VELOCITY EFFECTS AND TIIEIR MINIMIZATION IN MFL INSPECTION OF PIPELINES - A NUMERICAL STUDY G. Katragadda Y.S. Sun W. Lord S. S. Udpa and L. Udpa Department of Electrical and Computer Engineering Iowa State University Ames Iowa 50011 INTRODUCTION A wide variety of methods are used for the inspection of the 448000 kilometers of gas pipelines currently in operation in the United States. Speed and accuracy are the prime concerns in inspections of this magnitude. Magnetic flux leakage (MFL) inspection of pipelines [1] using a magnetizer moving at velocities up to 30 kilometers per hour is currently the most commonly used inline inspection method. At these velocities the leakage field signal is significantly distorted due to motion ally generated currents in the pipeline. Experimental measurements of the velocity effects is expensive and possible for only very limited choices of parameters such as geometry and dimensions of the probe defect etc. Analytical closed form solutions for electromagnetic (EM) non-destructive testing (NDT) problems including velocity effects can be found for only the simplest examples and are impractical for most NDT problems. Numerical analysis techniques for the modeling of velocity effects in a variety of EM areas are developing rapidly [2] [3]. In modeling the MFL inspection the numerical model is required to be capable of modeling non-uniform geometries in order to simulate defects. Also for accurate predictions nonlinearity in various regions of the geometry must be incorporated. A numerical model with these capabilities is an invaluable asset both in terms of studying in detail the total physics of the situation and also to aid in the magnetizer design. NUMERICAL MODEL Standard techniques such as the upwinding technique [4] can be used to model velocity effects in objects of uniform geometry. Since the geometry discussed (Figure 1) in this paper involves a defect in the tube wall and the position of the defect relative to the probe varies continuously with time the problem must be treated as one with a moving boundary. The transient nature of the process calls for the use of time stepping methods. Review of Progress in Quantitative Nondestructive Evaluation. Vol. 14 Edited by D.O. Thompson and D.E. Chimenti Plenum Press New York 1995 499
B shes Defect Spring mounted Sensor #2 =/==;~a~~~~~~~~~~:~:~~~:~~~l~~~~!== Pipe wall Sensor #1 Magnet(Leading pole piece) Magnet(Trailing pole piece) < Direction of motion (a) (b) Figure 1 Typical MFL assembly for pipeline inspection. (a) Cross-sectional side view (b) Cross-sectional end view. There are two different methods of handling the moving boundary problem. In the first method a moving coordinate system is used. In this method the observer is positioned on the object where the motion induced effect occurs. In this case there is no motion seen by the observer except that of the field source. The field problem can be expressed and solved by the ordinary Maxwell's equations provided that the continuously moving field source is defined. This method has the advantage that the VX B (motional induction) term responsible for spurious oscillations in other formulations does not appear in the governing equation. On the other hand this approach is laborious since the motion is taken into account by a moving mesh. This generally involves extensive work in mesh regeneration or local remeshing. Recently several methods to overcome this disadvantage have been suggested [4]. In the second method a fixed coordinate system is used. Here the observer has a fixed geometrical relation with the field source. A motion related term vx B appears in the governing equation which is given by: t7 1 (t7 -) - a A - t7 - vxj.1 vxa = Js-cry+crVXvXA (1) 500
to: For axisymmetric geometries discussed in this paper the governing equation reduces (2) It is found that among the several time stepping algorithms used to solve the above non-selfadjoint equation Leismann-Frind's method [5] provides a better solution in terms of the overall considerations of accuracy and stability for EM NDT applications. The Leismann-Frind method introduces an artificial diffusion term into the governing equation and uses different unknown time weighting factors for the individual terms in the governing equation (Eq. 2). The unknown artificial term and the time weighting factors are decided during the process of minimization of the errors. Expanding the resulting equation in a Taylor series results in the final Leismann-Frind equation which calculates the solution A n + 1 at the (n+l)th time step based on a known set of solution values at the nth time step An using (for simplicity a ID case is represented): [ ~ ll 0 crv2 ~t i]an+l _ {cr crv2~t a 2 a] n At - 2-2 2 - J s At + (1-0ar=---n-=-:.2 ::>z2 - crv... z A L.l Il az a az L.l 0 0 As is apparent from the equation the introduction of individual weighting terms results in a symmetric matrix. This method also eliminates spurious oscillations (by introducing the artificial reluctance term) that are possible in other formulations (Donea's method [6]). A 2D axisymmetric FEM code using quadrilateral elements is used to update A at each step. The coordinate system chosen assumes that the probe is fixed while the tube wall which may contain a defect is moving. The validity of the code is demonstrated by matching results with a series of experimental results. Figure 2 shows preliminary validation of the code for the static case by matching axisymmetric FEM axial scan results with results obtained from a 3D static FEM code and experimental scans with the magnetizer in air and with the magnetizer in pipe. These results are very encouraging considering the complexity of the geometry being modeled and the fact that the geometry being modeled is truly three dimensional in nature (Figure 1) whereas the 2D code approximates the geometry using axisymmetry. DATA ANALYSIS A numerical model gives us a convenient way of examining flux lines and induced currents which is not possible on a solely experimental platform. A study of the flux lines provides an understanding of the underlying physics of the problem. Figure 3 illustrates how velocity affects the flux lines in the vicinity of the magnetizer (unless otherwise indicated the defects studied are 10 X 10-2 meters long (axially) and 50% through wall). The effect is to distort the flux lines by dragging them behind as the tool moves forward. (3) 501
200 :ibrusil: :Brush: ' I. ~ I '~r ))~ '00 ~ c' ~ ;" ~ " 1 " I '"' 'Or "f " S : " -. - I!' j --9l1penment 'J.1. " so ----ZOF! " -----30 FE " ~ " \. -100' 0 0.01 0.02 0.03 O.Ooi 0.'" 0.06 0.07 Distance along the Magnetizer (meters) (a) 0.08 '!O[ '00 lf~>'-----0~':'\ ~.. -.. \;~ I~ 1 : ':r ) '\ r I' ; / / : --.IID11n~nt III '-r : -50r 'J / : ----20 FE I I -'-'-3DF!: \... 1. tco. a 0.01 0.02 O.a3 0.0.& J.05 oj.os 0.07 Distance along the MagnetIZer (meters) (b) J.e! ::[ - :f'\ A 30J 1.<\ /)i1 2OOr: I. '"~_". :"~; j ':t'.- -- :: -"""nm" b ":.:\:-~' 'JI' -100 I"' ----20F2: I ~'. " - '200[~'): '-'-'-30 FE : i; ~ f.joo j \. -~! ~~=-~~--~~~~ " 0.01 om a.oj 0.04 0.0.5 o.ca 0.07 DIstance along the Magnetizer (meters) (C) Figure 2 Comparison of the axisymmetric code with 3D and experimental results. (a) Magnetizer in pipe; wall thickness 0.635 X 10-2 meters (b) Magnetizer in pipe; wall thickness 0.864 X 10-2 meters (c) Magnetizer in air. 502
1. V = 1.25 meters/sec 2. V = 2.5 meters/sec 3. V = 3.75 meters/sec Figure 3 Effect of the magnetizer velocity on the flux lines.... 420 "'" g~~1... r _l'j~ ~ --Om...HC a:::rx:axaa 1.25/NC..-s1SK _2.50mefetSlMC _ 3.75 mefetwmc J... 1 J 0.2 0.4 0.6 0.8 '.2 4 1.6 Sensor Position (meters) " Figure 4 Effect of the magnetizer velocity on the leakage field signals. The effect of velocity on the axial component (Bz) of the leakage field signals is shown in Figure 4. Two distinct effects are observable one is the gross reduction of the magnetization level and the second is the distortion of the signal in the defect region. The reduction of the magnetization level is explained to be a result of motion induced distortion of the magnetizer field as shown in the flux plots of Figure 3. The analysis of the distortion in the defect region however requires a deeper study. Figure 5 shows a plot of the induced currents in the vicinity of the defect These currents are tenned "defect induced currents". The presence of the defect results in a radial component of the field at the edges of the defect The radial component of the field at the defect edges changes in magnitude as the magnetizer moves across the defect. This changing radial component of the field which is perpendicular to the direction of motion results in "defect induced currents". These currents are primarily the reason for the leakage field signal being distorted. Defect induced currents are in opposite directions at each edge depending on the direction of the magnetizer field. They have the effect of adding and subtracting from the leakage field signal and thus skewing it. 503
i I H r. Pipe wall (not to scale) Figure 5 Defect induced currents along the edges of a 4 X 10-2 meters long(axially) 50% through wall defect. TOOL DESIGN This paper investigates the possibility of optimizing the tool design for minimizing velocity effects. Traditional MFL defect characterization schemes examine the amplitude of the leakage field signals for depth infonnation. Since the amplitude of the signal varies with velocity these methods will loose accuracy without velocity compensation schemes. One of the schemes studied is the optimization of the sensor position. It is found that the shape of the leakage field signal is rendered invariant to velocity by moving the sensor from the center toward the leading pole piece. Also it is observed that the amplitude of the signal becomes invariant to velocity as the sensor position is moved closer to the trailing pole piece (Figure 6).... --0- uo l.1l r.:~ Sensor Position (mel.ers) (a).. '.' -..............! l ~ Sensor Position (me ten) (b) u '.'.. Figure 6 The influence of sensor position on velocity effects (a) Leakage fields corresponding to sensor #1 in Figure 1 (b) Leakage fields corresponding to sensor #2 in Figure 1. 504
CONCLUSIONS AND FUTURE WORK The results presented in this paper demonstrate the use of numerical modeling in predicting leakage field signals in analyzing the results and in tool design for pipeline inspection using MFL techniques. The velocity effects resulting from "defect induced currents" could provide valuable information about the defect Accurate isolation of the defect induced velocity effects is a difficult problem and will be investigated in detail in the immediate future by the authors. Results presented illustrate that it is possible to minimize velocity effects on the leakage field signals by intelligent tool design. A more thorough understanding of the effects of magnetizer velocity can be obtained by using a 3D model. Future work will include the development of a 3D code incorporating velocity effects. ACKNOWLEDGEMENTS This work is supported by the Gas Research Institute (Contract Number 509-271- 2563). The experimental results presented in this paper were obtained by the authors in cooperation with Battelle Columbus Ohio. The 3D simulations presented were carried out by Jiatun Si Iowa State University. REFERENCES 1. R.W.E. Shannon and L. Jackson "Flux leakage testing applied to operational pipelines" Materials Evaluation 46 pp. 1516-1524 1988. 2. Y. K. Shin and W. Lord "Numerical modeling moving probe effects for electromagnetic NDE" IEEE Trans. Mag. pp. 1865-1868 Mar. 1993. 3. Y. S. Sun W.Lord G. Katragadda and Y.K. Shin "Motion induced remote field eddy current effect in a magnetostatic non-destructive testing tool: a finite element prediction" IEEE Trans. Mag. pp. 3304-3307 Sep. 1994. 4. S.R.H. Hoole "Rotor motion in dynamic fmite element analysis of rotating electrical machinery" IEEE Trans. Mag. Vol. 21 pp. 2292-2295 1985. 5. H. M. Leismann and E. O. Frind "A symmetric - matrix time integration scheme for the efficient solution of advection - dispersion problems" Water Resources Research Vol. 25 pp. 1133-1139 1989. 6. J. Donea "A Taylor-Galerkin method for convective transport problems" Int J. Num. Meth. Engng. Vol. 20 pp. 101-119 1984. 505