Magnetic Flux Leakage Detector Development And Defects Quantitative Method Research For Tank Fillet Welds

A tank fillet weld plays an important role of connecting a tank wall and the tank bottom. Because of the particularity of its location and structure, a tank fillet weld bears an enormous alternating load in the daily production and run, making the material easy to produce fatigue defects. In order to ensure operational safety, it is very important to detect tank fillet welds.In this paper, author chooses tank fillet welds as the object of study and makes a research on the magnetic flux leakage(MFL) detection technology of tank fillet weld defects. Based on the basic principle of the MFL detection of the tank fillet welds, author uses the ANSYS software to establish a model tank fillet weld and makes a three-dimensional finite element simulation analysis of the tank fillet weld defects. Author has analyzed how a variety of factors, such as the permanent magnet size, the thickness of plating, the defect depth, the lift-off value, etc., have influenced on the MFL signals of the diagonal weld defects. On the basis of theoretical analysis and finite element analysis, author has developed a MFL detector for tank fillet welds and has made an experimental research on it. The MFL detector for tank fillet welds includes a magnetic system, a data acquisition system, a drive system, a positioning system and some subsidiary structures. Taking an advantage of the detector, author has made an experimental research on the MFL detection of the tank fillet welds in the laboratory. Through the analysis of the wave crests and troughs figure of the defect MFL signals and the optimal sampling point figure, we can conclude that the defect signal amplitude increases with the defect depth within a certain scope. The parameters of the defect size and position obtained in the experiment are consistent with the actual data of the prefabricated defect, which has verified the reliability of the MFL detector of tank fillet welds. Furthermore, comparing the MFL signals in the experiment with the ones in the finite element analysis, we find both are the same in terms of the laws of the MFL signals, verifying the results of the finite element analysis.Combining the artificial neural network with the simulation data in the finite element analysis, author extracts the corresponding characteristic quantity of the MFL signals to build the BP neural network for the quantization of the defect size after the test of the samples. After the test, we can know that the minimum prediction accuracy and the average prediction accuracy were 88.456% and 95.4182% respectively in terms of the BP neural network width, 87.74% and 94.1914% respectively in terms of the BP neural network depth, realizing the quantitative width and depth of defects.