This article presents research efforts made to assist a Fortune 500 glassmaker in reducing the variability of its automated visual inspection systems in assessing product quality. Through interviewing line workers, engineers and managers, and direct observations of the inspection process possible causes leading to large variability observed in the inspection process were identified. Responding to these findings, solutions related to equipment setup, job changes, and operational procedures were suggested in order to minimize the inspection process variations as well as to improve the replicability of the inspection stations. Additionally, a unique statistical experiment was conducted to analyze how three factors (namely, operator, defect size, and threshold value used) would affect the mean of the defective ratio. A robust design aiming at controlling these variations was then given to optimize the system's performance.
[1]
Feng Duan,et al.
Glass Bottle Inspector Based on Machine Vision
,
2008
.
[2]
Huimin Ma,et al.
A glass bottle defect detection system without touching
,
2002,
Proceedings. International Conference on Machine Learning and Cybernetics.
[3]
Paul P. Lin,et al.
Online Inspection for Glass Fiber Forming
,
2007
.
[4]
Wen Tan,et al.
Empty bottle inspector based on machine vision
,
2004,
Proceedings of 2004 International Conference on Machine Learning and Cybernetics (IEEE Cat. No.04EX826).
[5]
Domingo Mery,et al.
Image acquisition and automated inspection of wine bottlenecks by tracking in multiple views
,
2008
.
[6]
J. V. Grice,et al.
Design and Analysis of Experiment
,
2000,
Technometrics.