CN109685331A - A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning - Google Patents
A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning Download PDFInfo
- Publication number
- CN109685331A CN109685331A CN201811487973.7A CN201811487973A CN109685331A CN 109685331 A CN109685331 A CN 109685331A CN 201811487973 A CN201811487973 A CN 201811487973A CN 109685331 A CN109685331 A CN 109685331A
- Authority
- CN
- China
- Prior art keywords
- hidden layer
- neural network
- neuron
- moment
- layer
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0635—Risk analysis of enterprise or organisation activities
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Artificial Intelligence (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Educational Administration (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
Abstract
The invention discloses a kind of high-speed rail bogie sensor fault diagnosis method based on machine learning.Step of the invention includes: the mapping relations 1) established between input parameter and observed parameter, wherein each observed parameter corresponds to one group of input parameter；2) off-line training is carried out to neural network using observed parameter historical data and its corresponding input parameters history data, obtains a network observations device；3) fault real-time diagnosis is carried out to high-speed rail bogie using the trained network observations device.The present invention has well solved the weakness of sensor conventional failure monitoring, especially traditional shallow-layer neural network the disadvantages of there are gradient decaying, overfitting, Local Minimum, efficiently avoids the complicated processes of artificial extraction feature, improves fault diagnosis effect.
Description
Technical field
The invention belongs to the high-speed rail bogies of car, machine learning field, are related to a kind of method for diagnosing faults, are specially related to
A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning.
Background technique
Bullet train system shows the development trend to become increasingly complex, and bogie is the running gear of high-speed EMUs,
Have the function of carrying, vibration damping, guiding, traction and braking etc., determine train running speed and riding quality, in actual motion
In, frequently, use state is complicated and changeable for train turning, manipulates the environment unfavorable factor in use and maintenance in addition, causes to occur
Various failures, and important device of the sensor as acquisition of information provide guarantee for train operation is reliable and secure, and bogie passes
Sensor is usually operated under adverse circumstances, it is more difficult to be assessed with working condition of the conventional method to sensor, therefore accurately
Fault diagnosis is carried out to bogie sensor and has become the most important thing of safe train operation, conventional hardware redundancy method has been in recent years
The demand that fault diagnosis proposes instantly can not be adapted to, and machine learning neural network is not necessarily to establish because it associates, adaptively certainly
The advantages that accurate mathematical model, is widely used in fault diagnosis field.But how to be carried out using machine learning neural network
The sensor fault diagnosis of bogie, to guarantee operation security, improve maintenance efficiency and avoid unnecessary loss to be urgently
It solves the problems, such as.
Summary of the invention
For the technical problems in the prior art, the purpose of the present invention is to provide one kind to be based on machine learning network
The method for diagnosing faults of observer, this method establish depth confidence network as sensor signal observer, utilize a large amount of operations
Data carry out off-line training to network, obtain optimal network parameter.The estimated value of observer and event are utilized in on-line fault diagnosis
The measuring value of barrier sensor is compared, and analyzes residual error, is diagnosed fault.
The technical solution of the present invention is as follows:
A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning, step include:
1) mapping relations between input parameter and observed parameter are established, wherein the corresponding one group of input ginseng of each observed parameter
Number；
2) neural network is instructed offline using observed parameter historical data and its corresponding input parameters history data
Practice, obtains a network observations device；
3) fault real-time diagnosis is carried out to high-speed rail bogie using the trained network observations device.
Further, the neural network include sequentially connected third hidden layer h3, it is the second hidden layer h2, first implicit
Layer h1 and visual layers V；Wherein, in the neural network output of i-th layer of each neuron respectively as each mind of i+1 layer
It is mutually indepedent between each neuron of same layer through first input, there is a weight w table between the connected neuron of any two
Show its bonding strength, each neuron in the visual layers V is implied equipped with biasing a coefficient a, third hidden layer h3, second
The weight b that each neuron in layer h2, the first hidden layer h1 represents neuron itself weight equipped with one, third hidden layer h3,
Second hidden layer h2, the first hidden layer h1 and visual layers V correspondence are restricted Boltzmann machine.
Further, according to formulaOrCalculate each layer in the neural network
Neuron number；Wherein, a is adjustable constant coefficient, hkFor kth hidden layer, hk-1For -1 hidden layer of kth.
Further, the third hidden layer h3 includes 4 neurons, first comprising 3 neurons, the second hidden layer h2
Hidden layer h1 includes 4 neurons comprising 3 neurons, visual layers V.
Further, the energy function of the neural network are as follows:
Wherein, viIt is the stochastic regime of i-th of neuron of visual layers V, aiIt is viCorresponding biasing coefficient, wijIt is visual layers V i-th
J-th of interneuronal weight in neuron and the first hidden layer h1, hjIt is the random of j-th neuron in the first hidden layer h1
State, bjIt is hjCorresponding biasing；N is the number of neuron in visual layers V, and m is the number of neuron in the first hidden layer h1；
The optimized parameter { w, a, b } of the neural network is obtained by the training neural network.
Further, before being trained to the neural network, data are normalized first, then successively
Unsupervised training is carried out to Boltzmann machine is restricted, then the neural network is finely tuned with back-propagation algorithm, makes net
Network structure is optimal.
Further, the specific steps of the training neural network are as follows:
71) input data is inputted into the neural network, first of the training neural network is restricted Boltzmann
Machine reaches stable state；
72) joint probability distribution for being restricted Boltzmann machine study to external world's input for first is limited as second
The visual layers of Boltzmann machine processed input, until stable state；
73) step 72) is repeated, to the last one is restricted Boltzmann machine and reaches stable state；
74) using maximum likelihood function as objective function, joined with each layer that back-propagation algorithm finely tunes the neural network
Number, is optimal whole network.
Further, the observed parameter includes inclination angular speed, transverse acceleration, vertical acceleration, and the mapping is closed
System includes inclination angular speedTransverse acceleration Vertical acceleration Turn for t moment
To the inclination angular speed of frame, transverse acceleration, vertical acceleration, Pt-1For the speed of service at train t-1 moment, St-1For train t-
The percentage speed variation at 1 moment, θt-1For the inclination angle at train t-1 moment, Qt-1For the vertical deviation at train t-1 moment, Rt-1For column
The lateral displacement at vehicle t-1 moment, Vt-1For the wheel angular velocity at train t-1 moment, αt-1For the inclination angle speed of t-1 moment bogie
Degree, βt-1For the transverse acceleration of t-1 moment bogie, ht-1For the vertical acceleration of t-1 moment bogie.
Further, the method for fault real-time diagnosis being carried out to high-speed rail bogie using the trained network observations device
Are as follows: estimating for high-speed rail bogie items observed parameter is predicted first with the trained network observations device and Real-time Monitoring Data
Then evaluation calculates the residual error between the estimated value of each observed parameter and corresponding true value, then according to the residual of observed parameter
Difference threshold value corresponding with the observed parameter, which is compared, judges whether there is failure.
The beneficial effects of the present invention are:
The weakness of sensor conventional failure monitoring is well solved, there are gradients for especially traditional shallow-layer neural network
The disadvantages of decaying, overfitting, Local Minimum, so that the effect of fault diagnosis is had a greatly reduced quality, after machine Learning Theory proposition not
The problems such as addressing only gradient decaying, and possess the ability for automatically extracting feature, it efficiently avoids manually extracting feature
Complicated processes improve fault diagnosis effect.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention；
Fig. 2 is off-line training and real-time diagnosis flow chart；
Fig. 3 is the neural network structure figure with 3 hidden layers.
Specific embodiment
Technology contents of the invention in order to facilitate understanding by those skilled in the art, with reference to the accompanying drawing to the content of present invention into
One step is illustrated.
The present invention proposes a kind of neural network observation based on machine learning by taking motor train unit bogie sensor fault as an example
Device method for diagnosing faults utilizes a large amount of bogies by establishing observer of the machine learning neural network as sensor signal
Data carry out off-line training to neural network, obtain optimal network number, and in real-time diagnosis, using observer estimated value and
The measured value of fault sensor is compared, and is analyzed residual error and carried out fault diagnosis, and method flow is as shown in Figure 1.
Deep neural network observer is initially set up, essence is that the mapping established between input parameter and observed parameter is closed
System, i.e. y=f (W, i), wherein W indicates weight, and i indicates the input vector in bogie moving model, and y is output vector, defeated
Enter to contain all relevant parameters.The present invention is with the inclination angle speed alpha, transverse acceleration β, vertical in bogie operational process
For acceleration h etc., network observations device is established.
WhereinFor the inclination angular speed of t moment bogie, transverse acceleration, vertical acceleration.Pt-1For train
The speed of service at t-1 moment, St-1For the percentage speed variation at train t-1 moment, θt-1For the inclination angle at train t-1 moment, Qt-1For
The vertical deviation at train t-1 moment, Rt-1For the lateral displacement at train t-1 moment, Vt-1Wheel for the train t-1 moment is diagonally fast
Degree, αt-1For the inclination angular speed of t-1 moment bogie, βt-1For the transverse acceleration of t-1 moment bogie, ht-1For the t-1 moment
The vertical acceleration of bogie.
Off-line training is carried out to network observations device using historical data, using above-mentioned measurable data as input signal,
The signal for needing to diagnose carries out fault real-time diagnosis as output signal, then according to subsequent trained network observations device.That is root
According to the input signal of measurand, output signal training neural network, then according to the output of trained network observations device with
The residual analysis of reality output carries out fault detection, specific implementation such as Fig. 2.The input of visual layers V is data set in Fig. 3, h1,
H2, h3 indicate the first, second and third hidden layer.
This network stacks gradually and forms (specific as shown in Figure 3) by a series of Boltzmann machines that are restricted, and shows in figure
One contains the network of 3 hidden layers, input layer of the previous output layer as next unit, direction be from top to bottom,
So that the output of low layer for top layer provides the association that can be referred to, (i.e. the first hidden layer h1 is one layer of second hidden layer thereon
H2 provides an association that can be referred to, i.e. the second hidden layer h2 provides one for one layer of third hidden layer h3 thereon and can refer to
Association).As shown in figure 3, h3 includes 3 neurons, h2 includes 4 neurons, and h1 includes 3 neurons, and V includes 4 minds
Through member；It is mutually indepedent between same layer, there is a weight w to indicate its bonding strength between the connected neuron of any two, it can
Itself there is a biasing coefficient a depending on layer neuron, hidden neuron has expression its own weight b；Each layer of correspondence is restricted
Boltzmann machine.This network includes n visible elements altogether, m concealed nodes, and the connection between node only exists between layers,
Energy function indicates that there are an energy value, functions between each visible node and each hiding node layer are as follows:
Wherein viAnd hjIt is the stochastic regime of j-th of unit of i-th of unit of visual layers and hidden layer, a respectivelyiAnd bjIt is pair
The biasing answered, wijIt is the weight in i-th of unit of visual layers and hidden layer h1 between j-th of unit, the purpose of training network is exactly
In order to obtain optimal parameter { w, a, b }, specific training sees below continuous step one, two, three.
The neuron number of visual layers v can be determined by input data set, in order to ensure the optimization of network structure, led to
Traditional neural network the number of hidden nodes empirical equation is crossed, proposes a kind of derivation method of new the number of hidden nodes selection (after being detailed in
One group of formula), the empirical equation of traditional 3 layers of neural network is as follows,
Since one layer in deep neural network of output is next layer of input, kth hidden layer n can be obtainedi=hk-1, under
One of 2 formula are iterated to kth hidden layer by face,
Wherein, first layer inputs h0=ni, niFor node in hidden layer, a is adjustable constant coefficient.
Then before being trained using bogie operation data to neural network, place is normalized to data first
Reason.The training of neural network is divided into 2 parts, first is that successively to be restricted Boltzmann machine (each layer network is consistent) into
The unsupervised training of row, second is that being optimal network structure with back-propagation algorithm trim network.(energy function and section
Points are optimum network structures in order to obtain).Specific step is as follows, step 1: with v=h0(the number i.e. after normalized
According to) input training the 1st is used as to be restricted Boltzmann machine (i.e. the first hidden layer h1), reach stable state.Step 2:
The joint probability distribution that first is restricted Boltzmann machine study to external world's input is restricted Boltzmann machine as the 2nd
The visual layers of (i.e. the second hidden layer h2) input, and until stable state, step 3: repeating step 2, to the last one by
Limitation Boltzmann machine (i.e. third hidden layer h3) reaches stable state.Step 4: using maximum likelihood function as objective function,
Each layer parameter is finely tuned with back-propagation algorithm, is optimal whole network.
In first three step, the parameter and θ=(w that are restricted Boltzmann machine are completedij, ai, bj) optimization, wherein
aiIt is the biasing of i-th of node of visual layers, bjIt is the biasing of j-th of node of hidden layer, wijIt is i-th of node of visual layers and implicit
The connection weight of j-th of node of layer.The known Parameter Learning Algorithm for being restricted Boltzmann machine is as follows:
Δwij=ε (< vihj>data-<vihj>1)
Δai=ε (< vi>data-<vi>1)
Δbj=ε (< hj>data-<hj>1)
Wherein<>dataIndicate the probability distribution of hidden layer,<>1It is that the reconstruct that a gibbs sampler obtains is carried out to sample
Sample, ε be multiplication relationship in learning rate and parantheses, the step-length that numerical values recited representation parameter is adjusted every time usually takes 0.008
Between~0.3, ε can use suitable empirical value here.For overcoming the problems, such as that training process is easy to fall into local minimum, this hair
Bright introducing momentum term keeps parameter more new direction and gradient direction inconsistent, the method is as follows:
Wherein, m is momentum term, is a coefficient, and taking m=0.4, t herein according to empirical value is sample the number of iterations,
W when for the t times sample iterationijCorresponding value, θ are to maintain parameter identical with velocity vector direction.
Training can be started after work above by, which completing, is restricted Boltzmann machine, evaluated by reconstructed error by
Whether limitation Boltzmann machine trains completion, and reconstructed error is training data by being restricted Boltzmann machine progress successively lucky cloth
Difference after this sampling with former data, i.e.,Constantly reduce reconstructed error by iteration, until
It is all to be restricted Boltzmann machine training and finish, finally carry out global fine tuning in step 4, deep neural network last
Layer is used for parameter fitting, and therefore, the activation primitive of the last layer chooses hyperbolic tangent function, i.e. f (x)=1-1/ (1+e2x)。
After being trained by historical data to above-mentioned model, real-time estimation can be carried out, estimated value and true value are utilized
Between residual error judge whether failure, the specific method is as follows, and the residual error at certain a moment is described in detail below,
Typical sensor fault has stuck failure, deviation fault, constant gain failures.
Detection threshold value T is arranged to each parameter sensorsk.Compare the size of residual sum threshold value to determine whether there is event
Barrier.Dang ∣ e (k) ∣ < TkIt is determined as fault-free , Dang ∣ e (k) ∣ >=TkIt is determined as faulty, TkValue depending on design parameter,
When determine break down in some period when, according to sensor output value, can intuitively judge fault category, failure classes
The mathematic(al) representation of type are as follows:
Y(t)=ky(t)+ a, t >=T
Wherein k is permanent gain, and k ≠ 1, a are deviation, and a ≠ 0, T are Y at the beginning of failure occurs(t)It refers specifically to not
Know that the generation of failure claims, is exactly that different types of failure is referred to as.
In the case where certain section of time internal fault, the estimated value of observer is utilizedInstead of the true value y in above formula(t),
I.e.
Only it is to be understood that the value of parameter k and a, it will be able to distinguish deviation fault and constant gain failures etc., pass through simple letter
Number fitting obtains the estimated value of k and aWithIt can determine that deviation fault (k=1, ∣ a ∣ >=T according to following corresponding relationshipk),
Constant gain failures (the , ∣ a of k ≠ 1 ∣≤Tk).When detecting sensor fault, the processing such as signal isolation, reconstruct should be taken to arrange in time
It applies, and then guarantees the safe operation of bogie.
Contain the explanation of the preferred embodiment of the present invention above, this be for the technical characteristic that the present invention will be described in detail, and
Be not intended to for summary of the invention being limited in concrete form described in embodiment, according to the present invention content purport carry out other
Modifications and variations are also protected by this patent.The purport of the content of present invention is to be defined by the claims, rather than have embodiment
Specific descriptions are defined.
Claims (9)
1. a kind of high-speed rail bogie sensor fault diagnosis method based on machine learning, step include:
1) mapping relations between input parameter and observed parameter are established, wherein each observed parameter corresponds to one group of input parameter；
2) off-line training is carried out to neural network using observed parameter historical data and its corresponding input parameters history data, obtained
To a network observations device；
3) fault real-time diagnosis is carried out to high-speed rail bogie using the trained network observations device.
2. the method as described in claim 1, which is characterized in that the neural network includes sequentially connected third hidden layer
H3, the second hidden layer h2, the first hidden layer h1 and visual layers V；Wherein, i-th layer of each neuron in the neural network
The input respectively as each neuron of i+1 layer is exported, mutually indepedent between each neuron of same layer, any two are connected
Neuron between there is a weight w to indicate its bonding strength, each neuron in the visual layers V is equipped with biasing and is
Number a, third hidden layer h3, the second hidden layer h2, each neuron in the first hidden layer h1 are equipped with one and represent neuron itself
The weight b of weight, third hidden layer h3, the second hidden layer h2, the first hidden layer h1 and visual layers V correspondence are restricted Bohr hereby
Graceful machine.
3. method according to claim 2, which is characterized in that according to formulaOrCalculate each layer in the neural network of neuron
Number；Wherein, a is adjustable constant coefficient, hkFor kth hidden layer, hk-1For -1 hidden layer of kth.
4. method as claimed in claim 2 or claim 3, which is characterized in that the third hidden layer h3 includes 3 neurons, second
Hidden layer h2 includes 4 neurons comprising 3 neurons, visual layers V comprising 4 neurons, the first hidden layer h1.
5. method according to claim 2, which is characterized in that the energy function of the neural network are as follows:
Wherein, viIt is i-th of nerve of visual layers V
The stochastic regime of member, aiIt is viCorresponding biasing coefficient, wijIt is j-th in i-th of neuron of visual layers V and the first hidden layer h1
Interneuronal weight, hjIt is the stochastic regime of j-th of neuron in the first hidden layer h1, bjIt is hjCorresponding biasing；N is can
Depending on the number of neuron in layer V, m is the number of neuron in the first hidden layer h1；Institute is obtained by the training neural network
State the optimized parameter { w, a, b } of neural network.
6. method according to claim 2, which is characterized in that before being trained to the neural network, first to data
It is normalized, then successively carries out unsupervised training to being restricted Boltzmann machine, then calculated with backpropagation
Method finely tunes the neural network, is optimal network structure.
7. method as claimed in claim 6, which is characterized in that the specific steps of the training neural network are as follows:
71) input data is inputted into the neural network, first of the training neural network is restricted Boltzmann machine, makes
It reaches stable state；
72) joint probability distribution that first is restricted Boltzmann machine study to external world's input is restricted glass as second
The visual layers of the graceful machine of Wurz input, until stable state；
73) step 72) is repeated, to the last one is restricted Boltzmann machine and reaches stable state；
74) using maximum likelihood function as objective function, each layer parameter of the neural network is finely tuned with back-propagation algorithm, is made
Whole network is optimal.
8. the method as described in claim 1, which is characterized in that the observed parameter include inclination angular speed, transverse acceleration,
Vertical acceleration, the mapping relations include inclination angular speedIt is horizontal
To accelerationVertical acceleration For the inclination angular speed of t moment bogie, transverse acceleration, vertical acceleration, Pt-1For train t-
The speed of service at 1 moment, St-1For the percentage speed variation at train t-1 moment, θt-1For the inclination angle at train t-1 moment, Qt-1For column
The vertical deviation at vehicle t-1 moment, Rt-1For the lateral displacement at train t-1 moment, Vt-1For the wheel angular velocity at train t-1 moment,
αt-1For the inclination angular speed of t-1 moment bogie, βt-1For the transverse acceleration of t-1 moment bogie, ht-1Turn for the t-1 moment
To the vertical acceleration of frame.
9. the method as described in claim 1, which is characterized in that using the trained network observations device to high-speed rail bogie
The method for carrying out fault real-time diagnosis are as follows: predict high-speed rail first with the trained network observations device and Real-time Monitoring Data
Then the estimated value of bogie items observed parameter calculates residual between the estimated value of each observed parameter and corresponding true value
Then difference is compared according to the residual error of observed parameter threshold value corresponding with the observed parameter and judges whether there is failure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811487973.7A CN109685331A (en) | 2018-12-06 | 2018-12-06 | A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811487973.7A CN109685331A (en) | 2018-12-06 | 2018-12-06 | A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109685331A true CN109685331A (en) | 2019-04-26 |
Family
ID=66187305
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811487973.7A Pending CN109685331A (en) | 2018-12-06 | 2018-12-06 | A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109685331A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738996A (en) * | 2020-06-09 | 2020-10-02 | 交通运输部公路科学研究所 | Bridge health monitoring and early warning system based on machine learning |
| CN111879534A (en) * | 2020-07-29 | 2020-11-03 | 上海杰之能软件科技有限公司 | Performance detection method, system and equipment for urban rail vehicle braking system |
| CN114207336A (en) * | 2019-07-29 | 2022-03-18 | 西门子股份公司 | Diagnostic system for a valve that can be actuated by means of a control pressure |
| CN115326437A (en) * | 2022-08-19 | 2022-11-11 | 华东交通大学 | Embedded system device and method for monitoring and diagnosing bogie faults |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104616033A (en) * | 2015-02-13 | 2015-05-13 | 重庆大学 | Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine) |
| CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
| CN108089099A (en) * | 2017-12-18 | 2018-05-29 | 广东电网有限责任公司佛山供电局 | The diagnostic method of distribution network failure based on depth confidence network |
| CN108256556A (en) * | 2017-12-22 | 2018-07-06 | 上海电机学院 | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network |
| CN108304941A (en) * | 2017-12-18 | 2018-07-20 | 中国软件与技术服务股份有限公司 | A kind of failure prediction method based on machine learning |
-
2018
- 2018-12-06 CN CN201811487973.7A patent/CN109685331A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104616033A (en) * | 2015-02-13 | 2015-05-13 | 重庆大学 | Fault diagnosis method for rolling bearing based on deep learning and SVM (Support Vector Machine) |
| CN106769048A (en) * | 2017-01-17 | 2017-05-31 | 苏州大学 | Adaptive deep confidence network bearing fault diagnosis method based on Nesterov momentum method |
| CN108089099A (en) * | 2017-12-18 | 2018-05-29 | 广东电网有限责任公司佛山供电局 | The diagnostic method of distribution network failure based on depth confidence network |
| CN108304941A (en) * | 2017-12-18 | 2018-07-20 | 中国软件与技术服务股份有限公司 | A kind of failure prediction method based on machine learning |
| CN108256556A (en) * | 2017-12-22 | 2018-07-06 | 上海电机学院 | Wind-driven generator group wheel box method for diagnosing faults based on depth belief network |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114207336A (en) * | 2019-07-29 | 2022-03-18 | 西门子股份公司 | Diagnostic system for a valve that can be actuated by means of a control pressure |
| CN111738996A (en) * | 2020-06-09 | 2020-10-02 | 交通运输部公路科学研究所 | Bridge health monitoring and early warning system based on machine learning |
| CN111738996B (en) * | 2020-06-09 | 2023-04-07 | 交通运输部公路科学研究所 | Bridge health monitoring and early warning system based on machine learning |
| CN111879534A (en) * | 2020-07-29 | 2020-11-03 | 上海杰之能软件科技有限公司 | Performance detection method, system and equipment for urban rail vehicle braking system |
| CN111879534B (en) * | 2020-07-29 | 2022-04-15 | 上海杰之能软件科技有限公司 | Performance detection method, system and equipment for urban rail vehicle braking system |
| CN115326437A (en) * | 2022-08-19 | 2022-11-11 | 华东交通大学 | Embedded system device and method for monitoring and diagnosing bogie faults |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109685331A (en) | A kind of high-speed rail bogie sensor fault diagnosis method based on machine learning | |
| CN107941537B (en) | A kind of mechanical equipment health state evaluation method | |
| CN104712542B (en) | A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults | |
| Cao et al. | The fault diagnosis of a switch machine based on deep random forest fusion | |
| CN108931387B (en) | Fault diagnosis method based on multi-sensor signal analysis and capable of providing accurate diagnosis decision | |
| CN104792530B (en) | Deep-learning rolling bearing fault diagnosis method based on SDA (stacked denoising autoencoder) and Softmax regression | |
| CN101697079B (en) | Blind system fault detection and isolation method for real-time signal processing of spacecraft | |
| CN108569607A (en) | Elevator faults method for early warning based on bidirectional valve controlled Recognition with Recurrent Neural Network | |
| CN106649727A (en) | Database construction method used for fault detection of unmanned aerial vehicle flight control system | |
| CN109581871B (en) | Industrial control system intrusion detection method of immune countermeasure sample | |
| CN108332970A (en) | A kind of Method for Bearing Fault Diagnosis based on LS-SVM and D-S evidence theory | |
| CN113188794B (en) | Gearbox fault diagnosis method and device based on improved PSO-BP neural network | |
| CN104596780B (en) | Diagnosis method for sensor faults of motor train unit braking system | |
| US11780452B2 (en) | Method and system for fault diagnoses of intelligent vehicles | |
| CN110261109A (en) | A kind of Fault Diagnosis of Roller Bearings based on bidirectional memory Recognition with Recurrent Neural Network | |
| CN110705812A (en) | Industrial fault analysis system based on fuzzy neural network | |
| CN114297918A (en) | Aero-engine residual life prediction method based on full-attention depth network and dynamic ensemble learning | |
| CN113485302A (en) | Vehicle operation process fault diagnosis method and system based on multivariate time sequence data | |
| CN108427400A (en) | A kind of aircraft airspeed pipe method for diagnosing faults based on neural network Analysis design | |
| CN109443766A (en) | A kind of heavy-duty vehicle gearbox gear Safety Analysis Method | |
| Shang et al. | Fault diagnosis method of rolling bearing based on deep belief network | |
| CN104142680B (en) | Multiple-sensor fault diagnosis system and method based on robust input training neural network | |
| Chen et al. | Dynamic vision enabled contactless cross-domain machine fault diagnosis with neuromorphic computing | |
| CN114757365A (en) | High-speed railway roadbed settlement prediction and early warning method based on deep learning | |
| CN117219180B (en) | Method and system for monitoring, evaluating and dynamically regulating and enhancing mineralization effect of carbon dioxide |
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 |
Application publication date: 20190426 |
|
| RJ01 | Rejection of invention patent application after publication |