By Bennett Davis
Machine learning software created by researchers at MIT, Stanford University, and the Toyota Research Institute can cut design and development time for new batteries by as much as 98 percent, the developers say.
The program also can be used to slash development times in areas such as drug research and laser design.
The scientists focused on batteries for electric vehicles, for which long charging times are a barrier to widespread public adoption.
Normally, battery design is a hit-and-miss proposition: designs are conceived, prototypes are built and tested, then this or that component might be altered to see if the change improves the amount of power the battery can hold, how fast it charges, or how long it lasts. The process to work through a single design can take years and even then it might be scrapped.
The research group programmed the new software to look for battery designs that would fully charge in ten minutes or less while maximizing battery life.
First the scientists fed the program long-term performance data from batteries that were fast-charged over and over until they failed. The software detected patterns after only a few dozen simulated cycles that predicted how soon a new design would fail. The software then ignored those approaches.
The software also tested charging patterns and found that the best ones used the highest current in the middle of the charging cycle, not at the beginning or the end, which is intuitively what humans tend to assume.
The software turned in a design that maximizes battery life using fast charging cycles and did it in 16 days instead of two years or longer that the process normally could have taken.
The same program can make similar leaps in designs of other batteries, such as flow batteries that store renewable energy captured from grid networks.
The research team is applying the software to find other leaps in battery technology, from novel materials to manufacturing methods.
TRENDPOST: Machine learning and artificial intelligence already are beginning to revolutionize product development and fabrication. We will become accustomed to a swifter, steady flow of new and better products as researchers and businesses adopt AI more widely.