The Design Of Mic-motor Rotor Appearance Inspection System Based Deep Learning

With the rapid development of industrial automation in China,more and more manufacturing industries have realized the leap of industrial automation and intellectualization.As an indispensable basic product industry in the modernization of national economy and national defense,the defect detection of micro-motor industry’s rotor foot also depends on manual detection.Because of its low detection speed,unstable detection accuracy and high salary expenditure,manual monitoring has greatly restricted the development of the industry.Most of the foot defects of micro motor rotors can be divided into three categories: abnormal foot closure,skewed foot and sagging foot.These defects have the characteristics of various defects and high similarity of different kinds of defects.Traditional defect detection methods usually consist of feature extraction and classifier.It is necessary to extract features from the collected defect images,and then input them into various classifiers.This kind of algorithm requires researchers to set up feature extraction algorithms manually.The accuracy of the algorithm depends largely on people’s experience and the complexity of products.Its robustness is low,and it is not applicable to other classifiers.View defect detection.In view of the above situation,this paper proposes a defect detection system which integrates mechanical design,optical imaging,motion control,image processing and in-depth learning to solve this problem and realize the automatic defect detection of micro-motor rotor.The automatic feeding and sorting function of the rotor is realized by the designed mechanical automation structure.The angle correction algorithm of the rotor is designed by using image processing technology.Then the data set of foot defects is trained by the deep convolution neural network model ResNet,and the model parameters are optimized.The classification and recognition model of foot defects is obtained to realize the classification and detection of foot defects.In order to verify the reliability and superiority of the algorithm,the other six algorithms are used to carry out experiments,and then the experimental results are compared with ResNet.These six algorithms include K-Nearest Neighbor(KNN),Principal Component Analysis(PCA),Spatial Pyramid Matching(SPM),Multi-Layer Perceptron(MLP),LeNet and AlexNet.The final experimental results show that ResNet outperforms other algorithms and the recognition rate reaches 89%.