A Drexel University research team has developed a way to forecast
lithium-ion battery performance by analyzing microscopic images of
electrodes before cells are ever assembled, potentially transforming quality
control in an industry where cell assembly and formation account for more
than 60% of manufacturing costs.
Led by
Maureen Tang, PhD, associate professor of chemical and biological engineering and affiliate
faculty of mechanical engineering and mechanics, the study combines
electrochemical fluorescence microscopy with machine learning to predict
discharge capacity across multiple charge rates. The work appears in
ACS Applied Materials & Interfaces
and addresses a longstanding challenge in battery production: identifying
suboptimal electrodes before they enter the costly downstream manufacturing
steps.
The researchers analyzed nearly 200 fluorescence images from battery
electrodes fabricated at Argonne National Laboratory. Using a technique
called electrochemical fluorescence microscopy, the team mapped electronic
connectivity across each electrode's surface by introducing a special
fluorescent molecule that lights up only in regions where electrons can flow
freely.
"When we apply a small current, the molecule fluoresces only in regions
where continuous electronic networks exist," Tang explained. "This reveals
connectivity with high spatial resolution. Bright areas indicate strong
electronic pathways, while dark regions show electronically isolated
particles that won't contribute to battery performance."
From each image, the researchers extracted statistical patterns capturing
spatial organization and texture. A machine learning model trained on these
image features predicted battery capacity with high accuracy and errors
below 2%, using only a handful of key measurements. The most important
indicator turned out to be the count of disconnected dark regions, which
correspond to electronically isolated particles that cannot store or release
energy effectively.
"Design practitioners can use our compact equation to predict performance
from just a few image-derived metrics," said Karla Negrete, PhD ’25, the
paper’s lead author. "We achieve accuracy comparable to impedance-based
approaches while avoiding their reliance on post-formation data. That timing
difference is critical for manufacturing."
The framework offers advantages over conventional testing methods, which
measure electrical properties only after cells are fully assembled and
undergo initial charging cycles. Those measurements also tend to report
primarily on the best-connected pathways in an electrode, potentially
masking weak spots that fluorescence imaging reveals directly.
The practical implications extend beyond cost savings. Nominally identical
battery cells can diverge widely in capacity and lifespan, with the weakest
units limiting module safety and reliability in electric vehicles and
consumer devices. Early detection of electronic network defects could enable
proactive screening before faulty electrodes compromise entire battery
packs.
"This approach may expedite battery research and development by reducing the
number of cell experiments required to determine electrode viability," Tang
said. "For manufacturing quality control, we can detect defected electrodes
upstream of cell assembly for faster feedback and reduced scrap rates."
The team notes that scaling this approach will require collaboration among
national laboratories, academia and industry, establishing the foundation
for performance-guided optimization in electrode manufacturing. Ultimately,
data-driven manufacturing will rely on frameworks like this to connect
electrode fabrication, electronic structure and battery performance.