Using Simulation to Reduce Battery Testing Time and Cost
To predict the lifetime of a battery-powered product, engineers must understand how a battery will degrade over time. Popular methods of understanding battery aging usually rely heavily on physical testing, which is expensive or prohibitively time-consuming.
Simulation tools that use a physics-based approach to modeling Lithium ion cells, such as GT-AutoLion, enable engineers to decrease the amount of physical testing required to fully understand how Li-ion cells degrade over time.
Calendar and Cycle Aging
Standard experimental testing procedures for quantifying battery degradation include calendar aging and cycle aging formats. Calendar aging experiments store a Li-ion cell at various temperatures and states of charge for extended periods of time while the capacity of the cell is periodically checked. This data is generally visualized by showing the capacity retention (as a percentage of the beginning of life capacity) vs. the amount of days in storage. Cycle aging experiments cycle the Li-ion cell between 100% and 0% states of charge (or other SOC windows) at various temperatures and currents repeatedly. This data is generally visualized by showing the capacity retention vs. number of cycles the cell has run.
This data often comes from cell manufacturers, but sometimes cell buyers execute this testing themselves.
Depending on the number of cycles and current, cycle aging tests can take weeks or months to test. For example, if a cell is charged at discharge at a C-rate of 1 C, 500 cycles are completed in about 6 weeks.
Calendar aging, however, can take quite a bit longer. Depending on the projected life cycle of the product, different amounts of calendar aging may be required to properly test the degradation during the full life cycle of the product. For instance, cell phones may only be designed to last 2 years; whereas, battery electric vehicle may be designed to last 15 years. Unfortunately for the automotive industry, testing a Li-ion cell for 15 years is unfeasible because the standard development cycle is roughly 2-3 years. Additionally, battery technology changes very quickly – if 15 years of testing were done before the next generation BEV came out, the battery technology would be outdated by the time it was released.
Because of the great disparity between projected product lifetime and the product development cycle time, it’s not always feasible to rely solely on calendar aging or cycle aging data.
To help address this issue, physics-based aging models can be calibrated to available data and then used to project, or extrapolate, the degradation of cells beyond the available data.
Using GT-AutoLion to Predict Aging Beyond Measured Data
The calendar aging data presented earlier can be used to calibrate physics-based aging models in GT-AutoLion consisting of 4 parameters. In the image below, the 4 parameters were calibrated by using The Integrated Design Optimizer that comes with GT-AutoLion and GT-SUITE, which automatically varied the parameters to minimize the error between simulation and experimental data. The results of such a calibration shows good correlation between simulation and experimental data.
However, as you can see, this set of experimental calendar aging data was collected over an 870 day period, which is nearly 2 ½ years. What if you don’t have 2 ½ years to test the degradation of a Li-ion cell? The following images and discussion try to answer that question.
The images below show the power of physics-based aging models in GT-AutoLion by demonstrating how well they extrapolate after being calibrated to experimental data. Each plot has a portion with a white background, which is the data used to calibrate the model and a portion with a grey background, which is testing how well the calibrated model extrapolates into the future. For example, the image below assumes that only 450 days were available for testing the calendar degradation. Only the data in the white section (before the 450-day cutoff) was used to calibrate the model using GT’s Integrated Design Optimizer. The grey section illustrates how well the model extrapolates into the future for the other 420 days of available data.
This process was done for 750, 600, 450, 300, 100, and 50 days of data, and the results are shown in the image below.
As expected, the more data that is available for model calibration, the better the results will be. However, the results also show that even with a significant reduction in testing time, reliable physics-based models can be calibrated using GT-AutoLion. These calibrated aging models can then be used to predict how Li-ion cells will degrade in any system, which provides insight into how long a product will last and how well it will perform once it is aged. In my next blog, I’ll share more information on how physics-based simulations help engineers predict how batteries will degrade in real-world situations.
Written By: Joe Wimmer