Image detection method for truck brake chain tensioning fault

Abstract

The invention discloses an image detection method for a wagon brake chain tensioning fault, and belongs to the technical field of railway wagon fault detection, and the method comprises the steps: carrying out the instance segmentation of a brake chain part through a trained target segmentation model YOLO11-small, obtaining a brake chain segmentation image, carrying out the skeleton extraction of a segmented region, carrying out the quadratic curve fitting, and carrying out the segmentation of a brake chain part. And comparing the calculated and fitted curve parameter with a corresponding threshold value, so as to judge a result. According to the invention, the adopted YOLO11-small network model can automatically detect the brake chain part, the process is a process without personnel participation and is an automatic identification and determination process, and after the brake chain is accurately determined, the judgment on whether the brake chain has a tensioning fault is also an automatic identification and detection process, so that the detection efficiency is greatly improved. Therefore, the risk of missing detection due to visual fatigue caused by human factors is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of railway freight car fault detection, and in particular to an image detection method for freight car brake chain tensioning faults. Background Art

[0002] To ensure the safety of railway freight trains, railway stations are equipped with on-site train operation status detection equipment. This system can capture images of the side of the train through a high-speed camera array. Its ultimate goal is to detect and identify component faults in the acquired images through pattern recognition technology. After digital processing, the fault information is displayed on the information terminal of the detection center so that relevant train inspection personnel can conduct train inspections and maintenance based on the displayed fault information. Because a tightened freight train brake chain poses a safety hazard to driving, the current method of simply relying on the naked eye of train inspection personnel to determine whether the fault exists can easily cause visual fatigue, and thus easily lead to false detection and missed detection problems. Therefore, based on this system, an automatic detection method for freight train brake chain tightening faults is proposed to reduce the workload of train inspection personnel, improve the detection effect of faults, and ensure the safe operation of freight trains. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide an image detection method for truck brake chain tensioning failure, so as to solve the problem that the existing brake chain tensioning failure requires manual identification and judgment, which easily causes visual fatigue and thus easily leads to false detection and missed detection.

[0004] To achieve the above objectives, the method of the present invention uses a deep learning target segmentation network and a series of post-processing methods to determine whether the brake chain is tightened, thereby achieving the purpose of detecting the tension state of the brake chain during the operation of railway freight cars. Specifically, the method includes the following steps: 1) Capturing an image of the running gear of a railway freight car and segmenting the image using the trained object segmentation model YOLO11-small, so as to obtain a segmented image of the brake chain area when a brake chain is present in the running gear image; 2) Perform skeleton extraction on the segmented area at the brake chain area in the segmented image to obtain a skeleton image; 3) Perform quadratic curve fitting on the skeleton image to obtain a curve after quadratic curve fitting. Based on the parameters of the curve, determine whether there is a brake chain tensioning fault on the truck.

[0005] Its beneficial effects are as follows: the present invention uses the trained target segmentation model YOLO11-small to first perform target segmentation on the image to be detected, and can accurately determine whether there are brake chain components. If there are brake chain components, the segmentation result of the brake chain components can be obtained through the YOLO11-small network model, that is, the brake chain area can be accurately located, and the brake chain components can be accurately segmented. Therefore, the YOLO11-small network model used in the present invention can automatically detect the brake chain components, and this process does not require human participation. It is an automatic identification and determination process, thus avoiding visual fatigue caused by human factors and the risk of missed detection. After accurately determining the brake chain, the determination of whether the brake chain has a tensioning fault is also an automated identification and detection process. In the present invention, skeleton extraction of the detected brake chain components is performed to obtain a skeleton image including the brake chain components. The skeleton image of the brake chain is then used to perform quadratic curve fitting to obtain a curve diagram of the brake chain. This method is used in the present invention to avoid the influence of other areas in the image on the judgment process. Therefore, through the method steps of the present invention, only the curve diagram of the brake chain is retained before the final fault judgment. Then, the calculation of the specific parameters of the curve can accurately reflect the posture of the brake chain, and then the final result of whether the brake chain has a tensioning fault can be obtained based on the parameters. In summary, the present invention determines whether the brake chain is tensioned by using the target segmentation network in deep learning and a series of post-processing methods. This process is an automatic judgment process that does not require human participation and can accurately identify whether there is a brake chain tensioning fault. Therefore, the method of the present invention effectively avoids the need for manual identification and judgment of brake chain tensioning faults, which can easily cause visual fatigue and thus easily lead to false detection and missed detection problems.

[0006] Furthermore, in step 2), extracting the skeleton of the segmented region in the segmented image includes: 2.1) Obtaining a mask image of the segmented area in the segmented image to focus on the brake chain components in the segmented area; 2.2) Refine the mask image to obtain the skeleton image of the brake chain component.

[0007] Furthermore, in step 2.1), the brake chain component in the segmented area of interest is set to white and the remaining parts are set to black, so as to obtain a mask image of the segmented area.

[0008] Furthermore, in step 2.2), the skeleton image of the brake chain component is obtained by thinning the white area in the mask image.

[0009] In order to further improve the accuracy of the detection results, the present invention uses a mask image to perform the skeleton extraction process. The mask image can only highlight the information of interest and shield the rest of the information, so that the subsequent skeleton extraction and quadratic curve fitting processes will not be affected by non-interest points, thereby further improving the accuracy of the detection results.

[0010] Furthermore, in step 3), the parameters of the curve refer to the parameters at the lowest point of the curve.

[0011] Furthermore, in step 3), the parameter of the curve is a curvature value; Determining whether there is a truck brake chain tensioning fault includes: if the curvature value is less than a set first threshold, then there is a truck brake chain tensioning fault.

[0012] Furthermore, in step 3), the parameter of the curve is the radius of curvature; Determining whether there is a truck brake chain tensioning fault includes: if the curvature radius is greater than a set second threshold, then there is a truck brake chain tensioning fault.

[0013] In the present invention, when determining whether there is a brake chain tensioning failure, it is taken into consideration that the state of the brake chain can be used to determine whether there is a brake chain tensioning failure. Due to the influence of gravity, the brake chain is in a relaxed state under normal conditions, that is, the degree of curvature at the lowest position of the brake chain is the largest. Based on this, it can be determined that if the degree of curvature at this position (that is, the position where the degree of curvature of the brake chain is the largest during use) is too small, there is a risk of brake chain tensioning failure, and the degree of curvature can be reflected by the curvature radius and the curvature value. The greater the degree of curvature, the smaller the curvature radius and the larger the curvature value. That is, when the curvature radius is too large or the curvature value is too small, there is a risk of brake chain tensioning failure. Therefore, in the present invention, by using the parameters at the lowest point of the curve, which can be either the curvature value or the curvature radius, and performing corresponding judgment logic, it is possible to accurately obtain the result of whether there is a brake chain tensioning failure.

[0014] Furthermore, in step 1), the training method of the trained object segmentation model YOLO11-small includes: Divide the images in the running gear image dataset into a training dataset, a verification dataset, and a test dataset; Iteratively train the model YOLO11-small using stochastic gradient descent; Set the learning rate to decrease in sequence, and perform multiple trainings at each learning rate. Each time the set number of trainings is reached, the intermediate network model parameter file is saved. The trained network model parameter file is tested using the test data set, and the network model with the smallest error is selected as the optimal model parameter file.

[0015] In the present invention, during the model training process, the network model parameter files of each stage are recorded so that the network model parameter files can be tested with a test data set in the end to select the most accurate model parameter file suitable for brake chain tensioning fault detection, and then the model parameter file is applied to the network model, thereby ensuring the accuracy of the trained network model in actual use.

[0016] The method of the present invention has the following advantages: The proposed brake chain tensioning fault identification method can achieve a high fault detection rate and accuracy during truck operation even when there are few fault samples. The algorithm is stable and robust, and can adapt to changes in various external environments.

[0017] The application of this fault identification method can greatly reduce the workload of manual labor, shorten the maintenance time of brake chain components, improve maintenance efficiency, and reduce the probability of missed inspections, thereby better ensuring the safe operation of trucks.

[0018] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, which can be implemented in accordance with the contents of the specification, and to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following specifically cites preferred embodiments and describes them in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of the image detection method for a truck brake chain tensioning fault according to the present invention; Figure 2 Schematic diagram of the training process of the segmentation model YOLO11-small of the present invention; Figure 3 It is a fault determination flow chart of the present invention; Figure 4a is the acquired image of the truck running gear; Figure 4b yes Figure 4a The segmented image obtained after the segmentation model YOLO11-small; Figure 5a yes Figure 4b The corresponding mask map; Figure 5b yes Figure 5b The skeleton image obtained after thinning. DETAILED DESCRIPTION

[0020] The technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments. However, it should be understood by those skilled in the art that the embodiments described below are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0021] Embodiment of an image detection method for a truck brake chain tensioning failure The tension of the brake chain of a railway freight car is a type of fault that seriously endangers driving safety. The present invention uses the target segmentation network in deep learning and a series of post-processing methods to determine whether the brake chain is tensioned, thereby achieving the purpose of detecting the tension status of the brake chain during the operation of the railway freight car.

[0022] The method for detecting brake chain tension on trucks of the present invention is divided into two parts: performing instance segmentation of brake chain components from the captured image; performing skeleton extraction on the segmented area, and then performing quadratic curve fitting. The curvature value / curvature radius of the calculated curve is compared with a threshold to determine the result.

[0023] like Figure 1 As shown, the present invention specifically includes the following steps: 1) Capture an image of the running gear of a railway freight car and segment the image using the trained object segmentation model YOLO11-small. When a brake chain is present in the running gear image, a segmented image is obtained that shows the location of the brake chain area.

[0024] like Figure 4a as well as Figure 4b As shown in the figure, the instance segmentation of the brake chain parts is performed. This step mainly uses the target segmentation model YOLO11-small to obtain the segmentation image of the brake chain segmentation.

[0025] Figure 4a is the collected image to be detected, Figure 4b The segmented chain area position can be further processed later. In addition, the YOLO11-small model used in this embodiment can achieve the same performance, but with a faster speed, thereby improving recognition efficiency.

[0026] 2) Perform skeleton extraction on the segmented area at the brake chain area in the segmented image to obtain a skeleton image.

[0027] In this embodiment, skeleton extraction of the segmented region in the segmented image includes: 2.1) Obtaining a mask image of the segmented area in the segmented image to focus on the brake chain components in the segmented area; 2.2) Refine the mask image to obtain the skeleton image of the brake chain component.

[0028] Among them, in the mask image, the brake chain parts in the segmented area are set to white, and the rest are set to black. The mask image obtained in this embodiment is directly obtained based on the trained target segmentation model YOLO11-small. Specifically, the segmented area is obtained by the segmentation branch structure obtained by the YOLO11-small model, which is obtained by threshold judgment after sigmoid calculation. That is, the trained target segmentation model YOLO11-small in this embodiment obtains the segmentation image by performing instance segmentation on the brake chain parts of the image to be detected, and then continues to obtain the mask image, and the final output result is the mask image. Figure 5a The mask image is shown as follows. After the mask image is obtained by segmenting the area, the mask image is refined to obtain a skeleton image. The skeleton image of this embodiment is obtained by image erosion and opening operations using the OpenCV open source library, as shown in FIG. Figure 5b shown.

[0029] To improve the accuracy of the final detection conclusion, this embodiment utilizes a mask image for skeleton extraction. The mask image highlights only the information of interest while masking the rest. This prevents the subsequent image processing and final judgment from being affected by non-interest points, further improving the accuracy of the detection results. Furthermore, the skeleton image obtained in this embodiment accurately reflects the posture of the brake chain component while presenting it in a simplified form. This ensures accuracy in subsequent judgments, while also improving efficiency due to the simplified posture representation.

[0030] 3) Perform quadratic curve fitting on the skeleton image to obtain a curve after quadratic curve fitting. Based on the parameters of the curve, determine whether there is a brake chain tensioning fault on the truck.

[0031] This step is the overall process of judging the brake chain tensioning fault. After obtaining the mask image from the segmented area, the mask image is refined to obtain the skeleton image. The skeleton image is fitted with a quadratic curve. After obtaining the corresponding curve parameters, the threshold is compared to determine whether the fault has occurred.

[0032] Since the skeleton image of this embodiment can accurately reflect the posture of the brake chain component, a brake chain tensioning failure means that the brake chain is in an approximately straight line state. At this time, the curve of the skeleton image corresponding to the brake chain posture has a small curvature, and thus a small curvature value and a large curvature radius. Therefore, based on the curvature value or the size of the curvature radius, it can be accurately determined whether the brake chain is in a tensioning failure state, that is, the larger the curvature, the greater the curvature, and the smaller the curvature radius, the greater the curvature of the curve.

[0033] Therefore, in this embodiment, a quadratic curve fitting is performed on the skeleton image to obtain the curvature radius or curvature value of the curve, and then compared with the corresponding threshold value to determine whether a brake chain tensioning fault exists.

[0034] When using curvature values for judgment, such as Figure 3 As shown, the judgment process is: A quadratic curve is fitted on the skeleton image to obtain a curve after quadratic curve fitting, and a curvature value of the curve is determined. If the curvature value is less than a set first threshold, there is a truck brake chain tensioning fault. If the curvature value is not less than the set first threshold, the truck brake chain component is in a normal state.

[0035] When using the curvature radius for judgment, the judgment process is: A quadratic curve is fitted on the skeleton image to obtain a curve after the quadratic curve fitting, and the curvature radius of the curve is determined. If the curvature radius is greater than a set second threshold, there is a truck brake chain tensioning fault. If the curvature value is not greater than the set second threshold, the truck brake chain component is in a normal state.

[0036] The first threshold value and the second threshold value of this embodiment may be values obtained based on experience, or may be safety critical values of a truck brake chain tensioning failure determined through experiments.

[0037] Furthermore, when obtaining the corresponding parameters of the curve, this embodiment takes into account that the curve reflects the posture of the brake chain component. Due to the influence of gravity, the posture of the brake chain component is in a relaxed state under normal conditions, that is, the degree of curvature is greatest at the lowest position of the brake chain. Based on this, it can be determined that if the degree of curvature at this position (that is, the position where the degree of curvature of the brake chain is greatest during use) is too small, there is a risk of a brake chain tensioning failure. Therefore, this embodiment uses the parameters at the lowest point of the curve, which can be either a curvature value or a curvature radius. After performing corresponding judgment logic, it can accurately determine whether there is a brake chain tensioning failure.

[0038] The present invention uses the trained target segmentation model YOLO11-small to first perform target segmentation on the image to be detected, and can accurately determine whether there is a brake chain component. If a brake chain component exists, the segmentation result of the brake chain component can be obtained through the YOLO11-small network model, that is, the brake chain area can be accurately located, and the brake chain component can be accurately segmented. Therefore, the YOLO11-small network model used in the present invention can automatically detect the brake chain component, and this process does not require human participation. It is an automatic identification and determination process, thus avoiding visual fatigue caused by human factors and the risk of missed detection. After accurately determining the brake chain, the determination of whether the brake chain has a tensioning fault is also an automated identification and detection process. In the present invention, skeleton extraction of the detected brake chain components is performed to obtain a skeleton image including the brake chain components. The skeleton image of the brake chain is then used to perform quadratic curve fitting to obtain a curve diagram of the brake chain. This method is used in the present invention to avoid the influence of other areas in the image on the judgment process. Therefore, through the method steps of the present invention, only the curve diagram of the brake chain is retained before the final fault judgment. Then, the calculation of the specific parameters of the curve can accurately reflect the posture of the brake chain, and then the final result of whether the brake chain has a tensioning fault can be obtained based on the parameters. In summary, the present invention determines whether the brake chain is tensioned by using the target segmentation network in deep learning and a series of post-processing methods. This process is an automatic judgment process that does not require human participation and can accurately identify whether there is a brake chain tensioning fault. Therefore, the method of the present invention effectively avoids the need for manual identification and judgment of brake chain tensioning faults, which can easily cause visual fatigue and thus easily lead to false detection and missed detection problems.

[0039] The training process of the object segmentation model YOLO11-small in this embodiment includes: The images in the walking part image dataset are divided into training dataset, validation dataset and test dataset; the YOLO11-small model is iteratively trained by stochastic gradient descent; the learning rate is set to decrease successively, and multiple trainings are performed at each learning rate. Whenever the set number of trainings is reached, the intermediate network model parameter file is saved once; the trained network model parameter file is tested on the test dataset, and the network model with the smallest error is selected as the optimal model parameter file.

[0040] like Figure 2As shown, the training of the model in this embodiment collects a large amount of running gear image data through a camera, and uses image enhancement to expand the original image data set. The original data and the expanded data constitute the entire data set. 3 / 5 of the data set is randomly extracted from the data set as the training data set, and 1 / 2 of the remaining data images are randomly extracted as the verification data set. Finally, the remaining images are used as the test data set. The positioning model (YOLO11-small) of the rolling bearing is iteratively trained by SGD (stochastic gradient descent). The initial learning rate is 0.001, and the training is repeated 40,000 times. The learning rate is then adjusted to 0.0001 and the training is repeated 5,000 times. The learning rate is then adjusted to 0.00001 and the training is repeated 5,000 times. The intermediate network model parameter file is saved every 1,000 training times. Finally, the trained network model parameter file is tested on the test data set, and the network model with the smallest error is selected as the optimal model parameter file.

[0041] In the process of training the model, this embodiment records the network model parameter files at each stage so that the network model parameter files can be tested with a test data set in the end to select the most accurate model parameter file suitable for brake chain tensioning fault detection. The model parameter file is then applied to the network model, ensuring the accuracy of the trained network model in actual use.

[0042] The method of the present invention has the following advantages: The proposed brake chain tensioning fault identification method can achieve a high fault detection rate and accuracy during truck operation even when there are few fault samples. The algorithm is stable and robust, and can adapt to changes in various external environments.

[0043] The application of this fault identification method can greatly reduce the workload of manual labor, shorten the maintenance time of brake chain components, improve maintenance efficiency, and reduce the probability of missed inspections, thereby better ensuring the safe operation of trucks.

[0044] Although the present invention has been described in detail above using general descriptions and specific embodiments, it will be apparent to those skilled in the art that modifications and improvements may be made thereto. Therefore, such modifications and improvements, without departing from the spirit of the present invention, are intended to be within the scope of protection claimed herein.

Claims

1. An image detection method for truck brake chain tensioning failure, characterized in that: The steps include: 1) Capturing an image of the running gear of a railway freight car and segmenting the image using the trained object segmentation model YOLO11-small, so as to obtain a segmented image of the brake chain area when a brake chain is present in the running gear image; 2) Perform skeleton extraction on the segmented area at the brake chain area in the segmented image to obtain a skeleton image; 3) Perform quadratic curve fitting on the skeleton image to obtain a curve after quadratic curve fitting. Based on the parameters of the curve, determine whether there is a brake chain tensioning fault on the truck.

2. The image detection method according to claim 1, wherein: In step 2), skeleton extraction of the segmented region in the segmented image includes: 2.1) Obtaining a mask image of the segmented area in the segmented image to focus on the brake chain components in the segmented area; 2.2) Refine the mask image to obtain the skeleton image of the brake chain component.

3. The image detection method according to claim 2, wherein: In step 2.1), the mask image of the segmented area is obtained by setting the brake chain components in the segmented area of interest to white and the remaining parts to black.

4. The image detection method according to claim 3, wherein: In step 2.2), the skeleton image of the brake chain component is obtained by refining the white area in the mask image.

5. The image detection method according to claim 1, wherein: In step 3), the parameters of the curve refer to the parameters at the lowest point of the curve.

6. The image detection method according to claim 5, characterized in that: In step 3), the parameter of the curve is the curvature value; Determining whether there is a truck brake chain tensioning fault includes: if the curvature value is less than a set first threshold, then there is a truck brake chain tensioning fault.

7. The image detection method according to claim 5, characterized in that: In step 3), the parameter of the curve is the curvature radius; Determining whether there is a truck brake chain tensioning fault includes: if the curvature radius is greater than a set second threshold, then there is a truck brake chain tensioning fault.

8. The image detection method according to claim 1, wherein: In step 1), the training method of the trained object segmentation model YOLO11-small includes: Divide the images in the running gear image dataset into a training dataset, a verification dataset, and a test dataset; Iteratively train the model YOLO11-small using stochastic gradient descent; Set the learning rate to decrease in sequence, and perform multiple trainings at each learning rate. Each time the set number of trainings is reached, the intermediate network model parameter file is saved. The trained network model parameter file is tested using the test data set, and the network model with the smallest error is selected as the optimal model parameter file.

Images