CN112132824A - A method for automatic detection of truck axle box spring faults - Google Patents

Abstract

A method for automatically detecting a failure of a truck axle box spring belongs to the technical field of truck operation. The invention aims to solve the problems that the accuracy and the stability of fault detection of a truck axle box spring in a manual image detection mode are not high, and the driving safety of a truck cannot be ensured. The method comprises the following steps: 3D high-definition imaging equipment is built around the track, and after the truck passes through the equipment, a height image and a gray image are obtained; roughly positioning a shaft box spring component in the image by combining hardware wheel base information with priori knowledge; after the height image correction is carried out, the axle box spring is accurately positioned after the axle box spring vehicle type is identified by using an image processing method; carrying out fault analysis on the journal box spring by using an advanced image processing algorithm and a mode identification method, and judging whether the journal box spring has a fleeing-out fault or a breaking fault; and uploading an alarm to the axle box spring component with the fault, and carrying out corresponding processing by the staff according to the identification result. The invention is used for detecting the failure of the axle box spring of the truck.

Description

A method for automatic detection of truck axle box spring faults

technical field

The invention relates to an automatic detection method for the spring failure of a truck axle box. It belongs to the technical field of truck operation.

Background technique

The truck axle box spring plays a buffering and fixing role, which is used to avoid the instability of the vehicle's meandering motion within the operating speed range, ensure good guiding performance when the curve passes, reduce the wear and noise between the rim and the rail, and ensure safe and stable operation. smooth. The breaking or popping out of the axle box spring will endanger the driving safety. Therefore, the relevant departments of the railway attach great importance to the fault detection of the axle box spring. In the fault detection of the axle box spring, the fault detection is generally carried out by manually checking the image. However, due to the fact that the inspectors are prone to fatigue and inattention during the work process, and may also miss inspections or wrong inspections due to personal reasons, it is impossible to ensure the safety of trucks.

Image processing and pattern recognition technologies continue to mature, and automatic image recognition can improve detection efficiency and stability. Therefore, using image processing and pattern recognition to automatically identify axle box spring faults can effectively improve the detection accuracy and stability.

SUMMARY OF THE INVENTION

The present invention is to solve the problem that the fault detection accuracy and stability of the existing truck axle box spring by using the manual detection image method are not high, and the running safety of the truck cannot be ensured. An automatic detection method for a truck axle box spring fault is provided.

An automatic detection method for a truck axle box spring failure, comprising:

Step 1: After the truck passes through the imaging device, a large image of the axle box spring is obtained, and the large image of the axle box spring includes a height image and a grayscale image;

Step 2: Correct the height image in the large image of the axle box spring;

Step 3: Obtain the overall height sub-image and the gray-scale sub-image of the axle box spring respectively;

Step 4. Determine the fault type of the axle box spring according to the obtained grayscale image;

Step 5. After identifying the fault, calculate the position of the fault in the original image through the mapping relationship between the sub-image and the large image of the axle box spring, and the large image of the axle box spring and the original image, and display the fault through the fault display platform.

beneficial effect

1. The present invention uses the method of automatic image recognition to replace manual detection, improves detection efficiency and accuracy, and ensures the safety of trucks.

2. The image captured by 3D hardware equipment can directly locate the part according to the distance between the part and the camera, that is, the depth information of the part, and the subsequent identification is simple and efficient.

3. The system performs fault identification on the height and grayscale images collected by the 3D equipment. The 3D height image is the depth information of the detected object, so that the fault detection is not affected by rain, chalk, white paint and stains, so the system stability is enhanced .

4. The 3D height image is corrected first and then recognized, which reduces the complexity of fault detection later and has high recognition accuracy.

5. The image fusion of different detection stations and different vehicle models can obtain a normal and trouble-free template image of the spring, which improves the universality of the system.

6. First identify the fault of the head ring, then identify the fault of the tail ring, and finally identify the process of the left and right jumping out and the middle layer breaking, which improves the system detection efficiency.

Description of drawings

Fig. 1 is the flow chart of the present invention;

Fig. 2 is the specific fault identification flow chart of the present invention;

FIG. 3 is a diagram showing the relationship between depth information y and row position x.

Detailed ways

Embodiment 1: Referring to FIG. 1 to illustrate this embodiment, a method for automatic detection of a truck axle box spring failure in this embodiment includes:

Step 1. Build an imaging device around the track. The imaging device has a certain elevation angle with the side of the car body. After the truck passes through the imaging device, a large image of the axle box spring (that is, a rough positioning image is obtained, according to the wheelbase information and combined with the axle box spring on the bogie. The image containing the axlebox spring but the range is larger than the axlebox spring intercepted by the prior knowledge of the position in the middle), the large image includes the height image and the grayscale image;

Step 2: Correct the height image in the large image of the axle box spring;

Step 3: Obtain the overall height sub-image and the gray-scale sub-image of the axle box spring respectively;

Step 4. Determine the fault type of the axle box spring according to the obtained grayscale image;

Step 5. After identifying the fault, calculate the position of the fault in the original image through the mapping relationship between the sub-image and the large image of the axle box spring, and the large image of the axle box spring and the original image, and display the fault through the fault display platform.

Embodiment 2: The difference between this embodiment and Embodiment 1 is that in step 1, an imaging device is built around the track. After the truck passes through the device, a large image of the axle box spring is obtained, and the large image of the axle box spring includes a height image. and grayscale images; the specific process is:

The imaging device includes a camera acquisition unit, a magnetic steel unit, a 3D image acquisition industrial computer unit, a control industrial computer unit, and an image recognition unit, wherein the camera acquisition unit includes a camera and a compensation light module;

The camera acquisition unit captures and collects the images of the passing trucks, and the 3D image acquisition industrial computer unit stores the collected images; the magnetic steel unit transmits the signals of the near-end magnetic steel and the remote magnetic steel to the control industrial computer unit, and the control industrial computer unit passes the The obtained signal information of the near-end magnetic steel and the far-end magnetic steel calculates the vehicle speed and wheelbase information, and transmits the vehicle speed and the magnetic steel wheelbase information to the image recognition unit. The image recognition unit uses the obtained wheelbase information and image information, etc. to realize automatic identification algorithm;

Combining the magnetic steel wheelbase information obtained by controlling the industrial computer unit with the original image of the truck, the position of the axle box spring can be roughly estimated from the prior knowledge, and the large image of the axle box spring can be obtained, and the large image of the axle box spring includes a A corresponding height image and grayscale image.

Other steps and parameters are the same as in the first embodiment.

Embodiment 3: The difference between this embodiment and Embodiment 1 or 2 is that the step 2 corrects the height image in the large image of the axle box spring; the specific process is:

The side image of the truck is taken by a camera with an elevation angle, so the depth information of the same plane of the 3D image obtained at first is not the same. For example, the 3D image of the bearing is not an ideal cylinder, but the depth information value of the part far from the ground is relatively high. Small, the depth information value of the part close to the ground is larger;

For objects on the same plane, the ideal depth information value should be the same, but initially in the depth image, it is different in the images of different lines; there is a certain disturbance in the depth information, but generally the actual depth information is the same as the original depth information. The relationship between the difference value and the number of rows in the image is equivalent to the two right-angled sides of a right-angled triangle; the depth information of the current x row position is y, and the actual depth information is yr. The corresponding relationship is shown in Figure 3, and the formula is:

(yr-y) = k(x-x0) (1)

Among them, k represents the tangent value of the inclination angle; x0 represents the number of lines in the image where the depth value in the height image of the object on the same plane reaches the maximum value; the values of k and x0 are calculated and fitted through a large amount of collected data; Select the place where the actual object of the bogie side frame is the same plane and the size of the parts is large to collect data;

Since the truck itself is moving during the driving failure, the distance between each vehicle and the 3D high-definition imaging device varies. Therefore, the depth images of different vehicle models, different sites and passing vehicles at different times are collected to provide sufficient depth image correction. data;

For the large image of the axle box spring, yr is calculated by formula (1), and the correction of the height image is completed; at this time, the objects in the same plane are basically the same value, and the 3D image of the bearing is roughly a cylinder.

Other steps and parameters are the same as in the first or second embodiment.

Embodiment 4: The difference between this embodiment and one of Embodiments 1 to 3 is that the step 3 obtains the overall height sub-image and the gray-scale sub-image of the axle box spring respectively; the specific process is:

For the corrected height image, since the distance between the bearing and the camera and the distance between the axle box spring and the camera are similar, the height information value also uses similar prior knowledge to determine whether there is an axle box spring part in the current image;

If the components in the height image near the bearing height only have bearings, the current image does not contain the axle box spring components, so it is not in the recognition range; if the components in the height image near the bearing height include bearings and axle box springs , then follow-up identification is required; first perform global binarization on the axle box spring with the threshold th1; then find the connected area in the binarized image whose area is greater than the threshold th2; if more than one connected area is found, it is the axis containing the axis For models with box springs, follow-up identification is performed; if only one connected area is detected, it is a model that does not include axle box springs, and subsequent identification is not performed;

In the global binarized image, the connected area with the intermediate bearing removed is the position of the two axle box springs; through the position of the connected area, the overall height sub-image and the gray sub-image of the axle box spring are intercepted respectively; the head ring of the axle box spring is intercepted according to the proportion The height sub-image and the gray-scale sub-image of the upper adjacent part; the height sub-image and the gray-scale sub-image of the axle box spring tail ring and the lower adjacent part are intercepted according to the scale.

Other steps and parameters are the same as one of the first to third embodiments.

Embodiment 5: The difference between this embodiment and one of Embodiments 1 to 4 is that the step 4 determines and identifies the failure type of the axle box spring; the specific process is:

Axle box spring breakout is divided into three cases: head ring breakout, tail ring breakout and left and right tilt breakout; axle box spring breakage is divided into head ring breakage, middle layer spring breakage and crack-type breakage; axle box spring head ring breakout The exit and tail ring escape belong to the end spring, and both adopt the same identification method; the left and right inclined escapes are identified by other methods; the axle box spring is broken by two identification methods: head ring break and middle layer break; If the head ring escapes or is broken, the subsequent identification is not performed; if the tail ring is identified, the subsequent identification is not performed; further identification is performed on the left and right side of the axle box spring and the middle breakage;

Step 41. Template database establishment:

Collect grayscale images of axlebox springs at the end of different detection stations and adjacent components; perform multi-scale image fusion on the collected images, as normal and fault-free images, the fused images have more and more valuable information;

Step 42. Identification of the end protruding and breaking faults

Under normal circumstances, the positional relationship between the head ring of the axle box spring and the upper adjacent part is closely attached. After the head ring of the axle box spring is broken or jumped out, the head coil spring and the upper adjacent part are no longer closely attached, and the edge (in the image) The number of light and dark junctions) is more than the normal non-faulty image, and the texture is richer. By comparing the gray sub-image of the axle box spring head ring and the upper adjacent parts with the normal non-faulty image, it can be identified whether there is a fault; normal The positional relationship between the tail ring of the axle box spring and the lower adjacent part is that the tail ring is stuck in the groove of the lower adjacent part, and the tail ring of the axle box spring will ride on the lower adjacent part after it escapes. The image can be identified by comparing the image with the normal non-faulty image.

Match the grayscale sub-images of the axle box spring and adjacent parts at the current end with the normal non-faulty image in the previous step for similarity. When the similarity of the detected image is lower than a certain threshold, a fault alarm will be issued.

Step 43. Identify the fault of the middle layer spring

Identify whether the spring has a fault that the spring is inclined left and right by the overall inclination of the spring. If the fault has not been detected so far, the middle layer of the axle box spring is broken and the fault is identified;

The fault that the axle box spring is inclined to the left and right is identified by calculating the inclination of the external rectangle by the height image, and the OTSU binarization process is performed on the overall height sub-image of the axle box spring, and then the spring gap between adjacent springs is eliminated by the morphological opening operation. , and finally detect the largest connected area in the image; when the inclination angle of the smallest circumscribed rectangle of the largest connected area is greater than k, it is determined that the axle box spring has a left and right tilting fault;

For the fracture of the middle layer of the axle box spring, it can be identified by further analysis of each layer of spring; first, each layer of spring is segmented. This segmentation method can divide each layer of normal image spring into multiple small areas whose height is smaller than the normal spring diameter , if there is a connected area with a height greater than the height of a circle of springs, it proves that there is a broken fault of up and down dislocation, and a fault alarm is issued; if no fault is detected, but the width of the connected area is detected to be less than 80% of the normal width, it proves that there is There is a crack-type break in the spring, and a fault alarm is performed.

Embodiment 6. The difference between this embodiment and one of Embodiments 1 to 5 is that the segmentation method includes:

Get a spring mask image:

Perform global binarization on the original height image of the spring in the current column (the axle box spring has two columns of springs, after positioning each column (bar) sub-image, the subsequent fault detection of the axle box spring in each column should be performed separately), gray value If it is greater than th1, it is 255, and if it is less than th1, it is 0, and the mask image can be obtained;

Get a spring filtered image:

Assign all pixel values of the filtered image to 0, and the size is the same as the original image; for the (i, j) position, if it is not 0 in the mask image, perform the following operations: take (i, j) as the center, calculate The length of the original image is W and the width is H in the rectangular area (only the non-zero pixel position in the mask image is considered) as the pixel value in the filtered image;

Get a spring segmented image:

Set all the pixels of the segmented image to 0, and the size is the same as the original image; for the position (i, j), when the pixel in the mask image is non-0, perform the following operations: in the filtered image (i-h1, j) and The pixel values at the position (i+h1, j) are all different from (i, j). When the absolute value of the difference is greater than th, the pixel value of the segmented image is 255.

Claims (8)

1. A method for automatically detecting the failure of a freight car axle box spring is characterized by comprising the following steps:

the method comprises the following steps that firstly, after a truck passes through imaging equipment, a large image of the axle box spring is obtained, wherein the large image of the axle box spring comprises a height image and a gray level image;

step two, correcting a height image in a large image of the shaft box spring;

respectively acquiring an integral height sub-image and a gray sub-image of the axle box spring;

judging the fault type of the journal box spring according to the obtained gray level image;

and step five, after the fault is identified, calculating the position of the fault in the original image through the mapping relation between the sub-image and the axle box spring large image and between the axle box spring large image and the original image, and displaying the fault through a fault display platform.

2. The method for automatically detecting the axle box spring fault of the freight car according to claim 1, wherein in the first step, imaging equipment is built around a track, and after the freight car passes through the equipment, a large image of the axle box spring is obtained and comprises a height image and a gray level image; the specific process is as follows:

combining magnetic steel wheel base information obtained by processing of imaging equipment with an original image of a truck, estimating the position of an axle box spring component by priori knowledge to obtain a large image of the axle box spring, wherein the large image of the axle box spring comprises a height image and a gray image which are in one-to-one correspondence.

3. The method according to claim 2, wherein the step of correcting the height image in the large journal spring image; the specific process is as follows:

shooting a truck side image through a camera to obtain depth information of a 3D image, taking a relation between actual depth information and an initial depth information difference value and a line number in the image as a relation between two right-angle sides of a right-angle triangle, taking the depth information of a current x-line position as y, taking the actual depth information as yr, and taking a corresponding relation formula as follows:

(yr-y)＝k(x-x0) (1)

wherein k represents a dip tangent value; x0 represents the number of lines in the image of the same plane object when the depth value in the height image takes the maximum value; after the calculation of the formula (1), yr is obtained, and the correction of the height image is completed.

4. The automatic failure detection method for the axle box spring of the freight car according to claim 3, wherein the step three of obtaining the whole height sub-image and the gray sub-image of the axle box spring respectively; the specific process is as follows:

if only the bearing is arranged in the bearing height range in the height image, the image does not contain the axle box spring component and is not in the identification range; if the parts in the height image within the range near the bearing height are provided with the bearing and the axle box spring, the axle box spring is globally binarized by a threshold th 1; then searching a connected region with the area larger than a threshold th2 in the binary image; if more than 1 communication area is found, the vehicle type containing the axle box spring is identified, and subsequent identification is carried out; if only one connected region is detected, subsequent identification is not carried out;

removing the communication area of the middle bearing in the global binary image to obtain the positions of the two journal box springs; respectively intercepting an axle box spring integral height sub-image and a gray level sub-image through the position of the communication area; intercepting a height sub-image and a gray sub-image of the journal box spring head ring and the upper adjacent component according to the proportion; and (4) intercepting a height sub-image and a gray level sub-image of the tail coil of the journal box spring and the lower adjacent component according to the proportion.

5. The method according to claim 4, wherein the fourth step is to determine and identify the type of axle box spring failure; the specific process is as follows:

the shaft box spring play is divided into head ring play, tail ring play and left and right inclined play; the axle box spring is broken into head ring breaking, middle layer spring breaking and crack breaking;

when the head ring is identified to be broken or broken, the subsequent fault identification is not carried out; if the tail ring is identified to be shifted out, subsequent identification is not carried out; further identifying the left and right fleeing of the shaft box spring and the middle breaking of the shaft box spring;

fourthly, acquiring grayscale images of the axle box springs at the end parts of different detection stations and adjacent parts, and carrying out image fusion on the acquired images to obtain normal fault-free images;

step four, identifying the end fleeing and breaking faults:

the position relation of the head ring of the axle box spring and the upper adjacent part is in a normal condition of close attachment, when the head ring of the axle box spring is broken off or is shifted out, the head ring spring is not closely attached to the upper adjacent part any more, the edge is more than normal and fault-free images, and whether a fault occurs or not is judged by comparing the grey level sub-images of the head ring of the axle box spring and the upper adjacent part with the normal and fault-free images; the position relation of the tail ring of the axle box spring and the lower adjacent part is that the tail ring is normally clamped in a groove of the lower adjacent part, the tail ring of the axle box spring can ride on the lower adjacent part after coming out, and whether a fault occurs or not is judged by comparing gray level sub-images of the tail ring of the axle box spring and the lower adjacent part with normal fault-free images;

matching the similarity of the gray level subgraphs of the current end part axle box spring and the adjacent parts with a normal fault-free image, and performing fault alarm when the detected image similarity is lower than a certain threshold value;

step four and step three, identifying faults of the middle layer spring:

identifying whether the spring has a fault of left-right inclined jumping or not by the integral inclination degree of the spring, and identifying the break fault of the middle layer of the axle box spring if the fault is not detected;

identifying the left and right inclined fleeing faults of the axle box spring by calculating the gradient of an external rectangle through a height image, carrying out OTSU binaryzation processing on the whole height subimage of the axle box spring, eliminating the spring gap between adjacent springs through morphological opening operation processing, and finally detecting the largest communication area in the image; when the inclination angle of the minimum circumscribed rectangle of the maximum communication area is larger than k, the journal box spring is judged to have a fault of jumping out in a left-right inclined mode;

for the breakage of the middle layer of the axle box spring, each layer of spring in a normal image is divided into a plurality of small areas with the height smaller than the diameter of the normal spring by dividing each layer of spring, if a communicated area with the height larger than the height of one circle of spring appears, the breakage fault of up-down dislocation is proved to occur, and fault alarm is carried out; if the fault is not detected yet, but the width of the communication area is detected to be smaller than the normal width by a certain proportion, the existence of crack type fracture of the spring is proved, and fault alarm is carried out.

6. The automatic failure detection method for the axle box spring of the freight car according to claim 5, wherein the dividing method comprises:

(1) acquiring a spring mask image:

globally binarizing the spring original height image of the current column, wherein the gray value is 255 when the gray value is larger than th1, and 0 when the gray value is smaller than or equal to th1 to obtain a mask image;

(2) acquiring a spring filtering image:

assigning all pixel values of the filtered image to be 0, wherein the pixel values are the same as those of the original image; if the (i, j) position is not 0 in the mask image, the following operations are performed: taking (i, j) as a center, calculating a brightness mean value in a rectangular region with the original image length of W and the width of H (only non-0 pixel position in the mask image is considered) as a pixel value in the filtering image;

(3) acquiring a spring segmentation image:

setting all the pixels of the segmented image to be 0, wherein the size of the pixels is the same as that of the original image; for the (i, j) position, when the pixel is not 0 in the mask image, the following operations are performed: the pixel values at the (i-h1, j) and (i + h1, j) positions in the filtered image are all subtracted from (i, j), and when the sum of the absolute values of the differences is larger than th, the pixel value of the segmented image is 255.

7. The method for automatically detecting the axle box spring fault of the freight car according to claim 1, wherein an imaging device is built around the track, and the imaging device is a high-definition 3D imaging device.

8. The method according to claim 5, wherein the width of the communicating region is smaller than the normal width by a ratio of 80% of the normal width.

Images