CN108152059A - High-speed train bogie fault detection method based on Fusion - Google Patents
High-speed train bogie fault detection method based on Fusion Download PDFInfo
- Publication number
- CN108152059A CN108152059A CN201711385656.XA CN201711385656A CN108152059A CN 108152059 A CN108152059 A CN 108152059A CN 201711385656 A CN201711385656 A CN 201711385656A CN 108152059 A CN108152059 A CN 108152059A
- Authority
- CN
- China
- Prior art keywords
- channel
- signal
- fusion
- grader
- classification
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/08—Railway vehicles
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of high-speed train bogie fault detection methods based on Fusion, include the following steps：S1, the original data signal that the multiple sensors being mounted on bogie obtain is collected；S2, the original data signal of collection is cut into the individual signal of each channel；The corresponding spectrogram of signal of S3, each channel of generation；S4, by random forests algorithm, obtain the classification results of each channel frequency spectrum figure；S5, by voting method, the classification results of all channel frequency spectrum figures are merged, obtain fault type.Beneficial effects of the present invention are：Extraction bogie vibration signal characteristics process is simple and efficient, and by automatic Multichannel fusion method, substantially increase fault detect accuracy rate.
Description
Technical field
The invention belongs to train fault detection fields, and in particular to a kind of bullet train based on Fusion
Bogie fault detection method.
Background technology
In recent years, it is social with the continuous growth of China railways revenue kilometres and stepping up for train running speed
Transportation demand is gradually met.Meanwhile people also increasingly promote the safety requirements that bullet train transports.In this context,
More and more researchs are concentrated in the fault diagnosis of bullet train, and bogie is uniquely connected between train body and track
Unit, the degeneration of critical component is by serious threat to train operating safety.Therefore, high-speed train bogie fault detect has
Important research significance.
High-speed train bogie structure mainly include two groups of wheels to, framework, brake apparatus and vibration absorber etc..Bogie closes
Key member belongs to vibration absorber, mainly includes：Air spring, transverse damper and anti-snakelike damper, the situation of these components
Not only it is related to the operational safety of train, it is also closely related with the comfort level in train travelling process.Wherein, the event of air spring
Barrier, which is mainly manifested in the vertical stiffness caused by rubber layer ruptures, to be changed, and lateral damper and anti-snakelike damper are liquid-springing
Device, failure are mainly manifested in the variation of the damped coefficient as caused by oil leak.
The diagnostic method of existing high-speed train bogie failure is broadly divided into two classes：There is traditional signal analysis method to carry
The feature taken and the feature learnt automatically by neural network.The fault detection method has characteristic extraction procedure complexity, and carries
The feature got is unstable, the problem of needing manually to choose Reliability Channel, do not make full use of all channels.It needs to consume
Take a large amount of human and material resources and fault detect is carried out to it, and have the possibility for being unable to its fault type of accurate judgement.
Invention content
For above-mentioned deficiency of the prior art, the bullet train provided by the invention based on Fusion turns
It is solved to frame fault detection method and bogie vibration signal characteristics process complexity is extracted in existing fault detection method, and extracted
Signal characteristic it is unstable, the problem of needing manually to choose Reliability Channel.
In order to reach foregoing invention purpose, the technical solution adopted by the present invention is：One kind is based on Fusion
High-speed train bogie fault detection method, include the following steps：
S1, the original data signal that the multiple sensors being mounted on bogie obtain is collected；
S2, the original data signal of collection is cut into the individual signal of each channel；
The corresponding spectrogram of signal of S3, each channel of generation；
S4, by random forests algorithm, obtain the classification results of each channel frequency spectrum figure；
S5, by voting method, the classification results of all channel frequency spectrum figures are merged, obtain fault type.
Beneficial effects of the present invention are：Extraction bogie vibration signal characteristics process is simple and efficient, and by automatic
Multichannel fusion method substantially increases fault detect accuracy rate.
Further, the original data signal that multiple sensors obtain in the step S1 corresponds to different location on bogie
The vibration signal of different directions.
Above-mentioned further scheme has the beneficial effect that：FUSION WITH MULTISENSOR DETECTION collects the data of different location and different directions,
The diversity for collecting data is improved, improves the accuracy of bogie fault detect.
Further, in the step S2 different location sensor is mounted in each individual signal of channel and step S1
The original data signal of acquisition corresponds.
Above-mentioned further scheme has the beneficial effect that：The Dynamic data exchange of each sensor collection corresponds to a channel, is convenient for
The comparison of data processing.
Further, the specific method of spectrogram for generating each channel in the step S3 is：By the number of each channel
It is believed that number cutting into multiple segments according to its sample frequency, the spectrogram of channel signal is obtained.
Above-mentioned further scheme has the beneficial effect that：Spectrogram is a kind of combination time domain and frequency domain information of time varying signal
Effective visable representation, convenient for directly observing data processed result.
Further, in the step S4 random forest be decision tree combination, random forests algorithm mainly include it is following
Step：
S4-1, a sample set is randomly selected from the training set in data set；
S4-2, by randomly choosing a certain number of one decision tree of feature construction.For each split vertexes, according to base
The criterion of Buddhist nun or information gain selects best split values, until can not further divide, constructs a certain number of trees；
S4-3, count all decision trees as a result, obtaining the sample set in random forest by most voting methods
Final classification result.
Above-mentioned further scheme has the beneficial effect that：When the data set of processing has very much, have very relative to other algorithms
Big advantage；Its character subset is randomly selected；It is independent from each other between tree and tree during training；For unbalanced data
For collection, it can be with balance error；When there are a big chunk missing features, accuracy still can be maintained.
Further, voting method of the voting method in the step S5 for monolayer neural networks, the monolayer neuronal
The voting method of network be specially：
In S5-1, the classification results obtained according to step S4-3, there are L classification, M grader, the classification of M grader
As a result combine the matrix X for being expressed as a L × M, wherein each element xijMeet：
Wherein, SjRepresent the output classification of j-th of grader, xij=1 j-th of grader of expression thinks that the sample is i-th
Class, xij=0 j-th of grader of expression thinks that the sample is not the i-th class；Wherein i, j are respectively in the classification results of spectrogram
The number of grader and classification, i≤M, j≤L；
S5-2, using matrix X as the input of neural network, L × M neuromere of the input layer in Connection Neural Network
Point；
S5-3, weight and the initial value of biasing in neural network hidden layer are determined, the weight is with each grader and each
Classification is related, and biasing is as a correction amount；
S5-4, using cross entropy as the loss function of hidden layer in the neural network, sigmoid functions as activation letter
Number, and using stochastic gradient descent and back-propagation method update weight and biasing, obtain optimal network.
Description of the drawings
Fig. 1 is that the high-speed train bogie fault detection method based on Fusion realizes flow chart.
Fig. 2 realizes flow chart for random forests algorithm.
Fig. 3 realizes flow chart for monolayer neural networks voting method.
Fig. 4 is neural network structure schematic diagram.
Fig. 5 is the original image classification results of data set 1 and HOG tagsort result differential charts.
Fig. 6 is the original image classification results figure of data set 1.
Fig. 7 is the result inputted under the mixed running speed of data set 1 using HOG features and original image as random forest
Comparison diagram.
Fig. 8 is that block diagram is compared in the final result classification of data set 1.
Fig. 9 is random forest failure modes results contrast figure under 2 certain speeds of service of data set.
Figure 10 is that block diagram is compared in the final result classification of data set 2.
Figure 11 is different degrees of fault error rate line chart in data set 2.
Specific embodiment
The specific embodiment of the present invention is described below, in order to facilitate understanding by those skilled in the art this hair
It is bright, it should be apparent that the present invention is not limited to the range of specific embodiment, for those skilled in the art,
As long as various change in the spirit and scope of the present invention that appended claim limits and determines, these variations are aobvious and easy
See, all are using the innovation and creation of present inventive concept in the row of protection.
As shown in Fig. 1 to Fig. 3 institutes, the bullet train fault detection method based on Fusion, including following
Step：
S1, the original data signal that the multiple sensors being mounted on bogie obtain is collected；Wherein, multiple sensors obtain
The original data signal obtained corresponds to the vibration signal of different location different directions on bogie.
S2, the original data signal of collection is cut into the individual signal of each channel；Each individual signal of channel with
The original data signal that different location sensor obtains is mounted in step S1 to correspond.
The corresponding spectrogram of signal of S3, each channel of generation：By the data-signal of each channel according to its sample frequency
Multiple segments are cut into, obtain the spectrogram of channel signal；
Above-mentioned original data signal is cut into multiple segments is sample set in order to obtain, for training and testing classification mould
Type, it is assumed that the sample frequency of each sample is H, length 3H, then just has 3 frames after being converted to spectrogram.
S4, by random forests algorithm, obtain the classification results of each channel frequency spectrum figure；Wherein, random forests algorithm master
Include the following steps：
S4-1, a sample set is randomly selected from the training set in data set；
S4-2, by randomly choosing a certain number of one decision tree of feature construction.For each split vertexes, according to base
The criterion of Buddhist nun or information gain selects best split values, until can not further divide, constructs a certain number of trees；
S4-3, count all decision trees as a result, obtaining the sample set in random forest by most voting methods
Final classification result.
S5, by voting method, Decision fusion is carried out to the classification results of all channel frequency spectrum figures, obtains fault type；
Voting method is the voting method (NN) of monolayer neural networks, and a kind of weighting of voting method of monolayer neural networks is thrown
Ticket method specifically includes following steps：
In S5-1, the classification results obtained according to step S4-3, there are L classification, M grader, the classification of M grader
As a result combine the matrix X for being expressed as a L × M, wherein each element x ij meets：
Wherein, SjRepresent the output classification of j-th of grader, xij=1 j-th of grader of expression thinks that the sample is i-th
Class, xij=0 j-th of grader of expression thinks that the sample is not the i-th class；Wherein i, j are respectively in the classification results of spectrogram
The number of grader and classification, i≤M, j≤L；For each sample, a grader can only export one kind as a result, so matrix
For each row of X only there are one 1, remaining is 0 entirely.
S5-2, using matrix X as the input of neural network, L × M neurode in Connection Neural Network.
S5-3, weight and the initial value of biasing in neural network hidden layer are determined, the weight is with each grader and each
Classification is related, and biasing is as a correction amount；
Wherein, it is to take biasing initial value weight initial value is that average value is for 0 to use the initial method of weight and deviation
0th, variance be 1 just too be distributed in random value.
S5-4, using cross entropy as the loss function of hidden layer in the neural network, sigmoid functions as activation letter
Number, and using stochastic gradient descent and back-propagation method update weight and biasing, obtain optimal network.
In one embodiment of the invention, the voting method of above-mentioned monolayer neural networks is a kind of based on neural network
Voting method is a kind of Nearest Neighbor with Weighted Voting method based on reverse transmittance nerve network.Generally, (artificial) neural network is by inputting
Layer, hidden layer, three parts of output layer form, as shown in Figure 4.Wherein, each layer includes multiple neurodes, is connected to adjacent
Two layers of all neurons.The node of each hidden layer has corresponding weight vectors W and biasing b；The node of each neural network
There is corresponding activation primitive f, activation primitive is typically S type functions.For some node of a certain layer in hidden layer, if by him
N number of input be expressed as vectorial X, then its output can be expressed as scalar WX+b.
During training neural network, suitable cost function is selected, updates weight with reference to back-propagation algorithm
And biasing, so as to which cost function be made to reach minimum value, optimal network is obtained in the case of network not over-fitting.
In the voting method based on neural network of the present invention, if there is L classification, M grader, then for a reality
Example, the result of all graders can be combined the matrix X for being expressed as a L × M, wherein each element x { ij } meets：
Wherein, SjRepresent the output classification of j-th of grader, xij=1 j-th of grader of expression thinks that the sample is i-th
Class, xij=0 j-th of grader of expression thinks that the sample is not the i-th class；Wherein i, j are respectively in the classification results of spectrogram
The number of grader and classification, i≤M, j≤L；For each sample, a grader can only export one kind as a result, so matrix
For each row of X only there are one 1, remaining is 0 entirely.
Again using the matrix as the input of NN, L × M neurode of input layer is connected.In the nerve used in the present invention
Only there are one hidden layers for network, and the weight of each neurode is related to each grader and each class, similar in CE methods
Matrix W；And it biases and is then used as a correction amount.The Web vector graphic cross entropy is as loss function, and sigmoid functions are as sharp
Function living, and using SGD (stochastic gradient descent) and BP (backpropagation) method update weight and bias to obtain optimal network.
In one embodiment of the invention, the collection of data is real by carrying out emulation in SIMPACK dynamics simulation softwares
Test obtains, it is turned to the high-speed train bogie fault detection method provided by the invention based on Fusion
The accuracy of frame fault detect is verified.
The data of collection are divided into two datasets, as shown in Table 1 and Table 2, due to utilizing certain specific train model
With Beijing-Guangzhou Railway track spectrum, data set 1 and data set 2 have very strong realistic meaning.
Table 1
Status Type
Table 2
In preprocessing process is carried out to the data of two datasets, with 3 seconds for slice length, 2 seconds are the length that forward laps
It spends data serial section, then converts it into spectrogram.
Two parts are divided into for the experimental verification of data set 1：Failure modes and all speed mixing feelings under specific speed
Failure modes under condition correspond respectively to 7 × 20 samples and 6 × 7 × 200 samples.The data set is pressed 3：7 be divided into
The training set and test set in machine forest classified stage.The division proportion in ballot stage is training set：Verification collection：Test set=2：1：
2.Following two stages will be divided into the processing of data set 1：
First stage：Failure modes are carried out to the data of each channel using random forest.
The spectrogram that original size is 129 × 4 is compressed into 64 × 4 size, directly as the input of random forest, together
When, it is more more effective than with the HOG features of spectrogram directly using spectrogram to verify, one group of contrast test is designed, by spectrogram
Size is adjusted to 64 × 64 (being first compressed to 64, then each column replicates 16 times), and carries out normalizing according to the maximum value of training data
Change.Wherein, the parameter of HOG feature extractions is as shown in table 3：
Table 3
It obtains and the result of failure modes is carried out as shown in Figure 5 and Figure 6 to each channel using random forest, the results showed that：
It is directly more preferable using spectrogram effect, and HOG feature extractions cause useful information in most of channel to be lost.This result can be with
It is explained by the feature of spectrogram：Compared with other signals such as electrocardiogram and electroencephalogram, high-speed train bogie vibration signal
Spectrogram be more stable from time shaft, it is meant that its space characteristics of these sample statistics features are more prominent.Simultaneously
From the results, it was seen that the accuracy rate difference of different channels is very big.Specifically, the signal perpendicular to railway direction is corresponding logical
Road accuracy rate is higher, and the accuracy rate of plummet direction signal corresponding channel is relatively low.Therefore, the fault signature master of bogie critical component
It is embodied in oscillation crosswise, the main reason for oscillation crosswise is also train snakelike unstable motion.As a result it is also shown that when operation speed
When spending relatively low (being 40km/s in experiment), it is difficult to classify to failure.And the performance of different channels is in the different speeds of service
Under be similar, this just confirms the possibility classified to failure in the case where all speeds of service are mixed.In addition, letter
Number spectral transformation process largely have compressed data volume.For the single sample that original size is 243 × 3 × 58, phase
The spectral image answered only has the 1.84% of original size.This compression has filtered out redundancy well, remains stable event
Hinder feature, be also beneficial to accelerate training and test process.Above-mentioned experiment is repeated with the data set that all speed mix, as a result as schemed
Shown in 7, using original image still than using HOG characteristic effects more preferable.
Second stage：Using monolayer neural networks voting method failure modes result is obtained into row of channels fusion.
The result for the method that monolayer neural networks provided by the invention are voted is with using majority vote method (PV), classification entropy
The result that method (CE) and victor follow the example of (WTA) entirely is compared, to verify the high efficiency of this method.In view of CE and WTA are not needed to
Verification collection, PV only need test set, corresponding at its for the unwanted data set of ad hoc approach to ensure the fairness of comparison
Directly cast out in experiment.To each experiment, 20 experiments are carried out with one group of fixed random seed, to ensure the visitor of result
The property seen.In CE, control constant is arranged to 0.8.In the voting method of monolayer neural networks, when verification rate of accuracy reached arrives
100% or frequency of training when reaching maximum value 50, deconditioning network.The other parameter of the voting method of monolayer neural networks is such as
It is 100 to criticize size (batch size), learning rate 0.1.Failure modes result such as Fig. 8 of the data set 1 obtained and such as 4 institute of table
Show, even if classification results show in all methods in the case of most complicated all speeds of service mixing, individual layer god
The classifying quality that voting method through network obtains is best, and nicety of grading shows highest robustness close to 100%.
The result of CE and WTA is close, is fluctuated in the range of 93%-100%, wherein, the accuracy rate of CE is not always better than WTA.This
One conclusion shows WTA for the balance of the overall performance of Local Property and grader and the voting method phase of monolayer neural networks
It is poor.Since PV has ignored the greatest differences of different Channel Prediction abilities, its classification accuracy is minimum, and fluctuation range is also most
Greatly, it is differed from 79% to 96%.
Table 4
For data set 2, in the first stage using the data processing scheme identical with data set 1, due to having been verified that
HOG feature extractions are more worse than directly utilizing spectrogram effect, therefore are not tested on data set 2 in comparison of design.Obtain using with
The results are shown in Figure 9 for part classifying of the machine forest on data set 2, since different degrees of failure mixes, even if fault type
Less, classification accuracy is relatively low compared with data set 1, but high-precision channel number is still identical with the result on data set 1,
And the result of low velocity is more worse still than high speed.
Second stage, i.e. Multichannel fusion stage, by the voting method of monolayer neural networks by NN voting methods and CE,
The result of WTA voting methods compares.Wherein, except the controlling value in CE is set as 0.6, in NN, neural network is set
Other than deconditioning after 70 iteration (epochs), other parameter is as before.According to shown in Fig. 10 as a result, NN is thrown
The performance of ticket method is still substantially better than other methods, and accuracy rate is 90% or so.Figure 11 is shown under different degrees of failure
Error rate.In general, with the increase of degree of degeneration, error in classification reduces, because degeneration rate is higher to mean failure
Feature is distincter.
Summary experimental verification process, the high-speed train bogie provided by the invention based on Fusion
Fault detection method using the spectrogram of bogie vibration signal as feature, is merged using the voting method of monolayer neural networks
The corresponding multiple individual classification results of channel of multisensor, provide final failure detection result.This method is in two differences
Data set on tested, in various faults type, (seven kinds of failures, including a normal condition, three single parts are complete
Failure and three multi-parts it is entirely ineffective) in the case of and various faults degree mixing (fault degree is from 10% to 90%) feelings
It is proved to that there is higher classification accuracy under condition.
Beneficial effects of the present invention are：Extraction bogie vibration signal characteristics process is simple and efficient, and by automatic
Multichannel fusion method substantially increases fault detect accuracy rate.
Claims (6)
1. a kind of high-speed train bogie fault detection method based on Fusion, which is characterized in that including with
Lower step：
S1, the original data signal that the multiple sensors being mounted on bogie obtain is collected；
S2, the original data signal of collection is cut into the individual signal of each channel；
The corresponding spectrogram of signal of S3, each channel of generation；
S4, by random forests algorithm, obtain the classification results of each channel frequency spectrum figure；
S5, by voting method, the classification results of all channel frequency spectrum figures are merged, obtain fault type.
2. the high-speed train bogie fault detection method according to claim 1 based on Fusion,
It is characterized in that, the original data signal that multiple sensors obtain in the step S1 corresponds to different location different directions on bogie
Vibration signal.
3. the high-speed train bogie fault detection method according to claim 2 based on Fusion,
It is characterized in that, the original that each individual signal of channel is obtained with being mounted on different location sensor in step S1 in the step S2
Beginning data-signal corresponds.
4. the high-speed train bogie fault detection method according to claim 3 based on Fusion,
It is characterized in that, the specific method of spectrogram for generating each channel in the step S3 is：By the data-signal root of each channel
Multiple segments are cut into according to its sample frequency, obtain the spectrogram of channel signal.
5. the high-speed train bogie fault detection method according to claim 1 based on Fusion,
It is characterized in that, random forest is the combination of decision tree in the step S4, and random forests algorithm mainly includes the following steps that：
S4-1, a sample set is randomly selected from the training set in data set；
S4-2, by randomly choosing a certain number of one decision tree of feature construction, for each split vertexes, according to Geordie or
The criterion of information gain selects best split values, until can not further divide, constructs a certain number of trees；
S4-3, all decision tree of random forest is counted by most voting methods as a result, to obtain the sample set final
Classification results.
6. the high-speed train bogie fault detection method according to claim 5 based on Fusion,
It is characterized in that, voting method of the voting method in the step S5 for monolayer neural networks, the throwing of the monolayer neural networks
Ticket method is specially：
In S5-1, the classification results obtained according to step S4-3, there are L classification, M grader, the classification results of M grader
Combine the matrix X for being expressed as a L × M, wherein each element xijMeet：
Wherein, SjRepresent the output classification of j-th of grader, xij=1 j-th of grader of expression thinks that the sample is the i-th class, xij
=0 j-th of grader of expression thinks that the sample is not the i-th class；Wherein i, j are respectively the grader in the classification results of spectrogram
With the number of classification, i≤M, j≤L；
S5-2, using matrix X as the input of neural network, L × M neurode of the input layer in Connection Neural Network.
S5-3, weight and the initial value of biasing in neural network hidden layer, the weight and each grader and each classification are determined
Correlation, biasing is as a correction amount；
S5-4, using cross entropy as the loss function of hidden layer in the neural network, sigmoid functions as activation primitive,
And using stochastic gradient descent and back-propagation method update weight and biasing, obtain optimal network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711385656.XA CN108152059B (en) | 2017-12-20 | 2017-12-20 | High-speed train bogie fault detection method based on multi-sensor data fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711385656.XA CN108152059B (en) | 2017-12-20 | 2017-12-20 | High-speed train bogie fault detection method based on multi-sensor data fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108152059A true CN108152059A (en) | 2018-06-12 |
| CN108152059B CN108152059B (en) | 2021-03-16 |
Family
ID=62464084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711385656.XA Active CN108152059B (en) | 2017-12-20 | 2017-12-20 | High-speed train bogie fault detection method based on multi-sensor data fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108152059B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109580260A (en) * | 2018-11-27 | 2019-04-05 | 南京康尼机电股份有限公司 | A kind of inferior health diagnostic method of track vehicle door system |
| CN111024527A (en) * | 2019-12-06 | 2020-04-17 | 西安理工大学 | Crack propagation monitoring method based on multi-sensor data fusion |
| CN111127448A (en) * | 2019-12-26 | 2020-05-08 | 西南交通大学 | Model and method for detecting air spring fault based on isolated forest |
| CN111516711A (en) * | 2019-02-02 | 2020-08-11 | 株洲中车时代电气股份有限公司 | Safety monitoring method and device for running gear of motor train unit |
| CN111874813A (en) * | 2020-08-04 | 2020-11-03 | 上海海事大学 | Shore bridge crane reducer state detection method based on improved decision tree algorithm |
| CN112133941A (en) * | 2020-10-10 | 2020-12-25 | 西南交通大学 | Rapid fault diagnosis method for locomotive proton exchange membrane fuel cell system |
| CN112598287A (en) * | 2020-12-25 | 2021-04-02 | 武汉烽火信息集成技术有限公司 | Rail transit health assessment method and system |
| CN112991347A (en) * | 2021-05-20 | 2021-06-18 | 西南交通大学 | Three-dimensional-based train bolt looseness detection method |
| CN113128555A (en) * | 2021-03-09 | 2021-07-16 | 西南交通大学 | Method for detecting abnormality of train brake pad part |
| CN113255591A (en) * | 2021-06-25 | 2021-08-13 | 四川九通智路科技有限公司 | Bearing fault diagnosis method based on random forest and fusion characteristics |
| CN114187533A (en) * | 2022-02-15 | 2022-03-15 | 西南交通大学 | GB-InSAR (GB-InSAR) atmospheric correction method based on random forest time sequence classification |
| CN114397121A (en) * | 2022-01-21 | 2022-04-26 | 西南交通大学 | Bogie fault diagnosis method |
| CN114911789A (en) * | 2022-07-15 | 2022-08-16 | 武汉大学 | Water level data fusion method and device based on radar vision multi-sensor data |
| CN117892744A (en) * | 2023-12-27 | 2024-04-16 | 广州市弘宇科技有限公司 | Operation and maintenance method and system of RFID (radio frequency identification) identification equipment in civil aviation logistics field |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103335617A (en) * | 2013-06-19 | 2013-10-02 | 清华大学 | Vibrational signal based railway track geometric deformation detection method |
| CN103471865A (en) * | 2013-09-12 | 2013-12-25 | 北京交通大学 | Train suspension system failure isolation method based on LDA method |
| CN103728551A (en) * | 2013-01-30 | 2014-04-16 | 中国人民解放军海军航空工程学院 | Analog circuit fault diagnosis method based on cascade connection integrated classifier |
| CN104408468A (en) * | 2014-11-26 | 2015-03-11 | 西安电子科技大学 | Face recognition method based on rough set and integrated learning |
| CN104990709A (en) * | 2015-08-07 | 2015-10-21 | 杨玉娇 | Method for detecting locomotive bearing fault |
| CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
| CN105971901A (en) * | 2016-05-03 | 2016-09-28 | 北京航空航天大学 | Centrifugal pump fault diagnosis method based on complete ensemble empirical mode decomposition and random forest |
| CN106326929A (en) * | 2016-08-24 | 2017-01-11 | 北京工商大学 | Rail train running part rolling bearing fault diagnosis method |
| CN106503734A (en) * | 2016-10-14 | 2017-03-15 | 哈尔滨工程大学 | Based on trilateral filter and the image classification method of the sparse autocoder of storehouse |
| CN106653032A (en) * | 2016-11-23 | 2017-05-10 | 福州大学 | Animal sound detecting method based on multiband energy distribution in low signal-to-noise-ratio environment |
| CN107194600A (en) * | 2017-06-01 | 2017-09-22 | 国网山东省电力公司济南市历城区供电公司 | A kind of electric load Seasonal Characteristics sorting technique |
| CN107483267A (en) * | 2017-09-19 | 2017-12-15 | 中国人民解放军防空兵学院 | A kind of EIGRP routing failures recognition methods |
-
2017
- 2017-12-20 CN CN201711385656.XA patent/CN108152059B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103728551A (en) * | 2013-01-30 | 2014-04-16 | 中国人民解放军海军航空工程学院 | Analog circuit fault diagnosis method based on cascade connection integrated classifier |
| CN103335617A (en) * | 2013-06-19 | 2013-10-02 | 清华大学 | Vibrational signal based railway track geometric deformation detection method |
| CN103471865A (en) * | 2013-09-12 | 2013-12-25 | 北京交通大学 | Train suspension system failure isolation method based on LDA method |
| CN104408468A (en) * | 2014-11-26 | 2015-03-11 | 西安电子科技大学 | Face recognition method based on rough set and integrated learning |
| CN104990709A (en) * | 2015-08-07 | 2015-10-21 | 杨玉娇 | Method for detecting locomotive bearing fault |
| CN105868779A (en) * | 2016-03-28 | 2016-08-17 | 浙江工业大学 | Method for identifying behavior based on feature enhancement and decision fusion |
| CN105971901A (en) * | 2016-05-03 | 2016-09-28 | 北京航空航天大学 | Centrifugal pump fault diagnosis method based on complete ensemble empirical mode decomposition and random forest |
| CN106326929A (en) * | 2016-08-24 | 2017-01-11 | 北京工商大学 | Rail train running part rolling bearing fault diagnosis method |
| CN106503734A (en) * | 2016-10-14 | 2017-03-15 | 哈尔滨工程大学 | Based on trilateral filter and the image classification method of the sparse autocoder of storehouse |
| CN106653032A (en) * | 2016-11-23 | 2017-05-10 | 福州大学 | Animal sound detecting method based on multiband energy distribution in low signal-to-noise-ratio environment |
| CN107194600A (en) * | 2017-06-01 | 2017-09-22 | 国网山东省电力公司济南市历城区供电公司 | A kind of electric load Seasonal Characteristics sorting technique |
| CN107483267A (en) * | 2017-09-19 | 2017-12-15 | 中国人民解放军防空兵学院 | A kind of EIGRP routing failures recognition methods |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109580260A (en) * | 2018-11-27 | 2019-04-05 | 南京康尼机电股份有限公司 | A kind of inferior health diagnostic method of track vehicle door system |
| CN111516711A (en) * | 2019-02-02 | 2020-08-11 | 株洲中车时代电气股份有限公司 | Safety monitoring method and device for running gear of motor train unit |
| CN111024527A (en) * | 2019-12-06 | 2020-04-17 | 西安理工大学 | Crack propagation monitoring method based on multi-sensor data fusion |
| CN111127448A (en) * | 2019-12-26 | 2020-05-08 | 西南交通大学 | Model and method for detecting air spring fault based on isolated forest |
| CN111127448B (en) * | 2019-12-26 | 2022-05-10 | 西南交通大学 | Method for detecting air spring fault based on isolated forest |
| CN111874813A (en) * | 2020-08-04 | 2020-11-03 | 上海海事大学 | Shore bridge crane reducer state detection method based on improved decision tree algorithm |
| CN112133941A (en) * | 2020-10-10 | 2020-12-25 | 西南交通大学 | Rapid fault diagnosis method for locomotive proton exchange membrane fuel cell system |
| CN112598287A (en) * | 2020-12-25 | 2021-04-02 | 武汉烽火信息集成技术有限公司 | Rail transit health assessment method and system |
| CN113128555A (en) * | 2021-03-09 | 2021-07-16 | 西南交通大学 | Method for detecting abnormality of train brake pad part |
| CN113128555B (en) * | 2021-03-09 | 2022-05-31 | 西南交通大学 | Method for detecting abnormality of train brake pad part |
| CN112991347A (en) * | 2021-05-20 | 2021-06-18 | 西南交通大学 | Three-dimensional-based train bolt looseness detection method |
| CN113255591A (en) * | 2021-06-25 | 2021-08-13 | 四川九通智路科技有限公司 | Bearing fault diagnosis method based on random forest and fusion characteristics |
| CN114397121A (en) * | 2022-01-21 | 2022-04-26 | 西南交通大学 | Bogie fault diagnosis method |
| CN114187533A (en) * | 2022-02-15 | 2022-03-15 | 西南交通大学 | GB-InSAR (GB-InSAR) atmospheric correction method based on random forest time sequence classification |
| CN114187533B (en) * | 2022-02-15 | 2022-05-03 | 西南交通大学 | GB-InSAR (GB-InSAR) atmospheric correction method based on random forest time sequence classification |
| CN114911789A (en) * | 2022-07-15 | 2022-08-16 | 武汉大学 | Water level data fusion method and device based on radar vision multi-sensor data |
| CN117892744A (en) * | 2023-12-27 | 2024-04-16 | 广州市弘宇科技有限公司 | Operation and maintenance method and system of RFID (radio frequency identification) identification equipment in civil aviation logistics field |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108152059B (en) | 2021-03-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108152059A (en) | High-speed train bogie fault detection method based on Fusion | |
| CN104792530B (en) | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression | |
| CN102944418B (en) | Wind turbine generator group blade fault diagnosis method | |
| CN107194433A (en) | A kind of Radar range profile's target identification method based on depth autoencoder network | |
| CN109829236A (en) | A kind of Compressor Fault Diagnosis method based on XGBoost feature extraction | |
| CN104849650B (en) | One kind is based on improved analog-circuit fault diagnosis method | |
| CN106895975A (en) | Method for Bearing Fault Diagnosis based on Stacked SAE deep neural networks | |
| CN108285071A (en) | A kind of elevator Gernral Check-up method based on Bayesian network | |
| CN109572706B (en) | Driving safety evaluation method and device | |
| CN108228716A (en) | SMOTE_Bagging integrated sewage handling failure diagnostic methods based on weighting extreme learning machine | |
| CN106897826A (en) | A kind of street accidents risks appraisal procedure and system | |
| CN106897738A (en) | A kind of pedestrian detection method based on semi-supervised learning | |
| CN106682704A (en) | Method of disease image identification based on hybrid convolutional neural network fused with context information | |
| CN102663264B (en) | Semi-supervised synergistic evaluation method for static parameter of health monitoring of bridge structure | |
| CN1811367A (en) | Integrated supporting vector machine mixed intelligent diagnosing method for mechanical fault | |
| CN106197999A (en) | A kind of planetary gear method for diagnosing faults | |
| CN109597401A (en) | A kind of equipment fault diagnosis method based on data-driven | |
| DE102021124047A1 (en) | Open quantity fault diagnosis method for high speed EMU bearings | |
| CN105572492B (en) | A kind of municipal rail train subordinate inverter trouble-shooter | |
| CN101271625A (en) | Method for detecting freeway traffic event by integration supporting vector machine | |
| CN104809230A (en) | Cigarette sensory quality evaluation method based on multi-classifier integration | |
| CN107024331A (en) | A kind of neutral net is to train motor oscillating online test method | |
| CN109389171A (en) | Medical image classification method based on more granularity convolution noise reduction autocoder technologies | |
| CN105606914A (en) | IWO-ELM-based Aviation power converter fault diagnosis method | |
| CN106769030A (en) | A kind of bearing state tracking and Forecasting Methodology based on MEA BP neural network algorithms |
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 |