Research On Data-driven Defect Recognition And Diagnosis Of Weld Eddy Current Signals

With the development of modern industry,welding technology is an indispensable means of industrial technology in mechanical processing industry,and plays an important role in national economic construction.In the narrow overlap resistance welding process,the inspection of weld quality needs to shutdown first,and then observing the weld morphology or doing the cup convex experiment by sampling.But these methods are not real-time.In this paper,relying on the real-time detection project of narrow overlap resistance welding,and establishing the eddy current detection equipment following the welding machine during welding,then eddy current detection signal data and welding process data can be collected in real time.Based on these data,the intelligent identification model and cause diagnosis model of weld defects are established to realize the real-time detection of weld defects and improvement of welding quality,the main research works are as follows:(1)Establishing the eddy current detection equipment on the welder to collect real-time ECT signals of narrow overlap welds,and proposing an improved wavelet threshold algorithm to de-noise the original ECT signals of welds.The simulation results show that the improved wavelet threshold denoising algorithm is feasible and effective.It can filter the noise in the ECT signal better,and has stable denoising effect.(2)Based on the ECT signal after noise reduction,using the traditional method and the deep learning method to identify the weld defects.For the traditional method,EMD is used to decompose the ECT signal of narrow overlap weld,and the feature vector is constructed on the basis of the obtained IMF,then PCA is used to reduce the dimension of the feature parameters,finally SVM is applied to classify,and the defect recognition accuracy is 86.7%.The deep learning method combines wavelet time-frequency diagram with CNN to identify weld defects.Firstly,the original ECT signal is transformed into time-frequency feature map by continuous wavelet transform,and then defect recognition is carried out by using the CNN model of VGG16.In order to ensure the real-time recognition of weld defects,a two-stage defect recognition method based on CNN is proposed.Experiments show that the accuracy of this method reaches 96.94%,which is nearly 10% higher than the traditional method.The average time consumed in actual detection is only 2.4s,which can meet the online detection requirements of enterprises.(3)On the basis of identifying the weld defects,combined with the data of welding process parameters,the association rules algorithm is used to mine the relationship between different weld defects and these parameters,so as to diagnose the causes of different defects.The results show that the association rule model can improve the knowledge discovery of welding experience,thus providing guidance for the improvement of welding quality.In this paper,the research object is the narrow overlap resistance welding seam,and the intelligent identification model and the cause diagnosis model of weld defects are established based on data.The real-time on-line detection of weld seam is realized,which provides guidance for the improvement of welding quality and has good application prospects.