Research On Detection And Recognition Method Of Railway Freight Bearing Defects Based On Image

At present,black magnetic particle(BMP)testing technology is used in the maintenance of railway freight car bearings,This kind of investigation method has high labor cost,low work efficiency,and serious false detection and missed detection.In recent years,deep learning technology has been widely used in the field of surface defect detection and defect identification,which makes it have leapfrog improvement in detection accuracy and generalization ability.Based on the construction of railway freight car bearing defect database,data preprocessing and data enhancement are carried out,and a new improved convolutional neural networks model is proposed.By comparing with several traditional models,it is found that this model has certain advantages in recognition rate.The main contents of this paper are as follows:(1)The photo of bearing flaw detection of railway freight car is collected on the spot of bearing detection,and the label map corresponding to the original drawing is established by means of image processing.The original and label drawings are preprocessed by data cutting,data screening,etc.Because of the small number of original samples,the defect data set is enhanced by rotation,translation,filling and other data enhancement processing,which expands the data scale.90% and 10% of the data is used for training and detecting respectively.(2)Three convolutional neural network models,including UNet,SegNet and Attention UNet,are used to train and verify the flaw detection data set of bearing,and the advantages and disadvantages of the three convolutional neural networks are analyzed.(3)The Residual network,DenseNetwork and global maximum pooling are introduced.By improving the structure of UNet network,an improved convolutional neural network model is established.Train and test the defect data set of railway freight car bearing,and compare the analysis results with the results of three convolution neural network models.The results show that: compared with the traditional methods such as UNet network,Segnet network and Attention UNet network,the improved convolution neural network model has a certain advantage in recognition accuracy,which can reach98.8%.The research results of this paper have certain theoretical significance for realizing the automatic identification of railway freight car bearing defects.