Research On Hub Defect Detection Technology Based On Deep Learning

As an important part of the car driving process,the wheel hub is very important for people's travel safety.In the casting process of wheels,there are unavoidably defects that are difficult for human eyes to discern,which seriously affects the safe operation of cars and the safe travel of people.Therefore,it is very necessary to regularly inspect the defects of the hub.Most of the work of traditional detection methods is done manually,which requires repeated manual labor,thereby reducing the intelligence of detection.Aiming at the need to establish intelligent non-destructive testing of hubs and improve the accuracy of hub defect detection,this paper aims at the limitations of traditional inspection methods,based on hub X-ray images,combined with deep learning technology,using a deep learning based hub defect segmentation method.The main work of this article is as follows:1.Briefly introduce the research background and significance of this article,the overview of wheel hub defect detection,the current status of deep learning and the application of deep learning in image recognition.An automatic segmentation method for wheel defects based on deep learning algorithms is used.The structure of the convolutional neural network and the nonlinear characteristics of the radial basis function neural network are used to use the deep learning network structure to simulate human visual perception.2.This article introduces the theoretical basis of deep learning and convolutional neural networks.The structure of the convolutional neural network includes convolutional layers,pooling layers,activation functions,etc.Then,the forward and backward propagation of the convolutional neural network and the stochastic gradient descent algorithm are introduced.3.Apply the convolutional neural network to the segmentation of wheel defects.Defect the hub image with defects,perform data enhancement and expansion,and normalize it tobuild a hub defect image training library.For the first time,use U-Net convolutional neural network to train the hub defect segmentation model,and Simulate the human brain level perception system based on the region of interest,which can identify the gray pixels of the region of interest,and extract the intrinsic features of the defect area layer by layer through deep learning layered network and convolutional neural network,thus Realize automatic segmentation of wheel hub defects.4.In order to improve the detection accuracy,this paper improves the U-Net network,replacing the maximum pooling operation of the original U-Net model with a convolution operation to reduce the loss of feature information,and adding a Dropout layer to optimize the network.Reduce the complex co-adaptability between neurons and improve the reliability of the model.The results show that the improved U-Net network structure performs well on the segmentation of wheel defects,and the indicators of the data set are superior to the original U-Net network,in which the DICE coefficient is 0.8554 and the SSIM coefficient is 0.96,which is higher than the original U-Net network(0.7932 for DICE and 0.9176 for SSIM)shows that the improved network has a better ability to segment wheel hub defects and can meet the needs of the automated non-destructive testing of wheel defects.