HybridoNet-Adapt: A domain-adapted framework for accurate lithium-ion battery RUL prediction.

Area of Science:

• Battery Health Management
• Machine Learning for Engineering
• Predictive Maintenance

Background:

• Accurate Remaining Useful Life (RUL) prediction is crucial for lithium-ion battery safety and reliability.
• Current RUL models struggle with generalization when test data differs from training data distributions.

Purpose of the Study:

• Introduce HybridoNet-Adapt, a novel domain-adaptive framework for RUL prediction.
• Address the generalization gap in RUL models by bridging labeled source and unlabeled target domains.

Main Methods:

• Employ Maximum Mean Discrepancy (MMD) to learn domain-invariant representations by minimizing feature distribution differences.
• Utilize parallel source and target domain predictors with learnable trade-off parameters for dynamic output weighting.
• Integrate LSTM, multi-head attention, and Neural ODE blocks for advanced temporal feature extraction.

Main Results:

• HybridoNet-Adapt significantly outperforms non-adaptive baseline models on two public battery datasets.
• Achieved up to a 152-cycle Root Mean Square Error (RMSE) reduction under domain shifts.
• Demonstrated superior robustness and real-world applicability compared to existing methods.

Conclusions:

• Domain adaptation is a key strategy for enhancing the robustness of RUL prediction models.
• HybridoNet-Adapt offers a significant advancement in battery health management through its adaptive approach.
• The proposed framework shows substantial potential for improving the reliability of battery systems in diverse operational environments.