Train wheel ultrasonic automatic calibration and verification method and device and readable storage medium

Abstract

The invention relates to a train wheel ultrasonic automatic calibration and verification method and device and a readable storage medium, relates to the technical field of steel rail flaw detection, and aims to solve the problem of low flaw detection precision of the train wheel ultrasonic automatic calibration and verification method. The train wheel ultrasonic automatic calibration and verification method comprises the following steps: B, carrying out scale scanning correction; phased array channel loss and classification loss are fused; predicting a defect sequence; and gain compensation. According to the method, the amplitude of each defect in a phased array channel is extracted, the gain difference between the transverse through hole defect and the flat-bottomed hole defect under the same depth is calculated, and the sensitivity difference between flat-bottomed hole defect calibration and transverse through hole defect calibration under the same depth is compensated. In this way, the problems that in the prior art, a transverse through hole calibration mode is different from standard requirement flat-bottom hole calibration, and the flaw detection sensitivity is different are solved, and therefore the precision and quality of on-site flaw detection are improved.

Description

Technical Field

[0001] The invention relates to the technical field of rail flaw detection, and in particular to a method, a device and a readable storage medium for ultrasonic automatic calibration and verification of train wheels. Background Art

[0002] The existing wheel flaw detection standards are all formulated for manual testing. Before the flaw detection task is carried out, the sensitivity of the probe will be calibrated, and the calibration defects are flat-bottom hole calibration defects of the same size and different depths set according to the needs. During manual calibration, the probe is placed close to the surface to be inspected, and the probe is freely moved on the detection surface to find the highest position of the calibration defect signal at each depth, and the defect signal amplitude is adjusted to a suitable height. This sensitivity is used as the reference sensitivity, and then the reference sensitivity is compensated as the detection sensitivity. Finally, the detection area is inspected by manual scanning.

[0003] For automated equipment, the probes are usually relatively fixed in position. Since the tread detection surface is non-planar, in order to ensure that all probes fit the detection surface, the probe group can only move in a circular direction. It is impossible to freely move the probe group to calibrate flat-bottom holes for all probes.

[0004] Therefore, the existing automatic calibration and verification methods for train wheels with ultrasonic waves of different depths are all calibrated using transverse through hole defects instead of flat-bottomed hole defects, which results in the existing transverse through hole calibration method being different from the standard flat-bottomed hole calibration method, and there is a problem of different flaw detection sensitivity. This affects the accuracy and quality of on-site flaw detection, and easily leads to the problem of low flaw detection accuracy.

[0005] That is to say, the ultrasonic automatic calibration and verification method for train wheels in the prior art has the problem of low flaw detection accuracy. Summary of the invention

[0006] The present invention provides a train wheel ultrasonic automatic calibration and verification method, device and readable storage medium, which are used to solve the problem of low flaw detection accuracy existing in the train wheel ultrasonic automatic calibration and verification method.

[0007] The present invention provides a method for ultrasonic automatic calibration and verification of train wheels, which comprises:

[0008] Step S10, correcting the B scan of the wheel calibration test block according to the incident angle of the probe and the position of the defect of the wheel calibration test block;

[0009] Step S20, establishing a train wheel calibration model integrating phased array channel loss and classification loss;

[0010] Step S30, predicting the defect sequence of the wheel calibration test block using the train wheel verification model according to the corrected B-scan image;

[0011] Step S40, compensating the sensitivity difference between the flat-bottom hole defect calibration and the transverse through hole defect calibration of the wheel calibration test block at the same depth by using the train wheel verification model.

[0012] In one embodiment, step S10 specifically includes the following steps:

[0013] Step S101, calculating the straight-line distance from the probe to the defect according to the known defect position, and calculating the vertical distance from the probe to the defect by using the incident angle information and the straight-line distance;

[0014] Step S102, calculate the vertical distance and divide it by the straight-line distance to obtain the correction coefficient;

[0015] Step S103 , scaling the B-scan image by interpolation by multiplying the original B-scan height by the correction coefficient to obtain the corrected B-scan height.

[0016] In one embodiment, in step S102, the zoom factor is calculated using a zoom factor calculation formula, and the zoom factor calculation formula is as follows:

[0017] d1=L1×cos(θ1);

[0018] d2=L2×cos(θ2);

[0019] d3=L3×cos(θ3);

[0020] Among them, the refraction angles θ1, θ2, and θ3 are known, and L1, L2, and L3 are calculated according to the echo return time and the material sound velocity. L1, L2, and L3 are the horizontal distances between the three probes and the corresponding three different defect locations, and d1, d2, and d3 are the vertical distances between the three probes and the corresponding three different defect locations.

[0021] In one embodiment, in step S103, the corrected B-scan height is calculated using a sound path correction scale calculation formula, and the sound path correction scale calculation formula is as follows:

[0022]

[0023] Wherein, H is the B-scan height after correction, h is the B-scan height before correction, L1 and L3 are the horizontal distances between the two probes and the corresponding two different defect locations, and d1 and d3 are the vertical distances between the two probes and the corresponding two different defect locations.

[0024] In one embodiment, step S20 specifically includes adding a channel loss function to the yolo classification loss function so that the model can learn defects of various shapes under different channels, thereby avoiding the problem that the network cannot converge well due to the similarity between the radial plate hole and the defect shape.

[0025] In one embodiment, the channel loss function is calculated as follows:

[0026]

[0027] L total_class =L class +λL channel ;

[0028] Where N is the total number of phased array channels, x i is the predicted value of the current channel, c n is the probability after the activation function, is the actual channel type, L class is the classification loss, L channel is the phased array channel loss, and λ is the trade-off parameter between classification loss and channel loss in the YOLO model.

[0029] In one embodiment, step S30 specifically includes the steps of improving the position loss function and adding the deviation value of the next defect in the data annotation, wherein the calculation formula of the improved position loss function is as follows:

[0030]

[0031] Among them, x i ,y i is the coordinate of the previous defect, B is the absolute distance between the previous defect and the next defect, x i ′、y i '、 is the predicted defect sequence.

[0032] In one embodiment, step S40, gain compensation specifically includes the steps of extracting the amplitude of each defect in the phased array channel, calculating the gain difference between the side through hole and the flat bottom hole defects at the same depth, and compensating for the sensitivity difference between the flat bottom hole defect calibration and the side through hole defect calibration at the same depth.

[0033] The present invention also provides a train wheel ultrasonic automatic calibration and verification device, which adopts the above train wheel ultrasonic automatic calibration and verification method, and comprises:

[0034] A correction module, which is used for B-scan scale correction; and

[0035] a fusion module, which is electrically connected to the scale correction module, and is used for fusing the phased array channel loss and the classification loss; and

[0036] A prediction module, which is electrically connected to the fusion module, and is used for predicting a defect sequence; and

[0037] The compensation module is electrically connected to the prediction module, and is used to compensate for the sensitivity difference between the flat bottom hole defect calibration and the side through hole defect calibration at the same depth.

[0038] The present invention also provides a readable storage medium on which a program or instruction is stored. When the program or instruction is executed by a processor, the above-mentioned train wheel ultrasonic automatic calibration and verification method is implemented.

[0039] Compared with the prior art, the present invention has the advantage that by extracting the amplitude of each defect in the phased array channel, the gain difference between the defects of the side-through hole and the flat-bottom hole at the same depth is calculated, and the sensitivity difference between the calibration of the flat-bottom hole defect and the calibration of the side-through hole defect at the same depth is compensated. This avoids the problem that the calibration method of the side-through hole in the prior art is different from the calibration method of the flat-bottom hole required by the standard, and there is a different problem of flaw detection sensitivity, thereby improving the accuracy and quality of on-site flaw detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Hereinafter, the present invention will be described in more detail based on embodiments and with reference to the accompanying drawings.

[0041] Figure 1 is a method flow chart of the method for ultrasonic automatic calibration and verification of train wheels in embodiment 1 of the present invention;

[0042] Figure 2 It is a schematic diagram of the module composition of the train wheel ultrasonic automatic calibration and verification device in the second embodiment of the present invention.

[0043] Figure 3 is a schematic diagram of a first flat-bottom hole in Embodiment 1 of the present invention;

[0044] Figure 4 is a schematic diagram of a second flat-bottom hole in Embodiment 1 of the present invention;

[0045] Figure 5 is a schematic diagram of a third flat-bottom hole in the first embodiment of the present invention;

[0046] Figure 6 is a schematic diagram of a first transverse through hole in Embodiment 1 of the present invention;

[0047] Figure 7 is a schematic diagram of a second transverse through hole in the first embodiment of the present invention;

[0048] Figure 8 is a schematic diagram of a third transverse through hole in the first embodiment of the present invention;

[0049] Fig. 9 It is a schematic diagram of a phased array in Embodiment 1 of the present invention.

[0050] Reference numerals:

[0051] 100. Wheel calibration test block; 101. First flat bottom hole; 102. Second flat bottom hole; 103. Third flat bottom hole; 104. First transverse through hole; 105. Second transverse through hole; 106. Third transverse through hole; 107. Tread; 200. Fusion module; 300. Prediction module; 400. Compensation module; 500. Correction module. DETAILED DESCRIPTION

[0052] The present invention will be further described below in conjunction with the accompanying drawings.

[0053] Embodiment 1

[0054] like Figure 1 As shown, the present invention provides a method for automatic ultrasonic calibration and verification of train wheels, and the method for automatic ultrasonic calibration and verification of train wheels specifically comprises the following steps:

[0055] Step S10, correcting the B scan of the wheel calibration test block according to the incident angle of the probe and the position of the defect of the wheel calibration test block;

[0056] Step S20, establishing a train wheel calibration model integrating phased array channel loss and classification loss;

[0057] Step S30, predicting the defect sequence of the wheel calibration test block using the train wheel verification model according to the corrected B-scan image;

[0058] Step S40, compensating the sensitivity difference between the flat-bottom hole defect calibration and the transverse through hole defect calibration of the wheel calibration test block at the same depth by using the train wheel verification model.

[0059] According to the above steps, by extracting the amplitude of each defect in the phased array channel, the gain difference between the side-through hole and the flat-bottom hole defects at the same depth is calculated, and the sensitivity difference between the flat-bottom hole defect calibration and the side-through hole defect calibration at the same depth is compensated. This avoids the problem of different flaw detection sensitivities due to the difference between the side-through hole calibration method in the prior art and the flat-bottom hole calibration required by the standard, thereby improving the accuracy and quality of on-site flaw detection.

[0060] It should be noted that if Fig. 9 As shown in the figure, the processing methods for each ultrasonic angle of train wheel ultrasonic flaw detection are as follows. Taking one angle as an example, a certain angle beam of the ultrasonic probe obtains the waveform of the wheel rotating one circle, thereby combining it into two-dimensional data in two dimensions, namely, the circumferential direction and the sound path, which is the B-scan data.

[0061] It should be noted that the original B scan is a combination of the circumferential A scan, and the B scan scale correction is to restore the elongation of the defect by the oblique probe, so as to make the defect size scanned by the oblique probe and the defect scanned by the straight probe as similar as possible, so as to facilitate the training of the model.

[0062] Specifically, in one embodiment, step S10 specifically includes the following steps:

[0063] Step S101, calculating the straight-line distance from the probe to the defect according to the known defect position, and calculating the vertical distance from the probe to the defect by using the incident angle information and the straight-line distance, so as to correct the defect scanned by the oblique probe to the defect scanned by the approximate straight probe;

[0064] Step S102, calculate the vertical distance and divide it by the straight-line distance to obtain the correction coefficient;

[0065] Step S103, the corrected B-scan height is calculated by multiplying the original B-scan height by the correction coefficient. Since the width remains unchanged, the B-scan image is scaled by interpolation to restore the influence of the ultrasound projection angle on the imaging size as much as possible.

[0066] Specifically, Fig. 9 As shown, d2 = L2 × cos (θ2), d3 = L3 × cos (θ3), the refraction angles θ1, θ2, θ3 are known, L1, L2, L3 are calculated according to the echo return time and the material sound velocity, so the sound path correction scale is:

[0067]

[0068] Wherein, H is the B-scan height after correction, h is the B-scan height before correction, L1 and L3 are the horizontal distances between the two probes and the corresponding two different defect locations, and d1 and d3 are the vertical distances between the two probes and the corresponding two different defect locations.

[0069] In one embodiment, step S20 specifically includes the fusion of phased array channel loss and classification loss, specifically including the following steps, specifically including the step of adding the channel loss function to the yolo classification loss function so that the model can learn defects of various shapes under different channels, and avoid the network from not being able to converge well due to the similarity between the spoke hole and the defect shape.

[0070] It should be noted that since different channels of the phased array cause the same defect to produce different forms, the data needs to be annotated in combination with the channel information during training. By adding the channel loss function to the yolo classification loss function, the model can learn defects of various forms under different channels, avoiding the network from not being able to converge well due to the similarity between the radial plate hole and the defect form. For example, there are C, L, E, N, T, and U channels in the B scan, and the defect forms and structural waves that appear in different channels have certain differences and similarities. Therefore, the six types of channel information and the six types of custom defect form information are combined to set a common classification loss, so as to achieve better convergence in the reverse gradient derivation in the network and avoid the degradation of network performance caused by the similarity between the structural wave and the defect.

[0071] Specifically, in one embodiment, the calculation formula of the channel loss function is as follows:

[0072]

[0073] L total_class =L class +λL channel ;

[0074] Where N is the total number of phased array channels, x i is the predicted value of the current channel, c n is the probability after the activation function, is the actual channel type, L class is the classification loss, L channel is the phased array channel loss, and λ is the trade-off parameter between classification loss and channel loss in the YOLO model.

[0075] Specifically, in one embodiment, step S30 specifically includes the steps of improving the position loss function and adding the deviation value of the next defect in the data annotation.

[0076] Among them, the calculation formula of the improved position loss function is as follows:

[0077]

[0078] Among them, x i ,y i is the coordinate of the previous defect, B is the absolute distance between the previous defect and the next defect, x i ′、y i '、 is the predicted defect sequence. By minimizing the position loss function, the model can quickly identify defects under fixed deviations and locate defects even when the amplitude is very shallow.

[0079] It should be noted that the traditional YOLO model is unable to learn useful information about the relative position of defects. The relative position of defects is fixed in the calibration verification, and the radial plate hole area may cause the model to have misframes. Therefore, by improving the position loss function and adding the deviation value of the next defect in the data annotation, it is possible to achieve the goal of identifying the position of defect No. 2 when identifying defect No. 1.

[0080] Specifically, in one embodiment, step S40 specifically includes the steps of extracting the amplitude of each defect in the phased array channel, calculating the gain difference between the side through hole and the flat bottom hole defects at the same depth, and compensating for the sensitivity difference between the flat bottom hole defect calibration and the side through hole defect calibration at the same depth.

[0081] It should be noted that the purpose of this embodiment is to provide a train wheel inspection equipment calibration block detection that combines fully automatic calibration verification and wheel rim detection, which can compensate for the sensitivity difference between flat-bottom hole defect calibration and transverse through hole defect calibration at the same depth.

[0082] Specifically, Figures 3 to 8 As shown, in this embodiment, the train wheel set detection equipment calibration block (wheel calibration block 100) has a rim portion similar in shape to the wheel to be tested, and includes: three flat-bottom holes and three transverse through holes, which include a first flat-bottom hole 101, a second flat-bottom hole 102, a third flat-bottom hole 103, a first transverse through hole 104, a second transverse through hole 105, and a third transverse through hole 106 distributed in the wheel flange area of ​​the rim portion, wherein the first flat-bottom hole 101, the second flat-bottom hole 102, and the third flat-bottom hole 103 are all arranged within a distance At the rolling circle at the 70mm position on the side, the length directions of the first flat bottom hole 101, the second flat bottom hole 102, and the third flat bottom hole 103 are all set to be radial, and the buried depth of the first flat bottom hole 101 from the tread 107 is designed to be 10mm, the buried depth of the second flat bottom hole 102 from the tread 107 is designed to be 20mm, and the buried depth of the third flat bottom hole 103 from the tread 107 is designed to be 30mm; the length directions of the first transverse through hole 104, the second transverse through hole 105, and the third transverse through hole 106 are all set to be axial. In addition, the buried depth of the first transverse through hole 104 from the tread 107 is designed to be 10mm, the buried depth of the second transverse through hole 105 from the tread is designed to be 20mm, and the buried depth of the third transverse through hole 106 from the tread 107 is designed to be 30mm. The first flat bottom hole 101 , the second flat bottom hole 102 , the third flat bottom hole 103 , the first transverse through hole 104 , the second transverse through hole 105 , and the third transverse through hole 106 are evenly distributed in the circumferential direction of the test block, and the circumferential angular spacing is 10°.

[0083] It should be noted that in order to compensate for the sensitivity difference between the calibration of flat-bottom hole defects and the calibration of transverse through hole defects at the same depth, the solution of this embodiment is to set transverse through holes (the first to third transverse through holes mentioned above) and flat-bottom holes (the first to third flat-bottom holes mentioned above) defects of the same depth on the test block at the same time. Before calibration, use a probe of the probe group to calibrate the flat-bottom hole defects of the first flat-bottom hole 101, the second flat-bottom hole 102, and the third flat-bottom hole 103, and record the gain values ​​in sequence respectively. Then, calibrate the transverse through hole defects of the first transverse through hole 104, the second transverse through hole 105, and the third transverse through hole 106, and record the gain values ​​in sequence respectively. By calculating the difference between the transverse through hole and the flat-bottom hole defects at the same depth, the sensitivity difference value caused by the different defect shapes can be obtained. Using this gain difference as sensitivity compensation can make up for the sensitivity difference caused by the different defect shapes.

[0084] It should be noted that the automatic calibration and verification method of this embodiment uses the yolo model to learn different types of defects in the test block, and matches defects of different forms with the phased array channel, so as to efficiently realize the preliminary selection of defects, and automatically calculate the compensation gain after selection. The same depth of deep through holes and flat-bottomed hole defects are set on the calibration block of the train wheelset detection equipment at the same time to make the automatic flaw detection sensitivity consistent with the manual flaw detection sensitivity. Setting all defects on the same wheelset detection equipment calibration block is conducive to saving material space and calibration time.

[0085] Embodiment 2

[0086] like Figure 2 As shown, the present invention provides a train wheel ultrasonic automatic calibration and verification device, which adopts the above-mentioned train wheel ultrasonic automatic calibration and verification method, and includes a correction module 500, a fusion module 200, a prediction module 300 and a compensation module 400. Among them, the correction module 500 is used for B-scan scale correction; the fusion module 200 is electrically connected to the scale correction module 500, and the fusion module 200 is used for the fusion of phased array channel loss and classification loss; the prediction module 300 is electrically connected to the fusion module 200, and the prediction module 300 is used for the prediction of defect sequences; the compensation module 400 is electrically connected to the prediction module 300, and the compensation module 400 is used to compensate for the sensitivity difference between the flat bottom hole defect calibration and the transverse through hole defect calibration at the same depth.

[0087] In the above settings, by setting up a compensation module 400, the compensation module 400 is used to extract the amplitude of each defect in the phased array channel, calculate the gain difference between the defects of the transverse through hole and the flat-bottom hole at the same depth, and compensate for the sensitivity difference between the calibration of the flat-bottom hole defect and the calibration of the transverse through hole defect at the same depth. This avoids the problem of different flaw detection sensitivities due to the difference between the calibration method of the transverse through hole in the prior art and the calibration of the flat-bottom hole required by the standard, thereby improving the accuracy and quality of flaw detection on site. A fusion module 200 is set up, which uses the yolo model to learn different types of defects in the test block, and matches defects of different shapes with the phased array channel, so as to efficiently realize the preliminary selection of defects, and automatically calculate the compensation gain after the selection.

[0088] Embodiment 3

[0089] The present invention provides a readable storage medium, and when the program or instruction is executed by a processor, the train wheel ultrasonic automatic calibration and verification method in the above-mentioned embodiment 1 is implemented.

[0090] Although the present invention has been described with reference to preferred embodiments, various modifications may be made thereto and parts thereof may be replaced by equivalents without departing from the scope of the present invention. In particular, the various technical features mentioned in the various embodiments may be combined in any manner as long as there are no structural conflicts. The present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for automatic ultrasonic calibration and verification of train wheels, characterized in that: It includes: Step S10, correcting the B scan of the wheel calibration test block according to the incident angle of the probe and the position of the defect of the wheel calibration test block; Step S20, establishing a train wheel calibration model integrating phased array channel loss and classification loss; Step S30, predicting the defect sequence of the wheel calibration test block using the train wheel verification model according to the corrected B-scan image; Step S40: compensating the sensitivity difference between the flat-bottom hole defect calibration and the transverse through hole defect calibration of the wheel calibration test block at the same depth by using the train wheel calibration model.

2. The method for automatic ultrasonic calibration and verification of train wheels according to claim 1, characterized in that: The step S10 specifically includes the following steps: Step S101, calculating the straight-line distance from the probe to the defect according to the known defect position, and calculating the vertical distance from the probe to the defect by using the incident angle information and the straight-line distance; Step S102, calculate the vertical distance and divide it by the straight-line distance to obtain the correction coefficient; Step S103, calculating the corrected B-scan height by multiplying the original B-scan height by the correction coefficient, and scaling the B-scan image by interpolation.

3. The method for automatic ultrasonic calibration and verification of train wheels according to claim 2, characterized in that: In the step S102, the zoom factor is calculated using a zoom factor calculation formula, and the zoom factor calculation formula is as follows: d1=L1×cos(θ1); d2=L2×cos(θ2); d3=L3×cos(θ3); Among them, the refraction angles θ1, θ2, and θ3 are known, and L1, L2, and L3 are calculated according to the echo return time and the material sound velocity. L1, L2, and L3 are the horizontal distances between the three probes and the corresponding three different defect locations, and d1, d2, and d3 are the vertical distances between the three probes and the corresponding three different defect locations.

4. The method for automatic ultrasonic calibration and verification of train wheels according to claim 3, characterized in that: In step S103, the corrected B-scan height is calculated using a sound path correction scale calculation formula, and the sound path correction scale calculation formula is as follows: Wherein, H is the B-scan height after correction, h is the B-scan height before correction, L1 and L3 are the horizontal distances between the two probes and the corresponding two different defect locations, and d1 and d3 are the vertical distances between the two probes and the corresponding two different defect locations.

5. The method for automatic ultrasonic calibration and verification of train wheels according to claim 3, characterized in that: The step S20 specifically includes adding a channel loss function to the yolo classification loss function so that the model can learn defects of various shapes under different channels, thereby avoiding the problem that the network cannot converge well due to the similarity between the radial plate hole and the defect shape.

6. The method for automatic ultrasonic calibration and verification of train wheels according to claim 5, characterized in that: The calculation formula of the channel loss function is as follows: THE total_class =L class +λL channel ; Where N is the total number of phased array channels, x i is the predicted value of the current channel, c n is the probability after the activation function, is the actual channel type, L class is the classification loss, L channel is the phased array channel loss, and λ is the trade-off parameter between classification loss and channel loss in the YOLO model.

7. The train wheel ultrasonic automatic calibration and verification method according to claim 1 is characterized in that: The step S30 specifically includes the steps of improving the position loss function and adding the deviation value of the next defect in the data annotation, wherein the calculation formula of the improved position loss function is as follows: Among them, x i ,y i is the coordinate of the previous defect, B is the absolute distance between the previous defect and the next defect, x i ′、y i '、 is the predicted defect sequence.

8. The method for automatic ultrasonic calibration and verification of train wheels according to claim 1, characterized in that: The step S40 specifically includes the step of gain compensation, which specifically includes the step of extracting the amplitude of each defect in the phased array channel, calculating the gain difference between the side through hole and the flat bottom hole defects at the same depth, and compensating the sensitivity difference between the flat bottom hole defect calibration and the side through hole defect calibration at the same depth.

9. A train wheel ultrasonic automatic calibration and verification device, characterized in that: The train wheel ultrasonic automatic calibration and verification device adopts the train wheel ultrasonic automatic calibration and verification method according to any one of claims 1 to 8, which comprises: A correction module, which is used for B-scan scale correction; and a fusion module, which is electrically connected to the scale correction module, and is used for fusing phased array channel loss and classification loss; and a prediction module, which is electrically connected to the fusion module, and is used for predicting a defect sequence; and A compensation module is electrically connected to the prediction module, and is used to compensate for the sensitivity difference between the flat bottom hole defect calibration and the side through hole defect calibration at the same depth.

10. A readable storage medium having a program or instruction stored thereon, characterized in that: When the program or instruction is executed by the processor, the train wheel ultrasonic automatic calibration and verification method as described in any one of claims 1 to 8 is implemented.

Images