Design Of Multi-probe Rotating Eddy Current Testing System

In recent years,stainless steel welded pipe has been widely used in automobile,oil,hydraulic pipeline and other physical industries,especially in the automotive field.In order to ensure the personal safety of automobile drivers,the quality of stainless steel welded pipes used in automobiles has become the top priority.How to accurately identify the defects of stainless steel welded pipes is becoming the focus of attention.The defects of welded pipe are mainly concentrated at the weld,and the causes of different defects are also greatly different.As we all know,eddy current testing technology is an important member in the field of nondestructive testing.Compared with other nondestructive testing methods,it has the advantages of high efficiency,accuracy,no coupling agent,and easy to realize automatic testing,which makes it outstanding in the quality testing of metal materials.Because of the low cost of eddy current testing,it is often used to detect the quality of conductive materials.However,the impedance plane analysis method of eddy current testing,which we often use,mainly detects the existence of defects,and cannot identify the types of weld defects of stainless steel welded pipe.Aiming at the defect classification and recognition of industrial stainless steel welded pipe,this paper proposes an effective method for the defect classification and recognition of stainless steel welded pipe based on eddy current detection.This method is applied to signal processing(short time Fourier transform)and machine learning(convolution neural network).First,STFT is used to preprocess the original eddy current signal into two-dimensional time-frequency map;Then,the two-dimensional time-frequency map is input into the input layer of the convolutional neural network(CNN).In order to improve the accuracy in the classification and recognition of stainless steel welded pipe defects,the following operations were carried out: input the two-dimensional time-frequency map into the input layer of two neural networks VGG-16 and Goog Le Net,compare the results under the same learning rate,and use the neural network training model with higher accuracy.The results show that the STFT-CNN method is more accurate than the traditional method in classification and recognition,and the classification effect of VGG-16 training model is superior to Goog Le Net training model at 0.0001 learning rate,which has certain reference significance for the classification and recognition of industrial stainless steel welded pipe defects.The experimental results show that the multi-probe rotary eddy current testing system designed in this project has the characteristics of low false alarm rate and high degree of automation,which is reasonably applied by the factory,and shows a strong effect,which has certain reference significance for nondestructive testing in the field of stainless steel welded pipe testing.