Abstract: The reconstruction of magnetic flux leakage (MFL) defect profiles means the reconstruction of defect profiles and parameters from MFL inspection signals. It is the key for the inversion of MFL inspection signals. The studies of MFL inversion problem mainly based on neural network and optimization method. But these two methods have certain shortages. The precision of neural networks may be influenced by noises, and the optimization method is computational demanding. To overcome these shortages, a reconstruction approach for solving such inversion problems based on Bayesian estimation method is proposed. It formulates the inversion problem as a classical discrete-time tracking problem with state and measurement equations. State-space model of defect profile and MFL signals is established, the proposed method is adapted to reconstruct defect profile and the comparison between neural network and proposed method under different SNR. Results indicate that the proposed method has high accuracy and robustness against noise, and it is an effective and feasible approach for solving inverse problems.

Key words: electromagnetic, magnetic flux leakage inspection, profile reconstruction, particle filter, resampling