Fast-charging batteries are considered critical for the spread of electric cars. Long waiting times at the charging station are very unattractive for buyers and thus thwart plans to establish electromobility on a broad basis. The mass storage of energy in batteries is also an important future technology for increasing the share of renewable energies in the energy supply.
However, the development of new batteries is costly and time consuming. For example, a fast-charging battery has to strike a good balance between charging speed and service life. But there are innumerable ways to modify chemical composition, structure and other factors – and an improvement in one area, such as the energy density, often leads to losses in other areas, such as the charging speed.
In practice, battery technology only slowly improves in small steps over a long period of time. This also slows down other technologies such as the aforementioned electric car or XR glasses which rely on advanced battery technology. Artificial intelligence could automatically optimize the countless changeable parameters in battery development and thus ensure faster progress.
Research results in a month instead of two years
In 2019, researchers from Stanford University, MIT, and the Toyota Research Institute developed an AI that could predict the performance of lithium-ion batteries over their lifetime without the batteries having to be tested experimentally. That was not possible before: batteries had to be charged and discharged countless times until their charging capacity decreased.
For the AI training, the researchers used a data set for which they discharged hundreds of batteries. The AI ‹ ‹learned from the degradation of the batteries’ performance to make predictions for other battery architectures.
At the beginning of the year, the researchers then showed a new AI discovery: the optimal method for charging a lithium-ion battery in ten minutes. In the publication in the journal Nature , the researchers led by materials scientist William Chueh write that thanks to the AI, the optimized charging protocol was found in less than a month. Otherwise equivalent results would be produced in two years of research, says Chueh.
Automated search in the infinite material space
Aside from predicting the battery life or faster charging times, researchers at the American Argonne National Laboratory are trying to use AI to find new substances for electrolytes directly. To do this, the researchers are building a database of molecules that artificial intelligence searches for suitable electrolyte candidates for better battery performance. To do this, the researchers start with small molecules and rely on the AI ‹ ‹learning on this basis to assess more complex molecules.
Our goal is to use AI methods to search the virtually infinite space of possible materials, says Ian Foster, Director of the Argonne National Laboratory. So far, this process has been more of a trial-and-error method.
The team has not yet discovered any new material; research is just beginning. But as soon as the AI ‹ ‹has discovered an electrolyte candidate, a battery should be built and tested. The data obtained in this way should further improve the AI.
Competition makes cooperation difficult
Cheuh and Foster see a major problem in the lack of data: So far, battery data has not been freely exchanged between researchers and companies – the information is too valuable for a possible product.
However, with their 2019 research, Cheuh and colleagues published their complete data set. It is the largest set of battery performance data to date.
Foster and other battery researchers are looking for a way beyond open source: The Data Station platform is intended to help researchers train AI models together without having to share the valuable data. Each team keeps its internal data to itself and all researchers still benefit from it.
Data Station allows researchers to train an AI model with a collection of data sets from different groups without one group having to give direct access to their data. For this purpose, an AI model is uploaded to the platform, trained with the collected data and returned to the individual groups fully trained. The researchers do not know the exact content of the training data, but can test whether the AI ‹ ‹has gotten better at its task.
First AI successes in research and industry
Already there are early successes of AI assistance in battery research: AI has helped new stabilizers for lithium metal anodes to develop potential cathode layers to discover, battery management systems to improve or mathematical models to design batteries.
The Slovak company InoBat uses AI to develop complete batteries instead of individual improvements. The company relies on an AI-based research platform that designs rapid prototypes with new chemical compositions and predicts their performance.
This method is ten times faster than the traditional way, says InoBat’s managing director Marian Bocek. InoBat recently presented a first intelligent battery . It should increase the range of current electric vehicles by up to 20 percent.
InoBat plans to start the first production facility for the AI-designed batteries by the end of next year. It will take some time before the first AI battery can be found in many electric cars: the planned system has around half a percent of the production capacity of Tesla’s Gigafactory in Nevada. A larger factory should then go into operation in five years.
