A deep learning-based steel rail crack quantitative eddy current detection method, device and equipment

Abstract

The application discloses a kind of based on deep learning's steel rail crack quantitative eddy current detection method and device and equipment, comprising: acquisition crack eddy current response, establish first signal curve;Calculate crack inclination, and symmetrization conversion first signal curve, output second signal curve;Second signal curve is introduced vertical crack deep learning model, based on analysis result reconstruction oblique crack and determine oblique crack depth and width.For the deep learning model trained by vertical crack signal curve data set can be used for the quantitative detection of oblique crack, asymmetric oblique crack signal curve is symmetrization converted into symmetric curve.After obtaining the symmetrization signal curve of oblique crack, the data of symmetric signal curve is inverted using the trained vertical crack deep learning model, and the profile contour curve of oblique crack is obtained, and then the width and vertical depth of oblique crack are calculated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rail crack detection, and in particular to a rail crack quantitative eddy current detection method and device based on deep learning and equipment. BACKGROUND

[0002] Rail, as the most important component of train track, is composed of rail head, rail waist and rail bottom. Rail mainly plays the role of supporting locomotive wheel and guiding the wheel to move forward, and provides continuous and small resistance rolling surface for the wheel, and directly bears the pressure from the wheel. The rail of high-speed railway needs to meet the requirements of high stability, smooth surface, good elasticity and easy maintenance. At present, it has been found that high-speed railway accumulates many damages in the process of operation for many years. According to statistics, nearly half of the high-speed railway damages occur on the surface of the rail, and these damages will cause the fracture of the rail after developing along the transverse direction, which is a hidden danger seriously endangering personal safety.

[0003] Over the years, this kind of large load and high contact frequency causes different degrees of damage to the rail tread. The superficial defects of the tread include cracks, chipping, scratches and rust, etc. Among them, the expansion of the crack is the most serious harm to the rail. Under the action of periodic large load, the continuous expansion of the crack will cause the tread to appear peeling and chipping, and even make the rail fracture. According to the different normal angles of the crack and the tread, the rail tread crack can be divided into vertical crack and inclined crack. Different angle cracks will cause different damage areas after expansion. In the process of rail crack detection, the detection instrument will collect a large amount of detection signals, and the deep learning technology can realize damage identification and quantitative detection, so as to help the detection personnel to take corresponding maintenance measures for the rail section with different damage levels according to the specific crack depth. Deep learning is based on a large amount of feature data, and the deep learning model (currently mainly deep neural network) is trained, so that the trained model can realize accurate classification of data or quantitative calculation of related parameters.

[0004] In the prior art, the deep learning is used to quantitatively evaluate the crack, and the deep learning models of vertical crack and inclined crack need to be established respectively for the quantitative evaluation of vertical crack and inclined crack. It takes a lot of time to establish two kinds of deep learning models, and if the deep learning model of vertical crack is directly used for the quantitative calculation of the width and depth of inclined crack, the correct crack width and depth values cannot be obtained. The applicant proposes a rail crack quantitative eddy current detection method and device based on deep learning and equipment which can solve the above problems. SUMMARY

[0005] Therefore, the present application provides a steel rail crack quantitative eddy current detection method and device and equipment based on deep learning. A deep learning model trained by a vertical crack dataset is used to inverse the converted oblique crack signal curve, and the oblique crack profile curve is obtained by combining the oblique crack angle, and then the width and depth of the oblique crack are obtained.

[0006] The signal curve obtained by detecting the vertical crack by the eddy current detection method is a symmetrical curve, and the signal curve obtained by detecting the oblique crack is an asymmetrical curve. In the present application, a symmetrization conversion formula is used to convert the signal curve of the oblique crack into a symmetrical curve, so that it is applicable to the vertical crack deep learning model.

[0007] In one aspect, the present application provides a steel rail crack quantitative eddy current detection method based on deep learning. The method comprises: collecting a crack eddy current response to establish a first signal curve; calculating a crack angle and symmetrizing and converting the first signal curve to output a second signal curve; introducing the second signal curve into a vertical crack deep learning model for calculation, and reconstructing an oblique crack and determining the depth and width of the oblique crack based on the calculation result.

[0008] In another aspect, the present application provides a steel rail crack quantitative eddy current detection device based on deep learning, comprising:

[0009] The acquisition module is used to collect a crack eddy current response to establish a first signal curve.

[0010] The symmetrization conversion module is used to calculate a crack angle and symmetrize and convert the first signal curve to output a second signal curve.

[0011] The deep learning module is used to introduce the second signal curve into a vertical crack deep learning model for calculation, and reconstruct an oblique crack and determine the depth and width of the oblique crack based on the calculation result.

[0012] Finally, the present application provides a steel rail crack quantitative eddy current detection equipment based on deep learning, comprising: at least one processor; and a memory in communication connection with the at least one processor. Wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steel rail crack quantitative eddy current detection method based on deep learning as described in any one of the above.

[0013] A computer readable storage medium is provided, which stores a computer program. The computer program is executed by a processor to implement the steel rail crack quantitative eddy current detection method based on deep learning as described in any one of the above.

[0014] It can be found that, in order to enable the deep learning model trained on the vertical crack signal curve data set to be used for quantitative detection of the inclined crack, the present application converts the asymmetric inclined crack signal curve into a symmetric curve by using a symmetrization conversion formula. After obtaining the symmetrized signal curve of the inclined crack, the data of the symmetrical signal curve is inverted using the trained deep learning model, and combined with the crack inclination calculated according to the signal curve before conversion, the profile curve of the inclined crack is obtained, which intuitively reflects the extension of the crack, and then the width and vertical depth of the inclined crack are calculated. Cracks are divided into vertical cracks and inclined cracks according to types, and in the prior art, deep learning is used for quantitative evaluation of cracks, and deep learning models for vertical cracks and inclined cracks need to be established for quantitative evaluation of vertical cracks and inclined cracks. The present application only needs to establish a deep learning model for vertical cracks, which can be used for quantitative evaluation of vertical cracks and inclined cracks at the same time, improving efficiency and saving time. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0016] Figure 1 is a flowchart of an embodiment of the present application based on deep learning for quantitative eddy current detection of rail cracks;

[0017] Figure 2 is a schematic diagram of a rail inclined crack phase curve;

[0018] Figure 3 is a vertical crack and inclined crack phase curve measured in an experiment;

[0019] Figure 4 is the symmetrization conversion result of the inclined crack phase curve. DETAILED DESCRIPTION

[0020] The present application will be described in further detail below in combination with the drawings and embodiments. It is particularly pointed out that the following embodiments are only used to illustrate the present application, but do not limit the scope of the present application. Similarly, the following embodiments are only some embodiments of the present application, not all embodiments, and all other embodiments obtained by those skilled in the art without creative labor are within the scope of the present application.

[0021] The application provides a steel rail crack quantitative eddy current detection method based on deep learning, which converts an asymmetric oblique crack signal curve into a symmetric signal curve, and the signal refers to a characteristic quantity for representing an eddy current detection result, such as an amplitude, a phase, a real part, an imaginary part and the like, and in the embodiment, the phase is extracted to establish a phase curve for illustration. After the symmetric phase curve of the oblique crack is obtained, the deep learning model trained by using the vertical crack data set is used to inverse the converted oblique crack phase curve, and the oblique crack profile curve is obtained in combination with the oblique angle of the oblique crack, and then the width and depth of the oblique crack are obtained.

[0022] Please refer to Figure 1 , Figure 1 is a flowchart of an embodiment of the steel rail crack quantitative eddy current detection method based on deep learning. It should be noted that the method of the application is not limited to the flow sequence shown in Figure 1 . As shown in Figure 1 , the method comprises the following steps:

[0023] S1, collecting a crack eddy current response phase signal to establish a first signal curve;

[0024] S2, calculating a crack angle and symmetrically converting the first signal curve to output a second signal curve;

[0025] S3, introducing the second signal curve into a vertical crack deep learning model for calculation, reconstructing an oblique crack based on the calculation result and determining the width and depth of the oblique crack.

[0026] The first signal curve of S1 (collecting a crack eddy current response phase signal to establish a first signal curve) can be a vertical crack phase curve or an oblique crack phase curve.

[0027] S2 (calculating a crack angle and symmetrically converting the first signal curve to output a second signal curve) specifically comprises the following steps:

[0028] S21, calculating an asymmetry rate of the phase signal curve,

[0029] The calculation formula of the asymmetry rate is

[0030]

[0031] In the formula, R represents the asymmetry rate, x max , x1 and x2 respectively represent a peak value of the signal curve and a minimum value of the signal curve plus the abscissa at , and represents the difference between the maximum value and the minimum value of the signal curve.

[0032] S22, calculating a crack angle based on the asymmetry rate,

[0033] The crack inclination angle calculation formula is

[0034] θ c = 160.0R-2.9

[0035] In the formula, θ c is the inclination angle calculation value corresponding to the experimental phase curve, and R is the asymmetry rate of the experimental phase curve.

[0036] S23, if the crack inclination angle is not equal to 0, the phase curve is symmetrized and converted, and a second signal curve is output, otherwise the original phase curve is output,

[0037] The calculation formula of the symmetrization conversion is

[0038] f(x max -x i ,R) = (-1.90*10 -2 R+2.20*10 -3 )cos[(-0.37R+0.43)(x max -x i )]+(1.00*10 -2 R-1.90*10 -4 )sin[(-0.37R+0.43)(x max -x i )]+(-6.90*10 -3 R+5.00*10 -4 )cos[(-0.74R-0.86)(x max -x i )]+(4.80*10 -3 R-2.70*10 -4 )sin[(-0.74R+0.86)(x max -x i )]-1.60*10 -2 R+1.00

[0039] In the formula, R represents the asymmetry rate, and x max -x i represents the difference between the peak point abscissa of the signal curve and the abscissa of the i-th point.

[0040] The principle of the S2 is as follows.

[0041] Firstly, finite element analysis needs to be carried out on the rail crack eddy current detection. The inclined crack is not the same as the vertical crack, so the three-dimensional finite element method is adopted to simulate the detection of the PCB probe on the vertical crack and the inclined crack of the rail. According to the simulation results, the conversion relationship between the phase curves of the vertical crack and the inclined crack of the rail is found, and the conversion formula is constructed.

[0042] The eddy current scanning of the rail with vertical crack and different angle oblique crack was carried out by the PCB probe for simulation. The phase curve of the vertical crack is symmetrical about the peak point, while the phase curve of the oblique crack is not symmetrical about the peak point. With the increase of the crack angle, the asymmetry of the phase curve gradually increases, the peak horizontal coordinate moves to the direction of the increase of the crack angle, and the peak value increases.

[0043] Secondly, the phase curve conversion method of the vertical crack and the oblique crack of the rail is analyzed. In order to convert the phase curve of the vertical crack and the oblique crack, the asymmetry rate is introduced to measure the asymmetry degree of the phase curve of the vertical crack and the oblique crack of the rail, the crack angle is calculated, and the conversion relationship between the phase curve of the vertical crack and the oblique crack is derived.

[0044] 1. Curve asymmetry rate and angle calculation

[0045] According to the crack phase curve obtained by experiment and simulation calculation, the vertical crack and the oblique crack both have two troughs, and only the signal between the two troughs is analyzed in the following. The phase curve of the oblique crack of the rail is shown in Figure 2 , in which represents the difference between the maximum and minimum values of the phase curve. In order to quantify the asymmetry degree of the phase curve of the rail crack, the asymmetry rate R is introduced. Let x max , x1, x2 represent the peak value of the phase curve and the horizontal coordinate of the minimum value of the phase curve plus , and the calculation formula of R is

[0046]

[0047] The larger R is, the greater the asymmetry degree of the crack phase curve is.

[0048] The relationship between the asymmetry rate and the angle of the oblique crack with the angle of 20°, 40° and 60° is fitted, and the asymmetry rate of the oblique crack with the angle of 10°, 30° and 50° is verified. The phase curve of the oblique crack with the angle of 20°, 40° and 60° measured by experiment and the phase curve of the vertical crack are calculated by formula (1), and the corresponding asymmetry rate is obtained. The fitting straight line formula of the angle-asymmetry rate of the above four angles is:

[0049] θ c = 160.0R-2.9 (2)

[0050] In the formula: θ c is the fitting value of the angle corresponding to the experimental phase curve, also known as the estimated value of the angle; R is the asymmetry rate of the experimental phase curve.

[0051] The asymmetricity of the crack phase curve with actual inclination angle of 0°, 10°, 20°, 30°, 40°, 50°, 60° is calculated by formula (2) to obtain the corresponding inclination angle estimation value. Then, the average value e of the error between the inclination angle estimation value and the actual value is calculated θ , e θ The calculation formula is:

[0052]

[0053] In the formula, n eθ is the number of phase curves participating in the calculation of the average value of the crack inclination angle estimation error; θ ci is the i-th crack inclination angle estimation value; θ ri is the i-th crack inclination angle actual value. The calculation results are shown in Table 1.

[0054] Table 1 Asymmetricity of phase curves of cracks with different inclination angles and inclination angle estimation values

[0055]

[0056] 2. Derivation of symmetrization conversion formula

[0057] The phase curve of the oblique crack is an asymmetric curve, and the phase curve of the vertical crack is a symmetric curve. Therefore, the conversion from the phase curve of the oblique crack to the phase curve of the vertical crack is to convert the asymmetric curve to the symmetric curve. In order to realize the symmetrization conversion, a symmetrization conversion function f is constructed, and f satisfies the following formula:

[0058]

[0059] In the formula, y represents the ordinate of the i-th point on the phase curve of the vertical crack; represents the ordinate of the i-th point on the phase curve of the oblique crack; x max represents the abscissa of the peak value of the phase curve.

[0060] Since each phase curve of the oblique crack has a peak value, the peak value is taken as the reference during the symmetrization conversion, and the difference between the peak value and the abscissa of each scanning point is taken as the independent variable. For simplicity of expression, let d i = x max -x i . Dividing both sides of formula (4) by , we obtain:

[0061]

[0062] In the formula, d iThe difference between the abscissa of the peak point of the representative phase curve and the abscissa of the ith point. By calculating the ratio of the vertical crack phase curve to each oblique crack phase curve, a ratio curve is obtained. The calculation results show that the shape of the ratio curve is close to a sine function, so the ratio curve is subjected to Fourier fitting by using a Matlab fitting toolbox. The order of the Fourier fitting is adjusted to find a suitable fitting formula. By weighing the complexity of the function construction and the fitting accuracy, the second-order Fourier fitting is performed on each ratio curve, and the definition of the second-order Fourier fitting is as follows:

[0063] f(x) = a0 + a1 cos(ωx) + b1 sin(ωx) + a2 cos(2ωx) + b2 sin(2ωx) (6)

[0064] In the formula, a0 represents a constant term; ω represents the angular frequency of the trigonometric function; a1, b1, a1, and b1 represent the amplitude coefficients of each trigonometric function in the formula, respectively; and x represents the independent variable of the Fourier fitting definition.

[0065] Three phase curves with crack inclination angles of 20°, 40°, and 60° are selected to calculate the coefficients of the symmetrization conversion formula. The effect of the symmetrization conversion is verified by using three phase curves with crack inclination angles of 10°, 30°, and 50°. Figure 3 As shown in FIG. 3, three experimentally measured oblique crack phase signals and a vertical crack phase curve are shown. By using formula (5) to calculate the experimentally measured vertical crack phase curve and the three oblique crack phase curves, three ratio curves can be obtained. The second-order Fourier fitting is performed on the ratio curve, and the coefficients of the fitting formula are shown in Table 2.

[0066] Table 2: Coefficient values of the symmetrization conversion formula corresponding to the experimentally measured oblique crack phase curves

[0067]

[0068] The relationship between the coefficient values in Table 2 and the asymmetry rate is subjected to polynomial fitting, and the following formula is obtained:

[0069] a0(R) = -1.60 x 10 -2 R + 1.00 (7)

[0070] ω(R) = -0.37R + 0.43 (8)

[0071] a1(R) = -1.90 x 10 -2 R + 2.20 x 10 -3 (9)

[0072] b1(R) = 1.00 x 10 -2 R - 1.90 x 10 -4 (10)

[0073] a2(R) = -6.90 x 10 -2 R + 5.00 x 10 -4 (11)

[0074] b2(R) = 4.80 x 10 -2 R - 2.70 x 10 -4 (12)

[0075] Substituting formula (7)-(12) into formula (6), the following formula (13) is obtained.

[0076]

[0077] Using formula (13), three oblique crack phase curves with inclination angles of 10°, 30° and 50° are symmetrically converted, Figure 4 The results of the symmetric conversion are shown in the figure. As can be seen from the figure, the experimental oblique crack phase curves after symmetric conversion are basically consistent with the vertical crack phase curves. After the experimental oblique crack phase curves are symmetrically converted, the symmetric curves can be introduced into the trained deep learning model to calculate the depth and width of the oblique crack.

[0078] S3 (introducing the second signal curve into the vertical crack deep learning model for calculation, reconstructing the oblique crack based on the calculation result and determining the width and depth of the oblique crack) specifically includes the following steps:

[0079] S31, introducing the second signal curve into the deep learning model to calculate the crack profile curve corresponding to the second signal curve;

[0080] S32, combining the crack inclination angle and the crack profile curve corresponding to the second signal curve to reconstruct the oblique crack profile curve;

[0081] S33, calculating the width and depth of the oblique crack based on the oblique crack profile curve.

[0082] The principle of the S3 is as follows.

[0083] First, the vertical crack of the rail tread is scanned by the PCB eddy current probe, the vertical crack characteristic signal dataset is established, and the dataset is preprocessed, and then divided into a training set, a validation set and a test set. The above established training set and validation set are introduced into the GRU-RNN deep learning model for training and optimization, and the test set is used to evaluate the GRU-RNN deep learning model after training and optimization. The GRU-RNN inverts the collected vertical crack characteristic data to obtain the crack profile.

[0084] Secondly, the experimental measured rail head inclined crack phase curves contain phase curves with inclination angles of 10°, 20°, 30°, 40°, 50° and 60°. After the symmetry conversion of the inclined crack phase curves is completed, the trained deep learning model is used to reconstruct each inclined crack. According to the calculated crack contour sequence data, the width and vertical depth of the inclined crack are obtained. Combined with the calculated inclined crack inclination angle, when the calculated contour point z coordinate is less than 0, the contour point is horizontally translated, and the new horizontal coordinate of the i th contour point is:

[0085]

[0086] wherein x inew is the horizontal coordinate of the i th contour point after horizontal translation; x i is the horizontal coordinate of the i th contour point before horizontal translation; h i is the contour point depth (i.e. the negative value of the contour point z coordinate); and θ c is the fitted value of the crack inclination angle.

[0087] Finally, the crack width is defined as the horizontal distance between the half downlink depth and the half uplink depth of the calculated crack contour curve, which is denoted by w c . The crack depth is defined as the vertical distance between the minimum value in the crack contour curve and the network calculation value of the uncracked area, which is denoted by d c . From each data subset corresponding to the crack size, 10 samples are selected and introduced into the GRU-RNN for calculation to obtain the corresponding 10 crack contour calculation curves, and then the crack depth and width calculation values corresponding to the 10 samples are obtained, and the average value of the errors of these calculation values and the label values is calculated.

[0088] The average value of the errors of the crack depth calculation values and the label values is denoted by e d ; and the average value of the errors of the crack width calculation values and the label values is denoted by e w . The calculation formulae of e d and e w are as follows:

[0089]

[0090]

[0091] wherein n ed and n ew are the numbers of samples participating in the calculation of the average values of the crack depth and crack width calculation errors, respectively; d ci and w ci are the i th crack depth calculation value and the width calculation value, respectively; d ri and w rirespectively, are the i-th crack depth label value and width label value. The calculation results of the depth learning model on the vertical crack depth and width of different data sets are shown in Table 3.

[0092] The width and depth of the oblique crack are obtained from the reconstruction result of the oblique crack profile, and the average error of the width and depth of the oblique crack is calculated according to formula (15)-(16), and the calculation result is shown in Table 4. The average error of the oblique crack angle has been embodied in Table 1.

[0093] The above is the principle of step S3. In order to make the depth learning model trained on the vertical crack phase curve data set be used for quantitative detection of oblique cracks, the asymmetric oblique crack phase curve is converted into a symmetric curve by using a symmetrization conversion formula. After obtaining the symmetrization phase curve of the oblique crack,

[0094] Table 3 Calculation results of the depth learning model on the vertical crack depth and width of different data sets

[0095]

[0096] Table 4 Calculation results of the oblique crack depth and width

[0097]

[0098] The trained depth learning model is used to inverse the data of the symmetrical phase curve to obtain the profile curve of the oblique crack (which helps to intuitively reflect the extension of the crack), and then the width and vertical depth of the oblique crack are calculated. The cracks are divided into vertical cracks and oblique cracks, and in the prior art, the depth learning is used for quantitative evaluation of the cracks, and the depth learning models of the vertical cracks and the oblique cracks need to be established for quantitative evaluation of the vertical cracks and the oblique cracks. The present application only needs to establish a vertical crack depth learning model, which can be used for quantitative evaluation of the vertical cracks and the oblique cracks, improve the efficiency, and save the time. The phase curve obtained by detecting the vertical crack by using the eddy current detection method is a symmetrical curve, and the phase curve obtained by detecting the oblique crack is asymmetric. In the case, the phase curve of the oblique crack is converted into a symmetrical curve by using a symmetrization conversion formula, so that it is suitable for the vertical crack depth learning model.

[0099] The application also provides a steel rail crack quantitative eddy current detection device based on depth learning, which comprises:

[0100] The acquisition module is used for acquiring a crack phase signal and establishing a first signal curve.

[0101] The symmetrization conversion module is used for calculating a crack angle and symmetrizing and converting the first signal curve to output a second signal curve.

[0102] The deep learning module is configured to import the second signal curve into a vertical crack deep learning model for calculation, reconstruct a diagonal crack profile based on a calculation result, and determine a diagonal crack width and depth.

[0103] Further, the symmetry conversion module can be configured to: calculate an asymmetry rate of the phase curve; calculate a crack inclination angle based on the asymmetry rate; if the crack inclination angle ≠ 0, convert the phase curve symmetrically, output a symmetrical phase curve, and otherwise output the original phase signal curve.

[0104] Further, the asymmetry rate, the crack inclination angle and the symmetry conversion calculation formula are as follows:

[0105] The asymmetry rate calculation formula is

[0106]

[0107] In the formula, R represents the asymmetry rate, x max , x1 and x2 represent the phase curve peak value and the minimum value of the phase curve plus the abscissa at x2, respectively, denotes the difference between the maximum value and the minimum value of the phase curve;

[0108] The crack inclination angle calculation formula is

[0109] θ c = 160.0R-2.9

[0110] In the formula, θ c is the inclination angle fitting value corresponding to the experimental phase curve, and R is the asymmetry rate of the experimental phase curve; and the symmetry conversion calculation formula is

[0111] f(x max -x i ,R) = (-1.90×10 -2 R+2.20×10 -3 )cos[(-0.37R+0.43)(x max -x i )]+(1.00×10 -2 R-1.90×10 -4 )sin[(-0.37R+0.43)(x max -x i )]+(-6.90×10 -3 R+5.00×10 -4 )cos[(-0.74R-0.86)(x max -x i )]+(4.80×10 -3 R-2.70×10 -4)sin[(-0.74R+0.86)(x max -x i )]-1.60×10 -2 R+1.00

[0112] In the formula, R represents the asymmetry rate, x max -x i represents the difference between the horizontal coordinate of the peak point of the signal curve and the horizontal coordinate of the i-th point.

[0113] Further, the deep learning module can be specifically used for:

[0114] introducing the second signal curve into a deep learning model to calculate a crack profile curve corresponding to the second signal curve; reconstructing a oblique crack profile curve in combination with the crack inclination angle and the crack profile curve corresponding to the second signal curve; and calculating the oblique crack width and depth based on the oblique crack profile curve.

[0115] The present application also provides a steel rail crack quantitative eddy current detection device based on deep learning, comprising:

[0116] at least one processor; and

[0117] a memory in communication connection with the at least one processor, wherein

[0118] the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steel rail crack quantitative eddy current detection method based on deep learning.

[0119] The present application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steel rail crack quantitative eddy current detection method based on deep learning.

[0120] The above only describes some embodiments of the present application, and does not limit the protection scope of the present application, and any equivalent device or equivalent process transformation, or direct or indirect application in other related technical fields based on the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A deep learning-based quantitative eddy current testing method for rail cracks, characterized by, The method comprises: collecting a crack eddy current response, establishing a first signal curve; calculating a crack inclination angle, and symmetrically converting the first signal curve to output a second signal curve; introducing the second signal curve into a vertical crack deep learning model for calculation, and reconstructing an inclined crack and determining the depth and width of the inclined crack based on the calculation result; The symmetrically converting the first signal curve to output the second signal curve specifically comprises the following steps: calculating an asymmetry rate of the signal curve; calculating the crack inclination angle based on the asymmetry rate; if the crack inclination angle ≠ 0, symmetrically converting the signal curve to output the second signal curve, otherwise outputting the original signal curve; The calculation formula of the asymmetry rate is wherein R represents the asymmetry, x max x1, x2 represent the peak value of the signal curve and the minimum value of the signal curve plus the abscissa at the point, the difference between the maximum and minimum values of the signal curve; The calculation formula of the crack inclination angle is θ c = 160.0 R - 2.9 In the formula, θ c is the calculated value of the inclination angle corresponding to the signal curve, and R is the asymmetry ratio of the signal curve. The calculation formula of the symmetric conversion is f(x max -x i ) = (-1.90 x 10 -2 R + 2.20 x 10 -3 ) cos[(-0.37 R + 0.43)(x max -x i )] + (1.00 x 10 -2 R - 1.90 x 10 -4 ) sin [(-0.37R + 0.43)(x max - x i )] + (-6.90 x 10 -3 R + 5.00 x 10 -4 ) cos [(-0.74 R - 0.86)(x max - x i )] + (4.80 x 10 -3 R - 2.70 x 10 -4 ) sin [(-0.74R + 0.86)(x max - x i )] -1.60×10 -2 R+1.00 where R represents the asymmetry, x max -x i represents the difference between the abscissa of the peak of the signal curve and the abscissa of the i-th point.

2. The deep learning-based steel rail crack quantitative eddy current detection method according to claim 1, wherein The second signal curve is introduced into a vertical crack deep learning model for calculation, and an inclined crack is reconstructed and the depth and width of the inclined crack are determined based on the calculation result, specifically comprising the following steps: introducing the second signal curve into a vertical crack deep learning model to calculate a crack profile curve corresponding to the second signal curve; reconstructing an inclined crack profile curve in combination with the crack inclination angle and the crack profile curve corresponding to the second signal curve; and calculating the width and depth of the inclined crack based on the inclined crack profile curve.

3. A deep learning-based steel rail crack quantitative eddy current testing device, characterized in that, Comprise: The acquisition module is used for collecting a crack eddy current response and establishing a first signal curve; The symmetric conversion module is used for calculating a crack inclination angle and symmetrically converting the first signal curve to output a second signal curve; The deep learning module is used for introducing the second signal curve into a vertical crack deep learning model for calculation, and reconstructing an inclined crack and determining the depth and width of the inclined crack based on the calculation result; The symmetric conversion module can specifically be used for: calculating an asymmetry rate of the signal curve; calculating the crack inclination angle based on the asymmetry rate; if the crack inclination angle ≠ 0, symmetrically converting the signal curve to output the second signal curve, otherwise outputting the original signal curve; The calculation formula of the asymmetry rate is wherein R represents the asymmetry, x max x1, x2 represent the peak value of the signal curve and the minimum value of the signal curve plus the abscissa at the point, the difference between the maximum and minimum values of the signal curve; The calculation formula of the crack inclination angle is θ c = 160.0 R - 2.9 In the formula, θ c is the calculated value of the inclination angle corresponding to the signal curve, and R is the asymmetry ratio of the signal curve. The calculation formula of the symmetric conversion is f(x max -x i ) = 2.20 x 10 -2 R + 2.20 x 10 -3 ) cos[(-0.37 R + 0.43)(x max -x i )] + (1.00 x 10 -2 R - 1.90 x 10 -4 ) sin [(-0.37R + 0.43)(x max - x i )] +(-6.90 x 10 -3 R + 5.00 x 10 -4 ) cos [(-0.74 R - 0.86)(x max - x i )] + (4.80 x 10 -3 R - 2.70 x 10 -4 ) sin [(-0.74R + 0.86)(x max - x i )] -1.60×10 -2 R+1.00 where R represents the asymmetry, x max -x i represents the difference between the abscissa of the peak of the signal curve and the abscissa of the i-th point.

4. The deep learning-based steel rail crack quantitative eddy current detection device according to claim 3, wherein The deep learning module can specifically be used for: introducing the second signal curve into a vertical crack deep learning model to calculate a crack profile curve corresponding to the second signal curve; reconstructing an inclined crack profile curve in combination with the crack inclination angle and the crack profile curve corresponding to the second signal curve; and calculating the width and depth of the inclined crack based on the inclined crack profile curve.

5. A deep learning-based steel rail crack quantitative eddy current testing device, characterized in that, Comprise at least one processor, and The memory is in communication with the at least one processor, wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the deep learning-based steel rail crack quantitative eddy current detection method according to any one of claims 1 to 2.

6. A computer readable storage medium storing a computer program, characterized in that, The computer program is executed by the processor to implement the deep learning-based steel rail crack quantitative eddy current detection method according to any one of claims 1 to 2.

Images