Adaptive Wiener process-based remaining useful life prediction method considering multi-source variability.

Area of Science:

• Reliability Engineering
• Predictive Maintenance
• Data-Driven Prognostics

Background:

• Existing remaining useful life (RUL) prediction methods struggle with uneven measurement intervals and inconsistent frequencies in degrading equipment.
• Current approaches often overlook the variability of adaptive drift in future degradation processes, limiting their applicability.
• There is a need for advanced RUL prediction models that account for multi-source variability and dynamic degradation patterns.

Purpose of the Study:

• To propose a novel nonlinear degradation method for RUL prediction based on the adaptive Wiener process.
• To develop a model that effectively handles multi-source variability and the randomness of parameters in nonlinear degradation functions.
• To improve the accuracy and applicability of RUL predictions for degrading equipment under challenging measurement conditions.

Main Methods:

• Constructed a nonlinear degradation model with multi-source variability using the adaptive Wiener process, incorporating parameter randomness.
• Employed the particle filter algorithm for real-time estimation of multiple hidden states and derived the RUL distribution via first hitting time.
• Utilized the expectation maximization algorithm for adaptive model parameter updates based on monitoring data.

Main Results:

• The proposed adaptive Wiener process model effectively captures nonlinear degradation with multi-source variability.
• Real-time state estimation and RUL distribution were successfully achieved using the particle filter.
• Numerical simulations and lithium-ion battery experiments demonstrated significant improvements in prediction accuracy.

Conclusions:

• The developed nonlinear degradation method offers superior RUL prediction accuracy compared to existing approaches.
• The model's ability to handle uneven measurements and adaptive drift makes it valuable for practical equipment health monitoring.
• This approach shows strong potential for application in predictive maintenance and reliability engineering.