Over 250,000 electric vehicles (EVs) were sold every week last year globally, according to a recent survey from the International Energy Agency. The survey also found that consumer adoption is at a tipping point, with industry executives expecting EVs to account for 40% of car sales by 2030, largely due to EVs becoming cheaper.
The battery is the single biggest contributor to the cost of EVs—and a hotspot for concerns around safety and performance.
But new research from the University of Arizona shows that machine learning could help prevent EV batteries from exploding. Automakers can also use advanced algorithms to determine the specific chemistry, size and shape that leads to the best performance and more sustainable cars.
“Developing and perfecting these hypothetical batteries could unlock a billion-dollar opportunity,” said Benjamin Boeser, an innovation director at Mercedes-Benz, an IBM partner.
While it’s early days, AI is expected to increase the perceived value of EVs by more than 20%, according to a recent field survey from the IBM Institute for Business Value.
Safety first
The lithium-ion battery has been king of the global EV battery market for years, largely due to its exceptional energy density, extended lifespan and lightweight design.
Safety concerns persist with lithium-ion batteries, particularly related to the risk of “thermal runaway.” This occurs when the temperature of the batteries spikes unexpectedly, posing the risk that the batteries fail, catch fire or even explode.
Thermal runaway can occur because of various factors, including overcharging, over-discharging, high-temperature exposure, manufacturing defects and internal short circuits.
To tackle thermal runaway, a team at the University of Arizona has developed a machine learning model to predict and prevent temperature spikes in lithium-ion batteries.
Electric vehicles typically have a battery pack consisting of hundreds of closely connected cells packed into modules. If thermal runaway occurs in one cell, nearby cells are highly likely to heat up as well.
“It creates a chain reaction as the temperature in the battery accelerates in an exponential manner,” says Basab Ranjan Das Goswami, the key researcher on the project. “If this happens, the entire battery pack in the electric vehicle could explode.”
Inspired by weather forecasting frameworks that factor in time and location, the team developed an algorithm to predict when and where thermal runaway is likely to start. Using thermal sensors wrapped around battery cells, they fed historical temperature data into a machine learning algorithm to predict future temperatures.
“If we know the location of the hotspot or the beginning of thermal runaway, we can plan solutions to stop the battery before it reaches that critical stage,” Goswami said.
A potential future solution could include an early warning system that detects future hotspots and activates a safety switch to cut off electrical connections from the main pack.
Using AI to perfect EV battery chemistry
In addition to making batteries safer, advanced algorithms can also improve their performance and sustainability.
“In the future, low-energy consumption AI chips will help reduce the need for large batteries,” says Noriko Suzuki, a technology thought leader at IBM’s Institute for Business Value. “EV batteries are costly and very heavy, leading to more stress to road infrastructure.
IBM’s Research Lab in Almaden, California, has a dedicated project using AI and machine learning to develop more powerful, sustainable and energy-efficient batteries for EV vehicles.
Specifically, the team uses AI and machine learning workflows to speed up the discovery and optimization of electrolyte materials—a critical component impacting a battery’s safety, stability and efficiency. This workflow integrates automated simulation workflows and extensive domain-specific knowledge and datasets with purpose-built AI models, to discover new electrolyte formulations for high performance batteries.
“AI has the potential to supercharge the discovery of complex battery materials and processes, enabling faster charging, higher energy density and improved sustainability,” said Murtaza Zohair, an IBM research scientist in the Almaden lab.
The multi-modal AI foundation models, large chemistry models pre-trained on over 90 million molecules, can be fine-tuned with labeled battery datasets to predict the properties of complex materials, such as electrolyte formulations, to optimize battery performance. Deep search algorithms can quickly pull knowledge from tremendous volumes of existing scientific literature. The simulation toolkit can efficiently decide what kind of simulations to run to better understand materials and their functions. Finally, generative models can rapidly suggest new materials.
When GPS meets AI
While some researchers are focused on the internal battery chemistry, others are using AI to study the external factors that can help reduce the energy consumed by the EV battery.
A group of researchers at the Arab Academy for Science in Giza, Egypt, recently developed a new algorithm for the GPS system of EVs, which selects the shortest, fastest and most energy-efficient travel routes.
The researchers used the algorithm to examine the energy consumption of electric vehicles across different road topologies, evaluating the impact of various vehicle models, wind speed and direction on energy consumption.
The researchers discovered that wind speed and direction could conserve approximately 49% of the battery’s capacity during a short journey spanning approximately 50 km. They also found that road topology had a big impact on range prediction and energy consumption. Opting for a road with fewer inclines, for instance, could result in energy savings of about 46%.
Taking all these factors into consideration when choosing the optimum route for the EV could save money and increase the lifetime of the battery.
Beyond the battery
Optimizing the entire EV battery charging infrastructure will also be critical for paving the way for broad EV adoption. For example, AI will likely be integral for “smart charging,” where algorithms monitor battery health and status and suggest optimal charging points and timing to ensure maximum driving range and battery life. AI can also help with security monitoring and incident response for vehicles and charging networks alike.
Machine and deep learning algorithms, combined with sensors and cameras, are already providing in-car assistance to drivers looking to enhance their vehicle’s safety and efficiency. Volkswagen, for instance, has integrated generative AI into many of its cars so that drivers could eventually, for example, ask to re-route the car to the nearest EV charging station.
In the long run, we can expect AI to transform the EV market and transportation more broadly as it integrates driver assistance systems and autonomous driving, enables predictive maintenance and helps integration with the smart grid. Ultimately, the road ahead for electric vehicle innovation is boundless—and AI is only helping to make it a smoother journey.
Download the ebook: Your guide to GHG emissions accounting