CN102063109B - Neural network-based subway train fault diagnosis device and method - Google Patents
Neural network-based subway train fault diagnosis device and method Download PDFInfo
- Publication number
- CN102063109B CN102063109B CN201010563179A CN201010563179A CN102063109B CN 102063109 B CN102063109 B CN 102063109B CN 201010563179 A CN201010563179 A CN 201010563179A CN 201010563179 A CN201010563179 A CN 201010563179A CN 102063109 B CN102063109 B CN 102063109B
- Authority
- CN
- China
- Prior art keywords
- neural network
- data
- module
- subway train
- basis function
- 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.)
- Active
Links
Images
Landscapes
- Testing And Monitoring For Control Systems (AREA)
- Train Traffic Observation, Control, And Security (AREA)
Abstract
The invention discloses a neural network-based subway train fault diagnosis device and a neural network-based subway train fault diagnosis method. A lower computer classifies acquired subway train state information according to a functional unit, concentrates and summarizes data, and sends the data to an upper computer; the upper computer receives, processes and stores the subway train state data summarized by the lower computer; the upper computer comprises a data acquisition module and a neural network module; the data acquisition module finishes the classification of the acquired subway train state data; the lower computer acquires real-time data information of a subway train; and when the subway train fails or is to fail, the neural network module of the upper computer performs corresponding judgment and pre-judgment according to the output of a radial basis function neural network established during training and outputs fault information to a fault diagnosis result module. By the embodiment of the invention, the technical problems of insufficient acquired and analyzed data volume, low data processing efficiency and a complicated calculation process in the prior art can be well solved.
Description
Technical field
The present invention relates to a kind of trouble-shooter and method thereof; Especially a kind of trouble-shooter and the method thereof that can carry out classification processing to fault that is applied to subway train based on radial basis function neural network RBFNN (Radial Basis Function Neural Network, radial basis function neural network).
Background technology
Along with the fast development of urban rail transit in China, how to ensure that the safe operation of subway train becomes more and more urgent difficult point problem.China has disposed a large amount of city rail supervisory systems at present, has brought into play vital role at aspects such as ensureing operation security.But existing technology can't provide desirable solution to the aspects such as fault diagnosis of subway train, and this also becomes the bottleneck that the restriction urban rail transit in China continues development.
Existing built-in intelligent fault diagnosing device based on the data fusion pattern-recognition; Adopt neural network; In conjunction with sensor crowd information based on the data fusion pattern-recognition; By multiple neural network local diagnosis and decision level fusion pattern-recognition intelligent fault diagnosis mechanism, realize the function of data storage, graphic presentation and remote data transmission, accomplish the fault diagnosis task.
Existing built-in intelligent fault diagnosing device based on the data fusion pattern-recognition only is used to the information of autobiography sensor, and data volume is abundant inadequately.Underuse like the switching value information that characterizes the subway train running status etc., it is necessary that these information are made fault diagnosis often accurately, fast.
Existing built-in intelligent fault diagnosing device based on the data fusion pattern-recognition merges all data, rather than classifies by functional unit etc., causes efficient not high.
Existing built-in intelligent fault diagnosing device based on the data fusion pattern-recognition, the model that its fault diagnosis result module adopts is BP (Back Propagation, a backpropagation) model.Although this model investigation comparative maturity has tediously long iterative computation process in the algorithm, and is absorbed in local extremum easily.
Summary of the invention
The present invention provides a kind of subway train trouble-shooter and method thereof based on neural network; The collection analysis data volume that this invention can overcome the prior art existence well is abundant inadequately; Data-handling efficiency is not high; Computation process complicated technology problem; This invents the related data that described subway train trouble-shooter and method thereof can comprehensive collection subway train running statuses, carries out intellectual analysis after the classification processing again and handles, thereby can accurately export subway train fault that takes place and the fault that will take place.
The present invention provides a kind of embodiment of the subway train trouble-shooter based on neural network; A kind of subway train trouble-shooter based on neural network; Comprise the data acquisition bottom; Fault diagnosis result module, slave computer and host computer, slave computer according to functional unit to the collection of classifying from the subway train status information data of data acquisition bottom; And with gathering again after the data centralization to host computer; Host computer is responsible for receiving, handle and store the subway train status data that slave computer gathers, and host computer comprises data aggregation module and neural network module, and data aggregation module is at first accomplished the classification processing to the subway train status information data that collects;
In the training stage, neural network module classification processing the data after good utilize training sample to generate each required neural network of subway train fault diagnosis then as the training sample of each neural network;
In the application stage; Slave computer is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module of host computer is made corresponding judgment and anticipation according to the output of the neural network of setting up in the training stage, and exports the failure message result to the fault diagnosis result module.
As a kind of further embodiment of subway train trouble-shooter of the present invention based on neural network; Neural network module is the neural network based on RBF; Neural network module comprises input layer, hidden layer and output layer; Hidden layer produces the localization response to the excitation of input layer, and the characteristic function of hidden layer adopts nonlinear RBF, and output layer is to basis function output the carrying out linear combination of hidden layer.
As a kind of further embodiment of subway train trouble-shooter of the present invention based on neural network; Slave computer comprises gathers modular converter, analogue collection module, digital data acquisition module and bus interface module, gathers modular converter and gathers the data message that subway train installs sensor feedback additional; The analog quantity information of analogue collection module acquisition tables expropriation of land iron train operation state; The digital information of digital data acquisition module acquisition tables expropriation of land iron train operation state; Bus interface module is gathered the data message on the subway train bus.
As a kind of further embodiment of subway train trouble-shooter of the present invention based on neural network; Data aggregation module comprises system level data collection modules, central control unit data aggregation module, traction unit data aggregation module, brake unit data aggregation module and door control unit data aggregation module, data aggregation module with the sensing data that collects, analog data, digital data and subway train bus data according to functional unit category set synthesis system grade data, central control unit data, traction unit data, brake unit data and door control unit data.
As a kind of further embodiment of subway train trouble-shooter of the present invention based on neural network; Neural network module comprises system-level radial basis function neural network, central control unit radial basis function neural network, traction unit radial basis function neural network, brake unit radial basis function neural network and door control unit radial basis function neural network; System-level radial basis function neural network links to each other with the system level data collection modules; The central control unit radial basis function neural network links to each other with the central control unit data aggregation module; The traction unit radial basis function neural network links to each other with the traction unit data aggregation module; The brake unit radial basis function neural network links to each other with the brake unit data aggregation module, and the door control unit radial basis function neural network links to each other with the door control unit data aggregation module.
As a kind of further embodiment of subway train trouble-shooter of the present invention based on neural network; The fault diagnosis result module comprises system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module; System-level radial basis function neural network links to each other with the system level data collection modules; The central control unit radial basis function neural network links to each other with the central control unit data aggregation module; The traction unit radial basis function neural network links to each other with the traction unit data aggregation module; The brake unit radial basis function neural network links to each other with the brake unit data aggregation module, and the door control unit radial basis function neural network links to each other with the door control unit data aggregation module.
The present invention also provides a kind of embodiment of the subway train method for diagnosing faults based on neural network, and a kind of utilization is carried out the method for subway train fault diagnosis based on the subway train trouble-shooter of neural network, may further comprise the steps:
Slave computer installs sensing data, analog quantity information, digital information and bus interface information additional to the collection of classifying from the subway train status information data of data acquisition bottom according to subway train, and with gathering again after the data centralization to host computer;
The subway train status data that host computer receives, handles and the storage slave computer gathers, the data aggregation module of host computer is classified and overall treatment according to system level data, central control unit data, traction unit data, brake unit data and door control unit data to the subway train status information data that collects;
Neural network module is the neural network based on RBF; In the training stage; Data after classification of neural network module handle and overall treatment are good utilize training sample to set up each required radial basis function neural network of subway train fault diagnosis respectively as the training sample of system-level radial basis function neural network, central control unit radial basis function neural network, traction unit radial basis function neural network, brake unit radial basis function neural network and door control unit radial basis function neural network then;
In the application stage; Slave computer is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module of host computer is made corresponding judgment and anticipation according to the output of the radial basis function neural network of setting up in the training stage, and exports the failure message result to system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module respectively.
As a kind of further embodiment of subway train method for diagnosing faults based on neural network of the present invention, the subway train method for diagnosing faults is further comprising the steps of:
At learning phase, to the typical fault and the critical fault of subway train, produce dissimilar faults through simulation, the training sample vector that obtains is as the input of neural network;
Hidden layer neuron is learnt the input sample of neural network through the clustering algorithm process of unsupervised learning;
The output layer neuron is set up radial basis function neural network through the least square method learning algorithm process of supervision is arranged according to the optimum weight coefficient that calculates;
In the application stage; When the subway train breaks down maybe will break down the time; As the input of radial basis function neural network, the stable state output of radial basis function neural network is the failure mode that the failure mode that taken place maybe will take place after the real time data classification processing of gathering.
As a kind of further embodiment of subway train method for diagnosing faults based on neural network of the present invention, the clustering algorithm process of unsupervised learning may further comprise the steps:
Each cluster centre of initialization η
j, j=1,2 ..., L, L are the cluster number, x
N, iBe training sample vector, picked at random initialization cluster centre from the training sample vector of input;
The training sample vector of input is if satisfy condition
Then the training sample vector with input belongs to the q class;
After classification finishes, obtain the new cluster centre and the width of each classification, that is:
As a kind of further embodiment of subway train method for diagnosing faults of the present invention, have the least square method learning algorithm process of supervision may further comprise the steps based on neural network:
The mark output vector is y
N, k, then k output node had:
In the formula, ω
IkExpression hidden layer i unit is to the weights of output layer k unit, η
iThe cluster centre of doing a certain type of input can be got in the position of representing certain RP, and R () is the non-linear radial symmetry basis function of choosing, and the RBF of employing is the Gaussian function, || || expression is apart from certain some η
iDistance measure, L is the hidden layer unit number;
R in the formula
jBe hidden layer j neuronic output, y
jBe hidden layer j neuronic Gaussian function center, σ is the width of Gaussian function;
If sorter is output as:
The error function of define grid output is:
In the formula, d
kBe output layer k neuronic desired output, y
N, kBe actual output value, formula (4) substitution formula (7) can be got:
Find out and make error function E one group of weight coefficient ω hour
Ik, order:
Draw system of equations:
Find the solution optimum weight coefficient ω
Ikopt, optimum weight coefficient ω
IkoptAfter confirming, radial basis function neural network is set up and is finished.
Through using embodiment of the present invention described a kind of subway train trouble-shooter and method thereof based on neural network; Characteristics have been utilized based on radial basis function neural network is restrained fast, operand is little and non-linear approximation capability is strong; Related data that can comprehensive collection subway train running status remakes the input into neural network after the classification processing.Owing to introduced radial basis function neural network, judge accurately and predict thereby can make to fault that takes place and the fault that will take place.The data message amount is abundanter, and fault diagnosis is more accurate.Data message is through after the classification processing, and fault diagnosis efficiency has obtained further raising.Adopt the hierarchical structure of " bottom sensor-slave computer-host computer ", make system's clear in structure more.
Description of drawings
In order to be illustrated more clearly in the embodiment of the invention or technical scheme of the prior art; To do to introduce simply to the accompanying drawing of required use in embodiment or the description of the Prior Art below; Obviously, the accompanying drawing in describing below only is some embodiments of the present invention, for those of ordinary skills; Under the prerequisite of not paying creative work, can also obtain other accompanying drawing according to these accompanying drawings.
Fig. 1 is a kind of based on the subway train trouble-shooter of neural network and the structural principle block diagram of method thereof for the present invention;
Fig. 2 is a kind of based on the subway train trouble-shooter of neural network and the system architecture diagram of method thereof for the present invention;
Fig. 3 is a kind of based on the subway train trouble-shooter of neural network and the upper computer software process flow diagram of method thereof for the present invention;
Fig. 4 is the basic structure synoptic diagram of the radial basis function neural network of a kind of subway train trouble-shooter and method thereof based on neural network of the present invention;
Wherein, 1-data acquisition bottom, 2-data aggregation module, 3-neural network module, 4-fault diagnosis result module, 5-slave computer, 6-host computer, 7-input layer, 8-hidden layer, 9-output layer.
Embodiment
To combine the accompanying drawing in the embodiment of the invention below, the technical scheme in the embodiment of the invention is carried out clear, intactly description, obviously, described embodiment only is a part of embodiment of the present invention, rather than whole embodiment.Based on the embodiment among the present invention, those of ordinary skills are not making the every other embodiment that is obtained under the creative work prerequisite, all belong to the scope of the present invention's protection.
Embodiment as a kind of subway train trouble-shooter based on neural network of the present invention; As shown in Figure 2; Subway train trouble-shooter based on neural network comprises data acquisition bottom 1, fault diagnosis result module 4, slave computer 5 and host computer 6; Data acquisition bottom 1 comprises various sensors, and being mainly existing subway train does not have and the extra sensor that installs additional of system's needs.Slave computer 5 according to functional unit to the collection of classifying from the subway train status information data of data acquisition bottom 1; And with gathering again after the data centralization to host computer 6; The subway train status data that slave computer 6 gathers is responsible for receiving, handles and is stored to host computer 6; Host computer 6 comprises data aggregation module 2 and neural network module 3, and data aggregation module 2 is at first accomplished the classification processing to the subway train status information data that collects;
Wherein, neural network module 3 is the neural network based on RBF.In the training stage, neural network module 3 classification processing the data after good utilize training sample to generate each required radial basis function neural network of subway train fault diagnosis then as the training sample of each radial basis function neural network;
In the application stage; Slave computer 5 is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module 3 of host computer 6 is made corresponding judgment and anticipation according to the output of the radial basis function neural network of setting up in the training stage, and exports the failure message result to fault diagnosis result module 4.Wherein, fault diagnosis result module 4 can be external display screen or short message prompt mode.
Neural network module 3 comprises input layer 7, hidden layer 8 and output layer 9; The excitation of 8 pairs of input layers 7 of hidden layer produces the localization response; The characteristic function of hidden layer 8 adopts nonlinear RBF, basis function output the carrying out linear combination of 9 pairs of hidden layers 8 of output layer.Slave computer 5 comprises gathers modular converter, analogue collection module, digital data acquisition module and bus interface module, gathers modular converter and gathers the data message that subway train installs sensor feedback additional, like data such as acceleration, vibrations; The analog quantity information of analogue collection module acquisition tables expropriation of land iron train operation state is like voltage, electric current, temperature etc.; The digital information of digital data acquisition module acquisition tables expropriation of land iron train operation state is like switch closed condition, spacer assembly state etc.; Bus interface module is gathered the data message on the subway train bus, is primarily aimed at different subway train bus (like MVB, WTB etc.), through corresponding interface and agreement, and the subway train related data that collection needs.Data aggregation module 2 further comprises system level data collection modules, central control unit data aggregation module, traction unit data aggregation module, brake unit data aggregation module and door control unit data aggregation module, data aggregation module 2 with the sensing data that collects, analog data, digital data and subway train bus data according to functional unit category set synthesis system grade data, central control unit data, traction unit data, brake unit data and door control unit data.
Fault diagnosis result module 4 comprises system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module; System-level radial basis function neural network links to each other with the system level data collection modules; The central control unit radial basis function neural network links to each other with the central control unit data aggregation module; The traction unit radial basis function neural network links to each other with the traction unit data aggregation module; The brake unit radial basis function neural network links to each other with the brake unit data aggregation module, and the door control unit radial basis function neural network links to each other with the door control unit data aggregation module.
As shown in Figure 1; Embodiment of the present invention makes full use of the network topology relation of each equipment room of subway train; Various types of data is pressed the functional unit classification; Data like central control unit comprise subway train state, communications status, networking command, digital input/output signal and analog input signal etc.; The data of traction unit comprise traction duty, input instruction, motor speed, motor torque, current of electric, net pressure, motor temperature, level signal and load signal etc.; The data of brake unit comprise wheel footpath, load signal, rate signal, braking level, sensor states etc., and the data of door control unit comprise car door opening state, door contact interrupter feedback, door velocity feedback, door anti-pinch state, door excision state etc.According to the data of each unit mode, produce training sample again, utilize training sample to set up the corresponding radial basis function neural network in each unit, can make the fault diagnosis of each unit through fault simulation.Simultaneously; According to logical relation; With correlative information between the unit, extract same mode like subway train state, communications status, networking command, digital input/output signal, analog input signal, rate signal and level signal etc. through fault simulation; Produce training sample, utilize training sample to set up the radial basis function neural network that is used for subway train fault diagnosis (system-level) again.With comprehensive, improved fault diagnosis efficiency through classification.
As shown in Figure 4 is the basic structure of radial primary function network, and it can be realized by input vector X:{x
1..., x
NTo output vector Y:{y
1..., y
MMapping (or classification).The radial basis function neural network that the specific embodiment of the invention adopts is one type of three layers of special feedforward network, is suitable for function and forces into and classify.The characteristic function of its hidden layer unit adopts nonlinear RBF, can produce the localization response to the excitation of input layer, promptly only when import drop on a certain appointment of the input space among a small circle in the time, significant non-zero response just can be made in the hidden layer unit.The output layer node is then exported the basis function of hidden layer and is carried out linear combination.Compare with the BP net, structure is simpler, and pace of learning is faster.Not only avoided iterative computation process and the possibility that is absorbed in local extremum tediously long in the BP algorithm, and pace of learning is than high 3~4 one magnitude of BP algorithm.
As the embodiment of a kind of subway train method for diagnosing faults based on neural network of the present invention, as shown in Figure 3, may further comprise the steps based on the subway train method for diagnosing faults of neural network:
The subway train status data that host computer 6 receives, handles and storage slave computer 6 gathers, the subway train status information data that 2 pairs of the data aggregation module of host computer 6 collect is classified and overall treatment according to system level data, central control unit data, traction unit data, brake unit data and door control unit data;
In the training stage; Data after 3 classification of neural network module and overall treatment are good utilize training sample to set up each required radial basis function neural network of subway train fault diagnosis respectively as the training sample of system-level radial basis function neural network, central control unit radial basis function neural network, traction unit radial basis function neural network, brake unit radial basis function neural network and door control unit radial basis function neural network then;
In the application stage; Slave computer 5 is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module 3 of host computer 6 is made corresponding judgment and anticipation according to the output of the radial basis function neural network of setting up in the training stage, and exports the failure message result to system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module respectively.
As a kind of typical embodiment; Based on the subway train trouble-shooter of neural network at first to situation such as the typical fault of subway train and critical faults; Produce n class failure mode through simulation, the information such as related data on analog quantity, digital quantity and the subway train bus of the detection data of the autobiography sensor of collecting, sign subway train state and mobile unit.Then data message is classified and comprehensive (system-level) according to functional unit (central control unit, traction unit, brake unit, door control unit etc.), form and set up the required training sample vector x of each radial basis function neural network
N, iFor example: the training sample vector of central control unit is x
N, 1, the training sample vector of traction unit is x
N, 2, the training sample vector of brake unit is x
N, 3, the training sample vector of door control unit is x
N, 4, system-level training sample vector is x
N, 5Can confirm that through these training samples each neural network input layer number is n, the node number of output layer is log
2N.The clustering algorithm of unsupervised learning has been adopted in hidden layer neuron study; The output layer neuron adopts the least square method learning method that supervision is arranged.Training just obtains available neural network after accomplishing, and network parameter can deposit in the storer and store.The example that is established as with the central control unit radial basis function neural network specifies below.
For central control unit, after simulation produces n class fault, the training sample vector x that obtains
N, iComprise subway train state, communications status, networking command, digital input/output signal and analog input signal etc., this training sample will be as the input of neural network.
Adopt the clustering algorithm of unsupervised learning to the study of hidden layer neuron, this method is the most classical RBF (Radial Basis Function, RBF) learning algorithm.Concrete steps are following:
(1) each cluster centre of initialization η
j, j=1,2 ..., L, L are the cluster number.Initial cluster center can be from training sample picked at random;
(2) the input training sample is if satisfy condition
Then it is belonged to the q class;
(3) after classification finishes, can obtain the new center and the width of each classification, that is:
The least square method learning method of supervision is then adopted in neuronic study to output layer.The note output vector is y
N, k, then k output node had:
In the formula, ω
IkExpression hidden layer i unit is to the weights of output layer k unit, η
iThe cluster centre of doing a certain type of input can be got in the position of representing certain RP, and R () is certain non-linear radial symmetry basis function of choosing, || || expression is apart from certain some η
iDistance measure, L is the hidden layer unit number.Usually the RBF that adopts is the Gaussian function
R in the formula
jBe hidden layer j neuronic output, y
jBe hidden layer j neuronic Gaussian function center, σ is the width of Gaussian function.
If sorter is output as
If the error function of define grid output does
In the formula, d
kBe output layer k neuronic desired output, y
N, kBe actual output value.Formula (4) substitution following formula can be got
Make error function E one group of weight coefficient ω hour for finding out
Ik, can make
Can obtain a series of system of equations
Find the solution above-mentioned system of equations and can obtain optimum weight coefficient ω
Ikopt
Optimum weight coefficient ω
IkoptAfter confirming, the central control unit radial basis function neural network is just set up and is finished.
When the central control unit of subway train breaks down maybe will break down the time; As the input of central control unit radial basis function neural network, the stable state output of radial basis function neural network is the fault that the failure mode of the central control unit that has taken place maybe will take place after the real time data classification processing of gathering.
Set up traction unit, brake unit, door control unit and system-level etc. radial basis function neural network through identical method.When the unit of subway train or system break down maybe will break down the time; After the classification of real time data process and overall treatment of gathering; The fault that the failure mode that taken place of being the radial basis function neural network that trains of input, its stable state output maybe will take place.
The above only is a preferred implementation of the present invention; Should be pointed out that for those skilled in the art, under the prerequisite that does not break away from the principle of the invention; Can also make some improvement and retouching, these improvement and retouching also should be regarded as protection scope of the present invention.
Claims (9)
1. subway train trouble-shooter based on neural network; It is characterized in that: comprise data acquisition bottom (1); Fault diagnosis result module (4); Slave computer (5) and host computer (6); To the collection of classifying from the subway train status data of data acquisition bottom (1), and with gathering after the data centralization to host computer (6), host computer (6) is responsible for receiving, handle the subway train status data that storage slave computer (6) also gathers to slave computer (5) again according to functional unit; Host computer (6) comprises data aggregation module (2) and neural network module (3), and said data aggregation module (2) is at first accomplished the classification processing to the subway train status data that collects; Said neural network module (3) is the neural network based on RBF; Neural network module (3) comprises input layer (7), hidden layer (8) and output layer (9); Hidden layer (8) produces the localization response to the excitation of input layer (7); The characteristic function of hidden layer (8) adopts nonlinear RBF, and output layer (9) is to nonlinear RBF output the carrying out linear combination of hidden layer (8);
In the training stage, neural network module (3) classification processing the data after good utilize training sample to generate each required neural network of subway train fault diagnosis then as the training sample of each neural network;
In the application stage; Slave computer (5) is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module (3) of host computer (6) is made corresponding judgment and anticipation according to the output of the neural network of setting up in the training stage, and exports the failure message result to fault diagnosis result module (4).
2. a kind of subway train trouble-shooter according to claim 1 based on neural network; It is characterized in that: described slave computer (5) comprises gathers modular converter, analogue collection module, digital data acquisition module and bus interface module, gathers modular converter and gathers the data message that subway train installs sensor feedback additional; The analog quantity information of analogue collection module acquisition tables expropriation of land iron train operation state; The digital information of digital data acquisition module acquisition tables expropriation of land iron train operation state; Bus interface module is gathered the data message on the subway train bus.
3. a kind of subway train trouble-shooter according to claim 1 and 2 based on neural network; It is characterized in that: described data aggregation module (2) comprises system level data collection modules, central control unit data aggregation module, traction unit data aggregation module, brake unit data aggregation module and door control unit data aggregation module, data aggregation module (2) with the sensing data that collects, analog data, digital data and subway train bus data according to functional unit category set synthesis system grade data, central control unit data, traction unit data, brake unit data and door control unit data.
4. a kind of subway train trouble-shooter according to claim 3 based on neural network; It is characterized in that: described neural network module (3) comprises system-level radial basis function neural network, central control unit radial basis function neural network, traction unit radial basis function neural network, brake unit radial basis function neural network and door control unit radial basis function neural network; System-level radial basis function neural network links to each other with the system level data collection modules; The central control unit radial basis function neural network links to each other with the central control unit data aggregation module; The traction unit radial basis function neural network links to each other with the traction unit data aggregation module; The brake unit radial basis function neural network links to each other with the brake unit data aggregation module, and the door control unit radial basis function neural network links to each other with the door control unit data aggregation module.
5. a kind of subway train trouble-shooter according to claim 4 based on neural network; It is characterized in that: described fault diagnosis result module (4) comprises system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module; The system-level malfunction module links to each other with system-level radial basis function neural network; The central control unit malfunctioning module links to each other with the central control unit radial basis function neural network; The traction unit malfunctioning module links to each other with the traction unit radial basis function neural network; The brake unit malfunctioning module links to each other with the brake unit radial basis function neural network, and the door control unit malfunctioning module links to each other with the door control unit radial basis function neural network.
6. one kind is utilized that claim 1 is described carries out the method for subway train fault diagnosis based on the subway train trouble-shooter of neural network, it is characterized in that, may further comprise the steps:
Slave computer (5) installs data message on the digital information of the analog quantity information of the data message of sensor feedback, sign subway train running status that analogue collection module is gathered, sign subway train running status that the digital data acquisition module is gathered and the subway train bus that bus interface module is gathered additional to the collection of classifying from the subway train status data of data acquisition bottom (1) according to the subway train of gathering the modular converter collection, and with gathering again after the data centralization to host computer (6);
The subway train status data that host computer (6) receives, handles and storage slave computer (6) gathers, the data aggregation module (2) of host computer (6) is classified and overall treatment according to system level data, central control unit data, traction unit data, brake unit data and door control unit data to the subway train status data that collects;
Neural network module (3) is the neural network based on RBF; In the training stage; Neural network module (3) respectively as the training sample of system-level radial basis function neural network, central control unit radial basis function neural network, traction unit radial basis function neural network, brake unit radial basis function neural network and door control unit radial basis function neural network, utilizes classification and the data of overall treatment after good training sample to set up each required radial basis function neural network of subway train fault diagnosis then;
In the application stage; Slave computer (5) is constantly gathered the real time data information of subway train; When the subway train breaks down maybe will break down the time; The neural network module (3) of host computer (6) is made corresponding judgment and anticipation according to the output of the radial basis function neural network of setting up in the training stage, and exports the failure message result to system-level malfunction module, central control unit malfunctioning module, traction unit malfunctioning module, brake unit malfunctioning module and door control unit malfunctioning module respectively.
7. the method for subway train fault diagnosis according to claim 6 is characterized in that, said method is further comprising the steps of:
At learning phase, to the typical fault and the critical fault of subway train, produce dissimilar faults through simulation, the training sample vector that obtains is as the input of neural network;
Hidden layer neuron is learnt the input sample of neural network through the clustering algorithm process of unsupervised learning;
The output layer neuron is set up radial basis function neural network through the least square method learning algorithm process of supervision is arranged according to the optimum weight coefficient that calculates;
In the application stage; When the subway train breaks down maybe will break down the time; As the input of radial basis function neural network, the stable state output of radial basis function neural network is the failure mode that the failure mode that taken place maybe will take place after the real time data classification processing of gathering.
8. the method for subway train fault diagnosis according to claim 7 is characterized in that, the clustering algorithm process of said unsupervised learning may further comprise the steps:
Each cluster centre of initialization η
j, j=1,2 ..., L, L are the cluster number, x
N, iBe training sample vector, picked at random initialization cluster centre from the training sample vector of input;
The training sample vector of input is if satisfy condition
Then the training sample vector with input belongs to the q class;
After classification finishes, obtain the new cluster centre and the width of each classification, that is:
Wherein, m
jFor belonging to cluster centre η
jThe training sample number.
9. the method for subway train fault diagnosis according to claim 8 is characterized in that, said have the least square method learning algorithm process of supervision may further comprise the steps:
The mark output vector is y
N, k, then k output node had:
In the formula, ω
IkExpression hidden layer i unit is to the weights of output layer k unit, and x representes input vector, η
iThe cluster centre of doing a certain type of input is got in the position of representing certain RP, and R () is the non-linear radial symmetry basis function of choosing, and the non-linear radial symmetry basis function of employing is the Gaussian function, || || expression is apart from certain some η
iDistance measure, L is the hidden layer unit number;
R in the formula
jBe hidden layer j neuronic output, A is a scale factor, y
jBe hidden layer j neuronic Gaussian function center, σ is the width of Gaussian function;
If sorter is output as:
The error function of define grid output is:
In the formula, d
kBe output layer k neuronic desired output, y
N, kBe actual output value, formula (4) substitution formula (7) can be got:
Find out and make error function E one group of weight coefficient ω hour
Ik, order:
Draw system of equations:
Find the solution optimum weight coefficient ω
Ikopt, optimum weight coefficient ω
IkoptAfter confirming, radial basis function neural network is set up and is finished.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201010563179A CN102063109B (en) | 2010-11-29 | 2010-11-29 | Neural network-based subway train fault diagnosis device and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201010563179A CN102063109B (en) | 2010-11-29 | 2010-11-29 | Neural network-based subway train fault diagnosis device and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102063109A CN102063109A (en) | 2011-05-18 |
CN102063109B true CN102063109B (en) | 2012-09-05 |
Family
ID=43998418
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201010563179A Active CN102063109B (en) | 2010-11-29 | 2010-11-29 | Neural network-based subway train fault diagnosis device and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102063109B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104751229A (en) * | 2015-04-13 | 2015-07-01 | 辽宁大学 | Bearing fault diagnosis method capable of recovering missing data of back propagation neural network estimation values |
Families Citing this family (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102707713B (en) * | 2012-06-15 | 2014-04-16 | 长春工业大学 | Fault diagnosis system and method for automobile safety air bag assembly working procedure |
CN103207950B (en) * | 2013-04-16 | 2016-08-10 | 郑州航空工业管理学院 | A kind of intelligent method of transformer fault diagnosis based on RBF neural |
CN103777601A (en) * | 2014-01-17 | 2014-05-07 | 江苏新绿能科技有限公司 | Shield door door-machine controller neural network PID method and controller controlled through the same |
CN103914044B (en) * | 2014-03-26 | 2016-07-06 | 北方信息控制集团有限公司 | A kind of intelligent on-line monitoring protection device |
CN103914735B (en) * | 2014-04-17 | 2017-03-29 | 北京泰乐德信息技术有限公司 | A kind of fault recognition method and system based on Neural Network Self-learning |
CN104615122B (en) * | 2014-12-11 | 2018-04-06 | 深圳市永达电子信息股份有限公司 | A kind of industry control signal detection system and detection method |
DE102016008987B4 (en) | 2015-07-31 | 2021-09-16 | Fanuc Corporation | Machine learning method and machine learning apparatus for learning failure conditions, and failure prediction apparatus and failure prediction system including the machine learning apparatus |
JP6148316B2 (en) * | 2015-07-31 | 2017-06-14 | ファナック株式会社 | Machine learning method and machine learning device for learning failure conditions, and failure prediction device and failure prediction system provided with the machine learning device |
CN106959681B (en) * | 2016-01-12 | 2023-06-06 | 武汉科技大学 | FHN-based urban rail train door fault diagnosis method |
CN106628097B (en) * | 2016-11-22 | 2018-08-28 | 南京航空航天大学 | A kind of ship equipment method for diagnosing faults based on improvement radial base neural net |
CN108009574B (en) * | 2017-11-27 | 2022-04-29 | 成都明崛科技有限公司 | Track fastener detection method |
CN108398934B (en) * | 2018-02-05 | 2019-12-13 | 常州高清信息技术有限公司 | equipment fault monitoring system for rail transit |
CN110389570A (en) * | 2018-04-19 | 2019-10-29 | 株洲中车时代电气股份有限公司 | A kind of locomotive traction system trouble-shooter and method |
CN108647832B (en) * | 2018-05-18 | 2020-08-04 | 辽宁工业大学 | Subway operation interval time control algorithm based on neural network |
CN108985279B (en) * | 2018-08-28 | 2020-11-03 | 上海仁童电子科技有限公司 | Fault diagnosis method and device for MVB waveform of multifunctional vehicle bus |
CN109406117A (en) * | 2018-09-28 | 2019-03-01 | 东南大学 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
CN110135633A (en) * | 2019-04-29 | 2019-08-16 | 北京六捷科技有限公司 | A kind of railway service Call failure prediction technique and device |
CN111988192A (en) * | 2019-05-23 | 2020-11-24 | 北京交通大学 | Train communication network fault diagnosis method based on machine learning |
CN112109767A (en) * | 2020-09-21 | 2020-12-22 | 西南交通大学 | System and method for detecting pressure signal of bolt of urban subway rail |
CN112763091B (en) * | 2020-12-30 | 2022-04-15 | 西南交通大学 | Intelligent detection device and test method for temperature signals of subway bolts |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006000427A1 (en) * | 2004-06-25 | 2006-01-05 | Fev Motorentechnik Gmbh | Motor vehicle control device provided with a neuronal network |
CN1910434A (en) * | 2004-01-14 | 2007-02-07 | Abb公司 | Method and apparatus to diagnose mechanical problems in machinery |
CN101110155A (en) * | 2007-08-27 | 2008-01-23 | 北京交通大学 | Built-in intelligent fault diagnosing device based on data inosculating pattern recognition and method thereof |
CN101231673A (en) * | 2008-02-02 | 2008-07-30 | 湖南大学 | Analog circuit failure diagnosis method optimized using immune ant algorithm |
EP2053475A1 (en) * | 2007-10-26 | 2009-04-29 | Siemens Aktiengesellschaft | Method for analysing the operation of a gas turbine |
-
2010
- 2010-11-29 CN CN201010563179A patent/CN102063109B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1910434A (en) * | 2004-01-14 | 2007-02-07 | Abb公司 | Method and apparatus to diagnose mechanical problems in machinery |
WO2006000427A1 (en) * | 2004-06-25 | 2006-01-05 | Fev Motorentechnik Gmbh | Motor vehicle control device provided with a neuronal network |
CN101110155A (en) * | 2007-08-27 | 2008-01-23 | 北京交通大学 | Built-in intelligent fault diagnosing device based on data inosculating pattern recognition and method thereof |
EP2053475A1 (en) * | 2007-10-26 | 2009-04-29 | Siemens Aktiengesellschaft | Method for analysing the operation of a gas turbine |
CN101231673A (en) * | 2008-02-02 | 2008-07-30 | 湖南大学 | Analog circuit failure diagnosis method optimized using immune ant algorithm |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104751229A (en) * | 2015-04-13 | 2015-07-01 | 辽宁大学 | Bearing fault diagnosis method capable of recovering missing data of back propagation neural network estimation values |
CN104751229B (en) * | 2015-04-13 | 2017-05-24 | 辽宁大学 | Bearing fault diagnosis method capable of recovering missing data of back propagation neural network estimation values |
Also Published As
Publication number | Publication date |
---|---|
CN102063109A (en) | 2011-05-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102063109B (en) | Neural network-based subway train fault diagnosis device and method | |
Kou et al. | Fault diagnosis for open‐circuit faults in NPC inverter based on knowledge‐driven and data‐driven approaches | |
Wu et al. | Differential variable speed limits control for freeway recurrent bottlenecks via deep actor-critic algorithm | |
CN202063165U (en) | Subway train fault diagnostic equipment based on neural network | |
CN106443297A (en) | Decision tree SVM fault diagnosis method for three-level inverter of photovoltaic diode clamp type | |
CN102944418B (en) | Wind turbine generator group blade fault diagnosis method | |
CN101464964B (en) | Pattern recognition method capable of holding vectorial machine for equipment fault diagnosis | |
CN104392071B (en) | A kind of bullet train system security assessment method based on complex network | |
CN102749573B (en) | Based on the analog-circuit fault diagnosis method of wavelet packet analysis and Hopfield network | |
CN104934968A (en) | Multi-agent based distribution network disaster responding recovery coordinate control method and multi-agent based distribution network disaster responding recovery coordinate control device | |
CN103914735A (en) | Failure recognition method and system based on neural network self-learning | |
CN107817404A (en) | A kind of Portable metering automatization terminal trouble-shooter and its diagnostic method | |
CN106004518B (en) | A kind of electric automobile energy management prognostic control method based on car networking | |
CN102938562B (en) | Prediction method of total wind electricity power in area | |
Ding et al. | Application of multi-SVM classifier and hybrid GSAPSO algorithm for fault diagnosis of electrical machine drive system | |
CN107153845A (en) | A kind of isolated island detection method of the adaptive grid-connected photovoltaic system based on machine learning | |
CN102288398A (en) | Momentum wheel fault detection device and method based on support vector machine | |
CN110389570A (en) | A kind of locomotive traction system trouble-shooter and method | |
CN105574604A (en) | Power network operation event-oriented monitoring, pre-judging and analyzing system | |
CN109860736A (en) | The big data system and method utilized for battery echelon | |
CN104677629A (en) | Fault detection method for vehicle transmission | |
CN116484299A (en) | Charging pile fault diagnosis method based on integration of gradient lifting tree and multi-layer perceptron | |
Li et al. | Early gear pitting fault diagnosis based on bi-directional LSTM | |
CN110635474A (en) | Power grid dynamic trajectory trend prediction method based on long-term and short-term memory network | |
Wang et al. | Fault diagnosis of electrical automatic control system of hydraulic support based on particle swarm optimization algorithm |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant |