A Framework for Model-Based Adaptive Training

Relating Task Features to Training Units

Designing or planning a training course generally involves the creation of a skeleton training plan. Creating a training plan or flow chart comprising training modules is usually performed by human trainers. In Proto-MOBAT, as in the MOBIT project, the modules in skeleton training plans are called training units. The specification details from the previous sections provide the information for a design specification of training units (see Table 4-2). Training units are the interactive means of teaching a trainee the tasks he/she is expected to learn (to perform). The system design specification is therefore focused on these training units as this determines both the required expert agent(s) and trainer agent capabilities. A variety of training units have been designed for Proto-MOBAT [MOBIT 1994, W5-W6]. Their main attributes are shown below:-

  • Training Unit Class – This is an indication of a general type of training unit. Although the choices for Proto-MOBAT application are not considered a complete set, a training unit can generally be chosen from presentation, demonstration, practice and assessment. These general training unit classes can be further broken down into a class hierarchy with subtypes. For example, an assessment training unit for Proto-MOBAT application splits into different question and testing training unit subtypes.
  • Training Strategy – Initially determined by trainee self-supporting level for the most effective training approach. Based on the trainee profile and the way training is proceeding during a training session, the strategy selection is performed by trainer tasks and trainer methods during run time. Chosen from tutoring, coaching, facilitating and discovery.
  • Goal Expertise Level – Specified by expertise classification. Chosen from skill-based, rule-based and model-based.
  • Preferred Learning Mode – Specified by trainee characterisation. Chosen from rote, inductive and deductive.
  • Problem Space - Specified by grouping related training goals into an overall structure of the training domain (see Figure 4-5).
  • Generic task – Specified by task decomposition. Chosen from interpretation, identification, prediction, decision and execution. It may take several training units to teach a domain specific task which is decomposed into a set of generic tasks.
  • DomainModel Type - The realisation of a training unit can be with a choice of “types of models”. The goal expertise level, available knowledge in the domain and preferred learning mode dictates the associated expert model to be used. Chosen from a procedural model, associative model and principled model. The available knowledge for Proto-MOBAT is represented with three types of models: the plant process is represented with a simulation model; circuit board diagnostic rules are represented with an associative model; and plant operations are represented with a procedural model.
  • Problem Solving Method – Methods used by the expert agent or the trainee which control the way to achieve a task in the problem space. See Section 4.9 domain expert realisation for a Proto-MOBAT problem solving method discussion.
  • Items of Expertise – The object(s) of expertise to be taught. In SOAR, these items are called chunks of knowledge in the form of productions (see Section 2.6). The items here can be principles, rules and procedures. The list of objects may also include items of supporting expertise such as text, images and video. The principles, rules and procedures are represented in the domain model and supporting expertise items may be stored in a training materials database.
  • Training Unit Script – A general control structure for training unit implementation. See Section 4.10 generic trainer realisation.
From an initial training problem specification, a set of training units is to be realised with the selection of appropriate software methods. Table 4-2 Proto-MOBAT Training Units shows a set of 24 power plant training units. This table lists the training unit types and a subset of domain specific tasks a trainee is expected to learn (to perform). The type of domain model and a model description for two training units (TU12 & TU24) is shown below in Table 4-3 Proto-MOBAT Model Types.
Proto-MOBAT Training Units

Table 4-2 Proto-MOBAT Training Units

The analysis in Proto-MOBAT provides an indication of the granularity of training tasks. However, as the power plant in this chapter is an artificial problem, it does not provide a good indication for the number of training units needed with a real industrial training problem. In parallel with Proto-MOBAT research, an analysis for a real manufacturing assembly domain indicated 12 actual training objectives would result in 476 training units [MOBIT 1994, W3 & W7]. All 476 training units for the manufacturing assembly domain were realised with presentation user interface screens using multimedia objects. This large number can require a significant specification effort to identify each training unit attribute for an ITS implementation. However, if support tools to automate the generation of training units from training objectives are developed, or if re-usability of training units from similar domains is possible, then the training unit design effort can be reduced significantly.

Proto-MOBAT Model Types

Table 4-3 Proto-MOBAT Model Types

© | | Sitemap