Nondeseructive Detection Of Fin Tube Based On Machine Vision

The air cooler belongs to large equipment in petrochemical factory.Timely detection of internal damage of the finned tube in the air cooler can avoid leakage accidents caused by damage.At present,in petrochemical factories,usually the industrial endoscope is used by workers to detect the finned tubes one by one manually,which is less automated and easily lead to failed detection.In view of this situation,this thesis took the air-cooler finned tube as the research object,and designed a nondestructive inspection system for finned tube based on machine vision to to replace the manual detection method for achieving automatic detection.The nondestructive detection system of finned tube was studied,and the main research contents are as follows:1.The limitations of current nondestructive detecting technology for finned tube were analyzed.According to the characteristics of the finned tube,the overall design of nondestructive inspection system based on machine vision was finished as well as the hardware modules had been chosen.Finally,the hardware system to collect images inside the finned tubes was constructed which can control the detected position of the nondestructive inspection system.2.A series of image pre-processing algorithms were designed to analyze the image inside the finned tube acquired by the camera,the functions of which included the image pretreatment,the ROI(Region of Interest)extraction,the polar coordinate transformation of image,the compensation for image brightness,the image splicing,image binaryzation by adaptive threshold,as well as image noise quick elimination and image segmentation.The image processing algorithm can be used to extract the damage characteristics of the inner wall of the fin tube from the original image captured by the camera completely.3.A damage detection and identification network based on Faster RCNN was designed.The structure and principle of the three generation target detection network were analyzed.The network model was optimized to make it more suitable for internal damage detection of finned tube according to the characteristics of internal damage of finned tube.The data enhancement method was used to expand the training data set of the network which was then used for completing the network training.The accuracy of the network was tested,and the test resultsshow that the optimized network model can identify the damage type and mark the damage area from the damage feature image.The network has achieved good classification accuracy for the test data set.4.The main functions and testing requirements of the nondestructive inspection system were analyzed.Based on QT,the operation software on upper computer of the system was developed to realize human-computer interaction.The application test in the workshop was carried out at the industrial site of petrochemical factory.In the laboratory,the detection accuracy of the inspection system reached 88.1%.The reliability of the test system were proved by analyzing the experimental results,and the improvement direction of the inspection system is clarified.A nondestructive detection technique based on machine vision was proposed,which combines image processing with deep learning technology and can be applied to detected finned tubes.According to the characteristics of the finned tube,the hardware structure of the detection system and the software of the upper computer were designed,thus ensuring high accuracy of detection and good applicability of the nondestructive inspection system.This thesis presents the solution for fine tube inspection under the conditional constraints and interference factors that may exist in the actual detection process.Finally,the detection system has high detection accuracy and achieves automatic detecting.To a certain extent the nondestructive inspection system of fin tube designed in this thesis can replace the manual inspection method.It has certain theoretical value and practical application significance.