A Machining State-Based Approach to Tool Remaining Useful Life Adaptive Prediction

Area of Science:

• Machine Learning
• Predictive Maintenance
• Industrial Engineering

Background:

• Traditional remaining useful life (RUL) prediction models lack adaptiveness, hindering their effectiveness in dynamic industrial environments.
• Accurate RUL prediction is crucial for optimizing maintenance schedules and preventing unexpected equipment failures.

Purpose of the Study:

• To develop an adaptive LightGBM-based RUL prediction method that incorporates process and machining states.
• To enhance the accuracy, generalization ability, and reliability of RUL predictions.

Main Methods:

• A multi-information fusion strategy was employed to reduce model error and improve generalization.
• A novel preprocessing method was developed for precise feature extraction with improved time granularity, avoiding dimensional explosion.
• An importance coefficient and a custom loss function, sensitive to process and machining states, were introduced.

Main Results:

• The proposed method demonstrated superior performance in RUL prediction compared to traditional approaches.
• Extensive validation using actual tool life cycle data and 25 contrast experiments confirmed the method's effectiveness.
• The approach significantly improved prediction accuracy and model adaptability.

Conclusions:

• The developed LightGBM-based RUL prediction method offers a significant advancement over traditional models.
• The integration of multi-information fusion, advanced preprocessing, and state-aware loss functions leads to robust and adaptive RUL predictions.
• This method provides a valuable tool for enhancing predictive maintenance strategies in industrial settings.