A New Method of Remaining Useful Lifetime Estimation for a Degradation Process with Random Jumps.

Area of Science:

• Engineering
• Reliability Engineering
• Data Science

Background:

• System degradation can exhibit complex behaviors, including random jumps, impacting reliability.
• Traditional models may not accurately capture these non-monotonic degradation patterns.
• Accurate Remaining Useful Life (RUL) estimation is crucial for predictive maintenance and operational safety.

Purpose of the Study:

• To introduce a non-homogeneous jump diffusion process model for systems with complex degradation paths.
• To develop a state-space representation for accurate degradation evolution prediction and RUL estimation.
• To validate the proposed model's effectiveness using real-world industrial data.

Main Methods:

• Formulation of a non-homogeneous jump diffusion process model.
• Translation into a state-space model for analysis.
• Application of Monte Carlo simulation with particle filtering for RUL prediction.
• Model identification using Maximum Likelihood Estimation (MLE) and Expectation-Maximization (EM) algorithms.

Main Results:

• The proposed model accurately captures random jumps in degradation paths.
• Particle filtering-based Monte Carlo simulation effectively predicts degradation evolution.
• The approach provides more accurate and robust RUL estimations compared to CNN and LSTM methods.
• Validation with real-world temperature sensor data from a blast furnace wall confirmed effectiveness.

Conclusions:

• The non-homogeneous jump diffusion model offers a superior approach for RUL estimation in degrading systems.
• This method enhances predictive maintenance strategies and operational safety for complex systems.
• The study contributes a robust framework for analyzing and predicting non-monotonic degradation patterns.