Research On Defect Detection Technology Of Carbon Fiber Rod Based On Multi-sensor Information Fusion

Carbon fiber rod is widely used in oil production sites because of its corrosion resistance,high temperature resistance,and high strength.How to perform online defect detection on carbon fiber rod has become an important problem that needs to be solved urgently.Therefore,this subject investigates the online defect detection technology for the common defects of carbon fiber rod,and proposes the online defect detection and localization technology for carbon fiber rod based on water immersion ultrasonic array and image recognition.In this thesis,according to the characteristics of multi-layer and irregular boundary of carbon fiber rod,a carbon fiber rod defect detection equipment and a water immersion ultrasonic array with 32 probes are designed to realize online and full coverage scanning of carbon fiber rod.The equipment is used to collect 150 sets of sample for each of the four common defects of carbon fiber rod,and 19200 reflection waveform information were obtained.The ultrasonic array information fusion imaging method is proposed to fuse and map the spatial angle information of each probe and ultrasonic reflection amplitude and time of flight into an image,which can visually reflect the internal boundaries and defects of carbon fiber rods from the images.In view of the feature that the image reflects the defect information as well as the boundary information of the carbon fiber rod outside the defect,the image enhancement should not only enhance the image quality but also enhance the boundary information.This thesis proposes an image enhancement method combining bilateral filtering and Laplace operator to determine the defect and the boundary of the carbon fiber rod.Aiming at the problem that the gray value of the back interface of the image is too small,the multilevel threshold Otsu method is used for image segmentation to ensure the effective retention of image back division interface information during image segmentation and to reduce the case of defects being misclassified as back division interface.Aiming at the characteristics that geometric features such as length,area and width cannot effectively characterize the defects of carbon fiber rod,a feature extraction method combining geometric features such as length sum,area sum and maximum width of defects and texture features is proposed,which can effectively characterize the defects and lay the foundation for classification and identification.A defect recognition method of carbon fiber rod based on deep belief network is proposed.Three geometric features and two texture features are taken as input,and four kinds of defects,such as small crack,transverse crack,hole and chapped defect,are taken as output.The DBN model is established to realize the classification of four kinds of defects,and the recognition rate reaches 88.50%.Finally,an encoder is used to locate the carbon fiber rod defects axially.