CN106441107A - Method for automatic detection of steel rail abrasion - Google Patents

Abstract

The invention discloses a method for automatic detection of steel rail abrasion, and belongs to the field of detection. The method is used for optimizing the detection and calculation process of the steel rail abrasion. The technical main points of the method comprises the steps: collecting a laser image of a steel rail, and comparing the laser image with a complete steel rail laser light band image, so as to judge whether the steel rail is abraded or not; selecting and extracting a laser image characteristic quantity related with the abrasion of the steel rail if the steel rail is abraded, so as to calculate the abrasion depth and/or width of the steel rail. The beneficial effects of the invention are that the method employs an idea of firstly judging the abrasion and secondly carrying out the quantitative calculation, and selects the characteristic quantity in the calculation process, so as to optimize the calculation process of the abrasion depth and width.

Description

Automatic detection method of rail wear

technical field

The invention belongs to the detection field, and relates to an automatic detection method for rail wear, in particular to an automatic detection method for rail wear that can effectively detect the wear depth and width of the rail head surface based on inline laser image processing and a microprocessor.

Background technique

Railway is the main artery of transportation. Compared with other transportation methods, heavy-haul railway transportation has developed rapidly all over the world due to its large volume and low cost. In the track equipment, the rail is the most important component, which directly bears the load of the train and guides the running of the wheels. Whether the technical state of the rail is in good condition directly affects whether the train can run safely, smoothly and uninterruptedly at the specified speed. The railway locomotive transmits the driving force and braking force through the friction between the wheel and the rail, and the friction between the wheel and the rail will cause the wear of the rail. With the high-speed, heavy-load, and high-density operation of locomotives, the wear of rails will increase rapidly, especially the wear on the inner side of the outer rail with small radius curves is particularly serious.

The detection technology of rail wear has gone through a process from simple visual inspection to ruler and gauge tool inspection and digital instrument inspection. At present, in my country, there are main methods such as contact fixture measurement, eddy current detection, and optical triangulation measurement in rail wear detection in China. The detection results often depend on the attitude of the detection workers and the experience of using the instrument. These methods have low detection efficiency and low detection accuracy. Many problems such as high speed can no longer meet the current development needs of high speed. Although there is now a detection device that uses lasers to detect the surface wear of rails, it cannot achieve accurate step-by-step detection, and other detection methods are only at the level of theoretical research.

Contents of the invention

In order to optimize the detection and calculation process of rail wear, the present invention proposes an automatic detection method for rail wear, the technical points of which are: collecting the laser image of the rail, and comparing the image with the laser light band image of the complete rail to judge the detection of the rail Whether there is wear or not, if the rail is judged to be worn, select and extract the laser image feature quantity related to the wear amount of the rail to calculate the wear depth and/or width of the rail.

Beneficial effects: the laser image of the rail collected by the present invention is compared with the complete image to judge whether there is wear on the rail in the collected image, and when it is judged to be worn, the feature quantity is further selected to calculate the wear, and the wear is qualitatively judged first. The idea of quantitatively calculating the wear depth and width, and in the calculation process, the feature quantity is selected to optimize the calculation process of wear depth and width.

Description of drawings

Fig. 1 is the structural block diagram of rail wear automatic detection device described in embodiment 2;

Figure 2 is a schematic diagram of a wear-free laser image;

Figure 3 is a schematic diagram of an abrasive laser image;

Fig. 4 is a schematic diagram of brightness curve and disk diameter;

Fig. 5 is a schematic diagram of labeling of data points and feature quantities.

detailed description

Embodiment 1: A method for automatic detection of rail wear, which collects the laser image of the rail and compares the image with the laser light band image of the complete rail to determine whether there is wear on the detected rail. If it is judged that the rail is worn, the collected laser image Perform laser image processing, the laser image processing includes image preprocessing and image edge extraction, after the laser image processing, select and extract the laser image feature quantity related to the rail wear amount to calculate the rail wear depth and width. Wherein: the extracted laser image feature quantity related to the rail wear amount is more than one of the following feature quantities:

1) the lengths l _A and l _B of the two straight line parts of the laser image;

2) The width difference e of the two straight line laser images;

3) The longitudinal position difference z of the two straight line laser images;

4) The length l _C of the transition section between the two straight line laser images;

5) The inclination angle θ of the transition section between the two straight line laser images;

The wear width and wear depth are collectively referred to as the feature quantity of rail wear, and one or more of the above laser image feature quantities are selected to calculate the depth and width of rail wear;

When calculating the depth and width of rail wear, not all the above-mentioned laser image feature quantities are selected for calculation. In order to optimize the calculation process, the combination of laser image feature quantities is selected as the basic calculation data for calculating the depth and width of rail wear. The method of combination is: first determine the laser image feature quantity related to the wear feature, and select the preferred feature quantity from the laser image feature quantity, calculate the correlation coefficient between the remaining laser image feature quantities and the preferred feature quantity, and find its The average value is the threshold β for feature quantity selection. If the absolute value of the correlation coefficient between two laser image features |rTij|≥β, the two laser image features are related redundant features, and only One of them is selected as the laser image feature quantity for rail wear judgment. That is, when selecting the combination of laser image feature quantities for judgment, assuming that M is a set of feature samples collected at fixed points on the rail with wear, this set contains N fixed points of laser image feature quantities that reflect wear, and select the relevant The degree coefficient is used as a measurement parameter, which reflects the similarity between features. The two laser image features are related redundant features, and only one of them is selected as the laser image feature quantity for rail wear judgment, which is used to find the effective judgment of rail wear width. Combination of the minimum laser image features of depth and depth.

As an embodiment, the above-mentioned correlation coefficient is used as a measurement parameter to reflect the specific method of the correlation between features: use two sets of different laser image features: T _i ={t _ik ,k=1,2 ,...,n} and T _j ={t _jk ,k=1,2,...,n}, where k represents the kth test point, and there are n test points in total, then the correlation coefficient of two groups of laser image features is defined as follows :

In the formula, and are the mean values of two groups of features T _i and T _j respectively: and

The correlation coefficient rT _ij reflects the degree of correlation between two groups of features T _i and T _j . When the value of rT _ij is negative, it means that the two features are negatively correlated; when the value of rT _ij is positive, it means that the two features are positively correlated; when rT ij When _ij = 0, there is no correlation between the two wear features. When the absolute value of rT _ij is closer to 1, the correlation between the two laser image features is higher, and the redundancy generated is greater. In the laser image feature set, the correlation coefficient between the laser image features is used to set the threshold β. If the absolute value of the correlation coefficient between two laser image features |rT _ij |≥β, the two laser images The features are related redundant features, and only one of them is selected as the laser image feature quantity for rail wear judgment.

In another embodiment, the determination method for the above-mentioned threshold β is: for a single laser image feature quantity, select it as the preferred feature, and the method for judging the possibility that the rest of the laser image feature quantities are redundant features is : After the preferred feature is determined, the correlation coefficient between the rail wear width and depth correlation laser image feature set and the preferred feature quantity is obtained by calculation, and the mean value of the correlation coefficient data of this group is set as the threshold β, the determination method of the threshold β yes:

Where: in the formula, c is the number of feature quantities, l is the serial number of the preferred feature quantity, and j is the serial number of the optional feature quantity.

Therefore, in the above embodiment, the correlation coefficient between features is obtained to obtain the possibility of redundancy between features, and the mean value of the set of correlation coefficient data is set as the threshold β, and the threshold is used as the basis for judging whether the feature is redundant , so that when judging the redundancy of two features, only one of the redundant features is selected as the laser image feature for calculating the depth and width of wear, so as to optimize the calculation process and obtain the least combination of features.

As an embodiment, the calculation method of the wear depth and width is specifically disclosed: the length l _A of the two straight line parts of the laser image is used as the preferred feature for wear width detection, and the longitudinal position difference z of the two straight line laser images is used as the wear depth detection. The preferred feature; the length l _A and l _B of the two straight line parts of the laser image are calculated from the correlation coefficient of the two sets of laser image features as the feature quantity of the wear width detection, and the longitudinal position difference z of the two straight line laser images is used as the wear depth detection The characteristic quantity of the wear width is calculated as follows:

Among them, l is the width of the rail without wear;

The formula for calculating the wear depth is:

V=z·tan60°

As an embodiment, the image preprocessing includes the following steps:

First grayscale the image, draw the histogram of the grayscale image, and find out the range of grayscale concentration;

Then use the following formula to enhance the grayscale of the grayscale image to make the image clearer;

Among them: a, b are the left and right boundary points where the gray value is concentrated in the gray image histogram, and x, y represent the gray value before and after gray enhancement.

As an embodiment, the method for image edge extraction includes the following steps:

Take any median filter brightness curve that distributes pixels along the horizontal direction, take out the continuous points with the largest brightness gradient change on both sides of the maximum peak value of the curve, and take the midpoint p and q of the two groups of continuous points, between p and q The distance between them is used as the detection template diameter;

Let the brightness of the image be f(i,j), take a circle s(c,r) in the image field as the detection template, where c is the center of the circle, its coordinates are (i _c , j _c ), and r is the radius;

Define the set of pixels in s(c,r), and record the brightness sum of the pixels in the circle s as:

Make the center of the detection template move within a small range in the horizontal direction, calculate the brightness sum of each pixel in the detection template at each position, and the maximum brightness and center position of the template in this range is a pixel-level ridge edge point of the bright bar. The least squares method is used to fit a straight line, which is the center line of the one-line laser image, and the small range is an image interval of 2 times the radius around the center point of the circle, so as to obtain the length l of the two straight line parts of the laser image _A and l _B , the length l _C of the transition section between two straight-line laser images.

Embodiment 2: As a supplement to the technical solution of Embodiment 1, or as a separate embodiment: wear mainly occurs at the head of the rail, and wear includes top surface wear and side wear, and these two values must be detected at the same time during detection , to comprehensively judge the wear degree of the rail. In this embodiment, a high-intensity narrow-beam inline laser beam is used. The laser is at an angle of 60° to the plane where the inline beam is located and the surface of the rail to be tested. The high-resolution area array CCD image sensor is located directly above the laser image to capture the laser image. The beam image on the surface of the worn rail is bent, and the width and depth of the rail wear are determined by the position and degree of bending of the bending point.

The automatic detection device for rail wear includes: a linear laser, a CCD image sensor, a microprocessor, an execution unit, a display unit, an audible and visual alarm unit, and an interface unit. The CCD image sensor collects the laser image, and the obtained image information is transmitted to the microprocessor for analysis and processing, extracting the edge and center position of the image and fitting a straight line to form a complete rail laser light band image profile, and convert the image information into rail profile parameters. Store the feature quantity of the rail profile and compare it with the parameters of the complete rail to judge whether there is wear on the rail. If there is no wear, continue to the next point of detection; if there is wear, further determine the amount of wear, including the depth and width of wear. The execution unit receives the control signal from the microprocessor, controls the direction and speed of the detection device, and adjusts the orientation of the CCD image sensor. The output terminals of the microprocessor are respectively connected with the LCD display and the sound and light alarm system, and the LCD display is used to display the position of the rail. The current position and wear degree, the sound and light alarm system is used to prompt that the current position of the rail is worn and needs to be repaired. The interface unit is used to exchange information with the host computer, and the host computer can further refine the image of the wear position to determine the precise amount of wear.

Image preprocessing is the pre-processing stage of laser image edge extraction. Firstly, the image is grayscaled, and the histogram of the grayscale image is drawn to find out the concentration range of the grayscale. The left and right boundary points where the gray value is concentrated in the histogram, x and y respectively represent the gray value before and after the gray level enhancement) to enhance the gray level of the gray level image to make the image clearer.

The edge detection of the one-line laser image adopts the "roof-shaped" edge detection method. The edge detection method based on the brightness of a single pixel has poor anti-noise ability. In order to reduce the interference of image noise, the sum of the brightness of each pixel in a certain area is used as the basis for "roof-shaped" edge discrimination. Because the circle has isotropy and is not affected by the direction of the roof-shaped edge, the present invention adopts the "roof-shaped" edge detection method of the disk method. Move the disc detection template of appropriate size within a certain range on both sides of the line laser image. When the brightness and gradient change of each pixel in the template meets certain requirements, the center point of the template is the edge point of the roof shape.

Take any median filter brightness curve that distributes pixels along the horizontal direction, take out the continuous points with the largest brightness gradient change on both sides of the maximum peak value of the curve, and take the midpoint p and q of the two groups of continuous points, between p and q The distance between them is used as the diameter of the detection template, as shown in Figure 3.

Let the brightness of the image be f(i,j), take a circle s(c,r) in the image field as the detection template, where c is the center of the circle, its coordinates are ( _{ic ,j c )} , and r is the radius. Define the set of pixels in s(c,r):

And record the brightness sum of the pixels in the circle s as:

Make the center of the detection template move within a small range in the horizontal direction, calculate the brightness sum of each pixel in the detection template for each position, and the position of the center of the template with the largest brightness sum in this range is a pixel-level roof edge point of the bright bar. Use the least squares method to fit a straight line, which is the center line of the one-line laser image. The detected edge points and the fitted straight line are shown in Figure 4.

Further extract the feature quantity of the laser image related to the rail wear amount, including the selection method of the feature quantity and the determination of the threshold value.

The width and depth of the rail wear mainly involved in the present invention can be determined by the degree of curvature of the laser image, and the width and depth of the rail wear can be determined, and the following characteristic quantities can be used for selection:

1) the lengths l _A and l _B of the two straight line parts of the laser image;

2) The width difference e of the two straight line laser images;

3) The longitudinal position difference z of the two straight line laser images;

4) The length l _C of the transition section between the two straight line laser images;

5) The inclination angle θ of the transition section between the two straight line laser images.

One or more feature quantities can be selected for judging the depth and width of rail wear. When selecting a combination of feature quantities for judgment, it is required that different types of features have significant differences to avoid redundant features from interfering with judgment. Assume that M is a set of feature samples collected at fixed points on the rail with wear, and this set contains wear features of n fixed points. The correlation coefficient is selected as the measurement parameter, which can reflect the similarity between features, and is used to find the combination of the least feature quantities that can effectively judge the width and depth of rail wear. Suppose two groups of wear characteristics are: T _i ={t _ik ,k=1,2,…,n} and T _j ={t _jk ,k=1,2,…,n}, where k represents the first There are k test points and there are n test points in total, then the correlation coefficient of the two groups of features is defined as follows:

In the formula, and are the mean values of two groups of features T _i and T _j respectively: and

The correlation coefficient rT _ij reflects the degree of correlation between two groups of features T _i and T _j . When the value of rT _ij is negative, it means that the two features are negatively correlated; when the value of rT ij is positive, it means that the two features are positively correlated. When rT _ij =0, there is no correlation between the two features. Therefore, when the absolute value of rT _ij is closer to 1, it means that the correlation between the two features is higher, and the redundancy that may be generated at this time is greater.

In the feature set of rail wear, the correlation coefficient between each feature is used to set the threshold β. If the absolute value of the correlation coefficient |rT _ij |≥β between two features, it means that the two features are related. Redundant features, only one of them can be selected as the feature quantity for rail wear judgment.

For a single feature, the greater its direct relationship with the rail wear depth and width, the simpler the judgment method, the higher the feasibility for judging the wear amount, the greater the possibility of being selected, and the selected feature is the preferred feature. The possibility of a feature being a redundant feature is judged based on its correlation with the preferred feature. The higher the correlation, the greater the possibility of becoming a relevant redundant feature. After the preferred feature is determined, the correlation coefficient between the rail wear width and depth related feature set and the preferred feature quantity is obtained by calculation, and the mean value of this group of data is set as the threshold β, as shown in formula (4):

After the characteristic quantity is determined, the rail wear amount at the detection position is calculated, stored and displayed, and the sound and light alarm device is activated when there is excessive wear.

In the case of offline, the interface unit is used to exchange information with the host computer, and the host computer can further refine the image of the wear position to determine the precise amount of wear. Due to the adoption of the above technical solution, the automatic detection device for rail wear provided by this embodiment has such beneficial effects. Due to the use of image processing methods, under the control of the microprocessor, the device can be separated from the control of the PC. It runs automatically under the setting. The equipment has a certain degree of perfection and effectiveness, which is convenient for the use of testing personnel. It is not only easy to operate, accurate in testing results, but also low in manufacturing costs.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, any person familiar with the technical field within the technical scope of the disclosure of the present invention, according to the present invention Any equivalent replacement or change of the created technical solution and its inventive concept shall be covered within the scope of protection of the present invention.

Claims (10)

1. a rail wear automatic testing method, it is characterised in that gather rail laser image, and with complete rail laser Light belt image carries out image comparison, to judge whether detection rail has abrasion, it is judged that rail has abrasion, selects and extracts and steel The related laser image characteristic quantity of rail abrasion loss, to be calculated the rail wear degree of depth and/or width.

2. rail wear automatic testing method as claimed in claim 1, it is characterised in that described related to rail wear amount Laser image characteristic quantity is more than one in following characteristics amount,

1) length l of two sections of straight line portioies of laser image_AAnd l_B；

2) the stand out e of two sections of linear laser images；

3) lengthwise position difference z of two sections of linear laser images；

4) length l of changeover portion between two sections of linear laser images_C；

5) inclination angle theta of changeover portion between two sections of linear laser images.

3. rail wear automatic testing method as claimed in claim 2, it is characterised in that the characteristic quantity of rail wear includes mill Consumption width and the abrasion degree of depth, select one or more laser image characteristic quantity, for calculating the degree of depth and/or the width of rail wear Degree, when selecting the combination for the laser image characteristic quantity judging, it is assumed that M is to pinpoint to collect on the rail have abrasion Feature samples set, this set comprise N number of fixing point reaction abrasion laser image characteristic quantity, select correlation coefficient make For metric parameter, two laser image features are relevant redundancy features, only select one of them to judge as rail wear Laser image characteristic quantity, effectively being judged the combination of the minimum laser image characteristic quantity of rail wear width and the degree of depth, and It is calculated a certain characteristic quantity of rail wear with the combination of this feature amount.

4. rail wear automatic testing method as claimed in claim 3, it is characterised in that two laser image spies of described judgement Levy is that the method for relevant redundancy feature is：Select preferred features amount from laser image characteristic quantity, calculate remaining each laser image Characteristic quantity and the correlation coefficient of preferred features amount, and seek its mean value, this mean value is the threshold that laser image characteristic quantity selects Value β, if absolute value | rTij | >=β of the coefficient correlation wherein between certain two laser image feature, then this two laser images Feature is relevant redundancy feature, only selects one of them laser image characteristic quantity judging as rail wear.

5. the rail wear automatic testing method as described in claim 3 or 4, it is characterised in that calculate described correlation coefficient Method be：

If two groups of different laser image features are respectively：T_i={ t_ik, k=1,2 ..., n} and T_j={ t_jk, k=1,2 ..., N}, wherein k represents k-th test point, and total n test point, then the coefficient correlation of two groups of laser image features is defined as follows：

${\mathrm{rT}}_{ij}=\frac{\underset{k=1}{\overset{n}{\Σ}}(t_{ik}-\stackrel{\‾}{t_i})(t_{jk}-\stackrel{\‾}{t_j})}{\sqrt{\underset{k=1}{\overset{n}{\Σ}}{(t_{ik}-\stackrel{\‾}{t_i})}^2}\sqrt{\underset{k=1}{\overset{n}{\Σ}}{(t_{jk}-\stackrel{\‾}{t_j})}^2}}$

In formula,WithIt is respectively two stack features T_iAnd T_jMean value：With

Correlation coefficient r T_ijReflect two stack features T_iAnd T_jDegree of correlation, rT_ijValue for, when negative, representing two feature negatives Close；rT_ijValue be timing, represent two feature positive correlations；Work as rT_ijIt when=0, is incoherent between two wear characteristics, when rT_ijAbsolute value when being closer to 1, the degree of correlation of two laser image features is higher, and the redundancy of generation is bigger.

6. rail wear automatic testing method as claimed in claim 4, it is characterised in that the method calculating threshold value beta is：

$\mathrm{\β}=\frac{1}{c-1}\underset{j=2}{\overset{c-1}{\Σ}}\left|{\mathrm{rT}}_{1j}\right|$

Wherein：The quantity of the c amount of being characterized in formula, l is the sequence number of first-selected characteristic quantity, and j is the sequence number of alternative features amount.

7. rail wear automatic testing method as claimed in claim 4, it is characterised in that two sections of straight line portioies of laser image Length l_AAs the preferred features of abrasion width detection, lengthwise position difference z of two sections of linear laser images is as the abrasion degree of depth The preferred features of detection；Obtain the length of two straight line portioies of laser image by the Calculation of correlation factor of two groups of laser image features Degree l_AAnd l_BAs the characteristic quantity of abrasion width detection, lengthwise position difference z of two sections of linear laser images is as abrasion depth detection Characteristic quantity, abrasion width calculation formula be：

$W=l_B+\frac{2}{3}|l-(l_A+l_B)|$

Wherein, l is the width not wearing away rail；

Abrasion depth calculation formula is：

V=z tan60 °.

8. rail wear automatic testing method as claimed in claim 2, it is characterised in that the laser image of the rail of collection, Before extracting the laser image characteristic quantity related to rail wear amount, there is the step that laser image is processed, described laser image Process includes Image semantic classification and Edge extraction.

9. rail wear automatic testing method as claimed in claim 8, it is characterised in that described Image semantic classification includes as follows Step：

First by image gray processing, draw the histogram of gray level image, find out gray scale and concentrate scope；

Then use following formula, grey level enhancement is carried out to gray level image, make image become apparent from；

$y=255\×\frac{(x-a)}{(b-a)}$

Wherein：A, b are respectively the left and right boundary point of gray value integrated distribution in gray level image histogram, and x, y represent gray scale respectively Gray value before and after enhancing.

10. rail wear automatic testing method as claimed in claim 8, it is characterised in that use described Edge extraction Method, comprise the steps：

Appoint the medium filtering brightness curve taking a distribution image vegetarian refreshments in the horizontal direction, take respectively in this curve peak-peak both sides Go out the maximum continuity point of brightness step change, take the spacing of midpoint p and q, p and q of this two groups of continuity points as detection template Diameter；

If the brightness of image be f (i, j), take in picture field a round s (c, r) as detection template, wherein c is the center of circle, its Coordinate is (i_c,j_c), r is radius；

Definition s (c, r) in the set of pixel, and remember the brightness of pixel in round s and be：

$X=\underset{(i,j)\∈X}{\Σ}f(i,j)$

Move in making the detection template center of circle little scope in the horizontal direction, calculate each pixel intensity in the detection template of each position With in the range of being somebody's turn to do, brightness and maximum template home position, be a Pixel-level roof edge point of this bright wisp, utilizes minimum Square law fitting a straight line, this straight line is a wordline laser image center line, and described little scope is about putting centered on the center of circle The image of each 2 times of radiuses is interval, to obtain length l of two sections of straight line portioies of laser image_AAnd l_B, two sections of linear laser images Between length l of changeover portion_C.