The Research Of The Surface Crack Of Nuclear Fuel Pellet Based On Machine Vision

With the continuous development of China's economic level,the people's quality of life continues to improve,and energy consumption is also growing,especially electric energy,as an indispensable part of daily life,its importance is self-evident.As a clean energy source,nuclear power generation is crucial in the field of power generation in China.Nuclear fuel pellets,as the core components of nuclear reactors,can cause radiation leakage and cause huge property losses once the surface is damaged.At present,neutron photography and DR/CT photography are used at home and abroad to strictly control the quality of the pellets.However,since these techniques require imaging of the overall structure of the pellet,when the inspection task is only for the surface of the pellet,these methods are costly,technically difficult and low efficiency.Therefore,it is of great significance to study an efficient and convenient fuel cell surface quality inspection system.In recent years,the crack detection method based on machine vision has been widely used in pavement,wall,concrete and metal surfaces due to its high precision,non-contact and fast detection speed.Based on machine vision,this paper conducts an in-depth study on the method of detecting surface cracks in fuel pellets.The main research contents of the article are as follows:1)A set of core surface image acquisition devices was built.By this device,we can get the image of the surface of the pellet quickly and the quality of this image is high which satisfies the requirements of subsequent crack detection processing.2)Implement and test several mature crack detection algorithms.Traditional crack detection algorithms such as histogram analysis,gray threshold,edge detection operator and filtering.Machine learning based methods such as support vector machines.And based on the existing images,it evaluates the performance and analyzes the advantages and disadvantages of the algorithm.3)In view of the problems faced by the above methods,an algorithm for CNN-guided Beamlet to implement crack detection.Firstly,the convolutional neuralnetwork and window sliding technique are used to identify the cracked regions in the image to remove the interference of pseudo-cracks and background.Then the Beamlet algorithm is used to extract the crack features for the crack-containing regions.Finally,the interference is removed according to the morphology and the crack detection is obtained.In this method,the CNN model is used to extract the crack-containing regions,which overcomes the effects of gray unevenness and pseudo-cracks in the image itself.At the same time,the CNN model guides the Beamlet algorithm to detect crack-containing regions,compared with the traditional Beamlet algorithm.The image is detected,which greatly reduces the calculation amount of the algorithm,and the calculation speed of the algorithm is greatly improved.4)A method for proposing an adaptive threshold,and determining a threshold according to the gray mean value of each sub-image block itself,and filtering the Beamlet base of the sub-image block.The crack detection effect is greatly improved compared to the method using a fixed threshold.5)According to the actual situation,the types of end face cracks in this project are divided into four types: no cracks,local cracks,through cracks and reticular cracks,and the degree of damage to the imitations increases.The crack type classification model GradeCNN was trained and tested for the classification of crack types,which provided some reference for subsequent processing and analysis.