JP4235648B2 - Eddy current flaw detection signal processing method - Google Patents

Description

  The present invention relates to an eddy current flaw detection signal processing method for nondestructively measuring natural defects inside a metal structure on the surface of an inspection object.

  The eddy current flaw detector is used for nondestructively measuring a natural defect inside a metal structure on the surface of an inspection object from the outside. The apparatus places a transmission coil for passing a high-frequency current on the surface of a metal piece of an inspection object, and generates an eddy current on the surface of the metal piece by a magnetic field generated by the transmission coil. The response of the magnetic field changed by the eddy current is measured by the same coil (sending / receiving) or another receiving coil of the apparatus. If the metal piece is defective, the generated eddy current changes, and the response of the receiving coil changes (amplitude change and phase change). This response is generally expressed in XY space and is defined as X amplitude and Y amplitude. In addition, in this specification, a natural defect means the weld crack, stress corrosion crack, fatigue crack, etc. which generate | occur | produce naturally at the time of manufacture of a structure, and a service state.

In actual nondestructive inspection, this transmitter / receiver coil (hereinafter referred to as “eddy current flaw detection probe”) is usually a two-dimensional scan with a structure made of a metal material (in the direction parallel to the defect or perpendicular to the defect based on the length of the defect). And is measured and displayed as vector data in a two-dimensional space. Hereinafter, the display thus obtained is referred to as a C scope.
Further, changes in the X amplitude and the Y amplitude when the probe is scanned in a certain direction are plotted on the XY plane, and the waveform is obtained by connecting the plots. Hereinafter, this waveform is referred to as a Lissajous waveform.

In the conventional technology, this C scope is represented by a gray scale image by representing either the X amplitude, the Y amplitude, or the scalar quantity of the total amplitude (sqrt (X * X + Y * Y)), and the shape is defective. The presence or absence of is determined.
However, in reality, it is difficult to determine the presence / absence of a defect from only a single scalar quantity due to a change in the metal material or the presence of a weld. For this reason, it is the actual situation that it is determined whether or not it is a defect from the pattern of the Lissajous waveform. This defect pattern has a Lissajous waveform having a similar pattern with reference to a Lissajous waveform obtained from a calibration test piece having an artificial defect (EDM slit) having a known length and depth prepared in advance. The part is determined as a defect. As an example, the literature “A. Dogandzic & P. Xiang,“ A Statistical Model for Eddy-Current Signals from Generator Tubes, ”Review of Quantitative 4”, Review of Quantitative. The signal obtained from the above is expressed by a statistical model, and the presence or absence of a defect is determined based on the statistical model. In the present specification, an artificial defect is a simulated crack that is artificially created in a metal piece using a processing device or the like and has a known length, depth, and width.
A. Doganzic and P.A. Co-authored by King (A. Dogandzic & P. Xiang), “Review of Quantitative Nondestructive Evaluation; Paper title: Statistical model of eddy current signal from steam generator (A Statistical Modal Modal Current Signals from Generator Tubes) ", Vol. 23, 2004, p. 605-612.

  However, since the statistical distribution of signals obtained from defects varies depending on the defect shape and probe characteristics, it is difficult to use in a universal form. Therefore, it becomes a problem to provide a more objective eddy current flaw detection signal processing method.

Another problem is that since the eddy current flaw detection probe has a certain range of sizes, the defect is measured in a spatially spread form depending on the size of the probe. This means that the defect width and length are measured in accordance with the point response function (Point Spread Function (PSF)) or the line response function (Line Spread Function (LSP)) of the probe.
As a method for correcting such inaccuracies and restoring and measuring the original defect shape, there is a method called deconvolution. Convolution (convolution) is obtained by multiplying a defect shape by a spatial response function, whereas deconvolution processing requires inverse matrix operation corresponding to differentiation processing. Therefore, deconvolution is generally considered difficult to calculate.

  As an example, the document “P. Xiang, et.al.,“ Automated Analysis of Rotating Probe Multi-Frequency Eddy Current Current Generator Generator Tubes. & L.E. p. 151-164 (2001) ”introduces a method called“ Blind Devolution ”and proposes a method for simultaneously estimating both the spatial response function and the original defect shape from the measured two-dimensional space vector data. However, the general view is that it is difficult from a practical point of view to estimate both the spatial response function and the original defect shape from data contaminated with actual observation noise at the same time. Also, the obtained spatial response function has a larger error than the true value, and is not practical.

  Furthermore, although the spatial response function is defined in the form of PSF and LSF, the problem with eddy current flaws is that the addition of PSF does not become LSF. Actually, even if the measured values of the spatial response functions obtained from the point-like defects are added, the measured values of the spatial response function obtained from the linear defects are not obtained. The reason is that the generation of eddy current greatly depends on the defect shape, and the response characteristics are different between the central portion and the end portion of the linear defect. For this reason, it is necessary to define a spatial response function in consideration of the shape of the defect. Here, in this specification, the spatial response function is a general concept including a line response function and a point response function.

Accordingly, an object of the present invention is to provide an eddy current flaw detection signal processing method capable of automatically identifying a defective portion and accurately restoring a defect shape.
[Means for Solving the Problems and Effects of the Invention]

According to the eddy current flaw detection signal processing method of the present invention, based on the data of the defective portion and the non-defective portion taught from the vector data in the multidimensional space composed of the eddy current flaw detection signal for each flaw detection position of the inspection target, Calculating a discrimination boundary for identifying a part and a non-defect portion, projecting the vector data onto a normal vector orthogonal to the discrimination boundary, and converting the vector data into an amplitude value for each flaw detection position. And a step of setting a region above the threshold as a defective portion.
Therefore, the defect determination conventionally performed by the qualitative determination of the Lissajous waveform becomes objective and quantitative, and there is an advantage that the defect can be automatically identified.
Furthermore, since an S / N ratio as a measure of detection performance can also be defined, it is possible to mutually compare detection performance of different eddy current flaw detection probes.

Preferably, the discrimination boundary is a boundary determined by a linear discriminant function. Therefore, there is an advantage that the defective portion and the non-defective portion can be objectively classified and identified based on the statistical criteria.
Preferably, the method further includes a step of performing threshold filtering that leaves only the signal of the defective portion by setting the amplitude value equal to or less than a predetermined threshold to zero. Therefore, in addition to improving the visibility by setting the portions other than the defective portion to zero, there are advantages of preventing unnecessary noise from being mixed in the deconvolution calculation described later.
Preferably, from the vector data obtained by flaw detection of a predetermined calibration specimen, a step of obtaining a spatial response function of the eddy current flaw detection probe in a form depending on the longitudinal direction of the artificial defect and two directions orthogonal thereto. The method further includes deconvolution of the vector data of the inspection object using the spatial response function to obtain shape restoration data of a natural defect of the inspection object in which the spatial response characteristic is corrected.
Therefore, the original defect shape can be accurately restored, and there is an advantage that the accuracy of the flaw detection can be improved as compared with the conventional flaw detection signal processing method.

Preferably, the spatial response function is obtained in the form of an equivalent PSF. Therefore, by applying this equivalent PSF to the defect signal extracted using the linear discriminant function, it is possible to restore the defect shape stably and accurately, excluding the influence of noise included in the measurement data, There is an advantage that it is possible to accurately evaluate the length of a complex defect and restore a branch shape.
Preferably, the scanning pitch of the probe with respect to the inspection object is set to a predetermined value using the spatial response function so that the defect detection probability of the inspection object is not less than a predetermined value. Therefore, in addition to contributing to shortening of the inspection time, there is an advantage that it can be used for standardization of inspection procedures, such as setting flaw detection conditions with the same detection probability for flaw detection probes having different characteristics.
Further preferably, each of the vector data obtained by flaw detection on the inspection object and the calibration test piece is an X amplitude and a Y amplitude for each flaw detection position and a flaw detection signal output value in the vicinity of the flaw detection position. It consists of X amplitude and Y amplitude. Therefore, by using not only the X amplitude and Y amplitude of the flaw detection position but also the X amplitude and Y amplitude in the vicinity thereof, information equivalent to the pattern of the Lissajous waveform can be obtained. There is an advantage that a closer determination can be automatically made.
According to another eddy current flaw detection signal processing method of the present invention, a step of providing a calibration test piece to which an artificial defect having a known length and depth is provided, and an eddy for each flaw detection position of the calibration test piece. Calculating a discrimination boundary for identifying a defective portion and a non-defective portion based on data of a defective portion and a non-defective portion taught from vector data in a multidimensional space composed of current flaw detection signals; and flaw detection of the inspection object Obtaining vector data in a multidimensional space composed of eddy current flaw detection signals for each position; projecting the vector data of the inspection object onto a normal vector orthogonal to the discrimination boundary to obtain an amplitude value for each flaw detection position; Converting, and setting a region where the amplitude value is a predetermined threshold value or more as a defective portion.
The advantages of this method are substantially the same as those described in paragraph 0010.

A first embodiment of the eddy current flaw detection signal processing method according to the present invention will be described with reference to FIGS.
In FIG. 1, a flaw detection apparatus 1 is an apparatus for nondestructively measuring a natural defect 4a (for example, a linear shape) inside a tissue of a metal piece 4 as an inspection object from the outside. The eddy current flaw detection probe 2 and the information processing device 3 are each connected to the same device 1.
The flaw detection probe 2 is scanned 4b on the surface of the metal piece 4 in parallel with the longitudinal direction of the defect 4a, and an eddy current flaw detection signal output value for each flaw detection position is measured. Here, of the eddy current flaw detection signal output value, the amount corresponding to the inductance and resistance (resistance) of the coil constituting the probe, or the amount corresponding to the phase rotated thereof is referred to as X amplitude and Y amplitude, respectively.

Data of X amplitude or Y amplitude is processed by the information processing device 3, and the C scope 3b is displayed on the display 3a of the same device 3 by displaying in gray on a two-dimensional space corresponding to the position of the flaw detection probe 2. In the C scope 3b, the plus region is represented by a warm color (the warm color gradually darkens as the value increases), and the minus region is represented by a cool color (the cold color gradually darkens as the value decreases). The same). Instead of X amplitude or Y amplitude, all amplitude values defined by the square root of the sum of squares of X amplitude and Y amplitude may be displayed in gray.
On the same display 3a, a locus of X amplitude and Y amplitude when the probe 2 is scanned in a horizontal direction and a vertical direction with respect to the longitudinal direction of the defect 4a at a specific position is displayed as a Lissajous waveform 3c (FIG. 2). (Expanded view of C scope 3b and Lissajous waveform 3c displayed on display 3a).

FIG. 3 shows the procedure of the first embodiment of the eddy current flaw detection signal processing method according to the present invention.
After obtaining the C scope and the Lissajous waveform as described above, the two-dimensional space vector data is combined with data (X amplitude and Y amplitude) composed of output values in the vicinity of the target (flaw detection) position. Convert to data. Since the X amplitude and the Y amplitude are the same as the signals in the complex space, they are expressed as Yre (x, y) and Yim (x, y) by considering them as real and imaginary signals. Lowercase (x, y) is the spatial coordinate position of the inspection object.

Subsequently, a defective part and a non-defective part are taught from the multidimensional space vector. In general, a signal obtained from a test piece (calibration test piece) containing an artificial defect prepared separately from an inspection object is taught as a defective portion. When the defect to be found is clearly known by human vision, the defect signal of the inspection object may be taught as the defect portion.
Next, a linear discriminant function for calculating a discrimination boundary is calculated. By taking the inner product of the normal vector orthogonal to the linear discriminant function and the multidimensional space vector, the multidimensional space vector is projected onto the normal vector and converted into an amplitude value for each flaw detection position. Since this amplitude value is a scalar quantity, a C scope image can be generated. By using the C scope image obtained in this way, a defect area can be identified by setting an area having a predetermined threshold value or more as a defect area.
That is, the conventional typical flaw detection method determines whether or not the defect is a defect based on an inspector's empirical rule for each flaw detection probe from the C scope image and the Lissajous waveform, whereas the eddy current according to the present invention. The flaw detection signal processing method is different in that a defect is automatically identified using a linear discriminant function.

  The conventional defect determination based on the experience of the inspector depends on the characteristics of the Lissajous waveform. That is,

Formula 1

The defect is recognized by the shape of the pattern vector. When n> 0, m = 0, the defect is determined by the x-direction Lissajous waveform, and when n = 0, m> 0, the defect is determined by the y-direction Lissajous waveform. When n> 0 and m> 0, the defect is determined by the Lissajous waveform in both the x and y directions.

The automatic defect identification method using the linear identification function will be described in more detail.
In FIG. 4, reference numeral 3d indicates a defective part vector on the C scope 3b, and reference numeral 3e indicates a non-defective part vector on the C scope 3b.
In FIG. 5, symbol 3f indicates a typical Lissajous waveform of a defective portion corresponding to the defective portion vector 3d (see FIG. 4), and symbol 3g represents a typical Lissajous corresponding to the non-defective portion vector 3e (see FIG. 4). Waveform is shown. The horizontal axis of the graph in FIG. 5 is the voltage value (V) corresponding to the X amplitude in the output of the flaw detection probe, and the vertical axis is the voltage value (V) corresponding to the Y amplitude in the output of the flaw detection probe. is there.
In this embodiment, a total of 11 Lissajous waveforms of −5 to +5 points in the y direction are defined at each flaw detection position. Therefore, Z in Equation 1 constitutes a 22-dimensional vector. Of course, the number of points taking the Lissajous waveform is not limited to 11. That is, the number of points may be plural, and is appropriately determined depending on conditions such as the length and depth of natural defects.

  In the conventional discrimination method,

Formula 2

Rough defect identification is performed with such a quantity. Therefore, accurate determination of the defect edge depends on qualitative determination of the Lissajous waveform pattern.

On the other hand, in this embodiment, defect identification is performed using a multidimensional space vector defined by Equation 1. Various methods already exist for identifying two groups (defective part and non-defective part) defined in a multidimensional vector space. The most basic method is based on the linear discriminant function by Fisher. This linear discriminant function is defined as follows.
First, the measurement signal Z (x, y) of Equation 1 is divided into the following two groups.
1) Defect signal portion ω _S : {Zs (x, y) | (x, y) εdefect portion}
2) Noise signal part ω _N : {Zn (x, y) | (x, y) ∈non-defect part}
The average value of these two groups
And dispersion
Then, the optimum linear discriminant function for discriminating between these two groups is given by Fisher as follows.

Formula 3

This means a projection in the direction connecting the centers of the classes, and a plane that passes through the midpoint between the centers and is orthogonal to the natural axis P becomes an identification boundary. If g (Z) is used, the characteristic of the Lissajous waveform is defined as a scalar quantity at the flaw detection position of the inspection object.

When Equation 3 (Fisher's linear discriminant function) is applied to the eddy current flaw detection signal processing method according to the present invention, the following is obtained.
First, vector data Z in a multidimensional space consisting of eddy current flaw detection signals for each flaw detection position of the inspection object is obtained.
Next, the following formula
g (Z) = 0
To get the discriminant boundary. That is, the dimension of the measurement signal Z, which is a multidimensional space vector, is reduced by one dimension using the function g of Equation 3 to obtain a discrimination boundary. For example, when the measurement signal Z is a plane, the discrimination boundary can be obtained as a line, and when the measurement signal Z is a three-dimensional space, the discrimination boundary can be obtained as a plane.
Next, a function value t obtained by substituting the measured value Z _m (x, y) at the (x, y) position into Z of the function g is obtained by the following equation.
t = g (Z _m (x, y))
This is equivalent to projecting the measurement signal Z _m (x, y), which is a multidimensional space vector, onto the normal vector orthogonal to the discrimination boundary to obtain the numerical value t. Here, the numerical value t means an amplitude value for each flaw detection position. By obtaining the maximum value of t = g (Z _m (x, y)) for all scanning positions (x, y), the maximum amplitude value t _max of the defect signal portion and its maximum position (x _max , y _max ), that is, the position of the maximum defect depth. In addition, a predetermined threshold value (for example, a threshold value corresponding to −12 dB) is obtained on the basis of the maximum amplitude value _tmax , and a portion having a value larger than that can be automatically identified as a defective portion.
Of course, the magnitude of this signal also depends on the definition of the midpoint,

Formula 4

Then, the maximum signal strength is defined with the center of the noise part being zero, and the result such as the −12 dB evaluation length is different.

  According to the present invention, not only the defect part is automatically identified, but also a conventional concept such as an S / N ratio that defines defect detection performance can be defined for a Lissajous waveform that is multidimensional space vector data. It can also be used for mutual evaluation of measurement performance of flaw detection probes.

Here, the defect signal portion is taught using the measurement data itself of the actual inspection object, but the calibration test piece is different from the inspection object provided with the artificial defect having a known length and depth. It is also possible to use data measured under the same conditions.
That is, from the multidimensional space vector data consisting of eddy current flaw detection signals for each flaw detection position obtained by flaw detection of the calibration test piece with the eddy current flaw detection probe that has detected the inspection object, the defect signal portion in equation (3) is obtained. M _S and Σ _{S of} are obtained. A linear discriminant function is obtained using m _S and Σ _S of the defect signal portion. Then, it becomes possible to apply the above method to an unknown inspection object.
Such a method of using a test specimen for calibration has an advantage that a defect can be clearly and automatically identified even under a condition in which a clear defect signal is not seen in an actual inspection object. In addition, since the test piece for calibration is used as standard in eddy current flaw detection, there is no restriction on implementation.

Subsequently, an example of a result of using the eddy current flaw detection signal processing method according to the present invention will be described with reference to FIG.
In the figure, the X-amplitude C scope 5a and the Y-amplitude C scope 5b are the results of eddy current flaw detection on a metal piece to be inspected. The unit of the scale attached to each C scope is cm, and indicates the spatial position of scanning of the probe.
The metal piece has a welded portion and a rectangular artificial defect having a length of 10 mm and a depth of 2 mm parallel to the weld line in the vicinity of the welded portion. A horizontally long portion 5c (in the C scope 5a, which is drawn in a particularly dark cold color) seen at the center of the X amplitude C scope 5a corresponds to an artificial defect.
However, the defect cannot be clearly identified due to the influence of the horizontally long portion 5d (drawn in the C scope 5a, particularly in a dark warm color) corresponding to the noise signal of the welded portion that crosses the metal piece in the horizontal direction.

From the eddy current flaw detection results, two locations of a representative portion of a defect and a representative portion of a non-defect are taught. As a representative portion of the defect, the horizontally long portion 5c that is an artificial defect is designated (in the X scope C scope 5a, the horizontally long portion 5c is surrounded by a surrounding portion 5e). As a non-defective representative portion, a portion 5f surrounded by a square is designated in the C scope 5a of X amplitude.
7 is an enlarged view of the surrounding portion 5e, and FIG. 8 is an enlarged view of the surrounding portion 5f.

Subsequently, two discriminating groups corresponding to the C scope 5a of X amplitude and the C scope 5b of Y amplitude are used as teaching signals, and a linear discriminant function is obtained based on Expression 3. At this time, a Lissajous waveform of ± 5 points in the y direction of the flaw detection position is used as the multidimensional space vector.
FIG. 9 represents the value of this linear discriminant function as a C scope 6. In the C scope 6, the defective portion represents a plus region 6a (displayed in a warm color in the C scope 6), and the non-defective portion represents a minus region 6b (a region displayed in a cool color other than 6a in the C scope 6). To express.
Therefore, as is clear from the figure, the defective portion can be clearly identified.

FIG. 10 shows the C scope 7 after threshold filtering is performed by setting, for example, 25% of the maximum signal amplitude value, that is, −12 dB, setting the value equal to or less than −12 dB to zero, and leaving only the signal of the defective portion. .
In the C scope 7, the defective portion represents the plus region 7a (displayed in warm color in the C scope 7), and the non-defective portion represents the minus region 7b (region displayed in cold color other than 7a in the C scope 7). To express. As is apparent from FIG. 10, by using the C scope 7 after the threshold filtering, the influence of the noise signal 5d of the welded portion is eliminated, and the defective portion is automatically identified more clearly. Is possible.
When performing the flaw detection signal processing, the processing may be performed based on only the two-dimensional space vector data without converting the two-dimensional space vector data into the multi-dimensional space vector data as described above.

Next, a second embodiment of the flaw detection signal processing method according to the present invention for more accurately obtaining the defect length from the defect signal obtained in the first embodiment will be described.
Since the eddy current measurement signal has a finite probe size, the eddy current measurement signal has spatially spread response characteristics. Therefore, as in the following Expression 5, the defect signal W (x, y) is observed in a form smoothed by a point response function (Point Spread Function, or PSF (x, y)). Equation 5 is in the form of a convolution integral and is usually called convolution. Note that x and y in W (x, y) and PSF (x, y) indicate the spatial position of the probe.

Formula 5

  In order to accurately evaluate the shape and length of the defect signal itself, it is necessary to estimate the defect signal W (x, y) from H (x, y) which is the value of the observed eddy current measurement signal. . When PSF (x, y) is known, the method for obtaining the original defect signal W (x, y) from the observed value H (x, y) based on Equation 5 is called deconvolution. That is, conceptually, from Equation 5,

Equation 6

It is a method of restoring by.

  However, the inverse matrix of PSF generally has a bad structure and the inverse matrix diverges, and it is often impossible to obtain Equation 6 with high accuracy. For this reason, in the field of image analysis, the Richardson-Lucy algorithm (WH H. Richardson, “aysian-basic Iterative Method of Image Restoration”) that performs deconvolution using the maximum likelihood estimation method using Bayes's posterior probability. J. Opt. Soc. Am, Vol. 62, No. 1, pp. 55-59 (1972)) is often used. This is a technique for restoring the original signal using an algorithm like the following Expression 7. Equation 7 is an algorithm when PSF is known.

Equation 7

  On the other hand, when the PSF is unknown, there is an algorithm called Blind Signal Devolution that also estimates the PSF at the same time. For example, the document “P. Xiang, et.al.,“ Automated Analysis of Rotating Probe Multi-Frequency eddy Current Data From Steam Generator Tubes / In .. .151-164 (2001) ”, the Blind Signal Devolution is applied to eddy current flaw detection to simultaneously obtain the PSF (x, y) and the original defect signal W (x, y).

  However, there is a precision limit to obtain both the defect signal W (x, y) and the point response function PSF (x, y) from the measurement value H (x, y) of a single eddy current measurement signal. is there. Further, the shape of PSF (x, y) is different from general image analysis, and in the case of eddy current flaw detection, there is a defect shape dependency such as depending on the longitudinal direction of the defect. In other words, the defect shape dependency is obtained by performing line integration on the PSF (x, y), which is a response function to a point defect, to a line spread function (hereinafter referred to as “LSF” as appropriate). ) Is not. Therefore, it is necessary to consider deconvolution.

Most of the natural defects are linear. In this embodiment, an equivalent PSF (Equivalent Point Spread Function, or EPSF) that equivalently expresses LSF (x, y) is obtained.
In the signal processing method according to the present invention, the relationship between this equivalent PSF (x, y) and LSF (x, y) is defined by the following equation.

Equation 8

The feature of the equivalent PSF is that PSF having anisotropy with different spreading widths (R _x , R _y ) in each direction when the direction parallel to the defect is taken as the x-axis and the direction perpendicular to the defect is taken as the y-axis. LSF is defined as the integral of. This spread width can be determined by the least square method so that the measured data of the linear defect having a known length and depth matches the result of Equation 8.

FIG. 11 and FIG. 12 are LSFs when a linear artificial defect having a length of 10 mm and a depth of 2 mm is detected.
In FIG. 11, reference numeral 8a indicates a C scope diagram of LSF (x, y) of actual measurement values (total amplitude). Reference numeral 8b shows a C scope diagram obtained from Expression 8 after fitting (optimizing) the spread width (R _x , R _y ) of the LSF (x, y) of the actual measurement value. The units of the scales attached to the C scopes 8a and 8b are cm and correspond to the spatial position of the probe.
By the above fitting, the spreading width is x with the defect parallel direction as x.
It is obtained.
In FIG. 12, reference numeral 8 c indicates the LSF (x, y) defect parallel direction distribution, and reference numeral 8 d indicates the LSF (x, y) defect orthogonal direction distribution (in any distribution, the solid line graph indicates the actual measurement value). The broken line graph shows the fitting value). In both distributions 8c and 8d, the profiles of the measured values and fitting values in the defect parallel and orthogonal directions are in good agreement, and it can be seen that a reasonable LSF is obtained.

Reference numeral 9 in FIG. 13 shows a C scope diagram of an equivalent PSF corresponding to the above-mentioned fitted LSF. It can be seen that the spread in the horizontal direction in the figure, that is, the spread width in the defect parallel direction, is smaller than the spread in the vertical direction in the figure, that is, the defect orthogonal direction. Here, the longitudinal direction of the defect corresponds to the horizontal direction in FIG.
In the conventional method, the PSF is obtained by isotropic approximation of the fitted LSF. On the other hand, in the method according to the present invention, anisotropy can be introduced, and there is an advantage that a more accurate defect shape can be restored.

FIGS. 14 and 15 show the result of restoring the original defect shape from the measured C scope by the method of Equation 7 using the equivalent PSF (see FIG. 13).
In FIG. 14, reference numeral 10 a represents an actual measurement C scope, and reference numeral 10 b represents a defect C scope after deconvolution (after restoration).
The code | symbol 11 in FIG. 15 shows LSF (x, y) defect parallel direction distribution. In the figure, a solid line graph 11a indicates an actual measurement value, and a broken line graph 11b indicates a fitting value. In this distribution, the horizontal axis corresponds to the probe scanning position (unit: mm), that is, the defect length, and the vertical axis represents the voltage value (V) of the full amplitude.
Referring to the C scope 10a before restoration in FIG. 14 and the C scope 10b after restoration, the graph 11a of the actual measurement value (before restoration) and the graph 11b of the fitting value (after restoration) in FIG. It can be seen that the defect shape is sharper and closer to the original defect shape (linear defect having a length of 10 mm and a depth of 2 mm).
In FIG. 15, the fitting value graph 11 b extends from 9 to 19 on the horizontal axis, and the length after restoration is about 10 mm. On the other hand, the measured value graph 11a covers 8-20 on the horizontal axis, and the length before restoration (actually measured value) is about 12 mm. That is, the measured value (before restoration) graph 11a overestimates the length of the defect. Therefore, this method is reconstructed and evaluated with very high accuracy.

  Since this deconvolution operation is the same processing as the differentiation operation, if the observation data contains noise, an accurate result is often not obtained. However, this method has an advantage that stable and accurate restoration is possible by combining with a linear discriminant function.

Next, as a third embodiment of the flaw detection signal processing method according to the present invention, a method for determining the flaw detection pitch of the flaw detection probe based on the false detection probability will be described below.
When detecting a minute defect, the detection probability of missing a defect can be reduced as the spatial scanning pitch is reduced.
However, an excessively small scanning pitch is not efficient because it takes a long inspection time. Furthermore, in consideration of the spatial response characteristic depending on the size of the probe, an excessively small scanning pitch may not contribute to an improvement in detection probability. As a conventional technique, a method for determining a scanning pitch based on an actually measured LSF has been proposed (T. Ray, et.al, “Index size determination with varying coiled incidents intensified detection of impacted injections”). Review of Quantitative Nondestructive Evaluation, Vol. 24, p.767-774 (2005)).

In the present embodiment, the scanning pitch is appropriately set based on the LSF (x, y) obtained by Expression 8 of the second embodiment described above.
In FIG. 16, reference numeral 12 indicates a C scope diagram of LSF (x, y) obtained in the second embodiment, and reference numeral 13 is a graph for showing a distribution (profile) 13a in the Y-direction cross section. Yes, reference numeral 14 is a graph showing the distribution (profile) 14a of the X-direction cross section.
In the graph 13, when dy is set as the scanning pitch in the defect orthogonal direction (Y direction), a point of dy / 2 centered on the maximum position y0 in the Y-direction cross-sectional profile at the X coordinate center position x0 of the LSF. That is,

Equation 9

Is the smallest observable amplitude value.

  Since the initial value of the Y scanning pitch is considered to be determined randomly with respect to the defect position, in the best case, the maximum value of LSF (x, y), that is,

Equation 10

However, in the worst case, an amplitude value shifted from the maximum peak position by ½ of the scanning pitch, that is, dy / 2, is measured.
From the ratio between the two, the defect peak detection probability when the scanning pitch dy is determined is

Equation 11

Can be defined. When the scanning pitch is sufficiently small, α = αmax, so that the peak detection probability is 100%.
Conversely, when the defect peak detection probability Pd is determined to be a predetermined value,

Formula 12

The scanning pitch dy that satisfies the above is determined. The same applies to the determination of the defect parallel scanning pitch (X direction scanning pitch).

In this way, by obtaining the LSF based on the linear defect, it is possible to determine the scanning pitch at which the defect detection probability is a certain value or more. Since most normal natural defects are linear, the LSF according to the present invention is applicable to a wide range. This method of determining the scanning pitch based on the LSF has the advantage that the scanning pitch can be increased after ensuring the defect detection probability to a predetermined value or more, which can contribute to the improvement of inspection efficiency.
In addition, it is possible to standardize the flaw detection procedure by confirming that the scan pitch is not unnecessarily small and by setting a scan pitch with the same detection probability even when flaw detection using probes of different sizes is performed. There are benefits that can be useful.
Furthermore, since LSF (x, y) in Expression 8 is expressed by an analytic function called integral of equivalent PSF, LSF (x, y) depends on the length of the linear defect that can be defined as a required specification at the time of inspection. Can be obtained by calculation. Therefore, there is an advantage that flexibility in actual use can be increased. Furthermore, since different spread widths can be taken into account according to the longitudinal direction of the linear defect, there is an advantage that the scanning pitch can be evaluated with higher accuracy.

1 is a schematic diagram of an eddy current flaw detector for explaining a first embodiment of an eddy current flaw detection signal processing method according to the present invention. FIG. It is a figure which shows C scope and a Lissajous waveform as a result of having detected the test target object with the eddy current flaw detector in FIG. It is a step figure for explaining the procedure of the 1st example of the flaw detection signal processing method concerning the present invention. FIG. 3 is a diagram for showing a defective part vector and a non-defective part vector in the C scope in FIG. 2. FIG. 5 is a diagram for illustrating Lissajous waveforms respectively corresponding to a defective portion vector and a non-defective portion vector in FIG. 4. FIG. 4 is a diagram for showing a C scope of X amplitude and Y amplitude as an example of a result of using the first embodiment shown in the procedure in FIG. 3. It is an enlarged view of the defective part of the X amplitude C scope in FIG. It is an enlarged view of the non-defect part of the X amplitude C scope in FIG. It is the figure which expressed the value of the linear discriminant function obtained by the 1st Example of the flaw detection signal processing method concerning this invention on C scope. It is the figure which displayed on the C scope the value after filtering the value of the linear discriminant function obtained by the 1st Example of the flaw detection signal processing method which concerns on this invention. It is the figure which displayed LSF at the time of flaw detection of the artificial defect by 2nd Example of the flaw detection signal processing method concerning this invention. It is the figure which displayed LSF at the time of flaw detection of the artificial defect by 2nd Example of the flaw detection signal processing method concerning this invention. FIG. 12 is a diagram showing an equivalent PSF corresponding to the fitted LSF in FIG. 11. It is the figure which displayed C scope for showing the result of having decompress | restored the original defect shape from the measurement C scope using the equivalent PSF in FIG. It is the figure which displayed the defect parallel direction distribution for showing the result of having decompress | restored the original defect shape from the measurement C scope using the equivalent PSF in FIG. It is the figure which displayed C scope and cross-sectional distribution (X direction and Y direction) for demonstrating the 3rd Example of the flaw detection signal processing method which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Eddy current flaw detection apparatus 2 Eddy current flaw detection probe 3b C scope 5a X amplitude C scope 5c Horizontal portion 6 corresponding to artificial defect 6 C scope 6a Plus region 8a corresponding to defect portion C scope of LSF (x, y) of measured value

Claims (7)

Natural defects inside the metallographic inspection object surface a eddy current signal processing method for measuring externally nondestructively,
A two-dimensional space vector composed of the X amplitude and Y amplitude of the eddy current flaw detection signal measured at the flaw detection position of the inspection object and the X amplitude and Y amplitude of the eddy current flaw detection signal measured at a position near the flaw detection position. Calculating a discrimination boundary for identifying a defective portion and a non-defective portion using a linear discriminant function based on data of a defective portion and a non-defective portion taught from vector data of a multidimensional space obtained together ; ,
An eddy current comprising: projecting the vector data onto a normal vector orthogonal to the discrimination boundary, converting the vector data into an amplitude value for each flaw detection position, and setting a region having the amplitude value equal to or greater than a predetermined threshold as a defect portion. In the flaw detection signal processing method ,
Providing a calibration specimen with an artificial defect having a known length and depth;
The spatial response function of the probe is obtained from vector data in a multidimensional space consisting of eddy current flaw detection signals for each flaw detection position obtained by flaw detection of the calibration specimen using the eddy current flaw detection probe that flaws the inspection object. Determining in a form dependent on the longitudinal direction of the artificial defect and the direction perpendicular thereto;
The method further comprises deconvolution of the vector data of the inspection object using the spatial response function to obtain shape restoration data of the natural defect of the inspection object in which the spatial response characteristics of the probe are corrected. An eddy current flaw detection signal processing method. The eddy current flaw detection signal processing method according to claim 1, wherein the linear discriminant function is a linear discriminant function by Fisher and is given by the following equation.
here,
Z (x, y) = {Yre (x + i, y + j}, Yim (x + i, y + j))
(i = -n to n, j = -m to m)
(However, Yre (x, y) and Yim (x, y) represent the X and Y amplitudes of the vector data in the multidimensional space as signals of the real part and the imaginary part. x, y) indicates the spatial coordinate position of the inspection object.)
The measurement signal Z (x, y) is divided into the following two groups:
The average value of these two groups
And dispersion
And The eddy current flaw detection signal processing according to claim 1, further comprising a step of performing threshold filtering that leaves only a signal of a defective portion by setting the amplitude value below a predetermined threshold to zero. Method. And obtaining the spatial response function in the form of equivalent PSF, eddy current signal processing method according to any one of claims 1 to 3. The scanning pitch of the probe with respect to the inspection object is set to a predetermined value using the spatial response function so that a defect detection probability of the inspection object is equal to or higher than a predetermined value. Item 5. The eddy current flaw detection signal processing method according to any one of Items 1 to 4 . Natural defects inside the metallographic inspection object surface a eddy current signal processing method for measuring externally nondestructively,
Providing a calibration specimen with an artificial defect having a known length and depth;
A two-dimensional space vector composed of the X amplitude and Y amplitude of the eddy current flaw detection signal measured at the flaw detection position of the calibration test piece, and the X amplitude and Y amplitude of the eddy current flaw detection signal measured at a position near the flaw detection position. A step of calculating a discrimination boundary for identifying a defective portion and a non-defective portion using a linear discriminant function based on data of a defective portion and a non-defective portion taught from vector data in a multidimensional space obtained by combining When,
Obtaining vector data in a multidimensional space consisting of eddy current flaw detection signals for each flaw detection position of the inspection object;
Projecting the vector data of the inspection object onto a normal vector perpendicular to the discrimination boundary, converting the vector data into an amplitude value for each flaw detection position, and setting a region having the amplitude value equal to or greater than a predetermined threshold as a defective portion; An eddy current flaw detection signal processing method comprising :
Providing a calibration specimen with an artificial defect having a known length and depth;
The spatial response function of the probe is calculated from vector data in a multidimensional space consisting of eddy current flaw detection signals for each flaw detection position obtained by flaw detection of the calibration specimen using the eddy current flaw detection probe that flaws the inspection object. Determining in a form dependent on the longitudinal direction of the artificial defect and the direction perpendicular thereto;
The method further comprises deconvolution of the vector data of the inspection object using the spatial response function to obtain shape restoration data of the natural defect of the inspection object in which the spatial response characteristics of the probe are corrected. An eddy current flaw detection signal processing method. The eddy current flaw detection signal processing method according to claim 6, wherein the linear discriminant function is a linear discriminant function by Fisher and is given by the following equation.
here,
Z (x, y) = {Yre (x + i, y + j}, Yim (x + i, y + j))
(i = -n to n, j = -m to m)
(However, Yre (x, y) and Yim (x, y) represent the X and Y amplitudes of the vector data in the multidimensional space as signals of the real part and the imaginary part. x, y) indicates the spatial coordinate position of the inspection object.)
The measurement signal Z (x, y) is divided into the following two groups:
The average value of these two groups
And dispersion
And