Development Of Detection And Measurement System For Surface Defects Of Spheroid

As the core component in a nuclear reactor,the safety of its use is critical.In order to ensure the safe operation of nuclear reactors,it is necessary to detect defects on the surface of nuclear fuel pellets to screen out those that do not meet safety regulations.The special shape of nuclear fuel spheres and the complex surface information make it difficult to obtain either surface images or defect detection and measurement.In this thesis,to address the above problems,a deep learning-based spherical surface defect detection algorithm and a surface mapping-based defect measurement algorithm are investigated to establish a spherical surface defect detection and measurement system to enhance the detection and measurement accuracy of nuclear fuel sphere surface defects.The main research and innovation points of the paper are as follows.(1)A spherical image imaging device based on a line array camera was designed to address the problem of difficulty in acquiring images of the surface of nuclear fuel spheres.First,according to the characteristics of the nuclear fuel sphere itself,the line array camera and its suitable lens and light source are selected,and the distance between the imaging devices is calculated by using the relevant parameters of the camera and lens.Secondly,introducing stepper motors according to the demand of imaging device and designing hardware circuits with stm32as the core to drive nuclear fuel spheres with line array cameras to capture images,as well as designing the structural diagram of the imaging device.Then,the software process is designed according to the hardware device,using the speed conversion formula to match the camera line frequency and stepper motor speed to ensure that no distortion occurs in the captured image.Finally,the acquired spherical image is reasonably cropped to leave a complete image of one week of rotation of the nuclear fuel sphere.(2)For the problem that the defects on the surface of the sphere are small,and the proportion of them is too small compared with the background information,the measurement accuracy is low.An improved U2-Netp target segmentation algorithm is designed.The algorithm first replaces the up-sampling method in the RSU module with bi-cubic linear interpolation to enhance the image smoothness after up-sampling.Secondly,constructing spatial-channel attention and multi-feature fusion modules between the encoder and decoder to improve the accuracy of acquiring information of interest to the model.Finally,the fused prediction output of Sigmoid and Threshold and the Log＿Dice＿focal loss function are designed to overcome the problems caused by the gradient disappearance phenomenon and the extreme inhomogeneity of positive and negative samples.(3)A surface mapping-based algorithm for measuring spherical surface defects is proposed for the problem that spherical surface defects have different and irregular shapes and are difficult to measure in terms of length and area.The algorithm first performs a connected-domain marking of the spherical image.After that,it measures the area information of the spherical defects.or the skeleton refinement and then the length of the defect is measured by the connected domain marker.Secondly,the sphere is similarly unfolded using the principle of column surface unfolding and the unfolded area is divided to map the length and area of the defects on the surface of the sphere,respectively.Finally,the mapped length and area are marked on the spherical image respectively.(4)A three-dimensional reconstruction algorithm based on texture mapping is designed for the problem that the spherical image obtained by using a line array camera varies greatly from the shape of the sphere,and the defect location and size are difficult to observe.First,plotting the body data of the sphere in 3D space using the VTK algorithm and converting the spherical image into a texture patch.Next,determine the mapping relationship between the texture patch and the body data of the sphere,and map the texture patch to the body data of the sphere.Final rendering of the objects in the scene and interaction design using the mouse.Based on the above,this thesis designs a system for detecting and measuring defects on the surface of a sphere and tests it.The system designed in this thesis is fully functional and stable in operation.The imaging device of this system can image completely within 15s,and the defect detection accuracy reaches 85.3%,while the error of defect length measurement on the sphere surface is less than 1mm,and the area measurement error is 3mm~2,and the sphere is not deformed after 3D reconstruction,and all the indicators meet the enterprise requirements.