Neural Network Battery Capacity Estimation via Nyquist Plot Discrimination
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for estimating the full charge capacity of secondary batteries, such as the AC impedance measurement method, lack accuracy in distinguishing between batteries with varying degrees of deterioration, affecting the recycling and reuse of battery assemblies in electric vehicles.
Innovation Solution
A battery information processing system utilizing a trained neural network model to analyze Nyquist plots of secondary batteries, determining whether a battery belongs to a group with a full charge capacity within a reference range, and estimating its capacity using specific feature values from the plot, optimizing the estimation process for batteries with capacities above the reference threshold.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AC impedance measurement method is used to estimate full charge capacity, then the evaluation process can be performed, but the accuracy in distinguishing batteries with varying degrees of deterioration is insufficient
Solution Approach 1:
The patent segments the battery evaluation process into two distinct stages: first, discriminant analysis is used to classify batteries into groups based on whether their full charge capacity is within or outside a reference range; second, a neural network model estimates the actual capacity value. This segmentation allows each method to be optimized for its specific function, improving overall accuracy and reliability in distinguishing deteriorated batteries.
2Measurement precision
If a general neural network model is used for all batteries, then estimation can be performed, but accuracy is reduced for specific battery groups with capacities above reference threshold
Solution Approach 1:
The patent applies local quality by training the neural network model specifically on data from batteries whose full charge capacity is within the reference range. This localized training approach allows the model to be highly adapted and accurate for this specific group, rather than using a generic model that must accommodate all battery types and conditions.
3Measurement precision
If discriminant analysis is used to classify batteries, then the estimation process can be optimized for specific groups, but the overall system complexity increases
Solution Approach 1:
The patent implements preliminary action by performing discriminant analysis first to classify batteries into appropriate groups before applying the neural network estimation. This preliminary classification step organizes the data in advance, allowing the subsequent estimation process to be more efficient and accurate for each specific group, rather than attempting to estimate all batteries simultaneously.
Data Source
AI summary
A trained neural network model is a neural network model which has been trained based on Nyquist plots of a plurality of modules of which full charge capacity is within a reference range. A processing system determines to which of a first group of modules of which full charge capacity is within the reference range and a second group of modules of which full charge capacity is out of the reference range a module belongs, based on discriminant analysis in which at least one feature value extracted from the Nyquist plot of the module is adopted as an explanatory variable. When the processing system determines that the module M belongs to the first group, the processing system estimates a full charge capacity of the module by using the trained neural network model.