Optimization under Uncertainty Tool for Modeling Porous Lithium-Ion Batteries

Lefei Zhang, Salar Safarkhani, Guang Qiang LinPurdue 2016

Abstract

The motivation of this tool is to optimize the performance of battery based on energy output. During the manufacturing process, several parameters such as cathode thickness, the volume concentration of cathode and radius of negative active materials are subject to uncertainty. To optimize battery performance, it is significant to quantify those uncertainties through electrochemical multiscale computer simulation. Hence, this tool will focus on the optimization of the performance of lithium-ion battery under different currents. This tool will consist of a module on visualized generator of uncertainty input, an electrochemical system simulator, a visualization of output optimization module. First, the uncertainty input generator provides the option for selecting one of several statistical models for the input parameter distributions. The method of moment matching and Gauss-Hermite quadrature formula are used to simulate distribution. Simulations are performed using an existing electrochemical system simulator that in turn uses the data obtained from the uncertainty input generator to simulate energy and power, which can be considered as a black-box function. The simulation results are quantified graphically through error bar plots that visualize the impact of the uncertainties. For the optimization part, the variation and optimization of power and energy densities as a function of current density of the battery electrode are presented using GPy package and the result are obtained and plotted under uncertain input parameters. Bayesian optimization will be utilized to determine the global optimization through the black-box function. Additional work may be needed to include more of the uncertain variables in this framework.

Introduction

In the high-precision world of lithium-ion battery manufacturing, the concept of a "perfect" specification is often a theoretical ideal rather than a practical reality. Even with advanced production techniques, inherent variability exists in critical parameters such as cathode thickness, the volume concentration of active materials, and the radius of negative active particles.

Standard optimization models typically treat these variables as fixed constants. However, if a battery design is optimized for a specific thickness but the manufacturing process produces a slightly thinner or thicker cathode, the actual performance may fall significantly short of expectations. This research introduces a computational framework designed to optimize battery performance while accounting for this manufacturing "noise," ensuring that designs are robust and high-performing in real-world conditions.

Methodology

The researchers developed a three-part modular tool to bridge the gap between statistical uncertainty and electrochemical performance. The framework is designed to be flexible, treating the underlying physics as a "black box" to allow for future updates or different simulation engines.

1. Uncertainty Input Generator

This module allows users to select statistical models for input parameter distributions. To make these distributions computationally manageable, the team employed moment matching and the Gauss-Hermite quadrature formula. These mathematical techniques allow the system to approximate complex probability distributions using a small set of discrete points. This approach significantly reduces the number of simulations required compared to traditional Monte Carlo methods, which would be computationally prohibitive for complex electrochemical models.

2. Electrochemical System Simulator

The tool utilizes an existing multiscale simulator to model the battery's internal physics. By treating this simulator as a black-box function, the optimization tool focuses on the relationship between inputs (such as current density) and outputs (energy and power density) without requiring a total overhaul of the underlying physics code.

3. Bayesian Optimization and Visualization

To find the global optimum efficiently, the framework utilizes Bayesian optimization via the GPy Python package. This approach is particularly effective for optimizing functions that are expensive to evaluate. The results are then visualized using error bar plots, which provide a clear graphical representation of how performance might fluctuate due to manufacturing uncertainties.

Key Insights

  • The Black-Box Advantage: The modular nature of the framework allows it to be paired with various simulation engines or updated physics models without changing the optimization logic.
  • Quantifying Reliability: The use of error bars in the output visualization does more than identify the best design; it quantifies the level of trust an engineer can place in that design given the manufacturing tolerances.
  • Computational Efficiency: By using Gauss-Hermite quadrature, the researchers successfully captured the essence of uncertainty distributions without the need for thousands of individual simulations.

Results and Discussion

The study successfully demonstrates a functional pipeline that maps the impact of manufacturing variability on battery performance. Key results include:

  • Optimization under Flux: The tool successfully plotted the variation of power and energy densities as a function of current density while accounting for uncertain inputs.
  • Global Optimization: Using Bayesian optimization, the tool navigated the simulation space to find global performance peaks that might be missed by simpler local search methods.
  • Uncertainty Visualization: The framework provides a clear visual link between manufacturing tolerances and performance risks, such as potential energy or power loss.

Limitations and Future Work

While the framework provides a robust foundation, the authors identify several areas for expansion. Currently, the tool focuses on a limited set of parameters (cathode thickness, concentration, and particle radius). Future iterations should incorporate a broader range of variables to better reflect the complexity of a full production line. Additionally, as more variables are added, the "curse of dimensionality" may require even more advanced sampling techniques. Finally, comparing these optimized models against physical battery cells produced in a laboratory setting remains a critical next step for validation.

Summary updated

Abstract

The motivation of this tool is to optimize the performance of battery based on energy output. During the manufacturing process, several parameters such as cathode thickness, the volume concentration of cathode and radius of negative active materials are subject to uncertainty. To optimize battery performance, it is significant to quantify those uncertainties through electrochemical multiscale computer simulation. Hence, this tool will focus on the optimization of the performance of lithium-ion battery under different currents. This tool will consist of a module on visualized generator of uncertainty input, an electrochemical system simulator, a visualization of output optimization module. First, the uncertainty input generator provides the option for selecting one of several statistical models for the input parameter distributions. The method of moment matching and Gauss-Hermite quadrature formula are used to simulate distribution. Simulations are performed using an existing electrochemical system simulator that in turn uses the data obtained from the uncertainty input generator to simulate energy and power, which can be considered as a black-box function. The simulation results are quantified graphically through error bar plots that visualize the impact of the uncertainties. For the optimization part, the variation and optimization of power and energy densities as a function of current density of the battery electrode are presented using GPy package and the result are obtained and plotted under uncertain input parameters. Bayesian optimization will be utilized to determine the global optimization through the black-box function. Additional work may be needed to include more of the uncertain variables in this framework.
BibTeX
@inproceedings{zhang2016optimization,
  author = {Lefei Zhang and Salar Safarkhani and Guang Qiang Lin},
  title = {{Optimization under Uncertainty Tool for Modeling Porous Lithium-Ion Batteries}},
  booktitle = {Purdue},
  year = {2016},
}
Figure