Supervisor: Azzeddine Bakdi, Ph.D, Battery data analyst, department of research & technology, Corvus Energy.
Overview:
Lithium-ion batteries make prominent energy storage solutions covering many applications, including portable devices, stationary storage, transport, maritime, and aerospace sectors.
Safety and lifetime of lithium-ion batteries are major topics of concern, and they are related to the ageing mechanisms, especially Lithium Plating (LP). LP is a slow process that develops under a combination of three stress conditions; it causes battery State of Health (SoH) degradation, and it raises safety issues if not detected and mitigated properly.
LP is difficult to identify, it requires costly and time-consuming destructive tests done by electrochemical experts (tear down battery cells and microscopic imaging). However, recent research has shown that advanced sensor data analysis has the potential to identify LP.
The proposed project aims at LP prediction using sensor data collected from lab tests and data labels collected from destructive tests. It targets method development using a specific battery cell model. The proposed method is based on Incremental Capacity Analysis (ICA), ICA feature extraction and modelling, and LP prediction.
Proposed method:
This project will basically cover sensor data analysis using lab test data (provided by Corvus) to parameterize the ICA curve (combination of nā5 bell curves), then use the extracted features to predict LP. The proposed solution is using skewed PDF estimation to extract (ā25) features from an ICA curve. Secondly, modelling the correlations among the features and operating conditions: temperature, current, and SoH from full cycles. The method should then reconstruct the full ICA curve if it is partially observed, allowing for SoH estimation from incomplete cycles. The main focus is to develop LP predictive models that correlate the extracted features to LP, trained using the LP labels (provided by Corvus, from destructive tests). Advanced objectives (6 & 7) are to validate the method on long-term cycling data (clean lab data, incomplete cycles, and different conditions); and possibly tests on field data (large data, limited data quality).
Project roadmap:
(1) Sensor data processing and ICA curve parameterization.
(2) ICA feature extraction and correlations.
(3) ICA curve reconstruction from incomplete cycles.
(4) SOH prediction using ICA reconstruction (optional).
(5) LP predictive modelling using ICA features and post-mortem test labels.
(6) LP prediction from incomplete cycles.
(7) LP prediction from field data.
In average, objective 5 should be reached.