Detecting And Calibration Of The Surface Defect With Electromagnetic Acoustic Testing Based On Bp Neural Network

Electromagnet Acoustic Transducer(EMAT)is an rising and promising technique to evalueate surface defect.It has many features such as non-couplant,non-contact and high temperature resisitance so it is widely used in various field.However because of its low efficiency,it's rarely used in quantization of defects.This paper focus on the quantization evaluation of surface defects and non-metal coating thickness evaluation using EMATs,and to achieve this goal we study to improve efficiency of EMATs and to increase accuracy of EMATs calibration process.First,analyse the process and mechanism of EMATs working on a aluminium plate.Based on basic Maxwell formula,we build FEM(Finite Element Method)model of EMATs working on the aluminium plate surface using COMSOL.Then using experiment results and Rayleigh wave's two-frequency feature to confirm the FEM model's feasibility and accuracy.Then,to improve efficiency of EMAT's generating Rayleigh waves,this paper utilize orthodox experiment method to achieve this goal.Analyse which parameters of tranducers are fit to be factors of orthodox experiments method.Compare frequency component and time component of each experiment's signals,to confirm different factors and level only change efficiency of the transducer,not other feature of the signal.Present a new iteration mentod of orthodox experiments to obtain a better efficiency of EMATs.Third,based on the fact that when using Rayleigh waves to assess the defect sizes calibration error is quite handsome,present a new calibration process based on BP neural network.On the condition of limited number of calibration experiments,to increase the number of neural network training set of sample,this paper used interpolation method.A comprasion between this new method and conventional polynomial fitting curve method to calibrate EMATs signal is also presented in this chapter.Last,a new method to evaluate thickness of non-metal coating of aluminium plate using EMATs is presented in this paper.This new method also used BP neural network to calibrate datasets.To expand sample space,this paper used a unique logarithm linear interpolation method.The accuracy of this method is studied using actual lift-off experiments results.