MFL Signal Process Based On Wavelet Analysis And Neural Network

Magnetic flux leakage is a nondestructive testing application based in electromagnetism, It has been used in steel and oil and chemical industry areas, with the development of computer technology, process of MFL signal and intelligent recognition of defects are two important parts in pipeline magnetic flux leak detecting ,The paper presents the significance of pipeline magnetic technology developing ,By comparing the different methods of signal process, wavelet transform and neural network are selected as the main studying tools for the project.Firstly, the basic theory of MFL detecting together with the character of magnetic flux field are introduced, particularly descript the structure and working procedure of pipeline MFL detecting equipment.Secondly, paper recommends the theory of wavelet transform and testing signal singularity. Connects with the characters of MFL signal, analyze the signal by wavelet transform, detect the position of signal break point exactly.Then, the magnetic flux leakage defect signals waveforms detected by Sensors in the time domain are very similar, we can't differ them ,but the harmfulness is different, the paper introduces wavelet package transform and its application .Wavelet paqkage transform has the excellence, is to adjustably.produce a best base as a token to signal according to the different signal, In term of this theory, we decompose some simulate signal and pipeline MFL signals (the cracks and pockmarks signals)by wavelet package transform. Then select the best wavelet package base from the all bases, and depict the best base's brickwork base on the relevant position in time-frequency map, showing the strong or weak of every coefficient. As a result, the time-frequency character structure has been depicted quite visually. Due to the dangers of different defects, it is necessary to detect the type of defects in the project, Analysis time-frequency characteristics of two defects, find out their feature to identify them, so we can identify accurately these two defects.Lastly, The algorithm of MFL signal feature extraction from wavelet coefficients and the method of signal auto recognition based on RBF neural network are proposed in detail. Through studying the signal sample of flaw database, Fuzzy clustering algorithm is used for looking for base function center, then, good structure network is formed. Use this network to identify these defects, compared with BP neural network, identification of the MFL results reaches good effect.