Automatic Detection Of Automobile Hub Internal Defects Based On Deep Convolution Neural Network

Wheel hub is an important component and main stress component of automobile.At present,the mainstream manufacturing process is low-pressure casting.In the production process,there may be internal defects such as porosity,shrinkage or porosity,which seriously affect the product quality.X-ray nondestructive testing is necessary before leaving the factory.The core work of X-ray inspection of automobile wheel hub is to accurately judge whether the inspected wheel hub has defects and the level of defects.The traditional manual judgment method has the disadvantages of high labor intensity,strong subjectivity of test results,difficult to quantify and so on.Using deep learning algorithm to realize the automatic judgment of defects has always been a research hotspot in the academic and industrial circles,but it has not yet appeared The automatic defect identification system completely replaces the manual judgment.The research content of this paper focuses on the automatic detection of internal defects of automobile wheel hub.On the basis of consulting a large number of relevant technical literature at home and abroad,the image detection algorithm of deep learning applied to wheel hub casting defects is studied.In the traditional image non-destructive testing method,the hub is imaged by X-ray,and then the defect is identified manually by threshold segmentation technology.This kind of work with high repeatability is time-consuming and laborious,the manual film evaluation standards are different,and the accuracy is poor in some special situations.The traditional machine vision defect recognition method needs to extract the features of the region of interest manually,but because some special features of the wheel defects are difficult to determine,the recognition rate is low,and the manual feature extraction is time-consuming,which requires a lot of human resources,and the degree of automation is difficult to improve.Therefore,it is necessary to study the automatic segmentation algorithm of hub defects,and realize the automatic segmentation of hub defects by software,so as to carry out auxiliary detection.The core idea of convolution neural network is to extract the feature information on the image through convolution kernel.The deeper the network structure is,the more complex the extracted information is,and the parameters of the model can be reduced.The parameters of convolution kernel can be obtained through the existing data training model.Finally,the results can be obtained by inputting the image to be tested into the model.On the basis of deep learning,this paper studies the model method of applying target recognition,semantic segmentation and instance segmentation to wheel hub defect detection R-CNN is improved to make it more suitable for the wheel defect segmentation task,and a wheel defect segmentation algorithm mask-m2 det model based on m2 det network structure is proposed to replace the complex method of traditional recognition method which requires artificial feature extraction,artificial feature description and artificial feature selection.The experimental results show that compared with the traditional method,this method can identify the types of hub defects faster and more accurately and segment them.