Research On Electrode Defect Detection Of Coiled Lithium Battery

The winding lithium battery has the characteristics of high space utilization,high production efficiency and good safety.But in the process of assembly coating is easy to produce the phenomenon of diaphragm wrinkle,electrode dislocation,resulting in low capacity,poor circulation and other problems.The X-ray image of lithium battery is generated by X-ray imaging system,and the analysis of X-ray image can effectively detect electrode defects to ensure the battery quality.According to the characteristics of lithium battery electrode defects in X-ray imaging,a method of detecting lithium battery electrode defects based on convolutional neural network is proposed.The main electrode area in the X-ray image of lithium battery was extracted and cut into a small image as a data set to complete the training of neural network.Then the features extracted from the complete picture through the convolutional neural network are fed into the support vector machine to complete the final classification prediction.Experimental results show that the success rate of defect detection can reach 99% and the detection speed is high.Specific research contents are as follows:(1)The X-ray image of lithium battery was processed to separate the effective area image of the electrode,and then the electrode image was clipped to make the data set of the training network.Each image in the dataset is 95 by 95 pixels in size.The data can be divided into two categories.The pictures containing the defect area are negative examples and the pictures without the defect area are positive examples.(2)The convolution neural network is used to train the data set to learn the ability of extracting image features.In the network,improved batch normalization algorithm and decontamination convolution kernel are used to optimize network performance,and dropout and learning rate attenuation are added to improve network generalization ability and suppress overfitting.The network model was trained repeatedly,and the relevant parameters were constantly adjusted according to the training situation.Finally,a set of models and parameters with the best training results were saved to extract image features.(3)The convolutional neural network is used to extract the features of the complete lithium battery image,and output the characteristic diagram of defect distribution in the obvious labeling diagram,that is,the confidence result diagram of the predictive classification.These feature graphs are used as input to train the support vector machine,and the classification result of support vector machine output is used as the final result of battery defect detection.