Detection Of Valve Seat Defects Of Automotive Injector Based On Deep Learning

The valve seat of the automobile fuel injector is a device that injects gasoline into the automobile cylinder,which plays a very important role in the control of the automobile fuel quantity.However,due to the small size of parts and the limitation of processing technology,there are inevitably different kinds of defects such as scratch,defect,rust spot and white spot in the production process of the valve seat of automobile fuel injector.The text will use the depth detection technology to complete the flaw detection of the injector seat in the visual inspection system.The main research work includes:1.Research on vision detection systemAt present,the defect detection of the valve seat of the automobile injector is still done manually.In view of the strong dependence of the manual detection and the poor effect,this paper will build a visual detection system which is consistent with the working environment.Firstly,design special light sources and lighting methods to complete the fill light to highlight the defects of the image;then,the key hardware,such as CCD camera,magnifying lens and image acquisition card,is used to enlarge and shoot the image;the machine arm,vibrating trays and other devices are combined to realize the automatic loading and unloading operation;finally,detection technology based on depth learning is used to complete the detection of valve seat defects.The results show that the detection accuracy of the system is about 10% higher than that of the manual.2.Data set enhancement and productionIn order to solve the problem that there is no data set that can be used directly in the detection of valve seat defects of automobile fuel injector.In this paper,a data set of automotive injector seat with four kinds of defects is constructed.Firstly,the collected data is processed by normalization before model training;secondly,data enhancement is realized by means of geometry and color space transformation to solve the over fitting problem caused by insufficient data;then data annotation is carried out to better use in model training;finally,the data set is divided into training set and test set according to the ratio of 3 to 1.3.Optimization and improvement of the depth flaw model of the valve seat of automobile injectorIn order to solve the problem of missing inspection caused by small target,multi-scale and inaccurate positioning,the algorithm model of depth detection is optimized and improved.Firstly,K-means clustering function is introduced in the data preparation stage to select a reasonable aspect ratio in advance,change the size and number of candidate boxes,and solve the problem of missing detection caused by small targets and redundant boxes;secondly,in the problem of target location,use the merge loss function to solve the problem of inaccurate positioning;finally,the idea of Inception and Res Net is introduced to improve the feature network to solve the problem of multi-scale and non obvious features;the improved algorithm model is compared with the original Faster-RCNN model on the data set of the fuel injector seat.It is found that the average accuracy of the improved model is 72.41%,which is improved by nearly 4%compared with the original model,and it can be better applied to the detection of fuel injector seat defects.