Multiform Informed Machine Learning Based on Piecewise and Weibull for Engine Remaining Useful Life Prediction.

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Machine learning (ML) models often lack interpretability and can produce physically implausible predictions.
• Incorporating domain knowledge into ML can address these limitations, particularly for equipment degradation and failure prediction.

Purpose of the Study:

• To develop an Informed Machine Learning (IML) framework for predicting Remaining Useful Life (RUL) by integrating equipment domain knowledge.
• To enhance the accuracy and interpretability of RUL predictions.

Main Methods:

• The proposed IML model involves three steps: identifying knowledge sources from device domain expertise, formally expressing knowledge using Piecewise and Weibull distributions, and integrating this knowledge into the ML pipeline.
• The method was evaluated on the C-MAPSS dataset.

Main Results:

• The IML model demonstrated a simpler, more general structure compared to existing ML models.
• It achieved higher accuracy and more stable performance across various datasets, particularly under complex operational conditions.
• The approach effectively addresses challenges posed by insufficient training data.

Conclusions:

• Integrating domain knowledge via IML significantly improves RUL prediction accuracy and interpretability.
• The proposed method offers a robust solution for equipment health monitoring, especially when training data is scarce.
• This work provides a valuable framework for researchers applying domain knowledge in ML for predictive maintenance.