Li-ion Computer Vision Solution Offers Significant Means to Improve Device Battery Life

6 months ago by Sam Holland

Researchers led by the SLAC National Accelerator Laboratory have developed a computer vision system that, unlike human eyesight, can closely identify lithium-ion battery (LIB) chemistry issues.

The scientists have determined that their machine learning solution may greatly improve how we understand Li-ion degradation.

 

Research Overview

The research was conducted at SLAC (the Stanford Linear Accelerator Center) and focused on a key element of battery LIB chemistry: nickel-manganese-cobalt (NMC) particles within a battery’s cathode.

In particular, the study concerned the use of computer vision, or CV (as well as quantitative X-ray phase-contrast nanotomography, which was achieved thanks to the collaboration of two facilities: the SSRL and the ESRF).

Such technology was used to monitor the said NMC particles, particularly when they detach from their requisite cathode—as this follows the theory of various past research that battery degradation is largely caused by the break-away of NMC particles from their LIB’s carbon matrix.

 

X-ray Tomography Scan NCM particles.

SLAC researchers' computer vision algorithm achieves a groundbreaking ability to track a lithium-ion battery's cathode degradation by monitoring nickel-manganese-cobalt particles and their movement. Pictured: an X-ray tomography scan of NCM particles. Image Credit: SLAC National Accelerator Laboratory.

 

The Advantages and Disadvantages of Computer Vision

 

Advantages

Of course, to monitor the signs of degradation in LIB cathodes, the SLAC researchers knew that, even with technological aids, the human eye alone would not be enough to observe the process of the NMC particles departing from their matrix.

Accordingly, the researchers used a CV system to ensure the process was reliable. Computer vision, not to be confused with its sub-category, machine vision (which is more concerned with image inspection, and therefore goes hand-in-hand with AI-based assembly line monitoring), is a fast-developing technology that is capable of image analysis.

That means the solution can not only inspect images, but identify attributes within those images, too—this being the crucial advantage needed to determine the movement and conditions of the NCM particles, which would otherwise be imperceptible to humans.

This enabled the researchers to determine that the aforementioned theory is in fact valid: NMC particle detachment does indeed facilitate LIB degradation—“at least under conditions [that] one would typically see in consumer electronics, such as smartphones”, write SLAC staff on their News page.

Nevertheless, SLAC’s CV solution did present the researchers with obstacles.

 

The Disadvantages and Their Solutions

The disadvantage of using computer vision was that CV algorithms rely on identifying shapes based on their light or dark lines. “[T]hey’d have a hard time differentiating between several small NMC particles stuck together and a single large but partially fractured one,” writes SLAC. This was, of course, a problem, as NMC particles can both cluster and crack.

The SLAC researchers’ solution was to use a type of algorithm that could achieve hierarchical object recognition*, thus ensuring that the CV software had a context-sensitive understanding of NMC particles (again, consider the importance of it being able to identify whether they are broken up, clustered, and so on).

Following the necessary human-to-machine training from the researchers, the algorithm led to the system having a ‘3D picture’ of a single NMC particle, and with this came the CV system’s ability to better identify when such nickel-manganese-cobalt particles detached from their cathode. “Older methods would mistake a single fractured particle for several different particles, while [this] new method can tell the difference,” write SLAC staff.

*For more information on the use of hierarchical object recognition, read up on convolutional neural networks in the SLAC researchers’ Nature Communications paper, as well as their chosen system, ‘Mask R-CNN’.

 

A workflow of SLAC researchers' machine learning-based segmentation.

A four-part workflow of SLAC researchers' machine learning-based segmentation of a Li-ion battery's nickel-magnesium-cadium particles. Image Credit: SLAC (the Standford Linear Accelerator Center).

 

The Implications of the Research

SLAC’s research supports a long-standing consensus that the detachment of NMC particles from a LIB’s cathode is a root cause of battery degradation. But, more than this, according to one of the senior authors, Yijin Liu, it also calls into question what now appears to be a misunderstanding in science: NMC-based Li-ion degradation can be curbed by making battery particles smaller. Such an assumption was based on the pre-existing belief that NMC particles are more prone to break away from a LIB’s cathode if such particles are large.

Now, however, SLAC’s computer vision results suggest that all manner of broken-off NMC particles, including the small ones, may be linked to battery degradation. This should certainly inform Li-ion manufacture and R&D—and therefore LIB problem-solving and optimisation. Referring to this as “very valuable information”, the researchers write in Nature Communications:

“small particles have a [high] degree of uncertainty in terms of … physical detachment from the carbon/binder matrix, which … could inform the engineering effort to optimise the electrode formation for fast charging applications”.

Ultimately, SLAC’s LIB and computer vision research reflects not only the importance of understanding how to best maintain Li-ion batteries, but also the extent to which computer vision has advanced in recent years. It appears that these two areas of research combined may bring a leap forward in lithium-ion battery research and development.

Comments