Quantitative MFL Testing Technique for Strand Rope Based on a Small Number of Discontinuity Samples

Magnetic flux leakage testing (MFL) technology is widely used for the in-line inspection of ferromagnetic steel ropes. A common MFL tool is composed of permanent magnets that magnetize a rope axially to saturation and an array of magnetic sensors installed around the circumference of a probe to sense the leakage flux caused by discontinuities (such as broken wire, corrosion, and abrasions) in the rope. During actual testing of strand rope, the MFL discontinuity signals are submerged in the strand waveform signal-the magnetic signal of the surface of magnetized strand rope-and are difficult to detect. The MFL signal distortion is caused by the traditional filtering technique or wavelet signal processing technique. The error associated with quantitative detection of MFL is caused by this factor. The authors propose a technique of eliminating the influence of the strand waveform signal and enhancing the MFL signal based on the structure of strand rope. Complex nonlinear relationships exist between the size of discontinuities and the characteristic values of the MFL signal during quantitative detection; these relationships are combined with a neural network. The problem with the quantitative algorithm is the classification of broken wire numbers and the different discontinuity sizes are considered to be different classes. These traditional techniques based on neural nets used in quantitative detection can determine the sizes of discontinuities in the parameter space for which the neural network is trained, so they require a variety of real samples to train the network in practical applications. In view of the aforementioned challenges, this paper proposes a new technique of quantitative analysis that trains the network without the need for a variety of real discontinuity samples, and obtains accurate results for actual discontinuities in strand rope.