Characterization of defects in Magnetic Flux Leakage (MFL) images using wavelet transform and neural network

The colossal and cramped pipelines are used in piping industries like petroleum industries, power plants and other industries. The buried pipes may experience deterioration or breach of the material which leads to wrecking, percussion etc. So a distinct method is brought in to explore the defect in such regions, known as Magnetic Flux Leakage (MFL) technique which is one of the Non Destructive Testing (NDT) Method. According to this technique, the flux leakage variations at defect regions can be fetched using Hall sensors. The defect can be characterized by utilizing the raw data, which is obtained from COMSOL Multi-physics 4.3a simulation software. Preceding characterization, raw data has to be pre-processed to improve the fidelity of the information related to the discontinuities. In this work, different wavelet based de-noising techniques are performed to remove noise from the raw data. The performance measures are evaluated to select the appropriate wavelet de-noising technique for filtering the Magnetic Flux leakage (MFL) data. Then the features are extracted such as Mean, Standard deviation, Variance, Wavelet energy and Maximum flux value. This reduces the size of the data, which is used to train the neural network. The Feed Forward network is designed prior and trained to evaluate the defect length, width and depth. The corollaries in network training and testing are used to quantify the defect. The training phase plays an important role in characterization process. Percentage error for the different wavelets is computed while characterizing the defect in MFL images.