Scalable prognostic models for large-scale condition monitoring applications

ABSTRACT

High-value engineering assets are often embedded with numerous sensing technologies that monitor and track their performance. Capturing physical and performance degradation entails the use of various types of sensors that generate massive amounts of multivariate data. Building a prognostic model for such large-scale datasets, however, often presents two key challenges: how to effectively fuse the degradation signals from a large number of sensors and how to make the model scalable to the large data size. To address the two challenges, this article presents a scalable semi-parametric statistical framework specifically designed for synthesizing and combining multistream sensor signals using two signal fusion algorithms developed from functional principal component analysis. Using the algorithms, we identify fused signal features and predict (in near real-time) the remaining lifetime of partially degraded systems using an adaptive functional (log)-location-scale regression modeling framework. We validate the proposed multi-sensor prognostic methodology using numerical and data-driven case studies.

KEYWORDS:

• Degradation modeling
• residual useful life
• functional (log)-location-scale regression
• functional principal components analysis
• signal fusion

Acknowledgments

The authors acknowledge, with gratitude, the helpful comment of the anonymous referees and the associate editor.

Funding

This research was sponsored by grants from the U.S. National Science Foundation (CMMI-1536555).

Additional information

Notes on contributors

Xiaolei Fang

Xiaolei Fang received his B.S. degree in Mechanical Engineering from the University of Science and Technology Beijing, China, in 2008 and an M.S. degree in Statistics from the Georgia Institute of Technology, Atlanta, in 2016. Currently, he is working toward his Ph.D. degree in Industrial Engineering at the H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta. His research interests are focused on data analytics on high-dimensional signals, including multi-stream functional data and image streams, with applications in system monitoring, diagnostics, and prognostics. He is a member of IIE and INFORMS.

Nagi Z. Gebraeel

Nagi Z. Gebraeel received M.S. and Ph.D. degrees in Industrial Engineering from Purdue University, West Lafayette, Indiana, in 1998 and 2003, respectively. Currently, he is an Associate Professor at the H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology. His research focuses on improving the accuracy of predicting unexpected failures of engineering systems by leveraging sensor-based data streams. His major research interests are in the areas of degradation modeling and sensor-based prognostics, reliability engineering, and maintenance operations and logistics. He is a member of IIE and INFORMS.

Kamran Paynabar

Kamran Paynabar is an Assistant Professor in the H. Milton Stewart School of Industrial & Systems Engineering at Georgia Tech. His research interests include statistical learning and modeling integrated with engineering knowledge. His current research focuses on the analysis of high-dimensional complex data including multi-stream signals, images, videos, point-clouds, and network data for system modeling, monitoring, diagnostics, and prognostics using semi-parametric and nonparametric approaches. He received his B.Sc. and M.Sc. in Industrial Engineering from Iran University of Science and Technology and Azad University in 2002 and 2004, respectively, and his Ph.D. in Industrial and Operations Engineering from The University of Michigan in 2012. He also holds an M.A. in Statistics from The University of Michigan. He was the recipient of the INFORMS Data Mining Best Student Paper Award, the Best Application Paper Award from IIE Transactions, and the Wilson Prize for the Best Student Paper in Manufacturing.