In order to detect the corroded defects of the deep-sea scientific equipment

a magnetic flux leakage（ MFL） signal measurement system which has the three-dimensional（ 3 D） Hall sensor linear array is developed firstly. For a series of artificial pits locally magnetized

3 D MFL signals are measured synchronously. 3 D MFL signals from the array are divided into primary MFL signal and secondary MFL signal. For the primary MFL

extract geometrical characteristics such as crest height

waveform width

envelop area

perimeter

centroid

the distance between crest and trough

etc. For the secondary MFL signal

threshold width and the circular differential width. Then linear fitting is carried out on crest and trough of the Y axis signal and the length of the defect.Finally

the BP neural network model is used to calculate the width and depth of the defects. The experimental results show that the method can mine the concealed features within the MFL signals and the quantization mean absolute errors of the length

width and depth are 0. 63 mm

0. 51 mm and 0. 85 mm

respectively.