The technique might lower the cost of battery development.
Consider a
clairvoyant informing your parents how long you would live on the day you were
born. Battery chemists who use new computational models to calculate battery
lifetimes based on as little as a single cycle of experimental data may have a
similar experience.
In a recent
study, researchers at the U.S. Department of Energy's (DOE) Argonne National
Laboratory used machine learning to forecast the lives of several battery
chemistries. The scientists can properly calculate how long various batteries
will continue to cycle by analysing experimental data collected at Argonne from
a collection of 300 batteries representing six distinct battery chemistries.
Scientists
educate a computer programme to make inferences on an initial collection of
data, and then use what it has learnt from that training to make conclusions on
another set of data in a machine learning process.
"Battery
lifespan is of vital relevance for every user for every different sort of
battery application, from cell phones to electric cars to grid storage,"
said Argonne computational scientist Noah Paulson, one of the study's authors.
"It can take years to cycle a battery thousands of times until it dies;
our technology offers a type of computational test kitchen where we can quickly
determine how different batteries will perform."
"Right
now, the only method to evaluate how a battery's capacity diminishes is to
actually cycle the battery," said Argonne electrochemist Susan
"Sue" Babinec, another research author. "It's incredibly
expensive and time-consuming."
According to
Paulson, determining a battery's lifetime might be difficult. "The fact is
that batteries do not live forever, and how long they last is determined on how
we use them, as well as their design and chemistry," he added. "Until
recently, there hasn't been a good method to predict how long a battery will
survive. People will want to know how long it will be before they have to spend
money on a new battery."
The study
was notable for relying on substantial experimental work done at Argonne on a
range of battery cathode materials, including Argonne's proprietary
nickel-manganese-cobalt (NMC)-based cathode. "We had batteries that
represented different chemistries, and they degraded and failed in different
ways," Paulson explained. "The benefit of this study is that it
provided us with indications that are typical of how different batteries
function."
More
research in this area, according to Paulson, has the potential to shape the
future of lithium-ion batteries. "One of the things we can do is train the
algorithm on a known chemical and then have it predict an unknown
chemistry," he explained. "Basically, the algorithm may push us in
the path of new and improved chemistries with longer lives."
Paulson
believes that the machine learning method might speed up the creation and
testing of battery materials in this way. "Assume you have a fresh
material that you have cycled a few times. You may use our method to anticipate
its lifespan and then decide whether or not to continue cycling it
experimentally."
"If
you're a researcher in a lab, you can find and test many more compounds in a
shorter period of time because you can analyse them faster," Babinec
noted.
The
publication, "Feature engineering for machine learning enabled early
prediction of battery lifespan," published in the online issue of the
Journal of Power Sources on February 25.
Other
authors of the study include Argonne's Joseph Kubal, Logan Ward, Saurabh
Saxena, and Wenquan Lu, in addition to Paulson and Babinec.
An Argonne
Laboratory-Directed Research and Development (LDRD) grant supported the
research.
0 Comments