제조품질 향상을 위한 데이터 전처리 프로세스

Abstract

Improved manufacturing data acquisition systems such as sensors and fast communication systems have made it possible to collect various types of data that were previously unavailable. However, manufacturing data may be contaminated with errors during the data collection process due to such problems as noise or process environment. Utilization of these error-contained data can waste resources and render analysis results useless. To help data scientists and quality engineers dealing with manufacturing quality data, a guideline is proposed for appropriate pre-processing of manufacturing quality data in six steps. Two case study data are used for illustration. The proposed approach is compared with six other methods and shows advantageous in terms of the F1 score. This paper is expected to help quality practitioners and data scientists applying machine learning methods to manufacturing quality data.