Adaptive OED for Battery Formation Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current battery optimization methods rely on unoptimized empirical testing and under-validated physiochemical models, leading to inefficiencies in lithium-ion battery development and deployment due to high dimensionality and manufacturing variance, with existing approaches being either costly or insufficiently accurate.
Innovation Solution
A method using adaptive optimal experimental design (OED) to efficiently probe a multidimensional parameter space of battery cell formation and cycling protocols, incorporating hyperparameters and early prediction models to optimize battery processes, reducing testing time and costs by up to two orders of magnitude.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If model optimization is used to probe the parameter space, then computational cost is reduced, but accuracy is insufficient to capture all relevant degradation modes and manufacturing variation
Solution Approach 1:
The patent introduces a surrogate model as an intermediary between the complex battery degradation system and the optimization process. This surrogate model captures key degradation modes and manufacturing variations while requiring significantly less computational resources than full-scale simulations, thus resolving the contradiction between computational efficiency and prediction accuracy
Solution Approach 2:
The patent transforms the high-dimensional parameter space into a reduced set of critical parameters that dominate degradation behavior. By identifying and focusing on these key parameters through sensitivity analysis and dimensionality reduction techniques, the system achieves accurate predictions with reduced computational cost
2Measurement precision
If grid search is used to experimentally test parameters, then accuracy is improved, but time and resource costs increase significantly
Solution Approach 1:
The patent performs preliminary screening of the parameter space using a surrogate model to identify promising regions before conducting expensive experimental tests. This preliminary action filters out unpromising parameter combinations, allowing focused experimentation on a reduced set of candidates and significantly reducing total testing time while maintaining accuracy
Solution Approach 2:
The patent implements an iterative feedback loop where experimental results from grid search are fed back to update and refine the surrogate model. This continuous refinement allows the system to progressively improve prediction accuracy while reducing the number of required experiments, as the model becomes more efficient at identifying promising regions
3Loss of information
If comprehensive data analysis is performed on all testing data, then insight accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent extracts and focuses analysis on the most informative subsets of data that contain the critical signals for degradation prediction. By identifying and isolating key data features and patterns that dominate degradation behavior, the system achieves high-quality insights while processing only the essential data portions, significantly reducing computational overhead
Data Source
AI summary
A method of probing a multidimensional parameter space of battery cell test protocols is provided that includes defining a parameter space for a plurality of battery cells under test, discretizing the parameter space, collecting a preliminary set of cells being cycled to failure for sampling policies from across the parameter space and include multiple repetitions of the policy, specifying resource hyperparameters, parameter space hyperparameters, and algorithm hyperparameters, selecting a random subset of charging policies, testing the random subset of charging policies until a number of cycles required for early prediction of battery lifetime is achieved, inputting cycle data for early prediction into an early prediction algorithm to obtain early predictions, inputting the early predictions into an optimal experimental design (OED) algorithm to obtain recommendations for running at least one next test, running the recommended tests by repeating from the random subset testing step above, and validating final recommended policies.