Research On Machine Vision Detection Technology Of Tourbillon Watch Mini Escapement Wheel

Miniature escapement wheels are a type of parts frequently used in precision instruments and equipment,the flatness and the tooth length error of which are closely related to the precision and accuracy of the overall equipment.Traditionally,manual detection is mainly used,which,however,is prone to fatigue and misjudgement and difficult to quantify.In order to realize standard and automatic testing and enhance detection efficiency,a detection system for micro parts based on machine vision was designed in this thesis.The main research work of the thesis is as follows.1.Methods for classifying miniature parts.Miniature parts have small volumes and similar shapes,which causes frequent errors in manual classification during production.To enhance categorization accuracy,two parts classification methods were proposed in this thesis.One is a support vector machine(SVM)classifier method based on histogram of oriented gradients(HOG)features.The principle of this method is to extract HOG features of parts images and train the SVM classifier,and the trained classifier is then used for classification.Experiments demonstrated that this classification method had an accuracy of 98.5%.The second approach is to classify with deep learning network based on ResNet50.Before the fully connect layer of the conventional ResNet50 network,batch normalization and a rectified linear unit(Re Lu)were added together with optimized configurations of network parameters.As shown in the experiment,the accuracy of the network reached 99% on the test set.2.Method for detecting surface flatness of miniature escapement wheels.It is easy for the surface of miniature escapement wheels to deform under the influence of external forces,which may cause excessive deviation in its surface flatness.To detect defects of this type,a surface flatness detection method was proposed in this thesis.The procedure of the detection method is as follows.First,the image is calibrated and then sub-pixel edge information is extracted from the image.Afterwards,projection mapping is adopted to accurately calculate the boundary information.Ultimately,the horizontal edge range of the part in multiple frames is used to determine whether the flatness indicator is up to the standard.The experiment showed that the detection algorithm delivered an accuracy rate of 99.5% with a detection precision of 2 ?m.Single frame detection took 35 milliseconds in this case.As multiple frames werecollected for calculation,the detection result could be obtained within 4 seconds for an individual part.3.Algorithm for detecting tooth length errors of miniature escapement wheels.In the process of assembling plate teeth and shaft teeth of miniature escapement wheels,they are easily worn or broken,causing unqualified tooth length.It is known that the number of teeth for the escapement wheel can be even or odd.In view of these two cases,two corresponding tooth length error detection methods were proposed.As shown by the experiments,the accuracy for the even and odd teeth detection method was 99% or 98.8%,respectively.Besides,the detection precision reached 2 ?m and the detection time for a single frame was 40 milliseconds.Therefore,a part could be detected within 5 seconds.At the end of the thesis,the design approach and functional modules of the miniature parts detection system were introduced.