CN104615121A - Method and system for train fault diagnosis - Google Patents

Abstract

The invention discloses a method and a system for train fault diagnosis. A data acquisition module is used for obtaining real-time status of the electrical characteristics of the system hardware, memory module must be collected to process data stored in the database, wait for processing and analysis, failure analysis module for data were analyzed using neural algorithm for establishing operational models to identify the various factors associated with the failure; the display module through the screen to show the system structure, intuitive operation expression system as a whole, do basic alarm and remote terminal provided by to customers for remote viewing, the display module in the data collected after the system threshold data can be divided, depending on the threshold level defined alarm system by voice, color multimedia approach to the police, through the terminal alarm device.

Description

A kind of train fault diagnostic method and system

Technical field

The present invention relates to a kind of method for diagnosing faults and system, more precisely, relate to the method for diagnosing faults of the trains such as a kind of high ferro, subway, be a kind of based on industry control image data system, data stored, analyzes, reality and mutual train fault diagnostic system.

Background technology

According to the structure of train control system system, have similar industry control acquisition system at uphole equipment layer and mobile unit layer, these systems are that thermoacoustic prime engine and safety provide data, and data are preserved history of forming data.At uphole equipment layer, signal microcomputer monitoring system act as the role of collection and warning and fault analysis.

Signal microcomputer monitoring system is railway special signal microcomputer monitoring equipment, can be used as the aid of electricity business maintenance management.Signal microcomputer monitoring system is the important travelling facility ensureing traffic safety, strengthen the management of signalling arrangement joint portion, monitor railway signals equipment utilization quality.Signal microcomputer monitoring system is the important component part of railway equipment modernization.It is latest modern technological technology, as technology such as sensor, fieldbus, computer network communication, database and soft projects, combine together, monitoring the main running status of tracer signal equipment, grasp the utilization quality of equipment for telecommunication and signaling branch and fault analysis provides scientific basis.Meanwhile, system also has mathematical logic arbitration functions, when the working condition of signalling arrangement departs from predetermined threshold or occurs abnormal and alarm, avoids because of equipment failure or operates the safety, the running on time that affect train against regulations.

Signal microcomputer monitoring system can carry out on-line monitoring to signalling arrangement in real time, dynamically, accurately, quantitatively, the application quality of reflected signal equipment, joint portion equipment state, and status information is stored, reset, inquire about, report to the police, operate against regulations for preventing, fault analysis and judgement, especially Hidden fault, transient fault and intermittent defect being found to analysis, providing important means and foundation, to guaranteeing that transportation safety plays an important role.Signal microcomputer monitoring system is railway special signal microcomputer monitoring equipment, can be used as the aid of electricity business maintenance management.

Along with improving constantly of high ferro all technical and automaticity thereof, increasing VLSI (very large scale integrated circuit) is applied in modern high ferro technology, the automatic Detection and diagnosis of high ferro fault is had higher requirement simultaneously, because in so complicated system, a trickle independent failure is just enough to whole system was lost efficacy.High reliability and maintainable closely-related core content---the research of method for diagnosing faults with system, also just become the focus that people pay close attention to.

Although microcomputer detecting system is the function that train control system provides partial fault analysis and diagnosis, to analyze further many signal faults or there is certain defect, the fault diagnosis of current microcomputer control system is mainly for some specific module, mainly qualitative analysis is carried out to image data, so just can manage some specific electric characteristic, in data dependence aspect, profound excavation is not carried out to the relation between data, but, some fault Producing reason is not necessarily gone wrong by specific mechanism and causes, the reason of fault may have a lot of aspect, between subsystems different so just there is correlativity in data, therefore, if in the face of complicated Analysis on Fault Diagnosis, just cannot meet the requirement of fault analysis.

Summary of the invention

The present invention, in order to solve problems of the prior art, provides a kind of train fault diagnostic method and system.

In order to realize above-mentioned object, technical scheme of the present invention is: a kind of train fault diagnostic method, comprises the following steps:

Step S1: the real-time status data gathering train apparatus electrical specification; And respectively stored in process database and historical data base;

Step S2: with the data in historical data base for foundation, sets up the fault model based on neural network structure, and this neural network structure is made up of input layer, hidden layer and output layer, is respectively:

1) input layer i, its export equal input xi (i=1,2 ..., n), control variable value is transferred to hidden layer;

2) hidden node j, it is input as hj, exports as Oj, is respectively:

$h_i=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i-{\mathrm{\θ}}_j=\underset{i=1}{\overset{n+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i$

$O_j=f\left(h_j\right)=\frac{1}{1+e^{-h_j}}$

3) output layer node k, its input hk, output yk are respectively:

$h_k=\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j-{\mathrm{\θ}}_j=\underset{j=1}{\overset{m+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j$

$y_k=f\left(h_k\right)=\frac{1}{1+e^{-h_k}}$

In under-stream period, the connection weights of each node immobilize, and the calculating of network is from input layer, and successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete;

In the learning period, the output of each node remains unchanged, and from output layer, oppositely successively node ground calculates the index word of each connection weights one by one;

If the network of output layer exports and differs larger with desired output, then start backpropagation, export the signal errors with desired output according to network, each connection weights between network node are modified, reduces the error of network output signal and desired output;

Step S3: by fault model to the real-time status data analysis in process database, failure judgement occurs and carries out fault analysis.

Preferably, described real-time status data comprise the data message on the digital information of train operation state, analog quantity information and train bus-line.

Preferably, also comprise the step of man-machine interaction, user defines the rank of warning according to the threshold value of data.

A kind of train fault diagnostic system provided by the invention, comprising:

Data acquisition module, gathers the real-time status data of train apparatus electrical specification;

And the process database, the historical data base that are used for the real-time status data that storage data acquisition module collects is connected respectively with data acquisition module;

And based on the failure analysis module of the neural network structure be made up of input layer, hidden layer, output layer, in under-stream period, the connection weights of each node immobilize, the calculating of network is from input layer, successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete; In the learning period, the output of each node remains unchanged, and from output layer, oppositely successively node ground calculates the index word of each connection weights one by one; If the network of output layer exports and differs larger with desired output, then start backpropagation, export the signal errors with desired output according to network, each connection weights between network node are modified, reduces the error of network output signal and desired output;

Failure analysis module is according to the real-time status data in process database, and failure judgement occurs and reason.

Preferably, described data acquisition module comprises:

Digital data acquisition submodule, gathers the digital information characterizing high ferro train operation state;

Analog acquisition submodule, gathers the analog quantity information characterizing high ferro train operation state;

Logical signal gathers submodule, gathers the data message on high ferro train bus-line.

Preferably, also display module is comprised.

Fault diagnosis system of the present invention, real time data is saved in database according to time width definition, thus history of forming data, by the function introducing fault diagnosis, historical data is analyzed, can find relevant to trouble spot has abnormal electrical specification, like this by such correlativity, system can send early warning especially extremely for some.By failure analysis module of the present invention, the interact relation between parameters can be drawn; Predict the value of inconvenient measurement parameter; Carry out off-line or online optimal control etc.

Accompanying drawing explanation

Fig. 1 shows the structure principle chart of fault fault diagnosis system of the present invention.

Fig. 2 shows the structural representation of neural network in Fig. 1.

Fig. 3 shows the schematic flow sheet of the study of neural network in Fig. 2.

Embodiment

The technical matters solved to make the present invention, the technical scheme of employing, the technique effect easy to understand obtained, below in conjunction with concrete accompanying drawing, be described further the specific embodiment of the present invention.

With reference to figure 1, a kind of train fault diagnostic system provided by the invention, comprising: data acquisition module 1, is used for gathering the real-time status data etc. of train apparatus electrical specification, as the foundation of fault diagnosis system of the present invention.

The real-time status of the electrical specification of hardware device in acquisition system, mainly comprise the voltage of equipment, electric current, frequency etc., these values can be gathered by PLC or DCS system, only need train control system to be supplied to our acquisition interface just passable, these data are all used for as warning, picture and Analysis on Fault Diagnosis service.Described data acquisition module 1 comprises digital data acquisition submodule 10, analog acquisition submodule 10 and logical signal and gathers submodule 12.Wherein, digital data acquisition submodule 10 collection characterizes the digital information of high ferro train operation state, as switch closed condition etc.; Analog acquisition submodule 11 collection characterizes the analog quantity information of high ferro train operation state, as voltage, electric current etc.; Logical signal collection submodule 12 gathers the data message etc. on high ferro train bus-line.

The present invention also comprises the process database 3, the historical data base 2 that are connected the real-time status data collected for storage data acquisition module 1 respectively with data acquisition module 1; That is real-time status data store two processes, one is that the data of collection are stored into process database 2, wait pending and analyze, second is be stored in historical data base 2, for fault diagnosis system provides the foundation of analysis.

Also comprise failure analysis module 4, after obtaining a large amount of real time datas, these data are saved in historical data base 2 by system, and by failure analysis module 4 pairs of data analysis, this failure analysis module 4 utilizes neuron algorithm to set up operational model, finds out the various factors relevant to fault.That is this failure analysis module 4 builds based on the neural network structure be made up of input layer, hidden layer, output layer, to complete the function of fault analysis and diagnosis, it is analyzed and knowledge excavation commercial production and device data, monitor technology and equipment performance, identify the reason of producing change, reduce fluctuation, optimize and produce.

By different types of historical data source, help client's rapid build model, analyze continuous, discrete, batch production process.Data encasement, visual check data can be carried out, can also rule model be set up.Draw the interact relation between parameters, predict the value of inconvenient measurement parameter, carry out off-line or online optimal control, the knowledge excavated from these models, help to process improve income assess, whole process also very simple and convenient, workload is little.Use this model system, the reason of production run fluctuation can be found out, and adjust, reach stabilized product quality, improve the target of output.

With reference to figure 2, being generally expressed as of fault model: this neural network structure is made up of input layer, hidden layer and output layer, is respectively:

1) input layer i, it exports and equals to input x _i(i=1,2 ..., n), control variable value is transferred to hidden layer;

2) hidden node j, it inputs h _j(formula 1), output O _j(formula 2) is respectively

$h_i=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i-{\mathrm{\θ}}_j=\underset{i=1}{\overset{n+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i$

$O_j=f\left(h_j\right)=\frac{1}{1+e^{-h_j}}$

3) output layer node k, it inputs h _k(formula 3), exports y _k(formula 4) is respectively:

$h_k=\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j-{\mathrm{\θ}}_j=\underset{j=1}{\overset{m+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j$

$y_k=f\left(h_k\right)=\frac{1}{1+e^{-h_k}}$

Wherein, Wij represents the weights between hidden layer i-th node to an input layer jth node;

θ J represents the threshold value of a hidden layer jth node;

φ K represents the excitation function of hidden layer;

This neural network structure has two stages, be respectively under-stream period and learning period: in under-stream period, the connection weights of each node of this phase Network immobilize, the calculating of network is from input layer, successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete.In the learning period, the output of each node of this one-phase remains unchanged, and e-learning is then from output layer, and oppositely successively node ground calculates the index word of each connection weights one by one, to revise the weights of each connection, until input layer.These two stages are also called forward-propagating and back-propagation process, as shown in Figure 3, in forward-propagating, if the network of output layer exports and differs larger with desired output, then start back-propagation process, the signal errors with desired output is exported according to network, each connection weights between network node are modified, the error of network output signal and desired output is reduced with this, neural network passes through the computation process of ongoing like this forward-propagating and backpropagation just, finally makes the output valve of network output layer and expectation value reach unanimity.

The node of input layer just carries out some process of simply standardizing to the sample data of input usually, real network calculations process terminates to output layer from the second layer, for the ease of calculating, usually the threshold value in each processing node is also connected weights as one.In order to accomplish this point, if by each processing node with one export perseverance be 1 dummy node be connected.The description of the computation process of forward-propagating and the processing procedure of network node is basically identical; The computation process of backpropagation i.e. the learning process of network, its main thought utilizes the gap between the actual output of each node in the output layer of network and corresponding desired output, from output layer, suitable adjustment is carried out to internodal weights be connected to each other each in network, gap between the actual output of network and desired output can progressively be reduced, the final output obtaining optimum.

The neural network representation of knowledge is a kind of implicit representation of knowledge, Knowledge representation is topology of networks and be connected weights, adopt the expert system of nerual network technique, because neural network is the network system that a kind of information Storage and Processing is unified, therefore in the expert system adopting nerual network technique, reasoning process in the storage of knowledge and problem solving process is all carried out in the neural network of system, is the unification of inference machine and knowledge base.Neural network adopts the Fault Diagnosis Strategy of data-driven forward reasoning, and namely from original state, till forward reasoning arrives dbjective state, its troubleshooting step is:

1) fault sample is inputed to each node of input layer, it is also this layer of neuronic output simultaneously.

2) obtained the output of hidden neuron by formula (2), and it can be used as the input of output layer.

3) the neuronic output of output layer is tried to achieve from formula (4).

4) the neuronic final Output rusults of output layer is judged by threshold function table.

We represent fault type with Fk, then fault type threshold determination function is

Failure analysis module 4 according to the real-time status data in process database 3, the generation of failure judgement and reason.That is failure analysis module 4 finds out contacting of correlative factor and fault by the input of current real-time data, and the relation finally derived between the reason that may cause different faults, and utilize such derivation result, greatly can improve the success ratio of early warning, thus can improve and optimizate system.

According to existing Computerized monitor system principle, this system has mainly been accomplished image data, warning and has been monitored specific electrical specification, the fault that this mode can cause in the face of simple factor processes, but in complicated failure factor, just not necessarily can find failure cause, because the fault that complication system produces often is caused by different factor interactions.

System of the present invention can also comprise display module 5, represents system architecture by picture, intuitively the ruuning situation of expression system entirety, accomplishes basic alarm, can also be supplied to client carry out remote browse by remote terminal.Report to the police, after collecting data, system can carry out threshold value division to data, and according to different threshold values, define the rank of warning, system passes through sound.

System of the present invention can also comprise interactive module, converge all real time datas, built by historical data base and converge real time data, geography information navigation can be realized: receiving end can be positioned by Geographic Information System, no matter the data be sent on Web or mobile terminal are all encrypted, and the transmission of safety that can be complete, by asset structure database, record and distinguish different relationship of assets, and provide structure to navigate, relationship of assets is different, and the data presented also are had any different.

Accordingly, present invention also offers a kind of train fault diagnostic method, comprise the following steps:

Step S1: the real-time status data gathering train apparatus electrical specification; And respectively stored in process database and historical data base;

Step S2: with the data in historical data base for foundation, sets up the fault model based on neural network structure, and this neural network structure is made up of input layer, hidden layer and output layer, is respectively:

1) input layer i, its export equal input xi (i=1,2 ..., n), control variable value is transferred to hidden layer;

2) hidden node j, it is input as hj, exports as Oj, is respectively:

$h_i=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i-{\mathrm{\θ}}_j=\underset{i=1}{\overset{n+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i$

$O_j=f\left(h_j\right)=\frac{1}{1+e^{-h_j}}$

3) output layer node k, its input hk, output yk are respectively:

$h_k=\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j-{\mathrm{\θ}}_j=\underset{j=1}{\overset{m+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{jk}}{o}_j$

$y_k=f\left(h_k\right)=\frac{1}{1+e^{-h_k}}$

In under-stream period, the connection weights of each node immobilize, and the calculating of network is from input layer, and successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete;

In the learning period, the output of each node remains unchanged, and from output layer, oppositely successively node ground calculates the index word of each connection weights one by one;

If the network of output layer exports and differs larger with desired output, then start backpropagation, export the signal errors with desired output according to network, each connection weights between network node are modified, reduces the error of network output signal and desired output;

Step S3: by fault model to the real-time status data analysis in process database, failure judgement occurs and carries out fault analysis.Fault can be reported to the police after occurring, and user defines the rank of warning according to the threshold value of data.Alert levels is divided into critical, senior, intermediate, rudimentary, normal condition and unknown accident.

Trouble analysis system of the present invention, the real-time status of the electrical specification of hardware device in data acquisition module acquisition system, the data of collection are stored into process database by memory module, etc. pending and analysis, failure analysis module is to data analysis, utilize neuron algorithm to set up operational model, find out the various factors relevant to fault; Display module 5 represents system architecture by picture, the ruuning situation of expression system entirety intuitively, accomplish basic alarm, and be supplied to client by remote terminal and carry out remote browse, display module is after collecting data, and system can carry out threshold value division to data, according to different threshold values, define the rank of warning, system is reported to the police by sound, the multimedia mode of color, is reported to the police by terminal device.

Native system is gathered by electrical specifications such as the digital quantity to major equipment in environment, analog quantitys, the real-time status that acquisition equipment runs, the threshold value good according to predefined to the data collected by the rule defined compares judgement, then reported to the police in real time and early warning by the mode such as local computer, network, shown the running status of each system electrical by the mode of icon and picture.

Simultaneously, real time data is saved in database according to time width definition, thus history of forming data, by the function introducing fault diagnosis, historical data is analyzed, can find relevant to trouble spot has abnormal electrical specification, like this by such correlativity, system can send early warning especially extremely for some.By failure analysis module of the present invention, the interact relation between parameters can be drawn; Predict the value of inconvenient measurement parameter; Carry out off-line or online optimal control etc.

Whole system can adopt distributed frame, each subsystem can be cooperated mutually by network mode, use graphical tool intuitively, provided more powerful configuration function and expanded function, quickly and easily for their production run creates high performance processing window.No matter be simple unit man-machine interface (HMI), or complicated multinode, many on-the-spot data acquisitions and SCADA, the needs of various application type and application scale can be met easily, substantially the needs that industry gathers can be met, can based on the collection system of similar microcomputer control system.

The present invention is by preferred embodiment having carried out detailed explanation.But, by studying carefully above, concerning the change of each embodiment with to increase be apparent for one of ordinary skill in the art.Being intended that these changes all and increasing of applicant has all dropped in scope that the claims in the present invention protect.

Claims (6)

1. a train fault diagnostic method, is characterized in that, comprises the following steps:

Step S1: the real-time status data gathering train apparatus electrical specification; And respectively stored in process database and historical data base;

Step S2: with the data in historical data base for foundation, sets up the fault model based on neural network structure, and this neural network structure is made up of input layer, hidden layer and output layer, is respectively:

1) input layer i, its export equal input xi (i=1,2 ..., n), control variable value is transferred to hidden layer;

2) hidden node j, it is input as hj, exports as Oj, is respectively:

$h_i=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_t-\mathrm{\θ}=\underset{i=1}{\overset{n+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_j{x}_i$

$O_j=f\left(h_j\right)=\frac{1}{1+e^{-h_j}}$

3) output layer node k, its input hk, output yk are respectively:

$h_k=\underset{j=1}{\overset{m}{\mathrm{\Σ}}}{\mathrm{\ω}}_k{O}_j-\mathrm{\θ}=\underset{j=1}{\overset{m+1}{\mathrm{\Σ}}}{\mathrm{\ω}}_k{O}_j$

$y_k=f\left(h_k\right)=\frac{1}{1+e^{{-h}_k}}$

In under-stream period, the connection weights of each node immobilize, and the calculating of network is from input layer, and successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete;

In the learning period, the output of each node remains unchanged, and from output layer, oppositely successively node ground calculates the index word of each connection weights one by one;

If the network of output layer exports and differs larger with desired output, then start backpropagation, export the signal errors with desired output according to network, each connection weights between network node are modified, reduces the error of network output signal and desired output;

Step S3: by fault model to the real-time status data analysis in process database, failure judgement occurs and carries out fault analysis.

2. diagnostic method according to claim 1, is characterized in that: described real-time status data comprise the data message on the digital information of train operation state, analog quantity information and train bus-line.

3. diagnostic method according to claim 1, is characterized in that: the step also comprising man-machine interaction, and user defines the rank of warning according to the threshold value of data.

4. a train fault diagnostic system, is characterized in that, comprising:

Data acquisition module, gathers the real-time status data of train apparatus electrical specification;

And the process database, the historical data base that are used for the real-time status data that storage data acquisition module collects is connected respectively with data acquisition module;

And based on the failure analysis module of the neural network structure be made up of input layer, hidden layer, output layer, in under-stream period, the connection weights of each node immobilize, the calculating of network is from input layer, successively node ground calculates the output of each node one by one, until each node calculate in output layer is complete; In the learning period, the output of each node remains unchanged, and from output layer, oppositely successively node ground calculates the index word of each connection weights one by one; If the network of output layer exports and differs larger with desired output, then start backpropagation, export the signal errors with desired output according to network, each connection weights between network node are modified, reduces the error of network output signal and desired output;

Failure analysis module is according to the real-time status data in process database, and failure judgement occurs and reason.

5. train fault diagnostic system according to claim 4, is characterized in that: described data acquisition module comprises:

Digital data acquisition submodule, gathers the digital information characterizing high ferro train operation state;

Analog acquisition submodule, gathers the analog quantity information characterizing high ferro train operation state;

Logical signal gathers submodule, gathers the data message on high ferro train bus-line.

6. train fault diagnostic system according to claim 4, is characterized in that: also comprise display module.