| The manufacturing quality of the commutator is very important to the stable and reliable operation of the motor,and the surface defect of commutator is the main quality problem.At present,manual inspection is mainly used,because the requirements of the automobile industry for full inspection,the manual inspecting method not only has high labor intensity and low efficiency,but also has high quality dispersion and high cost.Machine vision is the mainstream method to realize automatic detection,but the traditional machine vision detection method has poor adaptability to the change of defect mode caused by the change of working conditions in the production line,that is,the generalization ability is insufficient and the algorithm development cycle is long.In this study,Vi Di as a commercial visual defect detection software for machine learning,is used to realize the Two-Category detection of 25 kinds of defect images by building and optimizing specific models.This method not only has high precision and efficiency,but also is easy to deploy.The major work is as below:Firstly,it has put forward the general design scheme on commutator defect detection.By analyzing the features of detection targets and the difficulties of detection assignments,the overall detection scheme is proposed from the perspective of hardware and software.Hardware work includes: building detecting platform and image collection system.Software work includes: building and optimizing Vi Di Models,and deploying Vi Di Models.Secondly,this approach studies the strategy and method of commutator defect detection based on Vi Di,and realizes mixed defect detection.The defects of the commutator are many kinds and forms,which lead to the low accuracy of the single detection algorithm.This study proposed two methods of combined detection by machine position and mixed detection by defect characteristics,and constructed Vi Di Models.Through the prepared dataset and the proposed evaluation indexes,it was verified the effectiveness of the two detection methods.According to the size of the dataset in the end,the paper gives guidance for choosing two methods.Then,the Vi Di Models optimization methods of commutator defect detection are studied.In order to improve the accuracy of Vi Di Models,a set of optimization schemes is designed in this study: Vi Di Models selection based on 5-fold cross-validation,combined detector based on Ada Boost algorithm,image data enhancement based on Gaussian noise and affine transformation,and parameter optimization of Vi Di Models.Experimental tests showed that the recall rate after model optimization increased by more than 2.00%.Finally,deploy the Vi Di Models and design simulation experiments to test the detection effect.In order to achieve the rapid and high-precision automatic detection of commutator defects,this research used the C # language to complete the deployment of the Vi Di Models on the Visual Studio 2015 programming platform.The 1200 pieces of labeled image data collected on the spot were combined according to the camera position.The Vi Di Models constructed through combined detection by machine position and mixed detection by defect characteristics was simulated and tested.The experimental results show that the recall rates of the two methods is 99.17% and 98.17% respectively,and the detection time is 89.6ms/ frame and 72.8ms/ frame respectively,which all meet the technical indexes requirements of the current commutator defect detection task. |
