Predicting Battery Life Using Machine Learning
Researchers at Argonne National Laboratory used a Machine Learning tool to accurately predict the number of cycles at which a battery performs.
Artificial intelligence (AI) has been proven a great invention. It is being used as a research tool to improve sustainability and forecast future developments. Machine Learning (ML) is a subset of artificial intelligence in which a lot of data, “big data,” is collected, interpreted and analyzed to reach a certain goal.
Researchers at the Department of Energy’s Argonne National Laboratory used a machine learning algorithm to predict the life of batteries within a few cycles of their operation. This technique is a relief for researchers working on battery development projects. The months and sometimes years spent on producing performance degradation estimates and prediction of the end of life of a battery using experimental testing could be skipped by using the new method.
A battery degradation sketch. Image used courtesy of Argonne National Laboratory
“Having to cycle a battery thousands of times until it fails can take years; our method creates a kind of computational test kitchen where we can quickly establish how different batteries are going to perform,” said Argonne computational scientist Noah Paulson, an author of the study.
Previous Work and Challenges vs. the New Method
Traditional battery life testing techniques such as using capacitors and resistors and cycling the battery are still used to project the battery's degradation profile. There have been a lot of other trials and research such as using mathematical models, physics-based models, and even machine learning. But all the predictions produced were not accurate due to lack of data or batteries' different chemical properties.
The researchers took into consideration all the previous machine learning algorithms used in predicting the life of batteries and addressed the problem that all the algorithms faced. The main problem was the ability to address a wide variety of failure mechanisms in batteries. The researchers collected data from a decade of many research topics in battery failure mechanisms. With these, they created a big database that addresses all the relevant failure mechanisms.
The model was tested with 300 pouch lithium-ion cells with different cathode chemistries. The 300 cells were divided into 4 sets, a training set to train the machine learning model, a validation set which is also used in training the model by comparing and evaluating the machine learning algorithms, a test set used with batteries having similar chemical characteristics as the training set and test the accuracy of the model, and an unseen cathode set which have cells with different chemistries than the cells in the training set. The machine learning algorithms used offered an excellent predictive quality across different chemistries.
Graphical abstract of the article showing the whole process for predicting different battery cycle life. Image used courtesy of the Journal of Power Sources
Future Study and Uses
In the future, the researchers seek to create a machine learning model that takes known chemistry as its input dataset and produces an output predicting the life of a battery with unknown chemistry.
This could be used by other researchers working on a new battery development technology, the algorithm could be used prior to experimental testing of the battery cycles. The researcher then could make a decision whether to do completely experimental testing or not. This could spare a lot of lost time in the process of developing new batteries.
“Say you have a new material, and you cycle it a few times. You could use our algorithm to predict its longevity, and then make decisions as to whether you want to continue to cycle it experimentally or not,” Paulson said.
The study was funded by an Argonne Laboratory-Directed Research and Development (LDRD) grant.