Research On Measurement Method Of Engine Main Bearing Cap Based On Deep Learning

With the rapid development of the industrial manufacturing industry,the production and demand of parts are increased.In order to ensure the quality of the parts produced,it is necessary to measure the size.The traditional measurement method is low,the accuracy is not high,and the accuracy and efficiency requirements cannot be met.Because machine vision-based parts measurement technology has the advantages of high measurement accuracy and fast detection efficiency,it has been widely used in part size measurement.This article uses the engine’s main bearing parts as the research object.Based on the visual measurement technology of the machine,it studies and designed a deep learning based engine spherical cover parts measurement system.The main research content includes:First of all,before measuring the size of the parts,the parts need to be identified to determine the measurement surface of the part.By analyzing the existing posture identification algorithm,a deep learning based gesture recognition network is proposed to identify posture of parts image.The network can not only correctly identify each measurement surface of the parts,but also have a fast recognition speed.Secondly,the algorithm extracting algorithm of the edge of the parts image was studied.Aiming at the interference of light and parts of the parts,based on DSnet based segmentation networks,combined with the most peripheral constraint algorithm,an improved part edge contour extracting algorithm was proposed.The experimental results show that the proposed algorithm can not only extract a complete,single pixel edge contour,but also effectively prevent interference of influencing factors.Then study the size measurement algorithm.By analyzing the shape and feature point information of the edge of the parts,the ROI and shi-tomasi feature point extraction algorithm is proposed to obtain the feature point coordinate information of the edge contour of the parts,and then proposes to use the european style distance calculation method to calculate the pixel size between the corresponding feature points.Then calculate the actual size of the part according to the camera calibration coefficient.Finally,complete the design and debugging of the experimental system based on the content of this article.By comparing with standard size data,the final experimental results show that the absolute error is 0.097 mm,which meets the factory’s accuracy requirements(0.1mm).