Research And Implementation Of Lithium Battery Wrinkle Detection System Based On Deep Learning

Under the harmonious and stable world structure,the constant exchange and innovation of economy,culture and science and technology have made people's quality of life constantly improved.Computer hardware and software technology has achieved rapid development,and electronic products closely related to life and entertainment are emerging one after another.Lithium battery is the main energy source of these electronic devices,and the demand for lithium battery is increasing.In the production process of lithium battery,there are gaps between diaphragm and electrode plate when winding.Or the pressing of fingers on the electrode plate during manual detection.All these factors will cause lithium battery wrinkles.The presence of wrinkles can detract from the life of the lithium battery and can even pose a safety hazard.At present,in the detection of lithium battery wrinkles,many domestic and foreign lithium battery manufacturers still use manual detection methods,which is extremely inconvenient and inefficient.In addition,it will be affected by factors such as mood,fatigue and experience of the test personnel,which will reduce the quality and efficiency of the test.Therefore,in order to solve practical engineering problems as a starting point,this paper proposes a lithium battery wrinkle detection method based on depth learning.Convolution neural network is used to learn the wrinkle characteristics of lithium batteries.Predicting the data to be detected through the learned wrinkle characteristics of the lithium battery.Finally,a lithium battery wrinkle detection model based on depth learning is established and compiled into a callable library file.The specific work is as follows:First,the collection of data sets,X-ray irradiation of lithium batteries through X-ray machine to obtain X-ray images.The collected images are then cropped.After the cropped image is subjected to a series of pre-processing operations,an image with a sharp contrast and a size of 64×64 is finally obtained.Second,the pre-processed image is divided into two types by using a semi-supervised self-learning method.One is the image of a lithium battery without wrinkles,and the other is the image of a lithium battery with wrinkles.The two types of data are labeled separately and made into data sets.A convolution network model is constructed,and the convolution neural network is used to extract features of lithium battery images,and forward propagation and backward propagation are adjusted to obtain optimal weights and offsets.Finally,the global optimal solution is found.Save the trained parameters and models.Third,to meet the demands of the industrial production,the speed of lithium battery wrinkle detection based on depth learning is higher than that of manual detection.In order to improve the detection speed,the Caffe deep learning framework closer to the bottom layer is selected to build a lithium battery wrinkle detection model and package it into a library file for convenient system call.Through experiments,the accuracy of the lithium battery wrinkle detection system can reach more than 97%,and the detection time for each image is less than500 ms.It is proved that the method of lithium battery wrinkle detection based on deep learning can accurately distinguish the wrinkles of lithium battery.The experimental results show that the method is faster than manual detection time for lithium battery wrinkles and meets industrial production requirements.