Statistical Quality Control for Spatial Point Data

Spatial point data consist of a number of randomly distributed points on the plane or in space. The “points” may be any objects of negligible size relative to the study region, and the focus of interest is typically on the distributional properties of point locations. Spatial point data are frequently encountered in manufacturing processes. For example, defective pixels on display panels/camera sensors (CCD or CMOS), surface defects on semiconductor wafers, nanoparticles in composite materials, etc. These “points” exhibit complex spatial patterns. They are quality characteristics having a crucial impact on the final product quality and process yield. In addition, they hold valuable information about the product manufacturing processes. Exploitation of such information they contain can be utilized for process monitoring and fault diagnosis.Rapid development of advanced image-based inspection systems has made possible the in-line inspection of spatial point data in manufacturing processes. However, due to the lack of quality control methodologies in related areas, inspection of such data is mostly carried out in primitive forms: human visual inspection and simple control charts counting the total number of points. The rich information contained in spatial point data has largely been left untapped. Although some prior work on analysis of point data can be found in other application areas such as biology, ecology, epidemiology, astronomy, etc., they have very different focuses of interest and have no concern over the critical issues in the context of manufacturing quality control.This project aims to develop systematic statistical quality control methodologies for spatial point data in manufacturing, by fully exploiting the rich process information they contain. Advanced algorithms will be developed, tested and validated through real manufacturing processes. Quantitative characterization of point distributional behaviors will be proposed, upon which process monitoring procedures for variation reduction and fault pattern detection methods for root cause diagnosis will be investigated. In this project, we will also extend our analysis from traditional two-dimensional point distribution to a much more challenging three-dimensional quality inspection problem which has rarely been studied in exiting literature. The successful completion of this project will deliver novel techniques targeting at the pressing issues of statistical quality control for manufacturing processes with spatial point data as vital quality characteristics. The proposed methods will be validated and implemented in real applications.