A railway turnout fault diagnosis method based on multi-visual feature extraction

Abstract

The present invention relates to the field of railway maintenance technology, and in particular to a railway turnout fault diagnosis method based on multi-visual feature extraction. It includes obtaining the spatial coordinates of the sampling points, calculating the characteristic variables, substituting the characteristic variables into the rule function to obtain the current remaining life of the turnout, and outputting the monitoring results according to the current remaining life and the minimum remaining life. The present invention proposes a method for automatically detecting railway turnouts, which saves the manpower cost and time cost of railway turnout maintenance. For railway turnouts, 9 variable features associated with the service life are proposed, and these 9 variable features can be extracted by image processing technology, which is easy to implement; the 9 variable features are associated with the service life, and a rule function is constructed to represent the relationship between the variable features and the service life, and the relationship between the variable features and the service life is successfully fitted.

Description

Technical Field

[0001] The invention relates to the technical field of railway maintenance, and in particular to a railway turnout fault diagnosis method based on multi-visual feature extraction. Background Art

[0002] As an important transportation network, the safety of railways is related to the country's transportation, trade and economic activities. Good track infrastructure and specifications can effectively reduce the possibility of accidents and reduce the harm caused by accidents to passengers and staff. About 90% of railway accidents occur in the railway network. Railway infrastructure is a complex dependent system, in which the weakest element determines the safety level of the entire system. Due to the analysis of the threats caused by the increase in the speed of railway traffic, the key element of this system is the technical and maintenance quality of railway turnouts. Turnouts are the intersections of tracks on railway lines and an important part of the railway network. They affect safety and determine the maximum speed of railway traffic.

[0003] Current railway regulations describe the rules and deadlines for visual inspections and technical tests of railway turnouts. However, they do not take into account the many factors that occur during maintenance and have a significant impact on the durability of railway turnouts. The efficiency and reliability of railway turnouts are primarily affected by the component rails. Defects and damage to the rails, especially increased wear rates, directly affect the durability of turnout components and thus the life cycle costs.

[0004] Research on turnout detection mainly focuses on manual defect detection and detection vehicle detection. No existing research has studied how to detect turnout defects based on in-service vehicles. Since manual detection methods and track detection methods are costly and inefficient, they are not sufficient to detect turnout defects. It is also shown that the wear of turnout components depends on speed. With the increase in railway speed, more attention needs to be paid to the detection of turnout wear. Summary of the invention

[0005] The present invention discloses a railway turnout fault diagnosis method based on multi-visual feature extraction, and the specific method is as follows:

[0006] Preset several sampling points on the railway turnout to be monitored;

[0007] Get the spatial coordinates of each sampling point;

[0008] Calculate characteristic variables according to the spatial coordinates of several sampling points;

[0009] Substitute the characteristic variables into the rule function to obtain the current remaining life of the turnout;

[0010] The difference between the current remaining life and the minimum remaining life is taken as the turnout fault diagnosis result;

[0011] The rule function is obtained by fitting the changes of characteristic variables and remaining life.

[0012] Furthermore, the spatial coordinates of each sampling point are obtained as follows:

[0013] Light field camera calibration;

[0014] Acquire image information of several sampling points through a light field camera;

[0015] The PSP algorithm is used to calculate two sets of poses;

[0016] The EPI depth estimation algorithm is used to calculate the depth information of the sampling point;

[0017] The P3P algorithm is used to solve the extrinsic parameter matrix of the light field camera and obtain the spatial coordinates of the sampling points.

[0018] Furthermore, the variable features include any two or more of the following:

[0019] gauge, distance from the axle of the short rail to the pinnacle, distance from the short rail to the bow of the transom, distance from the spar to the front end of the transom, width of track throughout the turnout length, distance from the leading edge of the steering wheel to the nearer edge of the bow, width of the steering wheel slot, width of the groove at the crossing and difference in gauge height at the point where the gauge is measured.

[0020] Furthermore, the rule function is expressed as:

[0021] Let the rule function be expressed as:

[0022]

[0023] in, represents the predicted remaining life of the i-th sample, X1 to X9 represent the variable characteristics of the sample, to Represents the constant to be fitted.

[0024] Furthermore, the regular function is obtained by fitting the changes of characteristic variables and service life. The specific method is as follows:

[0025] Measure several turnout samples to obtain the characteristic variables and actual remaining life of each turnout sample;

[0026] Adjustment to The value of , makes the evaluation target minimum;

[0027] The evaluation target is expressed as regression residual:

[0028]

[0029] Among them, Y irepresents the actual remaining life of the i-th sample;

[0030] To minimize the squared residual, take the derivative of each parameter and set it to 0:

[0031]

[0032] Substituting the sample into the solution to

[0033] Furthermore, the difference in the minimum remaining life is calculated as follows:

[0034] Substitute the maximum required value of wear of the characteristic variable into the rule function to obtain the difference in the minimum remaining life.

[0035] Due to the adoption of the above scheme, the present invention has the following effects:

[0036] 1. A method for automatic detection of railway turnouts is proposed, which saves the manpower and time costs of railway turnout maintenance.

[0037] 2. For railway turnouts, 9 variable features associated with service life are proposed. These 9 variable features can be extracted through image processing technology, which is easy to implement.

[0038] 3. The 9 variable characteristics were associated with the service life, and a regular function was constructed to represent the relationship between the variable characteristics and the service life. The relationship between the variable characteristics and the service life was successfully fitted and can be used for life prediction.

[0039] 4. In the fitting process, regression residuals are used as evaluation targets, and a specific, effective and simple solution method is given. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings of the present invention are as follows.

[0041] Figure 1 Schematic diagram of the turnout structure in the embodiment.

[0042] Figure 2 Schematic diagram of light field camera spatial coordinate measurement in an embodiment.

[0043] Figure 3 Schematic diagram of P3P algorithm projection in the embodiment.

[0044] Figure 4 It is a flow chart of spatial coordinate calculation in the embodiment.

[0045] Figure 5 Schematic diagram of the monitoring process in the embodiment. DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0047] A railway turnout fault diagnosis method based on multi-visual feature extraction, such as Figure 5 As shown, the specific steps are as follows:

[0048] S1. Preset several sampling points on the railway turnout to be monitored.

[0049] S2. Get the spatial coordinates of each sampling point.

[0050] In step S2, Figure 2 As shown in the figure, P is a point in space, c1 and c2 are the optical centers of the right and left cameras respectively. According to the geometric relationship in the figure, the depth information of point P can be obtained.

[0051] The depth of point P is: f is the focal length of the camera; Δμ is the distance between the two optical centers, i.e. the baseline length; μ1 and μ2 are the light deflection distances to the right and left cameras, respectively. It can be expressed by the slope of the diagonal line. The depth estimation can be converted into the problem of obtaining the slope of the diagonal line in EPI, and the rotation parallelogram operator (SPO) is used to obtain the angle of the diagonal line. The SPO algorithm introduces confidence to evaluate the desired result, and assigns different weight values ​​to the horizontal and vertical EPI depth estimation results according to the confidence to make the result more accurate. The horizontal EPI and the vertical EPI perform depth estimation on the pixel point, and then define a confidence c to measure the reliability. Finally, the weighted sum of the two depth estimation results is obtained, and the result is the final accurate depth estimation result:

[0052]

[0053] is the distance mean, d max is the maximum distance. σ is an adjustable parameter.

[0054] Then the P3P algorithm is used, such as Figure 3 As shown, this algorithm only needs 3 pairs of 3D-2D matching points to solve the camera's external parameter matrix and find the spatial coordinates of the measured point. In triangles OAB and Oab, the cosine theorem gives:

[0055]

[0056] Similarly, consider the other triangles:

[0057]

[0058]

[0059] d OA dOB d OC d AB d BC d AC are the lengths of sides OA, OB, OC, AB, BC, and AC respectively. Let n = d OA / d OC ,m=d OB / d OC remember Arranged:

[0060] (1-p)y 2 -pn 2 -2cos<b,c> m+2pnmcos<a,b> +1=0

[0061] (1-r)n 2 -rm 2 -2cos<a,c> n+2rnmcos<a,b> +1=0

[0062] Eliminate the variables to find n, m, and then find p, q, and finally d OA d OB d OC At most 4 sets of solutions can be obtained, and then the 4th pair of matching points can be used for verification to obtain the only optimal solution.

[0063] Photos taken with just one exposure contain rich 3D information, which can be used to restore the 3D information of target feature points.

[0064] The spatial coordinate calculation process is as follows Figure 4As shown in the figure, a microlens array is added in front of the imaging sensor in the light field camera. The light field camera records the direction information of the light by recording the two intersection points of the light with the main lens plane and the microlens plane, and then obtains the final image column by integral calculation. Light from different directions at the target point in three-dimensional space converges on the center of a microlens through the main lens. Each microlens corresponds to a certain number of sensor pixels, and these photosensitive pixels constitute a macro pixel. The light field camera used is the second-generation Lytro camera. The microlens array of the camera is 643*425, and 15*15 pixels are distributed behind each microlens, that is, its angular resolution is 15pixel×15pixel, which can record 225 light rays in different directions emitted from a target point. EPI is a two-dimensional 2D slice of a four-dimensional 4D light field. By fixing the horizontal and vertical coordinates of the main lens plane and the microlens array plane in the 4D light field, the horizontal or vertical EPI can be obtained. The horizontal and vertical EPI (u, v) is calculated by the light field camera acquisition as the main lens plane, and (s, t) is the micro lens plane, that is, u and v control the angle resolution, and s and t control the spatial resolution. (u, v) and (s, t) together determine the position and direction of the light distribution in space.

[0065] S3. Calculate characteristic variables based on the spatial coordinates of several sampling points.

[0066] A turnout is a system consisting of two rails, a turnout blade and a crossbeam, which are arranged at a fixed distance from each other. The geometric accuracy of the turnout is evaluated as follows: Figure 1 The diagnostic tests are performed with the following basic values ​​specified:

[0067] Track gauge a, d, k;

[0068] The distance z from the short track axis to the top of the spire;

[0069] The distance from the short rail to the crossbeam bow is e;

[0070] The distance r from the wing spar to the front end of the crossbeam;

[0071] The track width b along the entire turnout length;

[0072] The distance between the leading edge of the steering wheel and the nearer edge of the bow i;

[0073] The width of the steering wheel groove, the width of the groove in which the steering wheel sits;

[0074] The width of the groove at the intersection;

[0075] The difference in track gauge height at the point where the gauge is measured.

[0076] S4. Substitute the characteristic variables into the rule function to obtain the current remaining life of the turnout.

[0077] The regular function is obtained by fitting the changes of characteristic variables and remaining life.

[0078] Classify the nine sampling points according to the characteristic values ​​measured at different times. Let the function be: Y i =β0+β1X1+β2X2+···+β9X9. Where Y is time, and X is the wear value taken at different feature points. The sample conditional mean Y of the explained variable is estimated as a linear function of each explanatory variable: We use the least squares method to estimate and determine the sample regression function. The regression residual is:

[0079] To minimize the squared residual, take the derivative of each parameter and set it to zero.

[0080]

[0081] Substituting the existing samples into it, we can solve for β and finally get Y i According to the current regulations on rail wear, the maximum wear requirement can be substituted into the function to predict its lifespan.

[0082] S5. The difference between the current remaining life and the minimum remaining life is taken as the turnout fault diagnosis result.

[0083] Specifically, the maximum required value of wear of the characteristic variable is substituted into the rule function to obtain the difference of the minimum remaining life.

[0084] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A railway turnout fault diagnosis method based on multi-visual feature extraction, characterized in that: The specific method is as follows: Preset several sampling points on the railway turnout to be monitored; Get the spatial coordinates of each sampling point; Calculate characteristic variables according to the spatial coordinates of several sampling points; Substitute the characteristic variables into the rule function to obtain the current remaining life of the turnout; The difference between the current remaining life and the minimum remaining life is taken as the turnout fault diagnosis result; The rule function is obtained by fitting the characteristic variable and the change of the remaining life; The variable characteristics include any two or more of the following: the gauge, the distance from the axle of the short rail to the pinnacle, the distance from the short rail to the bow of the transom, the distance from the spar to the front end of the transom, the width of the track throughout the length of the turnout, the distance from the leading edge of the steering wheel to the nearer edge of the bow, the width of the steering wheel slot, the width of the groove at the crossing and the difference in gauge height at the point where the gauge is measured; The rule function is expressed as: Let the rule function be expressed as: in, represents the predicted remaining life of the i-th sample, X1 to X9 represent the variable characteristics of the sample, to represents the constant to be fitted; The regular function is obtained by fitting the changes of characteristic variables and service life. The specific method is as follows: Measure several turnout samples to obtain the characteristic variables and actual remaining life of each turnout sample; Adjustment to The value of , makes the evaluation target minimum; The evaluation target is expressed as regression residual: Among them, Y i represents the actual remaining life of the i-th sample; To minimize the squared residual, take the derivative of each parameter and set it to 0: Substituting the sample into the solution to The difference in minimum remaining life is calculated as follows: Substitute the maximum required value of wear of the characteristic variable into the rule function to obtain the difference in the minimum remaining life.

2. The railway turnout fault diagnosis method based on multi-visual feature extraction as claimed in claim 1, characterized in that: Get the spatial coordinates of each sampling point. The specific method is as follows: Light field camera calibration; Acquire image information of several sampling points through a light field camera; The PSP algorithm is used to calculate two sets of poses; The EPI depth estimation algorithm is used to calculate the depth information of the sampling point; The P3P algorithm is used to solve the extrinsic parameter matrix of the light field camera and obtain the spatial coordinates of the sampling points.

Images