CN111311591A - Method for detecting lifting amount of high-speed railway contact net - Google Patents
Method for detecting lifting amount of high-speed railway contact net Download PDFInfo
- Publication number
- CN111311591A CN111311591A CN202010153910.9A CN202010153910A CN111311591A CN 111311591 A CN111311591 A CN 111311591A CN 202010153910 A CN202010153910 A CN 202010153910A CN 111311591 A CN111311591 A CN 111311591A
- Authority
- CN
- China
- Prior art keywords
- speed railway
- lifting amount
- contact net
- similarity
- detecting
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Abstract
The invention discloses a method for detecting the lifting amount of a high-speed railway contact net, which comprises the following steps: step 1: continuously acquiring vibration images of a high-speed railway contact net to form a sample set; step 2: constructing a similarity neural network, inputting a sample set into the similarity neural network, performing convolution processing on data and a detection object in the sample set, and evaluating the similarity between the data and the detection object; and step 3: taking the data in the sample set as samples, and training the similarity neural network; and 4, step 4: inputting the data sample into a similarity neural network to obtain the lifting amount of the contact net; the method can be used for acquiring the lifting amount of the contact net system of the in-service high-speed railway, the non-contact detection is realized by adopting the high-speed linear array camera as an acquisition element so as to avoid the influence on the normal operation of the high-speed railway motor train unit, the robustness of the adopted detection method is better, the method can be used for detecting the lifting amount of the contact net of the high-speed railway under different operation environments, and the accuracy of the acquisition result is improved.
Description
Technical Field
The invention relates to the technical field of high-speed railway contact networks, in particular to a method for detecting the lifting amount of a high-speed railway contact network.
Background
In an electrified railway, an electric locomotive is mainly in sliding contact with an overhead contact network by means of a pantograph arranged at the top of the locomotive to realize current taking. The contact network system is used as an important power supply device of a high-speed railway, and the dynamic characteristics of the contact network system directly influence the operation stability and safety of an electric locomotive. The high-speed railway contact network system has the characteristics of open-air arrangement, large span and high flexibility, has very complex dynamic characteristics under the action of external excitations such as moving pantographs and environmental wind loads, and has influence on good contact between pantograph nets. Especially, with the improvement of the running speed of the high-speed railway in China in recent years, the research on the dynamic characteristics of the contact net system in a complex environment becomes an important link for ensuring the safe running of the pantograph net system.
At present, most of domestic and foreign researches on the dynamics of the high-speed railway bow net system are focused on the aspects of model updating, simulation algorithm improvement or simulation phenomenon analysis. For example, Song Y and the like adopt a nonlinear cable-rod bow net coupling model to study the dynamic behavior of a bow net and the dynamic response of a bow net system under the excitation of environmental wind; gregori S and the like provide a nonlinear fast integral algorithm based on a finite element model of a pantograph system to solve the problem of large dynamic simulation calculation amount of the pantograph system; wang H and the like provide a method for extracting and identifying the structural wavelength of the overhead contact system based on a time-frequency domain analysis theory aiming at the phenomenon of the irregularity of the overhead contact system of the high-speed railway. Although the dynamic response of the overhead line system under different external stimuli can be obtained through numerical simulation by the model simulation, a theoretical basis is provided for the design of the overhead line system of the high-speed railway. However, in consideration of balance simulation precision and calculation amount, a certain difference still exists between a simulation result of the bow net system model and an actual dynamic response. Therefore, the actual dynamic response of the high-speed railway pantograph-catenary system is obtained in a field measurement mode, which is not only the requirement for correcting the simulation result of the model, but also the objective requirement for deepening the dynamic theoretical research of the high-speed railway pantograph-catenary system and perfecting the structural design of the high-speed railway pantograph-catenary system in China. Cho YH et al, PetterNavik et al, adopt contact acceleration sensors to measure and collect the lifting amount of the contact net, and analyze the dynamic characteristics of the contact net; li RP and the like are based on an edge detection method, and the lifting amount of the indoor contact net is measured in an image acquisition mode.
The feasibility of the above-mentioned dynamic behavior detection method of the contact network has been proved in the experiment, but the following problems still exist: firstly, the adopted contact sensor introduces extra concentrated quality points for a high-speed railway contact network system while acquiring data, and deviation between the detected data and actual response is possibly caused; secondly, the contact type measuring mode sensor is relatively complex to install, needs to be directly installed at a cable position of a contact network, and may influence the normal operation of the high-speed motor train unit; thirdly, the adopted non-contact detection method has poor robustness and cannot adapt to the detection of the lifting amount of the high-speed railway contact net under different operating environments.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method for detecting the lifting amount of a high-speed railway contact net, which can realize accurate detection of the lifting amount of the contact net in different operating environments and reduce measurement errors on the premise of ensuring non-contact detection.
The technical scheme adopted by the invention is as follows:
a method for detecting the lifting amount of a high-speed railway contact net comprises the following steps:
step 1: continuously acquiring vibration images of a high-speed railway contact net to form a sample set;
step 2: constructing a similarity neural network, inputting the sample set formed in the step 1 into the similarity neural network, performing convolution processing on data in the sample set and a detection object, and evaluating the similarity between the data and the detection object;
and step 3: taking the data in the sample set obtained in the step 1 as a sample, and training the similarity neural network in the step 2;
and 4, step 4: and (4) inputting the data sample into the similarity neural network obtained in the step (3) to obtain the lifting amount of the contact net.
Further, the similarity in step 2 is evaluated by the following formula:
in the formula: z is a detection object, and x is a detection sampleThe utility model relates to a novel water-saving device,
in order to adopt the neural network processing procedure,
in order to realize the purpose,
is to detect the similarity between the object z and the nth candidate window on the sample x.
Further, in the step 2, a response row with a constant number of channels is obtained by adopting depth separable convolution, and the response row is subjected to segmentation, regression and classification operations to generate a three-branch output structure.
Further, the loss function in the training process in step 3 is as follows:
L＝a·Lmask+b·Lscore+c·Lbox
in the formula: l ismask、Lscore、LboxThe loss functions of the mask branch, the score branch and the box branch are respectively expressed; a. b and c are weight coefficients.
Further, the process in the step 4 is as follows:
s11: continuously acquiring vibration images of the high-speed contact network;
s12: acquiring a first frame image, and selecting a target overhead contact system as a sample input in a network;
s13: searching and matching a target area;
s14: outputting an image recognition result of the contact network to obtain the lifting amount of the contact network;
s15: and judging whether the acquisition is finished or not, stopping the detection if the acquisition is the last frame, and returning to the step S13 if the acquisition is not finished.
Further, the step S13 is as follows:
inputting the next frame as an image to be searched in the network to obtain the predicted output of three branches of the Score, the Box and the Mask; and obtaining a predicted value of the target area by the Score branch, performing Mask output calculation at a corresponding position according to the predicted value, and obtaining an area to be searched corresponding to the next frame by the Box branch.
Further, in the step S14, a coordinate value corresponding to each frame of image is calculated and output as the lifting amount of the overhead line system.
The technical scheme adopted by the invention is as follows:
(1) according to the method, the samples obtained in real time are used, so that the influence of an external environment on the acquisition result is reduced, and the method is more suitable for the actual engineering environment of a contact net in the running process of a high-speed railway;
(2) the invention can record the time-course curve of the lifting amount change of the high-speed railway contact net under the excitation of a moving pantograph, environmental wind disturbance and the like, and provides data support for the theoretical research of the dynamic characteristics of the high-speed railway pantograph-catenary system.
Drawings
FIG. 1 is a structural framework diagram of the detection method of the present invention.
FIG. 2 is a flow chart of the lift detection in the present invention.
FIG. 3 shows the result of identifying the high-speed railway catenary under different conditions in the embodiment of the present invention; a is under the indoor illumination compensation condition, b and c are under the outdoor strong light condition, and d is under the outdoor weak light condition.
Fig. 4 is a curve of an actual time course of the lifting amount of the high-speed railway contact net in the embodiment of the invention.
Fig. 5 shows the lifting amount of the overhead contact system of the high-speed railway obtained by the simulation result of the method and the model provided by the invention on the first-order frequency.
Fig. 6 is a frequency distribution of the high-speed rail catenary lifting amount and the tested catenary line obtained by the method of the present invention in the embodiment of the present invention.
Detailed Description
The invention is further described with reference to the following figures and specific embodiments.
As shown in fig. 1, a method for detecting the lifting amount of a high-speed railway contact network comprises the following steps:
step 1: continuously acquiring vibration images of a high-speed railway contact net to form a sample set;
firstly, a device for detecting the lifting amount of a high-speed railway contact net is built, a high-speed linear array camera is used as an image acquisition element, and a set of equipment for acquiring the lifting amount of the high-speed railway contact net is built, wherein the equipment comprises the following parts: the system comprises a high-speed linear array camera and a matched optical lens (contact network image acquisition), an adjustable support (adjusting acquisition height and angle), a gigabit network card (acquiring data transmission) and a workstation (contact network image identification and lifting amount change time-course curve output).
Step 2: constructing a similarity neural network, inputting the sample set formed in the step 1 into the similarity neural network, performing convolution processing on data in the sample set and a detection object, and evaluating the similarity between the data and the detection object;
resnet _50 is used as a backbone structure of the neural network;
the similarity was evaluated using the following formula:
in the formula: z is a detection object, x is a detection sample,
in order to adopt the neural network processing procedure,
in order to realize the purpose,
is to detect the similarity between the object z and the nth candidate window on the sample x.
In order to reserve the information of the target object in the response graph to the maximum extent, depth-wise cross convolution depth separable convolution is adopted to obtain the response with the unchanged channel number, namely response of candidate window (row), and the segmentation, regression and classification operations are carried out on the response, so that a three-branch output structure comprising Score, Box and Mask is generated.
Wherein M isnIndicates Mask prediction value, m, associated with nth rowψRepresenting a two-layer neural network, psi is the learning parameter.
Box and Score behave identically, namely:
wherein, bε，sρThe parameters are related simple neural networks, and epsilon and rho are corresponding learning parameters.
And step 3: taking the data in the sample set obtained in the step 1 as a sample, and training the similarity neural network in the step 2;
and (3) taking the image acquired in the step (1) as a sample, and training in the similarity neural network model established in the step (2). In the training process, a binary label y is usednLabel the nth row by e { + -1 } to
Marking the mask pixel position (i, j) corresponding to the nth row, and then regarding the positive row, the loss function L ismaskCan be expressed in the following form:
in the formula: wh is cnThe size of (a) is (b),
the pixel coordinate position (i, j) of the Mask predicted value corresponding to the nth row;
Lscoreand LboxThe same as above;
the penalty function for Box branches is:
in the formula: delta (i) is the normalized distance, sigma is smooth L1A loss function parameter.
The loss function for the Score branch is as follows:
Lscore＝-log(pu)
in the formula, puThe probability corresponding to the real classification u can be calculated by softmax of the full connection layer.
For the proposed method for detecting the lifting amount of the three-branch high-speed railway contact net, the loss function in the training process is expressed in the following form:
L＝a·Lmask+b·Lscore+c·Lbox
in the formula: l ismask、Lscore、LboxThe loss functions of the mask branch, the score branch and the box branch are respectively expressed; a. b and c are weight coefficients.
And 4, step 4: and (4) inputting the data sample into the similarity neural network obtained in the step (3) to obtain the lifting amount of the contact net.
And respectively detecting the lifting amount of the contact net under the indoor illumination compensation condition, the outdoor strong light condition and the outdoor weak light condition by taking the standard contact line and the actual contact net line established according to the TB 10621-2014 standard as detection objects. The specific detection steps are described as follows:
s11: continuously acquiring vibration images of the high-speed contact network;
assuming that the device for detecting the lifting amount of the high-speed railway contact net described in the step 1 adjusts an optical lens to obtain a clear contact line image, records acquisition parameters including an acquisition distance, an acquisition angle and the like, and starts to continuously acquire vibration images of the high-speed railway contact net.
S12: acquiring a first frame image, and selecting a target overhead contact system as a sample input in a network;
in the workstation, the target catenary is framed in the first frame image using a rectangular frame, which is input as a sample in the catenary.
S13: searching and matching a target area;
inputting the next frame as an image to be searched in the network to obtain the predicted output of three branches of the Score, the Box and the Mask; and obtaining a predicted value of the target area by the Score branch, performing Mask output calculation at a corresponding position according to the predicted value, and obtaining an area to be searched corresponding to the next frame by the Box branch.
S14: outputting an image recognition result of the contact network to obtain the lifting amount of the contact network; and calculating and outputting a coordinate value corresponding to each frame of image as the lifting amount of the contact net.
S15: and judging whether the acquisition is finished or not, stopping the detection if the acquisition is the last frame, and returning to the step S13 if the acquisition is not finished.
The standard contact line and the actual contact line established according to the TB 10621-2014 standard are used as detection objects, and the method and the device for detecting the lifting amount of the high-speed railway contact line are used for verifying the performance. Firstly, the method provided by the invention is used for effectively verifying the detection result of the lifting amount of the high-speed railway overhead line system under different illumination conditions and different excitation conditions. As shown in fig. 3, under different conditions, the method of the present invention can achieve accurate identification of the images of the overhead contact system, and has good adaptability to changes in the geometrical shape of the clue image caused by different excitations during the dynamic acquisition of the lifting amount, and changes in the gray value of the clue image caused by changes in the illumination conditions. Secondly, the accuracy of the method for detecting the lifting amount of the high-speed railway contact net is verified, as shown in fig. 4-6, and as can be seen from fig. 5, the frequency distribution of the lifting amount of the contact net obtained by the method is basically consistent with the theoretical analysis value of the measured line. The first order frequency component of the catenary lifting obtained by the method is basically consistent with the theoretical analysis value of the measured line.
By verifying the performance of the method in an actual line, the detection method can be used for detecting the lifting amount of the high-speed railway contact net, and has better robustness; the influence of the detection process on the normal operation of the high-speed railway motor train unit can be avoided, the method is suitable for different operation environments of the high-speed railway contact network, and the accuracy of the detection result is improved.
Claims (7)
1. A method for detecting the lifting amount of a high-speed railway contact net is characterized by comprising the following steps:
step 1: continuously acquiring vibration images of a high-speed railway contact net to form a sample set;
step 2: constructing a similarity neural network, inputting the sample set formed in the step 1 into the similarity neural network, performing convolution processing on data in the sample set and a detection object, and evaluating the similarity between the data and the detection object;
and step 3: taking the data in the sample set obtained in the step 1 as a sample, and training the similarity neural network in the step 2;
and 4, step 4: and (4) inputting the data sample into the similarity neural network obtained in the step (3) to obtain the lifting amount of the contact net.
2. The method for detecting the lifting amount of the overhead line system of the high-speed railway according to claim 1, wherein the similarity in the step 2 is evaluated by adopting the following formula:
in the formula: z is a detection object, x is a detection sample,
in order to adopt the neural network processing procedure,
in order to realize the purpose,
is to detect the similarity between the object z and the nth candidate window on the sample x.
3. The method for detecting the lifting amount of the high-speed railway contact net according to claim 1, wherein in the step 2, a response row with a constant channel number is obtained by adopting a depth separable convolution, and is subjected to segmentation, regression and classification operations to generate a three-branch output structure.
4. The method for detecting the lifting amount of the overhead line system of the high-speed railway according to claim 3, wherein the loss function in the training process in the step 3 is as follows:
L＝a·Lmask+b·Lscore+c·Lbox
in the formula: l ismask、Lscore、LboxThe loss functions of the mask branch, the score branch and the box branch are respectively expressed; a. b and c are weight coefficients;
wherein: l ismaskThe following were used:
in the formula: y isnE { + -1 } is a binary label,
wh is cnThe size of (a) is (b),
the pixel coordinate position (i, j) of the Mask predicted value corresponding to the nth row;
Lboxthe following were used:
in the formula: delta (i) is the normalized distance, sigma is smooth L1A loss function parameter;
Lscorethe following were used:
Lscore＝-log(pu)
in the formula, puThe probability corresponding to the real classification u.
5. The method for detecting the lifting amount of the high-speed railway contact net according to claim 4, wherein the process in the step 4 is as follows:
s11: continuously acquiring vibration images of the high-speed contact network;
s12: acquiring a first frame image, and selecting a target overhead contact system as a sample input in a network;
s13: searching and matching a target area;
s14: outputting an image recognition result of the contact network to obtain the lifting amount of the contact network;
s15: and judging whether the acquisition is finished or not, stopping the detection if the acquisition is the last frame, and returning to the step S13 if the acquisition is not finished.
6. The method for detecting the lifting amount of the overhead line system of the high-speed railway according to claim 5, wherein the step S13 comprises the following steps:
inputting the next frame as an image to be searched in the network to obtain the predicted output of three branches of the Score, the Box and the Mask; and obtaining a predicted value of the target area by the Score branch, performing Mask output calculation at a corresponding position according to the predicted value, and obtaining an area to be searched corresponding to the next frame by the Box branch.
7. The method for detecting the lifting amount of the overhead line system of the high-speed railway according to claim 6, wherein the coordinate value corresponding to each frame of image is calculated in the step S14 and is output as the lifting amount of the overhead line system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010153910.9A CN111311591A (en) | 2020-03-07 | 2020-03-07 | Method for detecting lifting amount of high-speed railway contact net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010153910.9A CN111311591A (en) | 2020-03-07 | 2020-03-07 | Method for detecting lifting amount of high-speed railway contact net |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111311591A true CN111311591A (en) | 2020-06-19 |
Family
ID=71148140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010153910.9A Pending CN111311591A (en) | 2020-03-07 | 2020-03-07 | Method for detecting lifting amount of high-speed railway contact net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111311591A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113028999A (en) * | 2021-02-24 | 2021-06-25 | 河南辉煌科技股份有限公司 | Contact line lift measurement method and system based on convolutional neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201555658U (en) * | 2009-12-15 | 2010-08-18 | 中国铁道科学研究院机车车辆研究所 | Railway contact net vibration performance test device |
| US20170131337A1 (en) * | 2015-11-09 | 2017-05-11 | Alstom Transport Technologies | Method and system for monitoring a pantograph of a railway vehicle and railway vehicle |
| CN107437245A (en) * | 2017-06-26 | 2017-12-05 | 西南交通大学 | High-speed railway touching net method for diagnosing faults based on depth convolutional neural networks |
| CN110569841A (en) * | 2019-09-02 | 2019-12-13 | 中南大学 | contact gateway key component target detection method based on convolutional neural network |
-
2020
- 2020-03-07 CN CN202010153910.9A patent/CN111311591A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN201555658U (en) * | 2009-12-15 | 2010-08-18 | 中国铁道科学研究院机车车辆研究所 | Railway contact net vibration performance test device |
| US20170131337A1 (en) * | 2015-11-09 | 2017-05-11 | Alstom Transport Technologies | Method and system for monitoring a pantograph of a railway vehicle and railway vehicle |
| CN107437245A (en) * | 2017-06-26 | 2017-12-05 | 西南交通大学 | High-speed railway touching net method for diagnosing faults based on depth convolutional neural networks |
| CN110569841A (en) * | 2019-09-02 | 2019-12-13 | 中南大学 | contact gateway key component target detection method based on convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| QIANG WANG 等: ""Fast Online Object Tracking and Segmentation: A Unifying Approach"", 《HTTPS://ARXIV.ORG/ABS/1812.05050》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113028999A (en) * | 2021-02-24 | 2021-06-25 | 河南辉煌科技股份有限公司 | Contact line lift measurement method and system based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Advances of research on high-speed railway catenary | |
| CN109324604B (en) | Intelligent train comprehensive fault analysis method based on multi-source signals | |
| CN111144039B (en) | Train dynamic weighing system and weighing method based on deep learning | |
| CN102854191B (en) | Real-time visual detection and identification method for high speed rail surface defect | |
| CN111444940A (en) | Fault diagnosis method for critical parts of fan | |
| CN110533229A (en) | Orbital maintenance moment prediction technique and device | |
| CN110120036A (en) | A kind of multiple dimensioned tire X-ray defect detection method | |
| CN105335692A (en) | Tire X-ray image detection and identification method and system | |
| CN103745238B (en) | Pantograph identification method based on AdaBoost and active shape model | |
| CN109917213B (en) | Contact network detection fault early warning method based on dimensionality reduction fusion and factor analysis | |
| CN107833211B (en) | Infrared image-based zero value insulator automatic detection method and device | |
| CN111899216A (en) | Abnormity detection method for insulator fastener of high-speed rail contact network | |
| CN110427943A (en) | A kind of intelligent electric meter technique for partitioning based on R-CNN | |
| CN109360188A (en) | A kind of detection method of high iron catenary equipotential line defective mode | |
| CN110568082A (en) | cable wire breakage distinguishing method based on acoustic emission signals | |
| CN110909657A (en) | Method for identifying apparent tunnel disease image | |
| CN108508023B (en) | Defect detection system for contact end jacking bolt in railway contact network | |
| CN114757925A (en) | Non-contact type high-voltage circuit breaker defect detection method and system | |
| CN111311591A (en) | Method for detecting lifting amount of high-speed railway contact net | |
| CN110793501A (en) | Subway tunnel clearance detection method | |
| CN108765393B (en) | Method for detecting vibration behavior of high-speed railway contact net | |
| CN116881530B (en) | Device surface defect detection system based on deep learning | |
| CN112950562A (en) | Fastener detection algorithm based on line structured light | |
| CN112362214A (en) | Method and system for online identification of belt tension | |
| CN112132088B (en) | Inspection point missing inspection identification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200619 |