Measuring the convergence and accuracy of trainees' knowledge structures for TFT-LCD visual defect categorization

Despite a large automated portion of the thin film transistor-liquid crystal display (TFT-LCD) manufacturing process, inspectors' eyes are still the main resource for ensuring the display quality of TFT-LCD panels. Since recruiting and training new inspectors is a frequent process in most TFT-LCD manufacturers, and since it takes most novice inspectors at least a year to become proficient, an effective training program is crucial for TFT-LCD manufacturers to secure their process efficiency and product quality. The purpose of this study is to establish a set of methods for measuring the convergence and accuracy of trainees' knowledge structures on TFT-LCD visual defect categorization, so that the development of knowledge during the training can be monitored and further applied to the design of training programs. The card sorting technique was first used to elicit knowledge. The sort data were then converted to the measures of edit distances and Collective Dice Coefficients. Finally, statistical analyses were applied to measure the convergence and accuracy of trainees' knowledge structures. Results showed that the current training program did increase the convergence and accuracy of trainees' knowledge structures. However, there was room for improvement. The training program could be enhanced by explicitly introducing the experts' knowledge structures to trainees. In addition, the inspection process could be improved by redesigning inspection procedures to correspond to the experts' knowledge structures. With these methods, the development of knowledge can be examined promptly and efficiently, and the effectiveness of the training program can be assured. Relevance to industry: Methods applied in this paper can contribute to the research and applications in the TFT-LCD industry for the design of the inspection-training program and the supervision of trainee's learning progress.