An integration framework based on deep learning and CFD for early detection of lithium-ion battery thermal runaway
Year: 2025
Type: article
Abstract: • An integrated framework coupling CFD and CNN-LSTM models to early detect fire risks. • Apply CFD numerical results as the data inputs of a CNN-LSTM model for training. • A CFD model with an in-house written UDF to represent abnormal heat generation. Lithium-ion batteries (LIBs) have been extensively adopted in various fields, leading to a rapid increase in fire risks and accidents. Early detection of potential LIB thermal runaway has a profound influence on reducing fire risks. In this work, we proposed an integration framework based on computational fluid dynamics (CFD) and deep learning to detect the battery thermal runaway early enough. Firstly, a dataset of the temperature contours for the battery thermal runaway has been built. A coupled model with the convolutional neural network (CNN) and the long short-term memory neural network (LSTM) is applied to train the dataset and predict the potential fire risks of the battery pack by identifying the abnormal heat generation. The performance of the proposed model has been proven to have a maximum accuracy of 0.967. The trained model performed an 85.02 F1-score, and all the risks can be detected timely. This framework can further expand the LIB safety margin by detecting the battery thermal runaway quickly and accurately and reducing potential battery fire risks and accidents.
Source: Applied Thermal Engineering
Authors: Ao Li, Shadi Abpeikar, Min Wang, Terry Frankcombe, Maryam Ghodrat
Institutions: UNSW Canberra, University of Canberra
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,660
OpenAlex ID: W4409776671