Sensitivity Analysis: How to Rank the Importance of Battery Model Parameters Using Simulation

Written by Ryan Dudgeon

May 13, 2022
battery modeling sensitivity analysis simulation

What is Sensitivity Analysis?

Sensitivity analysis is the study of how uncertainty in a model output (or response) can be attributed to different sources of uncertainty in the model inputs (or factors). These analyses are powerful tools for modeling and simulation activities, as they allow the modeler to identify the most and least influential parameters on the key predicted outputs of interest. Not only does knowing the important parameters allow the modeler to focus on a select few inputs for design work, but perhaps more importantly, identifying the negligible parameters allows a modeler to omit them from difficult optimization problems by setting them to constant values, thereby simplifying the optimization task.

The data science literature contains several sensitivity analysis methods. A general characteristic among them is a trade-off between number of required model evaluations and amount of information that can be extracted. For example, the so-called Sobol method can quantify the first-order, second-order, and even factor-factor interaction effects on a response of interest, but it often requires multiple thousands of model evaluations to extract this type of detailed information. As a result, we usually only perform the Sobol method on fast-executing metamodels.


The Elementary Effects Method & Use of Simulation 
In contrast, this study focuses on the the Elementary Effects method (sometimes called the Method of Morris), which has an advantage of being computationally efficient by requiring a relatively low number of model evaluations. However, the outputs of the method are not as detailed. As a result, it is good at providing a general ranking of many parameters, and for this reason it is particularly useful for identifying parameters with negligible influence that can be discarded from subsequent optimization or Design of Experiments studies.

This method was successfully applied to over 40 parameters of a GT-SUITE engine friction simulation model to reduce the complexity of an optimization, as described in this paper, A Global Sensitivity Analysis Approach for Engine Friction Modeling, by Oleg Krecker and Christoph Hiltner.

The full study will similarly apply the method to a battery model prior to an optimization task.


Why Engineers are Finding the Elementary Effects Method Beneficial 
As a computationally efficient method for sensitivity analysis that requires a relatively low number of model evaluations, the Elementary Effects method can rank all the parameters, identify the most important ones, and identify the negligible ones and is a useful modeling tool to gain insights to how a large number of parameters are affecting a model.

Applying the Elementary Effects method to a battery model having numerous parameters prior to optimization can aid in identifying parameters that have a negligible impact on the voltage and temperature results of interest, therefore reducing the optimization problem’s complexity by working with fewer parameters. To learn more about this in detail, we invite you to read the read the full technical case study, click here.


The study, “Sensitivity Analysis and Factor Screening to Rank the Importance of 34 GT-AutoLion Battery Model Parameters,” highlights:

  • A brief background on sensitivity analysis and their different types
  • How GT-AutoLion‘s simulation battery model was used to characterize the voltage and temperature responses at four different operating conditions
  • Why the Elementary Effects Method was the chosen sensitivity analysis method
  • How 34 battery model parameters were tested and how the Elementary Effects Method ranks the importance of these parameters
  • How the elementary effects method was applied in GT-SUITE using a Design of Experiments (DOE) study

To read the full study, click here.