July 28, 2022
Researchers at the University of Michigan have developed a new approach that can test new electric vehicle battery designs up to four times faster.
Their optimization framework could significantly reduce the cost of evaluating the performance of battery configurations over the long term.
“The goal is to design a better battery, and traditionally the industry has tried to do that using trial and error testing,” said Wei Lu, professor of mechanical engineering at UM and chief from the research team behind the framework, published in Patterns-Cell Press. “It takes so long to assess.”
As electric vehicle (EV) battery makers grapple with range anxiety and concerns about charging availability, the optimization system developed by Lu’s team could reduce around 75 % the simulation and physical test time of new and better batteries.
This speed could give a major boost to battery developers looking for the right mix of materials and configurations to ensure consumers always have enough capacity to reach their destinations.
Parameters involved in battery design include everything from the materials used to the thickness of the electrodes to the size of the particles in the electrode and more. Testing each configuration usually means several months of fully charging and then fully discharging – or cycling the battery – 1,000 times to mimic a decade of use.
It is extremely time-consuming to repeat this test through the large number of possible battery models to discover the best ones.
“Our approach not only reduces test time, it automatically generates better designs,” Lu said. “We use early feedback to weed out unpromising battery configurations rather than recycling them to the end. is not a simple task because a battery configuration that performs poorly in the first cycles may perform well later, or vice versa.
“We systematically formulated the early termination process and allowed the system to learn from accumulated data to produce promising new configurations.”
To achieve a dramatic reduction in time and cost, UM engineers leveraged the latest advances in machine learning to create a system that knows both when to stop and how to improve over time.
The framework stops cycle tests that do not start promisingly in order to save resources by using the mathematical techniques known as the asynchronous successive halving and hyperband algorithm.
During this time, it takes data from previous tests and suggests promising new sets of parameters to study using Tree of Parzen estimators.
In addition to cutting out tests that lack promise, a key time-saving element in UM’s system is the way it generates multiple battery configurations to test at the same time, known as asynchronous parallelization. If a configuration ends the test or is rejected, the algorithm immediately calculates a new configuration to test without having to wait for the results of the other tests.
UM’s framework is effective at testing designs for all types of batteries, from those used for decades to power internal combustion automobiles, to the smaller products that power our watches and cell phones. But EV batteries may represent the technology’s most urgent use.
“This framework can be tuned to be more efficient when a performance prediction model is incorporated,” said Changyu Deng, a UM doctoral student in mechanical engineering and first author of the paper. “We expect this work to inspire improved methods that lead us to optimal batteries to make better electric vehicles and other life-enhancing devices.”
A recent survey conducted by Mobility Consumer Index showed that 52% of consumers are now considering an electric vehicle for their next vehicle purchase. Despite the evolution of mentalities, concerns remain about the autonomy of vehicles (battery capacity) and the number of charging stations available to drivers.
Battery performance therefore plays a central role in making electric vehicles available to the masses to offset the impacts of climate change.
“By dramatically reducing test time, we hope our system can help accelerate the development of better batteries, accelerate the adoption or certification of batteries for various applications, and accelerate the quantification of model parameters for battery management systems. battery,” Lu said.