Research On Intelligen Classification And Recognition System Of Automobile Wheel Hub Based On Machine Learning

In recent years,China's automobile production has continued to develop steadily.It has been the world No.1 for nine years,which has brought opportunities and challenges to automotive-related enterprises.As an important part of the automobile,the automobile wheel hub needs to be classified in the manufacturing process.The most of wheel hub intelligent recognition system apply the traditional machine vision algorithm.On the one hand,it needs to manually determine the extracted features,which is time-consuming,laborious and the feature's robust performance is poor.On the other hand,the system usually integrates highly and it is difficult to make deeper development.We design the automobile wheel hub intelligent identification system based on machine learning.Firstly,the scheme of intelligent classification and identification system for wheel hub is designed based on the research of related technologies at home and abroad.We propose the software and hardware about the hub intelligent classification system.Secondly,the traditional image feature extraction method and the convolutional neural network based feature extraction method are studied respectively.In the traditional image feature extraction,the methods of image denoising,image segmentation,hub radius,center hole and inertia feature extraction are mainly studied.The network structure and parameters are designed based on VGG network,and parallel computation is introduced in network training.Lastly,Two classification schemes are proposed respectively for standard wheel hub identification and specific different series of wheel hub identification.For standard identification,we compare traditional wheel hub image classification scheme and scheme of machine learning based convolutional neural network classification from four aspects:preprocessing complexity,algorithm efficiency,classification accuracy and algorithm robustness.In the traditional scheme,the radius of the hub,the central hole,the spoke,and the method of extracting the moment of inertia are studied.In scheme of machine learning,it is finally determined that the VGG network is used as the basic network for wheel classification by comparing three networks:Alexnet,VGG and MobileNets.We find that the classification accuracy rate can still reach 99.43%in the case of complex detecting environmental and variable image backgrounds.For the specific series of different wheel hubs in the same category,the four moments of inertia and contour radius are combined with the features of the convolutional neural network,which solves the problem that the convolutional neural network is insensitive to scale and plating feature classification when the data set is limited.Applying the machine learning algorithm to the hub detection solves the shortcomings of the traditional method,such as artificial feature extraction,poor anti-interference and poor generality.More importantly,it can take the secondary development of the hub detection system.With convolutional neural network,wheel hub identification can be integrated with other inspection requirements,such as wheel hub defect detection and wheel hub number identification.