🇺🇸▼
Lithium-ion batteries power the technology we use every day. They run electric cars, work in smartphones, and keep life-saving medical devices alive. Making better batteries is very hard. To test a new battery, engineers must charge it and let it drain many times until it stops working. This test takes a long time. It can take months or even years to finish. This long wait delays new and better batteries from reaching stores.
Now, scientists have found a faster way to solve this problem. They used machine learning to create a new method. This technique can predict how long a battery will last in just one week or less. Researchers led by Zhang and their team published these findings in the journal Nature. Their method uses data from only the first few times a battery is charged. They use this early data to guess how the battery will perform over a long time. By shortening the testing time, this method will help create better batteries for many important uses.
The most important question for any battery is its life. This measures how many times a battery can be charged and discharged before it loses too much power. Usually, a battery is considered done when it holds only 80% of its original power. This number is very important for selling electric cars and for safe devices like heart pacemakers. Unfortunately, guessing this number has always been very difficult.
The old way is simple but very slow. Engineers must charge and discharge batteries in a lab. They watch the power drop until the battery fails. For batteries meant to last many years, this test needs thousands of cycles. Even with fast machines, this process takes many months. This is a big problem. Researchers cannot test new materials or designs quickly because they must wait so long for results.
"The traditional approach is a huge bottleneck," says a battery researcher who was not part of this new study. "You might have a great new idea for a battery part, but you could wait a whole year just to see if it degrades too quickly." This slow feedback loop makes progress hard. We need fast improvements in energy storage to support clean energy and electric cars.
The new method from Zhang's team skips the long wait by using smart artificial intelligence. They do not need to wait for a battery to fail completely. Instead, their computer model learns to find hidden patterns. These patterns appear in voltage and power changes during the first few days of a battery's life. These signals are too small for humans to see, but they hold important clues about how the battery will age.
The researchers taught their computer using a public set of data. This data included hundreds of commercial batteries tested until they failed. The model looked at voltage data from the first 100 charge and discharge cycles. This is a very small part of a battery's total life. The computer learned to connect these early signs with the final number of cycles the battery could handle. It learned to predict when the battery would reach that 80% power limit.