Research On Recognition Technology Of Deep Hole Internal Surface Defects And Roughness Based On Industrial Endoscope

The surface roughness of the parts has an important influence on the wear resistance,corrosion resistance,contact stiffness,fatigue resistance,service life and reliability of the parts.With the continuous improvement of the requirements for manufacturing accuracy of products,the detection accuracy and eficiency of surface defects and roughness in deep holes have also raised higher and higher requirements.The contact measurement method will leave scratches on the surface to be measured and is inefficient.The principle of non-contact measurement methods based on sound,light and magnetism is complex,and the operating environment requires high.In recent years,the surface roughness based on machine vision and image processing technology Degree measurement method is widely used.This paper proposes a deep hole inner surface quality detection method based on endoscopic images and deep learning,which can solve the rapid detection and recognition of deep hole inner surface defects and roughness.In this paper,based on the German Karlstos SP100-Sl industrial endoscope,a self-designed centering device is used to obtain a large number of deep-hole endoscopic images of tube plate parts,and the deep holes are divided into spiral holes according to the surface image characteristics of the deep holes..Chatter hole,ring hole three typical types.After line cutting the ring hole,the three-dimensional shape of the surface of the specimen was obtained using the Leica DCM-3D laser confocal microscope,which provided sample data for the roughness identification of the ring hole.The image preprocessing is studied,and the template matching circle center detection algorithm is designed.The algorithm uses the gray circle with known circle center and radius as the matching template,and determines the image ring texture by matching it with the edge image to be detected.Center and radius,and use genetic optimization algorithm to improve the search efficiency of the matching algorithm,to achieve fast and accurate detection of the center of the deep hole endoscopic image.The endoscopic image was circularly intercepted according to the center of the image circle,and the intercepted image was interpolated and expanded using a bilinear interpolation method,and a neural network identification sample database for deep hole internal surface defects was obtained.A convolutional neural network was proposed to classify and identify deep hole defect types.The influence of convolutional neural network structure and parameters on classification and recognition performance were studied.A convolutional nerve based on deep hole endoscopic image classification of hole surface defect types was established.The network uses deep hole endoscopic unfolding image sample data for convolutional neural network training.The test sample classification results show that the established convolutional neural network can achieve accurate classification and recognition of surface defects in deep holes,with a classification accuracy of 93.8%.For the normal deep hole endoscopic image with ring texture,the gray level co-occurrence matrix was used to statistically analyze the unfolded images of the inner surface of the deep hole with different roughness,and the correlation between the characteristic parameters of the gray angle co-occurrence matrix with different angles and the roughness were studied.Four characteristic parameters with strong correlation are obtained.The BP neural network is used to establish the hole surface roughness recognition model.The training of the roughness parameter recognition BP neural network is completed.The test results show that the established BP neural network roughness recognition model can Realize the accurate identification of the roughness parameters of the inner surface of the deep hole,and the identification accuracy can reach 84%.The thesis research provides a new method for classification and roughness detection of deep hole inner surface defects.