Study On Electromagnetic Wave's Feature And Its Recognition Of Electrostatic Spark

The semiconductor integrated circuits with great improvement has become a powerful booster for the upgrading of modern industry,however,spark caused by electrostatic discharge can make industrial equipment failure which depends on the high precision integrated circuits.The electrostatic spark can radiate high frequency electromagnetic wave which has great energy that will damage the equipment or cause the fire and other serious accidents.In this paper,four kinds of generating models of electrostatic spark and four kinds of conventional testing methods are introduced,at the same time,the corona discharge,spark discharge,brush discharge and brush discharge with propagation form are introduced as the research objects.Lots of samples are got for study on electromagnetic wave's feature and its recognition of electrostatic spark by experiments.The waveform obtained is usually mixed with noise from environment.Then wavelet denoising and wavelet packet denoising algorithm are simulated and compared.The simulation results show that the wavelet packet denoising algorithm is more suitable for the electromagnetic wave of electrostatic spark and the simulation results are verified by the real signal denoising effect.The Hilbert-Huang transform is taken to extract the features,and finally the marginal spectrum energy of the six high frequency signal components which have the main energy of the signal is extracted as the feature parameters.Then,the recognition work is in progress.RBF neural network is used to recognize the electrostatic spark,and then the support vector machine based on the linear kernel function and radial basis kernel function is taken to recognition work where the particle swarm optimization algorithm is used to optimize parameters.There are 300 spark samples where the 220 samples are training samples and the others are test ones.The recognition rate is 91.3% by RBF neural network with 73 correct,5 error and 2 failure;The recognition rate is 92.5% by SVM based on linear kernel function with 74 correct and 6 error while the recognition rate is 95% by SVM based on radial basis function kernel with 76 correct and 4 error.The recognition rate of the brush discharge is much higher.So,the support vector machine(SVM)is more suitable to recognize the electrostatic spark.