Stanford, MIT And Toyota Collaborate To Predict Battery Cycle Life

- Aug 16, 2019-

If cell phone battery manufacturers can determine which batteries can be used for at least two years, they can sell them to handset manufacturers and sell other types of batteries to less demanding equipment manufacturers. According to foreign media reports, a new study by Stanford University, the Massachusetts Institute of Technology, and the Toyota Research Institute found that manufacturers can do this. And the technology can be used not only to classify batteries, but also to help new battery designs get to market faster. Scientists combined with comprehensive experimental data and artificial intelligence revealed the key to accurately predicting battery life before lithium-ion battery capacity begins to decline. Using hundreds of millions of battery charge and discharge data points, the researchers trained a machine learning model that predicts how many charge and discharge cycles each battery can last based on voltage drops and some other factors in the early cycle of the battery.

The predicted value is within 9% of the number of cycles the battery can actually sustain. In addition, the algorithm classifies the battery as a long-life battery or a short-life battery based on the first 5 charge and discharge cycles of the battery. Moreover, 95% of the predictions are correct. This machine learning method accelerates the development of new battery designs and other applications, reducing production time and costs. Researchers have exposed the data set, which is the largest in its class.

One of the key points of the project is to find a better way to fully charge the battery within 10 minutes, which may accelerate the popularity of electric vehicles. To generate the training data set, the team continued to charge and discharge the battery until the battery's useful life was exhausted, with the researchers defining the battery capacity loss as 20%. In the process of optimizing fast charging, the researchers wanted to find out if it was necessary to exhaust the battery all the time. Can the answer be found in the early charge and discharge cycle of the battery?

In general, the capacity of a lithium-ion battery is stable for a period of time before it drops sharply. Most consumers in the 21st century know that there will be big differences in the sharp drop points. In this project, the battery used can last for 150 to 2300 cycles. The difference in the number of cycles is partly due to the different fast charging methods tested, but the difference in battery manufacturing also causes the difference in the number of battery cycles.

There are many potential applications for new methods developed by researchers. For example, the verification time of a new type of battery can be shortened. Considering the rapid development of battery materials, it is very important to shorten the verification time. In addition, with this sorting technology, batteries with shorter life spans for electric vehicles can be used to power street lights or as a backup power source for data centers. In addition, companies that recycle batteries can also find batteries with sufficient capacity from second-hand electric vehicle battery packs for secondary use.

In addition, another potential application is to optimize battery manufacturing. The final step in battery manufacturing is "forming", which typically takes days to weeks. But the methods provided by the researchers can significantly shorten the process, thereby reducing production costs. Researchers are currently using their machine learning model to optimize the method of fully charging the battery in 10 minutes, which they say can reduce charging time by more than 10 times.