Research On Key Technologies Of Automatic Crack Detection On The Surface Of Nuclear Containment

With the development of nuclear power industry,the safety of nuclear power stations has been paid more and more attention at home and abroad.As the outermost protective barrier,the safety performance of nuclear containment is very important.The pressure test is the main way to check the strength of nuclear containment under pressure,in which the defects inspection is one of the most important tasks.Through the inspection,measurement,comparison and analysis of the defects before,during and after the pressure test,it can provide an important reference for the evaluation of the containment strength.The defects inspection on the surface of containment has been gradually developed from artificial to intelligent,and the telemetering methods can be used to obtain the images of the containment.However,the defects in the images,especially the cracks,often need to be manually identified,extracted and measured.There are many researches on the cracks detection,but mostly for pavement cracks.Due to the low contrast and complex background of the containment images,the traditional methods can not get good results for detecting the cracks with small size and poor continuity.On the basis of the high resolution image acquisition system of the containment,the key technologies of cracks detection on the surface of the containment are studied in this paper.The main research contents include the following aspects:(1)The external defects inspecting system of nuclear containment is introduced from the aspects of system framework,collection environment,work flow,hardware and software module.Aiming at the source and characteristics of the images,the research foundation for the subsequent crack detection is established.(2)A fast mosaic method for poor texture images on the surface of the containment is presented.The algorithm is based on the overlapping images whose texture feature is not obvious by using the fixed distance parameters and approximate conditions.Firstly,the SURF operator is used to carry out the feature matching between adjacent images.Then the images are normalized to the same scale.Finally,the images are cropped into the same size according to the degree of overlap.The mosaic method can effectively realize the panoramic image stitching on the surface of the containment,which is helpful for the subsequent crack extraction,measurement and defect browsing.(3)According to the crack characteristics of small size,poor continuity,narrow width and low contrast,an adaptive width template binaryzation algorithm is proposed.The algorithm can obtain the skeleton information of cracks by searching for the most reasonable width from both sides of the local gray minimum point according to the characteristics of the cracks in the shape and gray level.At the same time,an improved block iterative algorithm is proposed to overcome the limitation of the global threshold in traditional iterative algorithm.The proposed algorithms can preserve all the features of small cracks,but the background noise is difficult to suppress.(4)The denoising algorithm based on morphology is proposed aiming at the problem of too much background noise in binaryzation images.The results of the adaptive width template and the improved block iteration method are gotten intersection to reserve the same crack information and eliminate the different noise information.Then,the longitudinal expansion convolution kernel is used to enhance the crack features according to the morphological characteristics of cracks.In addition,the features such as length,area and centroid ratio are comprehensively considered to achieve the effective removal of the background noise.(5)In order to further enhance the crack morphology and achieve the target extraction,the algorithm of tensor voting using trend information of target points is studied.In order to improve the effectiveness and efficiency of the algorithm,a two degree tensor voting with linear saliency enhancement is proposed by using the first vote weights for the twice vote;a local area tensor voting method is proposed to screen the image regions which may contain cracks;a tensor voting with initial information is proposed by embedding the previously acquired initial direction and saliency information into a new voting field;and the tensor voting based on the expansion and filtration is proposed.The improved methods can reduce the number of voting pixels,improve the efficiency of the algorithms,and enhance the linear morphological characteristics of cracks.(6)Based on the saliency and direction trend information of the target points on the tensor voting probability graph,the extraction of the target is realized.First of all,the seed points of the probability map is sampled by the gray scale.Then,the seed points are calculated to obtain the minimum spanning tree,and the centerline of the crack is obtained by further pruning and connecting of branches and leaves.Finally,the crack center line is projected into the original image,and the crack points are crowded together according to the gray level gradient to obtain the true crack information.(7)In order to identify the true cracks,the linear fitting and gray level change are used to exclude the other non-crack linear targets.Based on the coding rules,a crack arrangement scheme is proposed for the same crack target across multiple images.The further measurement of the length and width of cracks is realized.(8)According to the research of the algorithms for the crack inspection on the surface of the containment,the experiments are carried out by using the collected images.The experimental results were qualitatively and quantitatively analyzed from the overall effect,the local performance and the compared methods.The validity of the algorithms are verified.