Research On Non-destructive Flaw Detection Method Of Eddy Current Thermal Imaging Based On Convolutional Neural Network

Metallurgical industry provides raw materials for capital construction,military equipment and other industries.Detecting whether the raw materials are damaged or not plays a vital role in the production efficiency of enterprises and the safety of equipment operation.How to detect metal damage quickly,accurately and intelligently has become a subject of extensive research in the field of nondestructive testing.In the process of metal smelting,all links may cause metal damage defects,and the location and type of these damage defects are different.Therefore,how to extract the characteristics of metal damage defects in complex industrial environment,and accurately identify the location and type of damage defects based on these characteristics,is a research direction with application value,and also has important significance in industrial applications.Therefore,this paper focuses on the feature extraction and type identification of defect damage of ferromagnetic materials,and solves the problems of defect damage feature extraction and classification of ferromagnetic materials thermal imaging,as well as the scientific problems of feature extraction based on convolution neural network,and finally completes the research and development of non-destructive defect detection system of metal materials based on eddy current thermal imaging.The main research contents include the following parts:(1)Aiming at the data structure of eddy current thermal imaging in pulsed eddy current nondestructive testing,the traditional Hough circle detection algorithm is improved.In order to improve the recognition accuracy of metal damage defect types under complex working conditions,an improved Hough circle detection algorithm is used to extract local features and segment images from metal eddy current thermal imaging data.The experimental results show that the improved Hough circle detection algorithm can accurately detect the position of the inner ring of the high-power excitation coil and extract its internal pixels.(2)The original image datas are segmented by the improved Hough circle detection algorithm,and then the convolution neural network is used to extract the features of the detected area image.According to the experimental results,the parameters of the network model are optimized,and the weight of the network istrained by migration learning to detect the metal defect damage.The experimental results show that the feature extraction method based on convolution neural network can accurately identify the metal defect damage in the area to be inspected.(3)A set of eddy current thermal imaging nondestructive testing system is designed and implemented in this paper.Based on the principle of eddy current heating and infrared thermal imaging,the system can detect defects at any position on metal surface according to actual needs.The experimental results show that the set of experimental device can produce expected physical phenomena when it acts on the metal specimens to be tested,and verify the feasibility and effectiveness of the set of experimental device.In this paper,a set of eddy current nondestructive testing system is designed and implemented.The improved Hough circle detection algorithm is used to extract the local features of the thermal imaging image and to segment the image.The convolution neural network is used to extract the features of the segmented image.The experimental results show that the improved theory,experimental equipment and algorithm can accurately identify the surface defects of metal specimens to be tested.The research results can be applied to on-line nondestructive testing of various metal pipes,rods,wires and wires,and provide an effective technical means for monitoring the quality and health of various metal materials.