Attention-Based Quantile Regression for RUL Uncertainty Prediction

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Remaining Useful Life (RUL) prediction is crucial for Prognostics and Health Management (PHM).
• Traditional point prediction methods lack uncertainty quantification, hindering critical decision-making.
• Accurate RUL prediction is vital for maintaining complex equipment and preventing failures.

Purpose of the Study:

• To develop a probabilistic prediction model for RUL that provides interval predictions and quantifies uncertainty.
• To improve the accuracy and reliability of RUL predictions in PHM applications.
• To offer a robust solution for Remaining Useful Life estimation in complex machinery.

Main Methods:

• A novel model integrating a Long Short-Term Memory (LSTM) network with an attention mechanism and quantile regression was developed.
• A bidirectional LSTM captured temporal dependencies, while a six-head self-attention mechanism focused on key degradation features.
• The model simultaneously predicted three quantiles (10%, 50%, 90%) to construct an 80% confidence interval.

Main Results:

• The proposed model achieved excellent point prediction accuracy (RMSE: 11.245) and interval prediction quality (NASA Score: 166.414).
• The model maintained high interval coverage (86.7%) with a reasonable interval width (30.398).
• Ablation studies confirmed the significant contribution of the LSTM and attention components to model performance.

Conclusions:

• The proposed probabilistic RUL prediction model effectively addresses the limitations of traditional methods by providing uncertainty quantification.
• The integration of LSTM-attention and quantile regression offers a powerful approach for accurate and reliable RUL interval predictions.
• This research provides a valuable tool for enhancing the operational safety and efficiency of complex equipment through improved RUL estimation.