A Semi-Supervised Approach with Monotonic Constraints for Improved Remaining Useful Life Estimation

Area of Science:

• Engineering
• Data Science
• Reliability Engineering

Background:

• Remaining Useful Life (RUL) estimation is vital for Prognostics and Health Management (PHM) and Predictive Maintenance (PdM).
• Accurate RUL aids in optimizing maintenance schedules, understanding component degradation, and preventing unexpected failures.
• Current RUL estimation methods can be improved with more robust health indicators.

Purpose of the Study:

• To develop a semi-supervised methodology for creating monotonic health index models.
• To enhance existing Remaining Useful Life (RUL) estimation models using these monotonic health indexes.
• To demonstrate the effectiveness of the proposed approach on bearing datasets.

Main Methods:

• A semi-supervised learning approach was employed to construct health index models.
• Ensured monotonicity of the developed health indexes to reflect component degradation accurately.
• Integrated the monotonic health indexes into RUL estimation models.

Main Results:

• The methodology successfully generated monotonic health indexes.
• Utilizing monotonic health indexes significantly improved Remaining Useful Life (RUL) estimation accuracy.
• The approach provided better insights into bearing degradation patterns.

Conclusions:

• The proposed semi-supervised method for creating monotonic health indexes is effective for Prognostics and Health Management (PHM).
• Monotonic health indexes enhance the accuracy and interpretability of Remaining Useful Life (RUL) estimations.
• This approach offers a valuable tool for advancing Predictive Maintenance (PdM) strategies.