Battery Capacity Prediction Using Physics-Based Model Surrogates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing battery capacity prediction methods struggle to accurately capture the complex nature of battery degradation due to diverse aging mechanisms, significant device variability, and varied operating conditions, particularly in lithium-ion batteries, with physics-based pseudo two-dimensional models being computationally demanding and unsuitable for real-time applications.
Innovation Solution
A method and system that generates a physics-based model (PBM) using input data, preprocessing, feature generation, and model selection, incorporating mechanistic and data-based models to predict battery capacity, with a framework comprising an offline module for simulation and an online module for real-time prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If physics-based pseudo two-dimensional models are used to predict battery capacity, then prediction accuracy is improved, but computational complexity and time consumption increase significantly
Solution Approach 1:
The patent segments the battery capacity prediction task into two distinct phases: an offline training phase where a data-driven model learns from simulated physics-based model data, and an online prediction phase where the trained model provides rapid predictions. This segmentation allows the computationally intensive physics-based model to be used only during offline training, while the lightweight data-driven model handles real-time predictions, thus resolving the contradiction between accuracy and computational complexity.
Solution Approach 2:
The patent performs preliminary action by pre-training the data-driven model using simulated data from physics-based models before deployment. This preliminary training phase captures the complex electrochemical behaviors in advance, enabling the model to provide accurate predictions during online operation without requiring real-time computation of complex physics-based models, thereby reducing computational complexity while maintaining accuracy.
2Measurement precision
If physics-based models are used to capture complex electrochemical reactions, then prediction accuracy is improved, but real-time application capability deteriorates
Solution Approach 1:
The patent creates a simplified copy of the physics-based model by training a data-driven model (such as a neural network or Gaussian process model) to replicate the input-output behavior of the physics-based model. This copied model captures the essential electrochemical degradation patterns without requiring complex physics-based computations, enabling real-time predictions while maintaining accuracy. The copy is trained offline using simulated data from the physics-based model.
3Productivity
If data-driven models are used for battery capacity prediction, then computational efficiency is improved, but ability to capture complex electrochemical mechanisms deteriorates
Solution Approach 1:
The patent introduces simulated data from physics-based models as an intermediary between the complex electrochemical mechanisms and the data-driven prediction model. The physics-based model generates training data that encodes the complex electrochemical degradation mechanisms, which then serves as training input for the computationally efficient data-driven model. This intermediary approach allows the data-driven model to learn mechanism-based patterns without requiring real-time physics-based computations, thus maintaining both computational efficiency and mechanism capture capability.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Challenges such as diverse aging mechanisms, significant device variability, and varied operating conditions of batteries, make it difficult to develop a generalized prediction model that can accurately capture the complex nature of battery degradation. The existing prediction methods often struggle to guarantee prediction accuracy due to the complex internal electrochemical reactions and external use conditions. In order to address these challenges, the method and system disclosed herein propose a mechanism for generating a Physics Based Model (PBM) for a battery being monitored, by creating a battery profile and further by selecting appropriate models that match the battery. The PBM, once generated, is used to generate prediction of a set of state variables representing degradation of the battery.