Feature Extraction And Classification Of Typical Internal Structure Of 18650 Lithium Battery Thermal Runaway

At present,there are many accidents of lithium battery explosion and fire in air cargo and passenger transportation,the main reason is lithium battery thermal runaway.The main reason for thermal runaway is the change of the internal structural state of lithium batteries,so the detection of internal structural state of lithium batteries has become a significant problem that airports need to solve urgently.Aiming at the problem that the internal structural state of lithium batteries is rarely detected in current air transportation security,two methods for lithium battery classification and identification are proposed in this thesis.Specific research includes the following.Two algorithms are proposed based on lithium battery image features.The internal structure images of 18650 lithium battery collected by Micro Computed Tomography Scanner machine are analyzed,and then the difference of jelly roll structure in different types of lithium battery images is got,and then the lithium battery image recognition processes using traditional algorithms and deep learning algorithms are designed.The lithium battery images are classified and recognized.The internal structure images of the lithium battery are pre-processed first,the images are grayed out,filtered,enhanced,cropped and drawn separately,then the preprocessed lithium battery images are trained and tested.The feature vectors formed by permutation and combination of the optimized features are send into SVM,MLP,GMM and KNN classifier for training in the traditional algorithm,and the pre-processed images are send into Alexnet,Compact,Enhanced and Resnet50 network models for training in the deep learning algorithm.Then a group of images are given for testing,the optimal results of the two algorithms are selected respectively,and then the algorithm with the best test results is selected out.The test results show that the recognition accuracy and precision of the traditional algorithm are 79.6875% and 100%,and the recognition accuracy and precision of the deep learning algorithm are 99.6875% and 100%.In the case of the same precision of the two algorithms,the accuracy of the deep learning algorithm is 20.0521% higher than that of the traditional algorithm,that is,the deep learning algorithm is superior to the traditional algorithm.Human-computer interface is developed.In order to make the recognition function of lithium battery images more intuitive,the 18650 lithium battery recognition system software is developed based on the deep learning algorithm Halcon program.Through the joint programming of Halcon,C# and SQL Sever,a man-machine interaction interface with the recognition records is designed in the Visual Studio platform.The man-machine interaction interface with recognition record is tested and analyzed,and the experiment proves that the function of the man-machine interaction interface meets the actual needs.