Research On Extracting Critical Features Of Magnetic Memory Signals Based On Fuzzy Clustering Analysis

Because of the influence and interference of various factors such as the external environment during the test process,the magnetic memory data will appear ambiguous and dispersive.Therefore,it is difficult to distinguish the normal state only by the magnetic memory magnetic field strength curve and the gradient value curve.The critical characteristics of these four stages are: sexual injury,crack-penetration,and crack penetration.According to the X-ray examination results,the characteristics of magnetic memory signals in different stages of injury are extracted to determine the critical signal feature state.The fuzzy c-means clustering algorithm can quantitatively classify and identify signals at different stages of injury.In order to study the variation law of magnetic memory signals at the welds of the welded specimens under fatigue loading and to extract the characteristic values of the critical state of the signals,two kinds of fatigue tensile experiments with welded flawed plate specimens were designed and recorded during the experimental test.A large number of magnetic memory signal data are analyzed and analyzed,and the signal change laws of these magnetic memory signals under different defects and different damage stages are summed up.Combined with X-ray detection results,basic data are provided for the quantitative identification of weld defect levels in the next step..The following six kinds of magnetic memory signal critical characteristic parameters were extracted.The normalized signal strength change rate ?Hpy /?x and the tangential region signal strength change rate ?Hpx /?x,the normal gradient Kwy and the tangential gradient Kwx,and the normal gradient limit state were selected.The coefficient My and the tangential gradient limit state coefficient Mx perform vector synthesis on the tangential and normal eigenvalue vectors of three kinds of magnetic memory signals to form a three-dimensional synthetic feature vector,which is used as a parameter for extracting critical features of the magnetic memory signal.The fuzzy c-means clustering algorithm is used for training samples of incomplete penetration test specimens and slag defect specimens to calculate the prediction results and mean errors under different fuzzy weighted exponent m.The fuzzy weighted exponent m is optimized,and then on this basis,two kinds of defect clustering centers ci are calculated separately,and a fuzzy c-means clustering quantitative identification model is established based on the cluster center ci as a benchmark.The four states of the welded specimens are classified and verified,and the predicted classification accuracy rates obtainedare respectively For 84.62% and 81.48%,this method is validated for predictive classification.