An image-based multivariate generalized likelihood ratio control chart for detecting and diagnosing multiple faults in manufactured products

Abstract

Image-capturing systems are increasingly being used in manufacturing shop floors since they can reliably capture important aesthetic information pertaining to the quality of manufactured parts in real time. State-of-the-art image-monitoring applications have focused on the detection of a single fault; however, the number of fault clusters per image in industrial applications can be numerous. To address this issue, we propose the use of a multivariate generalized likelihood ratio (MGLR) control chart for monitoring industrial products whose quality is described by a specific pattern (e.g. uniform patterns in LED screens or decorative patterns in textile products). Our method is specifically designed for greyscale images that are typical outputs of real-time industrial image-capturing systems. Extensive computer simulations show that the proposed method can detect the occurrence of single and multiple faults. We also present an experimental study to highlight how practitioners can implement and make use of the MGLR control chart in image-monitoring applications.

Keywords:

• functional data analysis
• high dimensional data
• machine vision systems
• phase II methods
• statistical process control
• surveillance

Acknowledgements

The authors would also like to thank Huw D. Smith, Undergraduate Researcher at Auburn University, for his assistance with the experimental study.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Natural Science Funds for Distinguished Young Scholar [grant number 71225006].