To get started, I've done a background search on existing papers relevant to machine learning for battery cathodes. In particular, I focused on machine learning for NMC cathodes in Li-ion batteries.

What I found after extensive research far surpassed what I had thought was available at a cursory glance when I first had the idea to try to embark on this project!

I used what I thought was a pretty neat tool (though there are like a million others on the internet as it turns out) that scrapes Google Scholar and can sort by citations: Sort Google Scholar. I'm sure there's tons of other options that do something similar, but this one worked.

Below are my thoughts and takeways and at the bototm, a list of relevant articles sorted by topic.

Some thoughts

  1. There's lots of machine learning papers around state of health or the state-of-lithiation, which I would assume is because:
    • This can be extended to almost all battery systems regardless of cathode, etc.
    • It's more relevant to industry.
    • There's a LOT of data for it (more on this later).
  2. Plenty of overview articles from the last several years on the subject—you can take your pick.
  3. Some of the microstructure ones are the most fascinating, but also the most data-limited because each data point requires a full experiment.
  4. There's just one good doped NMC article as far as I could tell from 2021 that uses 168 real data points. The bulk of that effort must have been just compiling the data. The actual ML algorithms are pretty trivial. Could this really be the most recent article? There's another from 2023, but they are using simulation to estimate data to feed to the algorithm.

Key Takeaways

Real experimental data is the bottleneck.
Real experimental data is the bottleneck, and one of the catches is that state of health is easier because if you have 10k data points for 10 cycles of a lithium ion battery, you can use every one of those datapoints, because all of them tell you something about state-of-health, that is the current voltage and capacity at least per point if not other relevant points such as current rate etc. and so just from a few batteries, you could have 10s of thousands of data points to feed to a neural net for instance.

Doping is different for instance, because for each dopant combination, there needs to be a unique "result" which would be either late cycle capacity or initial capacity. This means data is limited per battery

Synthesizing data using simulations (e.g., DFT or FEA) cannot solve this
Lots of papers are trying to avoid this data problem by synthesizing data using DFT or FEA. DFT is generally limited by interactions and computational power and as such can give ambiguous results so I'll try to scrape data instead

Literature Review

Overview

Click to expand
  1. 2020 - Machine learning assisted materials design and discovery for rechargeable batteries
  2. 2023 - Machine Learning in Lithium-Ion Battery Cell Production: A Comprehensive Mapping Study
  3. 2021 - How Machine Learning Will Revolutionize Electrochemical Sciences
  4. 2024 - Lithium-ion battery digitalization: Combining physics-based models and machine learning
  5. 2023 - Machine learning-inspired battery material innovation

Microstructure

Click to expand
  1. 2023 - Time-dependent deep learning predictions of 3D electrode particle-resolved microstructure effect on voltage discharge curves
  2. 2024 - A generative machine learning model for the 3D reconstruction of material microstructure and performance evaluation
  3. 2024 - Improved Rate Capability for Dry Thick Electrodes through Finite Elements Method and Machine Learning Coupling

State of Health

Click to expand
  1. 2022 - Battery degradation prediction against uncertain future conditions with recurrent neural network enabled deep learning
  2. 2024 - Machine Learning-Based Electrode-Level State of Health Estimation for NMC/Graphite Battery Cells
  3. 2024 - Integrating machine learning for health prediction and control in over-discharged Li-NMC battery systems
  4. 2024 - Chapter 13 - Battery state-of-health estimation using machine learning
  5. 2023 - A Convolutional Neural Network to Predict the State-of-Lithiation of Lithium-ion Batteries with the Nickel-Manganese-Cobalt-Oxide Chemistry

Thermal Runaway

Click to expand
  1. 2024 - Machine Learning and Data-Driven Analysis of Thermal Runaway Characteristics in Lithium-Ion Batteries
  2. 2023 - Thermal Degradation of Polycrystalline Ni-Rich Cathodes, New Insights from Total Scattering and Machine-Learning Assisted Tomography

Doping

Click to expand
  1. 2023 - Maximizing the energy density and stability of Ni-rich layered cathode materials with multivalent dopants via machine learning
  2. 2021 - Machine-Learning Approach for Predicting the Discharging Capacities of Doped Lithium Nickel–Cobalt–Manganese Cathode Materials in Li-Ion Batteries
batteries
Copied!