Abstract
Accurate remaining useful life (RUL) prediction is the foundation of predictive maintenance and provides an effective way to reduce operational and maintenance costs. However, time-varying operating conditions (TVOCs) and limited run-to-failure data at industrial sites significantly hinder the performance of traditional RUL prediction methods. To overcome these challenges, this paper proposes a novel self-data-driven RUL prediction method for rotating machinery (RM) under TVOCs. First, inspired by time-varying response systems in control theory, a two-stage hybrid state-space model (HSSM) is developed by combining the dynamic modeling capabilities of state-space models with the nonlinear fitting ability of neural networks. This model innovatively integrates TVOCs into the mathematical formulation of the dynamic degradation process of RM. It aims to describe the functional relationship among operating conditions, system states, and response signals during the entire life cycle of RM under TVOCs, thereby avoiding the impact on RUL prediction caused by the coupling of operating conditions and system states. Then, based on the extended Kalman filter algorithm and the monitoring data generated by the system itself, an alternating parameter update strategy is introduced to solve the system state within the HSSM, enabling real-time tracking of RM degradation. Furthermore, an adaptive model selection strategy is proposed to dynamically identify the optimal state degradation model. This method effectively overcomes the problem of insufficient historical run-to-failure data for offline model training. In addition, the design characteristics of the HSSM at different stages provide an integrated solution for both anomaly detection and RUL prediction in RM. The effectiveness and superiority of the proposed method are validated through both simulation and experimental studies.
| Original language | English |
|---|---|
| Article number | 111981 |
| Journal | Reliability Engineering and System Safety |
| Volume | 268 |
| DOIs | |
| State | Published - Apr 2026 |
Keywords
- Hybrid space-space model
- Remaining useful life prediction
- Rotating machinery
- Self-data-driven
- Time-varying operating conditions
Fingerprint
Dive into the research topics of 'Self-data-driven remaining useful life prediction of rotating machinery under time-varying operating conditions based on a two-stage hybrid state-space model'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver