Research On The Filtering Methods Of The Pineline Magnetic Flux Leakage Inspection Data

Pipeline is widely used in our country as the transport facilities of the medium, for example, gas and oil. But due to the ageing pinepine, the pipeline leakage occurred frequently which is caused by corrosion, swearing and accidental damage. Therefor, research on pinepine leakage detection is of great significance. Nondestructive testing is put forward at the background of detecting the running state of pipeline when it works normally. However, the signals by nondestructive testing often contain much noise. The existence of the noise has an effect on the recognifition of the real state signals. Also, a large amount of data was got in the detetction process, which put forward higher request to the operating capacity for computers. So how to filter the big data fast and effectively becomes the problem that we must solve.The characteristics of the MFL signal and noise which is contained in the MFL signal were analyzed first in the thesis. And the MFL signal was simulated based on the characteristics. Pulse interference noise, periodical interference noise and white Gaussian noise were added to the ideal MFL signal without signal. The frequency spectrum analysis on the MFL with noise was carried out.Applying the median filter to the simulated signal was put forward, which demonstrated the feasibility of eliminating the pulse interference noise. Because the frequency of periodic interference noise is single-valued, three methods of designing the trapper were put forward and their performance was compared. The IIR notch trapper based on all-pass filter with optimal performance was applied to the simulated signal, which demonstrated the feasibility of eliminating the periodic interference noise.An approach to wavelet filtering based on EMD was presented. And apply wavelet threshold filter, the EMD filter and it to the simulated signal respectively. By analyzing the filtering effects, wavelet threshold filter with better comprehensive performance was selected and applied to the multi-channel simulated MFL signals. The phase deviation of the signals is not apparent after filtering, which demonstrated the feasibility of eliminating the white Gaussian noise.Finally, how to seclect cut-off points when dealing with big data and how to eliminate the hopping when cut-off data after filtering were presented were solved. An adaptive cut-off metihod was proposed and improved to make that the adjacent data have shared data. The hopping was solved by transition.