Quantitative Detection Of Cracks In Mechanical Parts Based On Magnetic Flux Leakage Testing

As the most frequently used material in the world,the ferromagnetic material that takes steel as the main ingredient has been widely used in many fields,such as electromechanical construction,mining,petrochemical,municipal,hydropower,transportation,aviation,agriculture and animal husbandry.Due to the sophisticated circumstances,industrial technologies inevitably introduce heavy equipments with high precision.Note that,most devices are working under the conditions of the high speed and temperature,the large pressure,as well as a long running duration,which might lead to a myriad of injuries,equipment failure and even catastrophes.Thus,in the domain of mechanical assessment a reliable and effective evaluation of early damages inside equipments is desirable so as to reduce the risk of accidents.Nondestructive testing technology is often viewed as an important approach to guarantee safe operations of equipments.The currently predominant techniques include:eddy current nondestructive testing,magnetic flux leakage nondestructive testing,magnetic particle nondestructive testing,ultrasonic nondestructive testing and other non-destructive testing methods,where each method has its own technical characteristics and scope of application.Magnetic nondestructive testing technology has been widely used due to its congenital advantages for the detection of ferromagnetic materials.MFL nondestructive testing,referring to the use of magnetic field excitation on the test piece,is essentially a sort of local magnetization.If the test piece surface cracks,the magnetic flux guide area will decrease sharply,which is due to the fact that the leakage magnetic field leaks,and thus the MFL signal that is leaked out can be detected by the sensor.Subsequently,the defect state of the test piece is estimated on the basis of the change of magnetic field.Compared with other nondestructive testing technique,the MFL nondestructive testing technology has advantages of high detection efficiency,broad range,easy operation,and therefore it has been used in applications with high-speed dynamic loads,severe environment where equipments are difficult to disassemble,and the crack defect on the tiny surface of testing piece.This thesis is dedicated to the quantitative crack detection of mechanical components through the magnetic flux leakage testing.As a nondestructive testing,the magnetic flux leakage NDT method can not only detect the crack location of testing piece,but also offer an accurate quantitative analysis of the size of the crack defect,yielding a precise prediction and judgment in the process of crack detection.This thesis is organized as follows:The basic concepts and theories of magnetic flux leakage detection are first provided,subsequently,the relation between the defect of rectangular flaw and the signal intensity of magnetic flux leakage is analyzed in detail by using magnetic dipole method.By making use of the defect inspection of mechanical parts surface crack,this thesis proposes an electromagnetic multi module MFL nondestructive testing system,designs a magnetoresistive sensor for leakage detection which performs as a core component in a magnetic circuit series nondestructive detection system,and then accomplish a physical production series circuit.Finally,with a 3D sensor module,this thesis provides a set of MFL nondestructive testing system for the quantitative defect detection of mechanical parts.A BP neural network based defect recognition system is developed,where the neural network is optimized by the genetic algorithm.BP neural network is trained by samples which are collected by the magnetoresistive sensor.The BP neural network(for quantitative crack identification)is treated from a deep learning perspective,where crack parameters are mapped to the experimental sample signal so as to establish a forward model for the prediction of the width and depth of cracks.Compared with actual defect sizes,BP neural network achieves a satisfactory performance,i.e.,the predicted errors are lying within an acceptable range,implying that BP neural network is applicable to predict the width and depth of micro-crack,as well as the size of crack defects.