The Magnetic flux leakage (MFL) method is commonly employed for the non-destructive evaluation (NDE) of steel plates used in the construction of oil storage tanks and pipelines, where large areas have to be covered within a short time. Due to fundamental characteristics of the MFL method, very narrow and deep pipe-type defects can produce very similar signals to wide and shallow lake-type defects. This inherent uncertainty of MFL can cause dangerous misinterpretations of the measurement data as crucially deep defects could remain unnoticed.

This paper proposes a numerical method for the computation of the worst-case (WC) solution candidate to the inverse MFL problem in terms of defect depth, which produces MFL signals differing from the original data only within the limits of the measurement uncertainty. A fully nonlinear magnetostatic FEM model is used to create Taylor series expansions of the model error w.r.t. the multi-parametric surface profile in 2D, which is then iteratively changed into the WC estimate. The algorithm is explained and its effectiveness is illustrated for a particular simulation example, showing the resulting WC depths under different conditions.