Dr. Tanvir Tanim and his team at Idaho National Laboratory designed lithium-ion batteries that can be charged in 10 to 15 minutes at a roadside charging station. They developed a machine learning algorithm that detects undesirable lithium plating that could occur under these conditions.

Tech Briefs: How did the project start?

Dr. Tanvir Tanim

Dr. Tanvir Tanim: We got involved with the extreme fast charging program, which was sponsored by the Department of Energy Vehicle Technology Office in 2017. The goal was to enable extreme fast charging (XFC) in lithium-ion batteries — 10 to 15 minutes charging or so — basically to make the EV recharging experience comparable to gasoline vehicles’ refueling experience. That's one of the major considerations for electric vehicle consumers.

For that program, we were testing a lot of lithium-ion batteries at extremely fast charging rates. When you charge these batteries at high rates you encounter many issues — lithium plating is one of the major ones. Batteries contain a finite amount of lithium, so you don't want it to get lost during charging and discharging, you want it to cycle back and forth from anode to cathode. During fast charging, however, the cyclable lithium can be plated onto the anode surface, which is an undesirable parasitic reaction — the plated lithium cannot be recovered — during discharge it cannot return to the cathode.

There are other issues with plating. If you don’t detect it early on, it will keep on happening and could result in a catastrophic event. For example, the plated Li could take a dendritic form and grow like a needle from the anode and could puncture the separator, creating an internal short circuit.

So, we were testing a lot of lithium-ion batteries, doing design modifications, then retesting to see whether lithium plating was happening. At that time, however, we did not have a good way to detect the plating. After testing we would tear open the cell to see whether there was plating, or we would do additional post-testing. We didn't really have any solid electrochemical signature-based lithium-plating detection method but we were learning a lot.

Then we got involved in a different program, called physics-based machine learning, also funded by DOE. The goal for that program was to develop artificial intelligence/machine learning-based algorithms for obtaining a reliable projection of battery life along with identifying the underlying root causes of degradation. By that time, we had collected a lot of data and generated a comprehensive understanding of different degradation modes and mechanisms. We thought, since we have all that data, and we know electrochemical signatures that relate back to lithium plating, why don't we just formulate that into a machine learning problem. We could use the electrochemical signatures, bring machine learning into it, and see whether that could help us come up with a strategy to detect lithium plating.

One thing led to another, and we reached the conclusion that this could be a very nice method for detecting lithium plating early on, without tearing open the cell. Opening the cell and doing post-testing takes time, is expensive, and delays the battery lifecycle development.

Tech Briefs: So then, your method was to detect certain electrochemical signatures?

Tanim: By that time, we had tons of electrochemical data, and gained a good understanding of the physics behind it. We identified the key electrochemical signatures that could be related back to lithium plating. Then we ran with that data, brought machine learning into it, and solved the problem we had formulated at that time.

Tech Briefs: Could you tell me something about what particular kinds of data you used?

Tanim: We were primarily focused on electrochemical data because you can collect it easily — in fact, during testing we always look for electrochemical signatures to describe life and performance issues in lithium-ion batteries. Typical signatures are voltage, current, temperature, and so on. You can convert those signatures into different secondary variables. For example, we looked at discharge capacity and capacity fade trends, their linearity or nonlinearity. Also, the end of charge rest voltage, end of discharge voltage and how they change over cycling, and in addition, coulombic efficiency.

Tech Briefs: What does discharge capacity mean?

Tanim: If you are using the battery, you are taking energy out of it, that is what we call discharge. One of the measures of capacity is ampere-hours.

Tech Briefs: How about capacity fade?

Tanim: A new battery has a certain capacity — let's say one amp-hour. With cycling, that capacity is going to decrease. Capacity (or energy) fade is the percentage of decrease. In normal cases the decrease trend will be quite linear, particularly in the initial cycles. But with lithium plating, the trend is very nonlinear — higher rate of fade at the beginning and less fade later.

Lithium plating can occur under different conditions. In addition to fast charging, it can occur if you charge a battery at sub-zero temperature, or if there are aging-related imbalance issues in the battery. The signatures might not be equally sensitive for all those plating conditions. So, we identified the most sensitive signatures for fast charging and then used them to develop our machine learning algorithm.

Tech Briefs: What are the signatures you used?

Tanim: We found that the most sensitive signatures for fast charging were trends in how the cell capacity was fading, linearly or nonlinearly, end-of-charge rest voltage, and coulombic efficiency.

We also found that two prominent signatures others have reported, dQ/dV and dV/dt, were not very sensitive under fast charging conditions, unless there were very aggressive plating situations.

Tech Briefs: Can you explain coulombic efficiency?

Tanim: Coulombic efficiency is the percentage ratio of discharge capacity divided by charge capacity.

Tech Briefs: How do you measure capacity fade?

Tanim: You have to do some testing. On a lab scale, when we cycle the battery, we can choose a particular cycle to measure its charge or discharge capacity. From current and time, we can calculate the capacity in ampere-hours. Now if you repeat the same process as the battery degrades, you can find the capacity at an aged state and from there can calculate percent fade (the capacity fade with respect to fresh state).

A researcher analyzes data at INL’s Battery Test Center. (Image Credit: INL)

Tech Briefs: Have you experimented with different charging protocols?

Tanim: To avoid lithium plating you can modify the battery design in different ways: you can change materials, you can change electrode design or several other aspects of the battery design, such as the electrolyte.

You can also change the operating conditions or charging profiles. For example, you can try different charging protocols and compare them to a baseline. For example, instead of constant current/constant voltage, you can try multistep or other charging protocols. And you can also change temperature.

Our method for detecting lithium plating will be applicable irrespective of any design changes or charging protocols.

Tech Briefs: How do you see this being implemented?

Tanim: There are two scenarios by which this method will make a valuable contribution. The first is for use by research scientists in a lab. This method will more quickly tell us whether under the given operating conditions, lithium plating is either happening or not— we don't have to tear open a cell or do any other post-test analysis. Just the electrochemical signature will tell us whether lithium plating is happening for this particular design and operating condition. That can be done within 10 to 25 cycles. It lets us know whether we must modify the design of the batteries — we want to identify that as soon as possible. We can then go back, reiterate the design, and redo the test to see whether we are moving in the right direction.

Electrochemical signatures, with some modification and further verification, can also be implemented in a battery management system on-board an electric vehicle as well as in stationary applications where lithium-ion batteries are used. OEMs or battery manufacturers already collect most of these signatures. Using them as a baseline, we can, maybe not in an early lifecycle, maybe a few years down the road, alert the user that although the battery was fine at the beginning of its life, something has changed, and lithium plating has started to happen. “Since there is now lithium plating, this is an early warning that you should do something about the battery.”

Tech Briefs: Could you hazard a guess as to how soon this might be commercialized?

Tanim: We have a provisional patent on this and we are working on submitting the full patent very soon. We are also looking for collaboration opportunities to further develop it and we are seeing a lot of interest from private companies. Also, DOE has a technology commercialization fund where we can collaborate with other private industries and further improve it and demonstrate it for onboard applications, but I don't want to give you any particular guess.

An edited version of this interview appeared in the December 2021 issue of Tech Briefs.