If you enjoy this newsletter please give us a share and subscribe! For business enquiries, reach out to us.
To use the words of Steve Levine from The Mobilist, “The elevated mania around EVs has triggered a gusher of quality blogs, newsletters, podcasts and YouTube channels concentrating on batteries.”
The latest substack to join the mania is Roman Leventov’s Battery Discovery. It documents his journey into batteries by critically reviewing recent research publications from his perspective as a data and software engineer at Northvolt.
Roman has kindly shared a digest of his recent posts:
The authors trained an LSTM-RNN model to estimate the remaining (degraded) cell capacity using as inputs only time-stamped voltage readings during charging cycles (no current and temperature inputs).
The model works quite well (errors in estimated capacity generally within 1-2%), though it could have worked even better if it was trained against more realistic “reference” capacity values.
The weakness of the model is since it uses only output cell voltage as the input, the model cannot identify and compensate for sensor bias.
In real-world applications (or further research), would be interesting to have a neural network trained with voltage, current, and temperature measurements, learning how the sensor biases look like and relate to other measurements, and make unbiased state-of-health estimations from biased inputs completely by itself. Perhaps, we can integrate this idea within a stochastic cell parameter estimator.
Closed-loop optimisation (CLO) helps to find an optimal fast-charging protocol quickly, assuming only a limited number of experiments can run in parallel. The two pillars of CLO are the early outcome predictor and Bayesian optimization:
The authors found, “contrary to conventional battery wisdom”, that the optimal (minimising degradation) fast-charging protocol is when a cell is charged at a nearly constant ~4C rate between 0% and 80% SoC, rather than with decreasing current.
However, in the experiment cells were cycled non-stop and without active cooling. In such an experiment, the charging protocol that minimises the cell's equilibrium temperature may perform the best because accelerated parasitic reactions (such as the growth of the SEI layer on the anode) due to the temperature of about 40 °C might contribute more to cell's degradation than Lithium plating. A nearly constant charging rate minimises heat because of the quadratic I2R Ohmic law.
It's hard to train a physics-agnostic model to predict cell failures from the data because there are too few cell failures and the data to train from. The most practical ways of increasing the probability of cell failures during cycling are:
Strain the separator by dropping, bending, or pressing the cell.
Cycle the cell at a high or freezing cold temperature.
Apply a very high charge or discharge current to the cell.
However, cells can still fail only rarely even after such abuses, so to avoid wasting a lot of cells, short circuit induction should be an integral part of the general cell testing process at the end of the production line.
I think that separator failures should be characterised by a phase transition of one or several cell parameters (such as self-discharge rate and internal resistance).
Physics-based models can help cell engineers to uncover gaps in their understanding of the processes in the cell and inform better cell designs. However, in a battery management system, the most accurate available model should be used, even if it’s not really “physical” (e. g., a neural net).
The authors suggest applying the approach from this paper for probabilistic prediction of cell parameters by training a variational autoencoder (VAE) model after pre-training using the output of an electrochemical cell model.
Having a stochastic framework for estimating all parameters at once is essential because the Li-ion cell has just a single output signal: voltage. The changes to the voltage response can often be attributed to changes in different cell parameters (e. g., self-discharge rate, or the open-circuit voltage relationship, or charge capacity). Estimating all parameters in separation (e. g. using linear regression) will overreact to the changes in the cell by fully attributing them to several different parameters.
This post is a summary of a very big review paper and more materials about Lithium plating.
A physical precondition of Lithium plating is that the standard electrode potential of Lithium intercalating into graphite anode (–2.84) is close to the potential of Lithium metal formation (–3.04).
The risk of Lithium plating increases when charging either 1) at low temperature (see below), 2) with a strong current, 3) at a high state-of-charge level (esp. overcharging). These lead to an increased concentration of Lithium close to the surface of the anode.
Lithium plating can be recognised by a “shoulder” on the voltage relaxation curve:
Lithium deposition takes over SEI growth as the leading capacity fade mechanism at 1C rate and 25 °C. And at a lower charging rate at a lower temperature and at a higher charging rate at a higher temperature.
Small cells have more even internal temperatures and local state-of-charge (including due to smaller overall currents) than large cells. Therefore, local Lithium deposition is less likely in smaller cells.
Cell capacity fade curves often have a recognisable knee:
Researchers suggested that cell capacity fade accelerates when Lithium deposition becomes irreversible.
Frequent rest recovers capacity and prolongs a cell's life because Lithium plating is a self-reinforcing process, therefore, preventing it early leads to compounding benefits.
Roman Leventov is an engineer working on digitalising battery systems at Northvolt. Previously, he worked on distributed computing systems such as Apache Druid. "Battery Discovery", inspired by "The Morning Paper", is a blog consisting of summaries of papers in the battery domain. The goal is to help people to get the gist of the research results when they don't have time to read papers in full.
🌞 THANKS FOR READING
📧 For tips, feedback, or enquiries, please reach out!
About us: Andrew is a PhD researcher at the University of Oxford (@ndrewwang). Nicholas is a business manager at UCL Business and Venture Fellow with Berkeley SkyDeck (@nicholasyiu). Ethan is a battery scientist with experience at startups, research labs, and EV manufacturers across the world (@ethandalter).