A Framework for Model-Based Adaptive Training

Workmanship Domain Overview

The workmanship training application involves the creation of an adaptable training package, while at the same time being an on-line decision support and reference system. Training is primarily targeted at printed circuit board manufacturing defect classifications. Training specification starts with a definition of requirements. The training purpose here is ‘to manufacture products that are compliant with company standards’. Typically, a training purpose is further specified by a set of training objectives. An example of just two training objectives are ‘to inspect workmanship’ and ‘to check reliability’. An example of a subject specific task is ‘to identify 12 types of Surface Mount Component Manufacturing Soldering defects’.

A detailed training analysis is necessary for a computer-based realisation. The requirements analysis in the MOBAT framework is specified in terms of three core activities:

  1. each training objective is decomposed into the tasks the trainee is expected to learn (subject specific and generic task decomposition);
  2. the classification of expected trainee goal expertise and available knowledge to accomplish the required trainee behaviour; and
  3. trainee profile characterisation.

The requirements analysis leads to a design specification which specifies how to deliver the training with a skeleton training plan and training units. The skeleton training plan for Workmanship-MOBAT is shown in Figure 6-2.

Figure 6-2 Workmanship-MOBAT Skeleton Training Plan

Figure 6-2 Workmanship-MOBAT Skeleton Training Plan

As in Proto-MOBAT and Scheduling-MOBAT, the skeleton training plan simply reflects an overall input via the trainer interface. The training system dynamically (i.e., at run-time) generates and invokes training units. This results in delivering a sequence of training units tailored to the needs of the individual trainee. The trainer can generate 976 training units in Workmanship-MOBAT. The trainer agent for the training system has a separate “training unit generator” problem space which implements the task primitive “to generate” as shown in Figure 6-1. This generator problem space is using a set of procedures which implement the required training units. The training units are selected and maintained by the didactic and diagnostic problem spaces, which are implemented as associative models. These problem spaces use a set of trainer tasks and trainer methods which vary the distinct domain model properties as presented in Section 6.4. The training units are categorised as presentation, example, practice and assessment. Each of these classes are described further below.

There are 337 presentation training units which introduce the subject matter in a lecture format. These units are useful for presenting completely fresh material or introducing new topics to trainees at a beginner level. The presentations can be used when facts and concepts have to be put across, but the length of each session should be limited (e.g., 5 minutes maximum) since, because training is mostly passive, it may be difficult to hold trainee attention and it is difficult for the trainer agent to know whether the trainees are actually absorbing and understanding the information presented. When the initial concepts are presented a more interactive method for imparting knowledge is used with example training units. There are 223 example training units for printed circuit boards with expert advise and explanation. The trainer asks the trainee to interpret photographs of printed circuit boards. If required, the trainer can call upon the help of experts to demonstrate or explain aspects of the particular workmanship example. The examples are used for teaching trainees to analyse the factors leading to a problem. The aim is for trainees to isolate and study the significant factors which determine a point of view on particular workmanship examples.

There are 194 practice training units for the trainee to exercise the required skills for deciding workmanship quality. The practice training units are adapted to real needs and clearly focus a trainee’s attention on making a decision. Advanced trainees may skip the presentation or example units and start with practicing the skills to follow procedures and make responses similar to those expected in the real job. Practice training units are designed for a single task. A more general tool for practice, covering a range of tasks, has a simulation-based training unit. In simulation training the trainer provides an exercise model which attempts to imitate reality, so that the trainee can use this device for learning purposes. The more realistic the simulation, the greater the potential for learning. Simulations require the trainee to follow procedures and make responses similar to those expected in the real job. In the process of doing this, the trainee receives rapid feed-back or reinforcement to see the relevance of the training objective and practice the skills involved. A simulation-based training unit is implemented for a particular process step in the workmanship domain (e.g., the print screen process). For each of the main 19 process steps in printed circuit board manufacturing, a simulation-based training unit is identified (see Figure 6-5 Partial Workmanship-MOBAT Problem Space Map ).

In industrial training, one or more of several assessment methods are generally used before training is considered completed. To some extent a trainer can infer that learning is taking place by observing changes in trainee behaviour. Other methods for observing learning is with pre and post training testing. A trainee’s performance may be assessed before training is given and then, again, afterwards. For the workmanship domain, based on appropriate industrial requirements, only post training tests have been designed for achievement against objectives. A total of 221 assessments training units which test the ability by the trainee to make correct quality decisions. Each of these units focus on testing the skills learned for a single task.

© | | Sitemap