Li-Ion Battery State Estimation Using Neural ODE Degradation Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the state-of-charge and state-of-health of Li-ion batteries in electric aircraft are computationally expensive or trade accuracy for interpretability, and there is a lack of datasets that accurately reflect the eVTOL duty cycle, making it challenging to predict battery performance and degradation effectively.
Innovation Solution
A physics-informed machine learning approach using Neural Ordinary Differential Equations (NODEs) integrated with mechanistic models to predict battery performance and degradation, combined with a universal battery degradation model (UBDM) that utilizes a neural network to approximate the derivative of battery parameters, and a dataset specifically generated for eVTOL conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based models are used to estimate battery state of charge and state of health, then accuracy and interpretability are improved, but computational cost increases
Solution Approach 1:
The patent introduces neural networks as intermediary components that bridge physics-based models and data-driven approaches. The neural networks learn from training data to predict battery states while incorporating physics-based constraints and relationships, achieving accurate predictions without requiring full physics-based model computations at runtime.
Solution Approach 2:
The patent transforms the complex physics-based model parameters into simplified neural network parameters through training. By changing from explicit physics model parameters to learned neural network weights, the system maintains accuracy while reducing computational burden during battery state estimation.
2Productivity
If empirical and machine learning models are used to predict battery degradation, then computational speed is improved, but interpretability is lost
Solution Approach 1:
The patent uses physics-based relationships as intermediaries to guide the machine learning model structure. The neural networks are designed to respect physical constraints and relationships, allowing fast predictions while maintaining interpretability through the embedded physics knowledge.
Solution Approach 2:
The patent creates a universal degradation model that can predict multiple battery parameters (capacity, resistance, thermal behavior) simultaneously using a single neural network framework. This multi-functional approach maintains interpretability while achieving computational efficiency across different prediction tasks.
3Measurement precision
If detailed physics-based degradation models are used to account for battery aging, then prediction accuracy is improved, but model complexity increases
Solution Approach 1:
The patent segments the battery degradation model into separate neural network components, each handling specific degradation aspects (capacity fade, resistance increase, thermal changes). This segmentation allows accurate prediction of each parameter while keeping individual model components manageable and interpretable.
Solution Approach 2:
The patent creates a composite modeling approach that combines elements of physics-based models and machine learning. The hybrid model integrates physics-based constraints with data-driven predictions, achieving high accuracy while avoiding the full complexity of pure physics-based models.
Data Source
AI summary
Disclosed herein is a method for predicting the performance characteristics of a Li-ion battery over the course of the lifetime of the battery. A model based on integration of a set of ordinary differential equations is used to predict the voltage and thermal characteristics of the battery in a short period of time and a universal ordinary differential equation is used to predict the degradation of the battery by changing the parameters of the system of ordinary differential equations. The degradation of the battery is predicted in terms of change in various parameters of the battery (e.g., capacity, resistance, thermal behavior).