Toward a better monitoring statistic for profile monitoring via variational autoencoders

Abstract

Variational autoencoders have been recently proposed for the problem of process monitoring. While these works show impressive results over classical methods, the proposed monitoring statistics often ignore the inconsistencies in learned lower-dimensional representations and computational limitations in high-dimensional approximations. In this work, we first manifest these issues and then overcome them with a novel statistic formulation that increases out-of-control detection accuracy without compromising computational efficiency. We demonstrate our results on a simulation study with explicit control over latent variations, and a real-life example of image profiles obtained from a hot steel rolling process.

Keywords:

• Deep learning
• high-dimensional nonlinear profile
• latent variable model
• profile monitoring
• variational autoencoder

Additional information

Notes on contributors

Nurettin Dorukhan Sergin

Nurettin Dorukhan Sergin is a doctoral candidate at the Industrial Engineer program at Arizona State University. His current research is focused on out-of-distribution behaviors of deep neural networks and spatiotemporal modeling of urban mobility. During his master's, he did research on agent-based modeling and its application to computational social simulation problems.

Hao Yan

Hao Yan received his BS degree in Physics from the Peking University, Beijing, China, in 2011. He also received a MS degree in Statistics, a MS degree in Computational Science and Engineering, and a PhD degree in Industrial Engineering from Georgia Institute of Technology, Atlanta, in 2015, 2016, 2017, respectively. Currently, he is an Assistant Professor in the School of Computing, Informatics, and Decision Systems Engineering at ASU. His research interests focus on developing scalable statistical learning algorithms for large-scale high-dimensional data with complex heterogeneous structures to extract useful information for the purpose of system performance assessment, anomaly detection, intelligent sampling and decision making. Dr. Yan was also the recipient of multiple awards including best paper award in IEEE TASE, IISE Transaction and ASQ Brumbaugh Award. Dr. Yan is a member of IEEE, INFORMS and IIE.