Multimodal recognition and prognostics based on features extracted via multisensor degradation modeling

Abstract

Remaining useful life (RUL) prediction is an important issue in prognostics and health management (PHM). Numerous studies have been conducted to construct degradation models for RUL prediction. However, their models fail to handle the scenarios where multiple failure modes exist, especially when the failure modes are unknown (unlabeled) beforehand and need to be recognized. This paper develops a multimodal recognition and prognostic method based on features extracted via multisensor degradation modeling. Specifically, we assume the failure mode of a unit follows a multinomial distribution. Given the failure mode distribution, we characterize the degradation status of the unit via degradation models based on each sensor signal and a constructed health index (HI). Our innovative idea is to extract features as the derivatives of the degradation status to comprehensively utilize information from multiple sensors for more effective failure mode recognition and RUL prediction. We develop a fusion coefficient-integrated expectation-maximization (FCIEM) algorithm to estimate model parameters by using data from historical units. Finally, we recognize the failure mode and predict the RUL of in-service units based on their extracted features and degradation status. Numerical experiments and a case study of aircraft engines were conducted to evaluate the performance of our proposed method.

Keywords:

• failure mode recognition
• feature extraction via degradation mechanism
• RUL prediction under multiple failure modes
• unavailable failure mode data

Data availability statement

The data that support the findings of this study are openly available on the Intelligent Systems Division of NASA website at https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported in part by the National Science Foundation of China under Grant 72101148; in part by the National Science Foundation of Shanghai under Grant 22ZR1433000; in part by the Shanghai Sailing Program under Grant 21YF1420100; and in part by Shanghai Chenguang Program under Grant 21CGA12. The authors gratefully acknowledge the support provided by the funding agencies.

Notes on contributors

Di Wang

Di Wang received the B.S. degree in industrial engineering from Nankai University, Tianjin, China, in 2015, and the Ph.D. degree in management science and engineering from Peking University, Beijing, China, in 2020. She is currently an Associate Professor with the Department of Industrial Engineering and Management, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China. Her research interests include statistical modeling of spatiotemporal data and artificial intelligence of data fusion for process modeling, monitoring, and prognostics. Dr. Wang is also a member of IEEE, INFORMS and IISE.

Yuhui Wang

Yuhui Wang received a B.S. degree in industrial engineering from Zhejiang University, Hangzhou, China, in 2021, and an M.S. degree in industrial engineering and management from Shanghai Jiao Tong University, Shanghai, China, in 2024. His research interests include statistical learning and deep learning for process modeling, monitoring, and prognostics.

Ershun Pan

Ershun Pan received the B.S. and M.S. degrees in mechanical design and manufacturing from Northeastern University, Shenyang, China, in 1997, and the Ph.D. degree in mechanical engineering from Shanghai Jiao Tong University, Shanghai, China, in 2000. He is currently a Professor and the Head of the Department of Industrial Engineering and Management, Shanghai Jiao Tong University. His current research interests include quality control, reliability engineering and maintenance policy, production system planning and design, and lean manufacturing technology.