the training framework is called Workmanship-MOBAT. This research is extending the training framework shown in Figure 6-1 with qualitative modelling dimensions which are used by a set of generic trainer task primitives and trainer methods.

Figure 6-1 Workmanship-MOBAT Framework Overview

The Qualitative Modelling (QM) community has made considerable progress in identifying model dimensions for reasoning about physical systems. Powerful qualitative reasoning methods can capture the distinctions and laws which govern behaviour of physical systems. Early QM methods have perhaps not been practical because of restricted modelling viewpoints with the use of implicit or simple problem solving methods that avoid sophisticated inference techniques. Recent research in QM has focused on using a set of multiple modelling dimensions (e.g., [Weld 1990, White & Frederiksen 1990, ITSIE 1992]). In this research the specification of an industrial training problem with a set of qualitative modelling dimensions has been adapted from the European Community ITSIE project [ITSIE 1992, D7]. The objectives of Workmanship-MOBAT are:

• to extend the MOBAT framework with multiple Qualitative Modelling dimensions;
• to verify the MOBAT framework with a complex industrial problem.

This chapter starts with an overview of the training domain in Section 6.3. To avoid duplication of Proto-MOBAT and Scheduling-MOBAT it is not necessary to discuss all of the Workmanship-MOBAT specification methods. The emphasis in this chapter is on specification and realisation methods for a principled set of modelling dimensions. Each of the ITSIE qualitative modelling dimensions are introduced and extended in Section 6.4 with examples for the manufacturing workmanship domain. In Section 6.5, the issue of model selection and model adaptation within the training framework is presented.