Training for visual inspection

For the past several years the emphasis in visual inspection research has been on associating training intervention strategies with changes in performance. Although this approach has provided a wealth of empirical results, the training interventions have not been associated with factors that determine improvement in visual inspection performance. An understanding of which factors are responsible for improved performance would provide a sound basis for empirical results to be extended to similar and different tasks.;The objective of this research was to develop a rational basis for training in visual inspection. The research identified factors that affect visual inspection and related the effects of various training interventions to changes in these factors, which in turn were associated with changes in inspection performance. The task used was a simulation of an air-frame visual inspection task. By using a model-based task analytical approach, the tasks in visual inspection were identified and key variables selected for which training improvements would be required. A model relating these key variables to known training interventions was developed to structure the experimental series.;A series of training experiments was developed to understand the effects of different training interventions in the visual inspection domain. The first experiment evaluated the effect of visual lobe size training on visual search performance. It was found that the visual lobe was trainable and an increased visual lobe resulted in an improved visual search performance. An experiment on feedback training compared cognitive and performance feedback by relating the effects of feedback information to changes in process measures and performance measures. Subjects provided with performance feedback showed the greatest improvement in performance measures. An experiment in feedforward training evaluated the effect of prescriptive, descriptive and a combination of prescriptive and descriptive information on process and performance measures of visual inspection. The use of feedforward information was found to be task specific. Subjects used this information only if they felt that it reduced uncertainty. The final two experiments involved schema training and attribute training. A combined active and progressive parts training approach in classifying ill-defined faults into categories, and for classifying multi-attribute faults, respectively, proved to be most effective.;Overall, the predictions of the key variables/training intervention strategies were confirmed by experimental results. The general application of this model, as well as its specific application to aircraft structural inspection are discussed.