Basics of Battery Degradation
Today’s article will be a bit nerdy and wonky — but fret not! The goal is to take on a complex topic — battery degradation — and make it digestible for a layman audience (including myself).
Why do we care?
First of all, why is understanding battery degradation important? The most obvious reason is to be able to accurately capture performance degradation over time, so a customer knows what to expect out of their electric vehicle (EV) battery.
Understanding a battery’s state of health is also critical to accurately evaluating the residual values of an EV. Most of us can relate to this for private, passenger vehicle sales — if you buy a used internal combustion engine (ICE) vehicle, the price is typically determined by some combination of odometer reading, accident reports, and maintenance history. For EVs with fewer moving parts and service requirements, the importance of the odometer and maintenance pales in comparison to the battery’s condition. Residual value is even more relevant for commercial fleets, as most commercial fleets regularly buy new vehicles and sell them after a few years, or primarily buy used vehicles. Having an accurate estimate of residual value can make or break a fleet business.
Lastly, the lifetime of a battery will directly impact the amount of virgin minerals we have to extract, how much recycling capacity needs to be built, and the environmental impacts associated with processing these materials.
What do we care about?
When evaluating battery degradation, what metrics are most salient for companies and customers? After poring over a bunch of research papers and industry articles, it seems to come down to the following. The first two are fairly obvious to consumers:
- Capacity state of health: the % of energy capacity remaining, due to a loss of active material
- Power state of health: % of performance remaining, impacting charge and discharge rates, due to a rise in internal resistance of the battery¹
The remainder provide more nuance for companies to provide more accurate services:
- Cycle life remaining before hitting some end-of-life threshold, typically defined as 80% of rated energy capacity²
- Forecasting voltage-capacity curves for accurate range estimation
- Evaluating the likelihood of a more catastrophic failure rather than nominal degradation
Lastly, it’s important to be able to interpret these metrics at both cold and hot environmental extremes, as a battery’s capacity and performance varies across temperature.
What’s going on?!
But what is actually going on inside the battery that reduces capacity and performance? The key drivers of battery degradation, going from least to most impact (roughly), are:
Time
Even when a battery is just sitting there, a battery will very gradually degrade, although this is a relatively minor effect. This type of capacity and performance loss is typically called “calendar aging”, and is typically due to the growth of the solid electrolyte interphase (SEI), a necessary stabilizing layer that protects a battery’s anode from the electrolyte (just like the surface of an aluminum bar reacts with oxygen to form a protective aluminum oxide that prevents further reactions). The SEI layer doesn’t completely prevent reactions, however, and continued but much slower reactions lead to SEI growth, which both consumes usable lithium and impedes the movement of remaining lithium ions from the electrolyte to the anode, thus increasing internal resistance.
Cycling
Simply using the battery will also contribute to degradation, although less than the other factors below. While we tend to think of batteries as a solid-state, non-moving part of a car, a battery by definition requires atoms and molecules to move around. Extracting them from a cathode matrix, moving them through an electrolyte and the SEI layer, and shoving them into the anode in a process called intercalation physically causes expansion and contraction. These mechanical stresses can crack the SEI layer, anode, and cathode — which in turn leads to loss of electrical contact, loss of active material, and higher internal resistance. All of these effects are exacerbated by the factors below.
Heat
Heat increases reaction speed, so electrolyte decomposition, SEI growth, and other gradual effects become more pronounced. At very high temperatures, the SEI can decompose, and the reformation of the SEI consumes more active material in the anode and electrolyte. Furthermore, binder materials and transition metal cathode materials like manganese can react and dissolve into the electrolyte. These dissolved metals can precipitate on the anode surface, preventing Li ion diffusion and forming dendrites that may cause short circuits.
Extreme SOCs
At high SOCs, the anode has limited room for further lithium ion intercalation. This means the rate of diffusion from the surface to the center of the anode particle is reduced, so if too many Li ions arrive at the surface, they will preferentially bond with each other and form metal dendrites. These dendrites tend to be narrow and spiky, and in addition to consuming lithium, they may puncture the separator and cause a short circuit and thermal runaway.
High and low states of charge cause mechanical and electrical stresses. Between 0 and 100% SOC, for example, a graphite anode can expand up to ~13%. (Silicon-based anodes expand even more, to over 300%!) These mechanical stresses can cause the anode particles to separate from the electrically conductive binders that connect them to the metal current collectors, leading to increased electrical resistance.
At very high and low SOCs, the copper and aluminum current collectors can also become electrically unstable and dissolve. The metals are then transferred to the electrolyte before precipitating at the anode, also causing dendrites that increase resistance and may cause short circuits.
High current (discharging and charging)
High current also increases mechanical stresses because the battery tries to shove more Li ions around more quickly. Anode particles may crack and separate, causing a loss of active material and creating fresh anode surfaces for SEI growth. The cathode structure may also become disordered, reducing the number of sites where Li ions can intercalate and reducing diffusion rates through cathode particles.
Low temperature usage
The biggest risk of low temperature charging is lithium dendrite formation. Similar to what happens at high SOCs, low temperatures reduce the rate of diffusion from the surface to the center of the anode particle. If the rate of lithium ions arriving at the anode surface is greater than the rate of diffusion, lithium plating at the surface will occur, causing dendrites and potential short circuits.
Despite how scary the above sections may sound, a properly-designed battery management system will account for these factors and protect the battery from excessive usage. For example, at low temperatures the BMS will significantly throttle current into the battery to avoid lithium plating, and there is almost always additional buffer below and above the 0–100% SOC range visible to a consumer.
This EV-Tech Explained video and this Recurrent article give great overviews of battery degradation mechanisms. (And if you want a primer on how Li-ion batteries work, I can’t recommend The Limiting Factor strongly enough.)
All models are wrong, but some are useful
Now that we understand the mechanisms of battery degradation, how should we model their effect on the metrics we care about?
The most basic model for capacity fade is based on a combination of the Arrhenius Law to account for temperature and a power law to account for battery energy cycling. According to the paper, “the main limitation of this model is that it is not applicable at low temperatures, i.e. below 0 °C, and is recommended for applications between 15 and 60 °C”.
The paper illustrates additional capacity fade models that account for both cycling and calendar aging, as well as a variety of models for the change in internal resistance corresponding to the power fade. Note that these are cell degradation models, and the study assumes that all cells experience similar DoD and temperature, which may not be accurate; a more sophisticated model might consider a battery pack to be limited by its weakest battery module, and a battery module to be limited by its weakest battery cell. Additionally, driving behavior is quite spiky in terms of power draw, so to capture non-steady driving behavior, one can take a page from these authors and sample the C rate at a certain frequency to calculate the degradation at each timestep.
Ultimately though, battery degradation is difficult to model neatly in a closed, static form. To get more dynamic models, engineers have used online filters to update model parameters, such as the Kalman filter, particle filter, various machine learning algorithms, etc.
However, the model itself may need to be more complex to capture the various degradation modes; for example, lithium plating will cause much more rapid capacity degradation than cycling stresses. To address this, one could use methods like symbolic regression to simultaneously estimate both the model itself as well as its parameters, as NREL has demonstrated.
Some researchers have also proposed end-to-end machine learning, like leveraging a seq2seq model to forecast voltage-capacity curves.
Data, data, data
Of course, none of these techniques matter if you can’t extract the data to build the models. Unfortunately, collecting (and potentially transmitting) the data comes at a cost, so we need to be judicious in selecting the most important variables. Ultimately, the type of data we should capture should correlate with the drivers of battery degradation:
- Environmental: time, temperature (high and low)
- Usage: cycles, high current, time at high/low SOCs
- Battery performance: actual energy consumed before hitting voltage cutoff points to benchmark capacity, actual internal resistance
While some manufacturers like Tesla are closely monitoring their battery pack performance, how granular does that data collection get? For those who aren’t capturing degradation in-house yet, what does the landscape look like for obtaining that data? Are there options for creating a more public (or at least industry-wide) dataset of real-world battery degradation?
If you have insight into any of these questions, would welcome your thoughts!
[1]: A battery can be electrically modeled with two equivalent circuits: (1) a current source with an internal capacitance and self-discharge resistance, and (2) a voltage source with multiple parallel RC circuits in series with an internal resistance. This paper decided to use 3 parallel RC circuits in series as a balance between accuracy with empirical data and model complexity. The R/C values (through tau) can change based on current draw.
[2]: Why is a battery’s end-of-life defined at 80%? While a battery is certainly still usable beyond this point, and there is a small but robust second-life battery market, battery degradation typically hits a “knee” around this point. According to Qnovo, “tests have shown that the battery capacity loss tends to accelerate past 80%…this rapid capacity loss may be accompanied by an increased likelihood of lithium metal plating. In other words, a dead battery with its capacity past 80% becomes a serious fire hazard!” (The fact that the second-life battery market exists at all suggests that under much gentler usage than an automotive drive cycle the accelerated degradation and risk of lithium plating can be appropriately managed.)
