Research On Application Of Machine Vision In Surface Defect Detection Of Underwater Nuclear Power Equipments

Safety is always the foundation of nuclear energy development.Surface defects are an important factor affecting the safety of the nuclear power plants(NPPs),and surface defects of key equipment will cause the risk of leakage.Therefore,surface defect detection is very important,and it is also one of the research hotspots in the field of non-destructive testing(NDT).During the operational stage of a NPP,some equipments are only accessible on one side,which makes vision-based inspection method the only optional surface inspection method.However,due to the particularity of the inspection environment in NPPs,most of the inspections need to be carried out underwater,which makes the application of related technologies difficult.Aiming to solve the difficulties of application machine vision technologies in NPP inspection,the methods of underwater image quality improvement and image mosaic are proposed in the thesis.Moreover,a three-dimensional(3D)detection method based on photometric stereo(PS)is studied to cover the shortages of two-dimensional detection methods.The contents and achievements of the thesis are as follows:(1)An image quality improvement method combining Automatic Color Equalization(ACE)and Dark Channel Prior(DCP)algorithm is proposed to solve the problem of underwater image degradation.The degradation characteristics of underwater inspection images of NPPs is studied.The color richness of the images is improved with ACE,and the influence of fogging is eliminated with DCP.Furthermore,the insufficient brightness of DCP image is corrected by modifying the estimation formula of transmittance and background lights.Experimental results show that the underwater image quality evaluation index(UIQM)of pressure vessel is improved by 60%with the image enhancement technology.(2)A super-field image generation method is proposed to solve the problem of limited field of view(FOV)of camera in large scene.The field of view and pixel size of the camera are obtained by a laser,and the pixel changes between image frames in the video are calculated.Then,the corresponding number of pixels in the image frames are extracted,and a complete image is combined.The method overcomes the shortage of image stitching methods without feature extraction,and obtains a reliable and high precision super-field image.Experimental results show that the method provides a mean for super-field defect analysis.When the test parameters and camera parameters change,the relative error of measurement is less than 1%.(3)An estimation method of light direction and light intensity under the condition of unfixed installation of the camera and light sources is proposed to solved the application problem of photometric stereos in the scenario of unfixed installation of the camera and light sources.The distance between the camera and the surface to be inspected is obtained by a laser.Then,the positions of the light sources and the surface to be inspected are calculated,and the light directions under the illumination condition of the near-field LED are estimated.The luminous characteristic parameters of the LED are determined with iterative optimization algorithms.The normal vector errors caused by different light sources are eliminated to achieve accurate estimation of illumination intensity.Depth information of surface defects can be obtained,the depth measurement errors of scratches are less than 0.5mm.(4)An underwater PS method based on light model is proposed to solve the problems caused by changes of light intensity distribution and light paths underwater.According to the transmission of light refraction effects between different media path changes,combined with the transmission law of of light refraction,light converging effects in the water,brightness equations of the incident lights and the reflected lights,the transmission path of the reflected light entering the camera through different media are analyzed.A general model of underwater lights is deduced under the condition of point light sources with near-field LED.Furthermore,the light intensity distribution in the light field is determined to calculate high precision surface normal vectors underwater.Experimental results show that accurate information of small-scale features underwater can be obtained,which is helpful to improve the detection ability of micro-defects on the surface of nuclear power equimpents.The image quality improvement method is evaluated with public underwater image data set,LI test set and underwater inspection data set of NPPs respectively.The PS method is evaluated with natural defects and artificial defects of surface damage samples.In the underwater PS experiment,ping-pong balls and flat plates with approximate Lambert reflection,coins with rich texture details,and real detection objects(welds)were tested.Experimental results verify the effectiveness of the proposed methods.