Remaining Useful Life Prediction for Lithium-Ion Batteries Based on Gaussian Processes Mixture

Area of Science:

• Battery Engineering
• Machine Learning
• Prognostics and Health Management (PHM)

Background:

• Lithium-ion battery Remaining Useful Life (RUL) prediction is critical for operational safety and efficiency.
• Capacity degeneration trajectories exhibit multimodality due to self-charging and regeneration, challenging traditional models.
• Support Vector Machines (SVM) and Gaussian Process Regression (GPR) struggle to accurately model these multimodal behaviors.

Purpose of the Study:

• To develop a novel RUL prediction method capable of handling multimodal capacity degeneration trajectories.
• To improve the accuracy and reliability of Lithium-ion battery RUL prediction.
• To introduce a Gaussian Process Mixture (GPM) model for enhanced battery prognostics.

Main Methods:

• Proposed a Gaussian Process Mixture (GPM) model to address multimodality in battery capacity data.
• Applied GPM by fitting distinct segments of degeneration trajectories with separate GPR models.
• Validated the method using experimental data from commercial Type 1850 Lithium-ion batteries provided by NASA.

Main Results:

• The GPM model effectively processed multimodal capacity degeneration trajectories, revealing subtle differences.
• Experimental results demonstrated the GPM model's superior predictive accuracy compared to SVM and GPR.
• The GPM model provided a predictive confidence interval, enhancing prediction reliability.

Conclusions:

• The Gaussian Process Mixture (GPM) model offers a significant advancement in Lithium-ion battery RUL prediction.
• GPM's ability to model multimodality leads to more accurate and trustworthy battery health assessments.
• This approach provides a more reliable alternative to traditional RUL prediction methods.