Identification of Internal Short-Circuit Faults in Lithium-Ion Batteries Based on a Multi-Machine Learning Fusion

Guangying Zhu, Tao Sun, Yuwen Xu, Yuejiu Zheng, Long ZhouBatteries 2023 · DOI

TL;DR

This paper tackles the critical challenge of detecting internal short-circuit (ISC) faults in lithium-ion batteries (LIBs), a major safety concern. They propose a novel multi-machine learning fusion approach, combining ResNet-CNN with transfer learning, to accurately identify and grade ISC fault severity, achieving high accuracy on both simulated and experimental data.

What Problem Does This Paper Address?

Internal short-circuit (ISC) faults are a leading cause of thermal runaway in lithium-ion batteries (LIBs), posing significant safety risks, especially in electric vehicles. Early detection of these faults is crucial but challenging because their initial symptoms are subtle and easily missed. Conventional machine learning models often struggle with the limited availability of real-world ISC fault data and the complex, non-linear relationships between battery voltage and fault severity. This paper aims to bridge this gap by developing a robust method for identifying and classifying different levels of ISC faults, enabling timely warnings.

Methods

The authors introduce a multi-machine learning fusion method for ISC fault identification. The core of their approach involves:

  1. Data Acquisition: Both simulated and experimental data were collected. Simulation data was generated using a first-order RC equivalent circuit model of a battery pack, incorporating various ISC resistance levels. Experimental data was obtained from actual battery cells subjected to simulated ISC conditions.
  2. ResNet-CNN Pre-training: The simulated ISC fault data, after voltage normalization, was used to pre-train a ResNet-101 convolutional neural network (CNN). This step leverages the large volume of simulated data to learn general ISC fault characteristics.
  3. Transfer Learning and Fusion: To adapt the model to real-world experimental data and improve its performance on subtle faults, a transfer learning approach was employed. Key layers of the pre-trained ResNet-101 model were frozen, while the remaining layers were fine-tuned using the experimental battery data. This fusion creates a more robust model capable of identifying different ISC fault levels.
  4. Voltage Normalization: A crucial preprocessing step involved normalizing the battery voltage curves using the maximum-minimum normalization method. This ensures that the input data is consistently scaled, regardless of the absolute voltage values.

Datasets

  • Simulated Dataset: Generated using a first-order RC equivalent circuit model of a 96-cell battery pack. ISC faults were simulated across a range of resistance values (1-300 Ω), representing different fault severities.
  • Experimental Dataset: Obtained from laboratory experiments on ternary lithium-ion battery cells and modules subjected to simulated ISC conditions.

The paper does not explicitly introduce a new benchmark dataset but rather utilizes both simulated and experimental data to train and validate their proposed method.

Key Insights and Ideas

  • The Power of Fusion: Combining simulated data for initial learning with experimental data for fine-tuning via transfer learning proves effective in overcoming the limitations of scarce real-world fault data.
  • Voltage as a Key Indicator: Normalized voltage curves, even for subtle ISC faults, contain discernible patterns that deep learning models can learn.
  • Graded Warning System: The ability to not only detect but also classify the severity of ISC faults is crucial for implementing a graded warning system, enhancing battery safety.
  • Micro-Short Circuits are Tricky: The study highlights that identifying very subtle micro-short circuits remains a challenge, as their voltage signatures can closely resemble normal operation.

Key Results

  • Simulation Data Performance: The ResNet-CNN model achieved an impressive 99.9% accuracy in classifying different degrees of ISC faults on the simulated dataset.
  • Experimental Data Performance: The multi-machine learning fusion model, after transfer learning, achieved a 96.67% prediction accuracy on the laboratory battery dataset.
  • Accurate Identification: The fused model demonstrated the ability to accurately identify four distinct levels of ISC faults and distinguish them from normal battery states.
  • Improved Low-Level Fault Detection: Compared to the standalone ResNet-CNN on experimental data, the fused model showed improved performance in identifying faults at lower ISC resistance levels.

Limitations and Future Work

  • Micro-Short Circuit Challenge: The paper acknowledges that identifying micro-short circuit faults, particularly those with higher resistance (e.g., 300 Ω), remains a challenge due to their subtle voltage signatures.
  • Future Work Suggestions: The authors propose strengthening the model training for micro-short-circuit faults by incorporating more simulated and experimental data for these specific conditions. They also suggest collecting further real-world battery ISC fault data to validate the proposed method more comprehensively.

Summary updated

Abstract

Internal short-circuit (ISC) faults are a common cause of thermal runaway in lithium-ion batteries (LIBs), which greatly endangers the safety of LIBs. Different LIBs have common features related to ISC faults. Due to the insufficient volume of acquired ISC fault data, conventional machine learning models could not effectively identify ISC faults. To compensate for the above deficiencies, this paper proposes a multi-machine learning fusion method to predict ISC faults and to perform faults warning classification under multiple operating conditions using the input of voltage normalization. Firstly, learning data acquisition is captured by experiments and simulation. Secondly, the simulation data are inputted into the ResNet-convolutional neural network (CNN) for pretraining, followed by the transfer learning method to freeze parts of the model layers in the CNN, and part of the experimental data are also inputted into the CNN model for parameter fine-tuning to build a multi-machine learning model. Finally, the degree of ISC faults within the laboratory battery is predicted based on the multi-machine learning model. The results show that the CNN model had a 99.9% prediction accuracy on the simulated dataset, and the multi-machine learning fusion model after transfer learning had a 96.67% prediction accuracy on the laboratory battery dataset, which can accurately identify different levels of ISC faults in batteries and realize the graded warning of ISC faults.
BibTeX
@inproceedings{zhu2023identification,
  author = {Guangying Zhu and Tao Sun and Yuwen Xu and Yuejiu Zheng and Long Zhou},
  title = {{Identification of Internal Short-Circuit Faults in Lithium-Ion Batteries Based on a Multi-Machine Learning Fusion}},
  booktitle = {Batteries},
  year = {2023},
  doi = {10.3390/batteries9030154},
}
Figure