CN112233095A - Method for detecting multiple fault forms of railway wagon locking plate device - Google Patents
Method for detecting multiple fault forms of railway wagon locking plate device Download PDFInfo
- Publication number
- CN112233095A CN112233095A CN202011110722.4A CN202011110722A CN112233095A CN 112233095 A CN112233095 A CN 112233095A CN 202011110722 A CN202011110722 A CN 202011110722A CN 112233095 A CN112233095 A CN 112233095A
- Authority
- CN
- China
- Prior art keywords
- locking plate
- fault
- model
- segmentation model
- railway wagon
- 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.)
- Granted
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
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge 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]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
A method for detecting multiple fault modes of a railway wagon locking plate device belongs to the field of railway wagon operation. The invention aims to solve the problem that the locking plate device is easy to loosen and displace, so that the locking plate device cannot fix the cross rod and the driving safety is influenced. According to the invention, different vehicle images are collected by an image detection system built around a track to form a truck locking plate sample data set; establishing an improved YOLACT deep learning instance segmentation model; based on the acquired locking plate sample dataset and the improved Yolcat deep learning example segmentation model, the improved Yolcat deep learning example segmentation model is trained for multiple times, and the trained optimal model is used as a segmentation model of the locking plate component; and after the locking plate area is segmented by the example segmentation model, carrying out binarization on the locking plate area, and judging the specific fault type of the locking plate according to the segmented binarization result. The method is used for detecting the fault of the railway wagon locking plate.
Description
Technical Field
The invention relates to a method for detecting multiple fault modes of a railway wagon locking plate device. Belongs to the field of railway wagon operation.
Background
With the rapid development of railway industry in China, railway safety is particularly important, and under the large background that the technology of rail wagons is comprehensively improved, component faults occurring in the operation process of the wagons are urgently needed to be found in time, and serious consequences are avoided. The locking plate device is an important device for fastening the bottom cross rod of the truck, is a pivot of the bogie, plays a supporting role for the bogie, and can not play a fixing role for the cross rod because the cross rod continuously moves under stress in the operation process, the locking plate devices at two ends can also loosen, shift and even be damaged, deformed and lost by serious people, and the driving safety is influenced.
Under this background, the probability that the fault was discovered has been guaranteed better to the mode that adopts machine detection and artifical detection to combine together, can not only improve work efficiency, also can reduce time and the cost of artifical train inspection, greatly reduced the potential safety hazard.
Disclosure of Invention
The invention aims to solve the problem that the locking plate device is easy to loosen and displace, so that the locking plate device cannot fix the cross rod and the driving safety is influenced. A method for detecting multiple fault modes of a railway wagon locking plate device is provided.
A method for detecting multiple fault modes of a railway wagon locking plate device comprises the following steps:
step one, establishing a locking plate image sample library, and acquiring a locking plate sample data set;
step two, establishing an improved YOLACT deep learning instance segmentation model;
step three, further training the improved Yolact deep learning instance segmentation model for multiple times based on the acquired locking plate sample data set and the improved Yolact instance segmentation model; observing a loss error descending curve in the training process until an optimal model is trained and exported to be used as a segmentation model of the locking plate component;
and fourthly, after the locking plate area is segmented by the example segmentation model, binarizing the locking plate area, segmenting the locking plate displacement, the locking plate defect and the locking plate deformation condition, and judging the specific fault type of the locking plate according to the segmented binarization result.
Advantageous effects
1. The deep learning technology and the image processing technology are applied to the fault detection of the railway wagon, so that the automatic fault identification is realized, the fault detection efficiency can be improved, the train inspection time and the train inspection cost are reduced, and the risk of the crisis driving safety is also reduced.
2. By adopting the improved YOLACT example segmentation model, the segmentation precision of the model is greatly improved while the characteristic of high speed of the original YOLACT model is ensured, the speed requirement of real-time vehicle passing and real-time fault detection is met, and meanwhile, the segmentation precision can also meet the requirement of the locking plate device on the segmentation precision.
3. In the fault judging stage, the image processing technology is adopted to process the divided locking plate areas, so that the fault forms of the locking plate device can be subdivided, the fault alarm information is more accurate, the specific fault conditions can be further verified by manual work, and the potential safety hazard of the operation of the truck is greatly reduced.
Drawings
FIG. 1 is an overall flow chart of fault identification;
FIG. 2 is a schematic diagram of a deformable convolution kernel;
FIG. 3 is a diagram of a mask re-scoring branch structure;
FIG. 4 is a graph showing the result of the segmentation of the normal locking plate;
FIG. 5 is a graph showing the result of dividing the displacement region of the locking plate;
FIG. 6 is a diagram showing the result of the segmentation of the defective area of the locking plate;
FIG. 7 is a diagram showing the result of dividing the deformation region of the locking plate;
Detailed Description
The first embodiment is as follows: specifically describing the present embodiment with reference to fig. 1, the method for detecting multiple failure modes of a railway wagon locking plate device of the present embodiment includes:
step one, establishing a locking plate image sample library, and acquiring a locking plate sample data set;
step two, establishing an improved YOLACT deep learning instance segmentation model;
step three, training the improved Yolcat deep learning example segmentation model for multiple times based on the acquired locking plate sample dataset; observing a loss error descending curve in the training process until an optimal model is trained and exported to be used as a segmentation model of the locking plate component;
step four, after the locking plate area is divided by the example division model, the locking plate area is binarized, the locking plate displacement, the locking plate defect and the locking plate deformation condition are divided, the specific fault type of the locking plate is judged according to the divided binarization result, and the locking plate fault type comprises the following steps: shifting the locking plate, damaging the locking plate and deforming the locking plate;
and step five, through the judgment of the steps, if the locking plate is judged to be in fault, the locking plate is immediately uploaded to a fault display platform according to the judgment result.
The second embodiment is as follows: the first step is to establish a locking plate image sample library and obtain a locking plate sample data set; the specific process is as follows:
capturing a locking plate device area image through a TFDS-truck fault track edge image detection system, and forming a sample set by collecting different vehicle images in different time periods and different weather conditions; dividing a network by using an instance, and creating a target detection JSON annotation file in a COCO format; firstly, labeling a sample set file by using labelme, and labeling a locking plate area; txt files were then created, in the format:
__ignore__
_background_
Lockingplate
wherein Ignore and background are fixed formats in labels.txt files; lockingplate represents the lock plate sample label name;
and finally, generating a COCO-format locking plate sample data set through labelme2COCO. py code conversion, wherein the COCO-format locking plate sample data set comprises two generated png-format images and corresponding labels (the images are generated after being labeled by a labelme tool, each data set image corresponds to one label, and the labeled images and label files need to be converted into a COCO-format data set, and then a model is trained).
Other steps and parameters are the same as those in the first embodiment.
The third concrete implementation mode: the second step is to establish an improved yolcat deep learning instance segmentation model; the specific process is as follows:
the YOLACT example segmentation model is an example segmentation model based on a one-stage network, is slightly lower than the segmentation model of a two-stage network in two stages in accuracy, but is high in speed and capable of ensuring real-time performance; therefore, on the premise of ensuring the speed real-time performance, the method further improves the model, improves the segmentation accuracy and enables the model to be better applied to the segmentation of the locking plate area;
a one-stage network comprising: yolact, SDD, YoLO and the like, wherein an ancor frame generated by a one-stage network is only a logic structure or a data block, and only the data block needs to be classified and regressed; a two-phase (two-stage) network comprising: fast R-CNN, Mask-RCNN and the like, an nco frame generated by a two-stage network is mapped to a feature map region (except RCNN), and then the region is input to a full connection layer again for classification and regression, and each nco mapping region needs to be subjected to the classification and regression;
step two, establishing a CNN backbone network:
firstly, improving a backbone network, wherein a convolutional layer plays a key role in extracting image features, and in order to improve the accuracy of extracting the image features (a Yolact protoconvolutional layer is a traditional general convolutional kernel, the convolutional kernel needs to be changed into a deformable convolutional kernel, and the required component feature extraction precision is improved, as shown in figure 2, a common convolutional kernel extracts 3 × 3 standard feature points, more complex feature points on the right side can be extracted after improvement, the new structural feature points conform to the irregular 6-sided features of a locking plate), and the convolutional layer needs to be modified appropriately, so that the collection function of the backbone network on the features in different forms is improved;
the CNN main network (common Yolcat feature extraction network) introduces a Deformable convolution DCNv2(Deformable ConvNet v2) to change a convolution kernel with a fixed shape in the CNN main network into a variable convolution kernel, so that the sampling mode is changed, and a common convolution kernel with the size of 3 x 3 is arranged at each position point p to be convolved0All 9 positions are sampled, and are formulated as follows:
wherein p is0Representing each position point to be convolved, y0(p0) Representing the convolved output values at each position; r represents a sampling rule grid: expressed as: r { (-1, -1), (-1,0), ·, (0,1), (1,1) }; p is a radical ofnRepresents an enumeration of the positions listed in R; w (p)n) Represents pnConvolution kernel values corresponding to the positions; x (p)0+pn) Represents pnThe positions of the pixel values of the image to be convolved correspond to the positions of the pixels;
by introducing an offset quantity delta pnWhen the convolution kernel becomes variable, the 9 sampled positions are diffused to the periphery, and the formula is as follows:
wherein, y1(p0) Representing the output value of convolution at each position after introducing the offset;
the deformable convolution kernel is shown in FIG. 2;
by introducing an offset Δ pnBy introducing variable convolution, the most suitable offset (determined according to the effect of extracting the features after multiple convolution layers) is selected, and the features of the locking plate region can be extracted more intensively;
after the introduction of the deformable convolution, the geometric transformation of a target (an accurate locking plate area) can be better adapted, but the selection of the offset is too large, so that the sampling characteristic exceeds the target range, the characteristic is not influenced by the image content, and the weight Deltam is further introduced to solve the problemnThe region corrected by the offset is distributed with different weights, so that more accurate feature extraction is realized (after the offset is introduced, the offset also accords with the region feature of the locking plate, if the overlarge offset is selected at will, the image feature extraction is caused to be invalid), and the formula is expressed as follows:
wherein, y2(p0) Representing the output value of the convolution at each position of the introduced weight;
DCnv2 is applied to the last conv3 to conv5 layers, 3 x 3 convolution layers in each ResNet block are replaced by 3 x 3 deformable convolution layers, and the deformable convolution layers are added into a backbone network of Yolcat, so that the characteristic extraction effect is better, the extraction speed is not influenced, and the precision is greatly improved.
And step two, Anchor Anchor point hyperparametric optimization:
the locking plate is of an irregular hexagonal structure, so the locking plate is special in shape, and in the real-time running process of the truck, images shot by a camera can be stretched or compressed, so that the sizes of the locking plate areas are different;
the hyper-parameter anchor is used after an original picture passes through a series of convolutional layers, pooling layers and an activation function relu to obtain a feature map, window sliding is carried out on the feature map, and a region in the original picture corresponding to the hyper-parameter anchor is reversely deduced according to anchors with different length-width ratios and different areas, wherein the region is a required target region; according to a traditional Anchor selection mode, the robustness is low, so that the selection of a proper Anchor anchor point hyper-parameter is very important;
the anchors super-parameter comprises proportion and aspect ratio, the invention tries to set various anchors super-parameter according to the shape characteristic of the locking plate, for example, (1) the anchors size of each layer of characteristic of FPN is kept unchanged, and the aspect ratio quantity of anchors is increased; (2) keeping the aspect ratio unchanged, and increasing the proportion of the anchor size of each layer of characteristics of the FPN; the number of anchors is increased in the two modes, so that the extracted feature information is richer, and the feature extraction of locking plate areas with different sizes can be better adapted; the aspect ratio is kept unchanged through the test selection mode (2), the dimension proportion of the anchor is increased, the method is more suitable for the state rule of the locking plate, and the method is more suitable for extracting the regional characteristics of the locking plate; when the target is predicted, the target can be more accurately extracted according to the optimized parameters;
step two, adding mask re-scoring branches:
the prediction accuracy of the YOLACT model is lower than that of a two-stage network, so that in order to further improve the prediction accuracy, a mask re-scoring branch is added on the basis of the structure of the YOLACT model, a mask prediction result obtained by the YOLACT model is used as input, the characteristics are extracted by using 6 layers of convolution layers with Relu as an activation function, then a global pooling layer is connected, predicted IoU (intersection ratio of a predicted frame and a real frame) is output, the predicted IoU and a classification score are multiplied to serve as final scores, namely the re-evaluated optimal scores, and finally the YOLACT deep learning example segmentation model is obtained; the added network branch structure is shown in fig. 2, and by adding the network structure, the difference between the classification prediction and the mask segmentation quality is reduced, and the overall prediction precision is improved.
Other steps and parameters are the same as those in the first or second embodiment.
The fourth concrete implementation mode: the difference between the embodiment and one of the first to third embodiments is that the fourth step judges the specific fault type of the locking plate according to the divided binarization result; the specific process is as follows:
(1) detecting the lower edge of the locking plate: the lower edge (the lowest side) of the normal locking plate is horizontal to the rail, as shown in fig. 3, when the lower edge of the locking plate forms a certain included angle with the rail, as shown in fig. 3, the locking plate shifts, at the moment, the included angle between the lower line and the horizontal direction is detected by adopting a mode of changing a detection straight line by hough, and if the included angle is more than 10 degrees, the locking plate shifts; if the included angle is smaller than 10 degrees, other fault types are continuously judged;
(2) detecting the right side edge of the locking plate: the lower edge of the normal locking plate forms an angle of 90 degrees with the right edge, if the angle is abnormal, the situation of damage or deformation is considered to occur, at the moment, the right 1/3 image of the binarization region is intercepted as a judgment basis, the region outline and the area are calculated, the situation that the area is too small (the area is less than 50 percent of the area of the normal locking plate) and the outline point is sunken towards the left side is considered to be the defect of the locking plate; if the area is too small and the contour point is not sunken towards the left side, the locking plate is considered to be deformed;
(3) if the locking plate is not separated, the locking plate is considered to be lost;
(4) if the abnormal conditions do not occur, the locking plate is considered to be normal.
Other steps and parameters are the same as those in one of the first to third embodiments.
The fifth concrete implementation mode: the difference between this embodiment and one of the first to the fourth embodiments is that the optimal model in the third step is:
and selecting 1000 locking plate images as a test set, testing the robustness of the trained model, ensuring that the test result meets the requirement of accurately detecting and segmenting the locking plate region, and if the accuracy rate meets 100 percent, considering that the robustness is high, namely the locking plate region is the optimal model, and if the accuracy rate does not meet the requirement, continuing training until the segmentation accuracy rate is 100 percent.
Other steps and parameters are the same as in one of the first to fourth embodiments.
Claims (8)
1. A method for detecting multiple fault modes of a railway wagon locking plate device is characterized by comprising the following steps:
step one, establishing a locking plate image sample library, and acquiring a locking plate sample data set;
step two, establishing an improved YOLACT deep learning instance segmentation model;
step three, further training the improved Yolact deep learning instance segmentation model for multiple times based on the acquired locking plate sample data set and the improved Yolact instance segmentation model; observing a loss error descending curve in the training process until an optimal model is trained and exported to be used as a segmentation model of the locking plate component;
fourthly, after the locking plate area is segmented by the example segmentation model, binaryzation is carried out on the locking plate area, the locking plate displacement, the locking plate defect and the locking plate deformation are segmented, and the specific fault type of the locking plate is judged according to the segmented binaryzation result; the locking plate failure types include: shifting the locking plate, damaging the locking plate and deforming the locking plate.
2. The method for detecting multiple fault modes of the railway wagon locking plate device as claimed in claim 1, wherein in the first step, a locking plate image sample library is established, and a locking plate sample dataset is obtained; the specific process is as follows:
capturing the locking plate device area image through an image detection system, and forming a sample set by collecting different time periods, different weather conditions and different vehicle images; creating a target detection JSON annotation file in a COCO format by using an instance segmentation network; marking a sample set file by using labelme, marking a locking plate area, and creating a labels.txt file; through labelme2COCO. py transcoding, a lock plate sample dataset in the COCO format is generated.
3. The method for detecting multiple fault modes of the railway wagon locking plate device as claimed in claim 2, wherein in the second step, an improved YOLACT deep learning instance segmentation model is established; the specific process is as follows:
step two, introducing a deformable convolution DCnv2 into the CNN backbone network, changing a convolution kernel with a fixed shape in the CNN backbone network into a variable convolution kernel, and checking a normal convolution kernel with the size of 3 x 3 on each input y (p)0) All sampling 9 positions, the formula is:
wherein, y (p)0) Represents; p is a radical of0Representing the convolved output values at each position; r is a sampling rule grid, expressed as: r { (-1, -1), (-1,0), ·, (0,1), (1,1) }; pnRepresents an enumeration of the positions listed in R; w (p)n) Represents PnConvolution kernel values corresponding to the positions; x (p)0+pn) Represents PnThe positions of the pixel values of the image to be convolved correspond to the positions of the pixels;
introducing an offset quantity delta pnBecomes a variable convolution kernel, and the 9 sampled positions are diffused to the periphery, and the formula tableShown below:
wherein, y1(p0) Representing the output value of convolution at each position after introducing the offset;
further introducing a weight Δ mkAssigning different weights to the offset-corrected regions, the formula is as follows:
wherein, y2(p0) Representing the output value of the convolution at each position of the introduced weight; (ii) a
DCNv2 applies the last conv3 to conv5 layers, replacing the 3 × 3 convolutional layers in each ResNet block with 3 × 3 deformable convolutional layers, adding the deformable convolutional layers to the yolact backbone network;
secondly, keeping the aspect ratio of the Anchor unchanged, increasing the Anchor size proportion of each layer of characteristics of the FPN, and optimizing the Anchor hyper-parameter;
and step three, adding a mask re-scoring branch on the basis of the original structure of the model, taking a mask prediction result obtained by the YOLACT model as input, extracting features by using 6 layers of convolution layers with Relu as an activation function, connecting a global pooling layer, outputting a predicted IoU, and multiplying the predicted IoU and the classification score to obtain a final score, namely the re-evaluated optimal score.
4. The method for detecting multiple fault modes of a railway wagon locking plate device as claimed in claim 3, wherein the specific fault type of the locking plate is judged in the fourth step; the judging method comprises the following steps:
under normal conditions, the lower edge of the locking plate is kept horizontal to the rail, when a certain included angle is formed between the lower edge of the locking plate and the rail, the included angle between the lower line and the horizontal direction is detected in a hough conversion detection straight line mode, if the included angle is larger than 10 degrees, the locking plate is judged to be displaced, and if the included angle is smaller than 10 degrees, the fault type is continuously judged;
under normal conditions, the lower edge of the locking plate forms an angle of 90 degrees with the right edge, if the angle is abnormal, the locking plate is judged to be defective or deformed, at the moment, the image 1/3 on the right side of the binarization region is intercepted as a judgment basis, and the region outline and the region area are calculated; if the area is too small and the contour point is sunken towards the left side, judging that the locking plate is in defect failure; if the area is too small and the contour point does not sink to the left, judging that the locking plate has a deformation fault;
if the locking plate is not separated, the locking plate is not considered to exist, and the alarm locking plate is lost;
if the situations do not occur, the locking plate is judged to be normal;
when any fault occurs in the locking plate, the locking plate is immediately uploaded to a fault display platform through a TFDS truck fault detection system according to a fault judgment result.
5. The method for detecting multiple fault modes of the railway wagon locking plate device as claimed in claim 1, wherein the optimal model is as follows:
and selecting 1000 locking plate images as a test set, testing the robustness of the trained model, judging the model to be the optimal model if the test result meets the requirement of accurately detecting the accuracy of the locking plate area segmented by 100%, and continuing training until the segmentation accuracy reaches 100% if the test result does not meet the requirement.
6. The method of claim 2, wherein said labels.
__ignore__
_background_
Lockingplate
Wherein Ignore and background are fixed formats in labels.txt files; lockingplate denotes the lock plate specimen label name.
7. The method as claimed in claim 2, wherein the COCO format locking plate sample data set includes generated png format images and labels corresponding to each image.
8. The method as claimed in claim 2, wherein the detection system is a TFDS-wagon fault trail edge image detection system.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011110722.4A CN112233095B (en) | 2020-10-16 | 2020-10-16 | Method for detecting multiple fault forms of railway wagon locking plate device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011110722.4A CN112233095B (en) | 2020-10-16 | 2020-10-16 | Method for detecting multiple fault forms of railway wagon locking plate device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112233095A true CN112233095A (en) | 2021-01-15 |
| CN112233095B CN112233095B (en) | 2021-05-28 |
Family
ID=74118855
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011110722.4A Active CN112233095B (en) | 2020-10-16 | 2020-10-16 | Method for detecting multiple fault forms of railway wagon locking plate device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112233095B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112907560A (en) * | 2021-03-16 | 2021-06-04 | 中科海拓(无锡)科技有限公司 | Notebook appearance flaw segmentation method based on deep learning |
| CN117496191A (en) * | 2024-01-03 | 2024-02-02 | 南京航空航天大学 | Data weighted learning method based on model collaboration |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103500328A (en) * | 2013-10-16 | 2014-01-08 | 北京航空航天大学 | Method for automatically detecting deflection fault of railway wagon locking plate |
| CN106600581A (en) * | 2016-12-02 | 2017-04-26 | 北京航空航天大学 | Train operation fault automatic detection system and method based on binocular stereoscopic vision |
| CN108898574A (en) * | 2018-05-09 | 2018-11-27 | 江苏大学 | Train bogie cross-braced device head bolts looseness fault automatic testing method |
| CN109165541A (en) * | 2018-05-30 | 2019-01-08 | 北京飞鸿云际科技有限公司 | Coding method for vehicle component in intelligent recognition rail traffic vehicles image |
| CN109977962A (en) * | 2019-03-21 | 2019-07-05 | 国网山东省电力公司经济技术研究院 | A kind of Cable's Fault hidden danger automatic identifying method and system |
| CN111079747A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon bogie side frame fracture fault image identification method |
| CN111080603A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of shaft end bolt of railway wagon |
| CN111091544A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of side integrated framework of railway wagon bogie |
-
2020
- 2020-10-16 CN CN202011110722.4A patent/CN112233095B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103500328A (en) * | 2013-10-16 | 2014-01-08 | 北京航空航天大学 | Method for automatically detecting deflection fault of railway wagon locking plate |
| CN106600581A (en) * | 2016-12-02 | 2017-04-26 | 北京航空航天大学 | Train operation fault automatic detection system and method based on binocular stereoscopic vision |
| CN108898574A (en) * | 2018-05-09 | 2018-11-27 | 江苏大学 | Train bogie cross-braced device head bolts looseness fault automatic testing method |
| CN109165541A (en) * | 2018-05-30 | 2019-01-08 | 北京飞鸿云际科技有限公司 | Coding method for vehicle component in intelligent recognition rail traffic vehicles image |
| CN109977962A (en) * | 2019-03-21 | 2019-07-05 | 国网山东省电力公司经济技术研究院 | A kind of Cable's Fault hidden danger automatic identifying method and system |
| CN111079747A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Railway wagon bogie side frame fracture fault image identification method |
| CN111080603A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of shaft end bolt of railway wagon |
| CN111091544A (en) * | 2019-12-12 | 2020-05-01 | 哈尔滨市科佳通用机电股份有限公司 | Method for detecting breakage fault of side integrated framework of railway wagon bogie |
Non-Patent Citations (5)
| Title |
|---|
| DANIEL BOLYA ET AL.: "YOLACT++ Better Real-time Instance Segmentation", 《ARXIV:1912.06218V2》 * |
| JIFENG DAI ET AL.: "Deformable Convolutional Networks", 《ARXIV:1703.06211V3》 * |
| XIZHOU ZHU ET AL.: "Deformable ConvNets v2: More Deformable, Better Results", 《ARXIV:1811.11168V2》 * |
| 刘盛亚: "复杂环境下列车关键部件故障实时图像检测算法研究", <中国优秀硕士学位论文全文数据库 工程科技II辑> * |
| 刘翔: "基于机器视觉的货车典型故障图像识别算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112907560A (en) * | 2021-03-16 | 2021-06-04 | 中科海拓(无锡)科技有限公司 | Notebook appearance flaw segmentation method based on deep learning |
| CN117496191A (en) * | 2024-01-03 | 2024-02-02 | 南京航空航天大学 | Data weighted learning method based on model collaboration |
| CN117496191B (en) * | 2024-01-03 | 2024-03-29 | 南京航空航天大学 | Data weighted learning method based on model collaboration |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112233095B (en) | 2021-05-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112233095B (en) | Method for detecting multiple fault forms of railway wagon locking plate device | |
| CN111652295B (en) | Railway wagon coupler yoke key joist falling fault identification method | |
| CN111080608A (en) | Method for recognizing closing fault image of automatic brake valve plug handle of railway wagon in derailment | |
| CN111080609B (en) | Brake shoe bolt loss detection method based on deep learning | |
| CN111489339A (en) | Method for detecting defects of bolt spare nuts of high-speed railway positioner | |
| CN111080605A (en) | Method for identifying railway wagon manual brake shaft chain falling fault image | |
| CN111652846B (en) | Semiconductor defect identification method based on characteristic pyramid convolution neural network | |
| CN111080607A (en) | Rolling bearing oil slinging fault detection method based on image recognition | |
| CN113096085A (en) | Container surface damage detection method based on two-stage convolutional neural network | |
| CN110853091A (en) | Method and system for identifying winding defect image of engine fuse | |
| CN112733747A (en) | Identification method, system and device for relieving falling fault of valve pull rod | |
| CN115527170A (en) | Method and system for identifying closing fault of door stopper handle of automatic freight car derailing brake device | |
| CN113506269A (en) | Turnout and non-turnout rail fastener positioning method based on deep learning | |
| CN115995056A (en) | Automatic bridge disease identification method based on deep learning | |
| CN111080604A (en) | Image identification method for breakage fault of hook lifting rod of railway wagon | |
| CN113221839B (en) | Automatic truck image identification method and system | |
| CN111080599A (en) | Fault identification method for hook lifting rod of railway wagon | |
| CN112329858B (en) | Image recognition method for breakage fault of anti-loosening iron wire of railway motor car | |
| CN112102280B (en) | Method for detecting loosening and loss faults of small part bearing key nut of railway wagon | |
| CN117372677A (en) | Method for detecting health state of cotter pin of fastener of high-speed railway overhead contact system | |
| CN112258483A (en) | Coupler yoke pin inserting and supporting dislocation and nut loss fault detection method | |
| CN112132088B (en) | Inspection point missing inspection identification method | |
| CN115082650A (en) | Implementation method of automatic pipeline defect labeling tool based on convolutional neural network | |
| CN111833323B (en) | Image quality judgment method for task-divided rail wagon based on sparse representation and SVM (support vector machine) | |
| CN114187550A (en) | Bow net core part identification method based on improved YOLO V3 network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |