Lithium battery fault diagnosis by integrating improved EMD decomposition algorithm and 2DCNN

Abstract

With the development of science and technology, lithium batteries, as important energy storage devices, have become a research hotspot for fault diagnosis. In response to the shortcomings of traditional fault diagnosis techniques in processing complex signals and extracting key features, a high-performance lithium battery fault diagnosis model is constructed by combining the high-dimensional representation ability of a two-dimensional convolutional neural network (2DCNN) with the decomposition stability of an optimized empirical mode decomposition (EMD) method. The study first uses an improved EMD algorithm to process the voltage data of lithium batteries and extract fault features. Moreover, the processed data are input into 2DCNN model for training and classification. The experiment results showed that the fault feature information consistency of the research designed model was 98.7% when the iteration number reached 600. The feature recognition accuracy dropped to 99.2% when the image contained 70 features. The running time in all 7 data groups was significantly lower than other methods, with the highest being below 6ms. The results indicate that the designed lithium battery fault diagnosis model has improved the accuracy and efficiency of fault diagnosis. This can provide new technological means for lithium battery fault diagnosis and helping to promote the further development and application of lithium battery technology.