Research On Defect Recognition Based On Eddy Current Testing Technology

Eddy current testing is one of the five conventional non-destructive testing technologies.It does not need coupling agent and has the advantages of high speed,low cost and easy to realize automatic testing.It is one of the main means of testing various metal materials and a few non-metallic conductive materials such as graphite,carbon fiber composite materials and product quality.At present,most of the research on material defect identification based on eddy current detection technology is to extract the time domain or frequency domain features of eddy current detection signals,and then select appropriate data processing methods and recognition and classification algorithm to achieve the purpose of accurate detection of material defects.However,due to the difference of detection object,detection method and eddy current detection sensor system,data processing and defect recognition algorithm also need to be adjusted according to different data characteristics,resulting in complex defect identification process,high computational cost,unable to identify automatically,and does not have universal applicability.Aiming at the problems mentioned above,this paper proposes a method of eddy current detection defect recognition based on clustering algorithm,analytical model and machine learning,which achieves the purpose of defect recognition with low computational complexity,high efficiency and more universal applicability.The main research work of this paper is as follows:The graphical characteristics of impedance signal change trajectory are studied,and a new analytical model of the trajectory graph is presented as the basis for further study of impedance signal and defect recognition.In the eddy current testing process,the change track of impedance signal at material defect presents Lissajous figures on the complex plane.In this paper,a new analytical model is proposed to describe the Lissajous figures.By adjusting the parameters of the analytical model,a large number of artificial data sets close to the actual experimental data can be obtained,which solves the problem of lack of data under limited conditions.The geometric features of Lissajous figures are studied and a new method of extracting geometric features based on K-means algorithm for Lissajous figures is proposed.In order to obtain the geometric features(amplitude,width,angle and symmetry)of Lissajous figures,a new method of automatically calculating and extracting geometric features based on K-means algorithm is proposed,and four eigenvalues of Lissajous figures are defined,which solves the problem that traditional feature extraction methods lack universal applicability.Based on the analysis and feature extraction method of Lissajous figures metioned above,five machine learning algorithms are used to automatically identify steel plate defects.Five classificated models,namely,classification tree,discriminant analysis,naive Bayesian,K-nearest neighbor and support vector machine,are trained using experimental data and simulation data.Experiments results show that the proposed method based on K-means algorithm is effective in extracting geometric features,and the analytical model has strong flexibility and robustness.