Defect Detection Of Ferrite Parts Based On Machine Vision

Missiles,as precision-strike weapons,have gradually evolved into the leaders of modern warfare,and have become a key factor in accelerating the course of war and deciding the outcome of war.The control and navigation tasks of this type of weapon run through the entire process of weapon flight,so higher requirements are placed on the accuracy of the navigation system.The inertial navigation system has become the core equipment of missile and other weapon guidance systems with its full autonomy and comprehensive output parameters.The accuracy of the inertial navigation system is determined by the inertial instrument.The ferrite part studied in this article is a part commonly found in inertial instruments.Ferrite is an important non-metallic magnetic material.It is an indispensable part of automatic control and instrumentation.It has a wide range of applications in the aerospace field.Ferrite parts are manufactured by a roasting method similar to the ceramic process,with a hard texture and high brittleness.In the production and processing of ferrite parts,due to some reasons,the surface of the ferrite parts is prone to defects such as cracks,cracks or chipping.Under the vibration of the carrier,there is a risk that the defects gradually deepen,resulting in component failure,causing the inertial instrument to be stuck,and affecting the normal operation of the inertial navigation system.Therefore,the defect detection of ferrite parts is very important.Ferrite parts are small circular parts.At present,the defect detection method of ferrite parts is to detect whether the ferrite parts are defective under the microscope through manual detection,which has the problems of long detection time and low accuracy.When there are many batches of parts to be inspected and the batches are large,it will consume a lot of time and energy of the inspection personnel,and the long-term inspection will cause the inspector's eyes to fatigue,resulting in an increase in the probability of misdetection.For similar parts defect detection problems,researchers at home and abroad are committed to using digital image processing technology to detect part defects,but due to the complex background of ferrite parts and dark colors,there are a lot of noise pixels on the surface of the parts in the image;And when acquiring images,the edges of the circular parts will be collected,and the shallow boundary features are similar to the defect features.Therefore,traditional image processing methods for extracting shallow features cannot be used to detect defects in ferrite parts.With the development of convolutional neural networks and the emergence of semantic segmentation based on fully convolutional neural networks,it becomes possible for machines to automatically detect part defects.This type of method has strong robustness and generalization,which greatly improves the accuracy of part defect detection.The key and difficult point of ferrite part defect detection is how to improve the accuracy of part defect detection,improve the detection effect of the detailed information of the defect edge,and clearly distinguish the defect from the part edge.This paper uses the method of convolution neural network to ferrite parts Defects are accurately detected.The main steps involved are: collecting images of ferrite parts,processing the images and making defect detection data sets,image augmentation,establishing network models and initializing model parameters,loading training data sets and labels,training network models,and verifying whether the network model is over Fit,test using test images and visualize test results.The content of this paper mainly includes:(1)The Ferrite SDD,a defect detection dataset for the surface defects of ferrite parts,used for defect classification and semantic segmentation of defects,respectively.(2)Build a convolutional neural network for classification,use the Ferrite SDD1 surface defect detection dataset of ferrite parts for defect classification to train the network,combined with the sliding window sub-region scoring mechanism,for the ferrite Parts are inspected for area defects.(3)According to the characteristics of the parts,different types of semantic segmentation networks were built,and they were trained by using Ferrite SDD2,a surface defect detection dataset for ferrite parts used for defect segmentation.The network does not require any post-processing operations.Oxygen component defects are accurately detected pixel-wise and the positioning results of the component defects are visualized.The detection accuracy can reach 99.3%,the mean Intersection over Union can reach 88.3%,and the detection defect accuracy is 0.6 microns.