Study On Wire Rope Local Flaw Quantitative Testing Based On MFL Imaging Principle

Wire ropes have a wide application in many industrial fields at present, so it has a very important influence on society and economy that they are used in security. The ultimate aim of wire rope nondestructive testing (NDT) is to obtain the residual strength of wire ropes in service and then to decide their retirement dates, and the quantitative testing of wire rope defects is prerequisite to this aim. Most wire rope electromagnetic (EM) testing instruments available on the market produce the circumferentially integrated volume of magnetic flux leakage (MFL) along the rope, and are used to identify wire rope defects with a senior expert's experience generally. This test method with one-dimensional signal can detect almost all serious defects by now, however, it's far below a quantitative testing due to the complexity and diversiform of the MFL of defects and due to the loss of the circumferential information of it.In order to improve the accuracy of quantitative testing of wire rope local flaw, an approach of feature extraction and quantitative identification based on two-dimensional imaging by scanning the magnetic flux leakage around wire rope surface is proposed in this dissertation, and the main results of it are as follows:1. On the basis of the magnetic dipole model for the one-dimensional MFL of a typical flaw, a calculating method and a simulation model of the two-dimensional MFL of the flaw are presented, which can be a useful reference for the measured two-dimensional MFL of the actual flaw.2. On the basis of a study on two-dimensional MFL testing technique, a wire rope hardware inspection platform based on Hall sensor array is established. Synchronizing, spatial domain sampling and transmitting of the multi-channel signals from the Hall sensor array are achieved in the platform. The axial and circumferential MFL information is obtained, which is a precondition of the quantitative inspection of local flaws.3. An adaptive notch filtering algorithm is proposed to filter the strand-waveform which features in high energy and extremely narrow band. It can automatically search for the optimal frequency value of the strand-waveform and set it as the notch frequency. So the parameters of the filter are adaptive with respect to the structure of wire rope without manual adjustment.4. For the defect signal with a narrow axial span is distorted to some extent through the adaptive notch filter above, another noise suppression algorithm based on wavelet packets analysis is proposed. The strand-waveform and the high frequency stochastic noise are all suppressed through this method, and data compression is also achieved at the same time.5. As a preparation of the subsequent feature extraction, the defect signal obtained through wire rope signal denoising is smoothed and transformed to a corresponding gray-scale map, and then a gray-scale enhancement algorithm based on segmental linear transformation is proposed to compress the gray-scale range of noise and stretch that of the defect signal. As a result, the residue noise is attenuated and the defect information is enhanced in the gray-scale map.6. The feature extraction algorithm based on two-dimensional MFL imaging is studied, which includes size normalization, gray-scale normalization and feature extracting based on K-L transformation. The gray-scale map of each defect signal is normalized to a unit with a center corresponding to the center of the defect and with uniform size through size normalization. As to gray-scale normalization, the gray-scale maps of the same kind of defects are stretched to the same rang, and the maps of different kind of defects remain the ratios of amplitude among all the defect signals. Using K-L transformation, the correlation among all components of the gray-scale map of each defect is eliminated, on the basis of which the dominating characteristics are preserved and the minor ones are ignored.7. The defect recognition method based on artificial neural network (ANN) is studied. Experiments are carried out respectively using back propagation (BP) network based on pattern matching and learning vector quantization (LVQ) network based on clustering principle, both with the inputs of feature vectors obtained by K-L transformation, which can be a useful reference for selecting the optimal recognition method.