Physics-Informed Battery Models for Accurate RUL and SOH Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies struggle to accurately predict battery performance metrics such as remaining useful life, state of health, and state of charge, especially in devices with higher current demands and power needs, leading to inefficiencies and potential safety issues from premature battery replacement or failure.
Innovation Solution
A physics-informed machine learning model that utilizes both cycling data and electrodynamic parameters to predict battery performance, trained using a combination of measured and simulated data, with a loss function that balances accuracy and physical plausibility, incorporating electrodynamic parameters calculated from probing waveforms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional machine learning models are used to predict battery performance, then the model can be trained quickly with available data, but the prediction accuracy deteriorates due to lack of physical constraints and understanding of battery degradation mechanisms
Solution Approach 1:
The patent transforms the ML model by incorporating physics-based parameters and constraints into the model architecture and loss function. This changes the model parameters to include physical laws governing battery degradation, thereby improving prediction accuracy while maintaining manageable complexity through structured integration of physical principles
Solution Approach 2:
The patent introduces physics-based degradation mechanisms as intermediaries between the input cycling data and the output performance predictions. These physical principles act as mediators that guide the ML model's learning process, improving accuracy by ensuring predictions align with known battery behavior while the modular structure keeps complexity controlled
2Reliability
If physics-based models are used to predict battery performance, then the predictions are physically plausible, but the computational complexity and data requirements increase significantly
Solution Approach 1:
The patent merges physics-based models with machine learning approaches in a hybrid framework. This combination allows the system to leverage the physical plausibility of physics-based models while using ML to handle complex patterns in cycling data, achieving reliable predictions without excessive computational complexity through synergistic integration
Solution Approach 2:
The patent applies physics-based constraints locally to specific aspects of the prediction model rather than requiring complete physics-based modeling of all battery processes. This selective application of physical principles ensures physical plausibility in critical areas while keeping overall computational complexity manageable by not over-modeling all aspects
3Measurement precision
If more cycling data is collected to improve prediction accuracy, then the model performance improves, but the time required for testing and data collection increases
Solution Approach 1:
The patent performs preliminary action by incorporating physics-based knowledge and degradation mechanisms into the model before actual prediction. This pre-integration of physical principles allows the model to make accurate predictions with less training data, reducing the time required for extensive cycling tests while maintaining high prediction accuracy
Solution Approach 2:
The patent uses simulated cycling data generated from physics-based models as copies of real battery behavior. These simulated datasets serve as training examples that replicate actual battery degradation patterns without requiring extensive physical testing, thereby improving prediction accuracy while minimizing the time and resources needed for data collection
Data Source
AI summary
A system and method are provided for predicting battery-performance information (e.g., remaining useful life (RUL), state of health (SOH), and/or state of charge (SOC)) for battery based on cycling data. For example, the battery-performance information can be predicted using machine learning (ML) models that predict battery-performance information for the battery based on cycling data and electrodynamic parameters (EDPs) that are generated either by calculating the EDPs using probing waveform data or predicting the EDPs from cycling data. The ML model can have been trained using results from a physics-based model when calculating the loss function used for training the ML model.