Battery diagnostic technique that uses a machine-learning virtual reference electrode to detect lithium-plating risk early

How to apply this

Using only the voltage and current signals of a standard cell, one can build a virtual reference electrode model that estimates the anode potential and detects lithium-plating risk early.

1. Collect charge-discharge data and anode-potential labels from some cells equipped with a reference electrode.
2. Train a supervised regression model using standard-cell signals as input, and check overfitting with cross-validation.
3. Predict anode potential on independent test cells and confirm the RMSE and error distribution.
4. Apply a simple control rule that lowers charge current when the predicted potential approaches the plating-risk threshold.
5. Afterward, compare with reference-electrode remeasurement or electron microscopy to verify whether plating signs actually appeared.

Data needed

Operational records of battery voltage, current, temperature, and SOC during charge-discharge, along with anode-potential labels measured with a separate reference electrode in some cells, are needed.

Checking the result

Verify regression metrics such as RMSE using train-validation-test splits, and independently confirm whether the predicted plating risk matches reference-electrode measurements or subsequent electron microscopy observations in the actual cell.

What can go wrong

Direct extrapolation to commercial cells without a reference electrode can cause performance to fluctuate greatly, and the most common mistake is to generalize a model tuned only to test conditions such as low negative-to-positive capacity ratio.

이 가이드는 원문의 방법을 바탕으로 AI가 제안한 적용 방향입니다. 원문에 적힌 내용이 아니며 같은 결과가 재현된다는 보장도 아닙니다.