CN113641486B - Intelligent turnout fault diagnosis method based on edge computing network architecture - Google Patents
Intelligent turnout fault diagnosis method based on edge computing network architecture Download PDFInfo
- Publication number
- CN113641486B CN113641486B CN202110757298.0A CN202110757298A CN113641486B CN 113641486 B CN113641486 B CN 113641486B CN 202110757298 A CN202110757298 A CN 202110757298A CN 113641486 B CN113641486 B CN 113641486B
- Authority
- CN
- China
- Prior art keywords
- data
- cloud layer
- fault
- edge
- fault diagnosis
- 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
- 238000003745 diagnosis Methods 0.000 title claims abstract description 74
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000007781 pre-processing Methods 0.000 claims abstract description 22
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 13
- 238000003062 neural network model Methods 0.000 claims description 31
- 238000013500 data storage Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 11
- 238000012544 monitoring process Methods 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 238000011282 treatment Methods 0.000 claims description 9
- 230000001965 increasing effect Effects 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 7
- 230000005540 biological transmission Effects 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 5
- 230000009471 action Effects 0.000 claims description 4
- 230000002035 prolonged effect Effects 0.000 claims description 3
- 230000007704 transition Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 230000008676 import Effects 0.000 claims description 2
- 230000001502 supplementing effect Effects 0.000 claims 1
- 238000012423 maintenance Methods 0.000 abstract description 11
- 238000004422 calculation algorithm Methods 0.000 abstract description 7
- 238000010586 diagram Methods 0.000 description 4
- 238000012821 model calculation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000002194 synthesizing effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5061—Partitioning or combining of resources
- G06F9/5072—Grid computing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
-
- 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
-
- 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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/50—Indexing scheme relating to G06F9/50
- G06F2209/502—Proximity
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Test And Diagnosis Of Digital Computers (AREA)
Abstract
The invention discloses a turnout intelligent fault diagnosis method based on an edge computing network architecture, which comprises an access layer, an edge cloud layer and a center cloud layer. The access layer sensor receives turnout current curve data and sends the turnout current curve data to the edge cloud layer for fault diagnosis; the edge cloud layer receives the current curve data, performs data preprocessing firstly, and performs fault diagnosis by using a CNN fault diagnosis model transmitted by the center cloud layer; the central cloud layer generates a large number of data samples by means of SMOTE and GAN methods by means of a small amount of historical data, trains a CNN model, and transmits the CNN model to the edge cloud layer. The central cloud layer receives the preprocessing data of the edge cloud layer to replace a generated sample for training the CNN model; the method solves the problems of unbalanced data samples, complex calculation of an artificial intelligent neural network fault diagnosis algorithm model, occupied cloud computing bandwidth, insufficient computing capacity of edge nodes and the like in the prior art, and improves the stability and the high efficiency of the existing intelligent operation and maintenance system.
Description
Technical Field
The invention belongs to the technical field of rail transit informatization intelligence, and particularly relates to a turnout intelligent fault diagnosis method based on an edge computing network architecture.
Background
With the advent of the 5G age, high-density, high-strength and high-speed railway operation is brought, and how to ensure the safety and efficiency of train operation has become a research hotspot gradually. The turnout is an important component of a railway signal system, and the main function of the turnout is to change the running direction of a train. Because the turnout is frequently used and the working environment is bad, the turnout is vulnerable to the consumption equipment, and once the turnout fails, the normal operation of the railway is seriously affected. At present, the stability and the safety of the turnout system in China mainly depend on a simple alarm threshold device for real-time monitoring, namely, parameters such as working current, power and the like at a certain moment exceed a threshold value, so that the turnout is indicated to be abnormal. However, the method has no timeliness, accuracy and intelligence, can not discover faults timely and accurately, can intelligently obtain fault types and solutions, brings great inconvenience to maintenance personnel, and brings great hidden danger to railway safety operation.
With the continuous progress and development of modern signal processing technology, some switch fault feature extraction and diagnosis methods are proposed aiming at the historical data and working state of the switch machine. The method is characterized in that the method processes qualitative and quantitative aspects, and researches on turnout fault diagnosis by utilizing methods of feature extraction, data quantization, machine learning, neural network and the like which are popular at present. However, two problems in the existing method cannot be solved well, firstly, the sample data are unevenly distributed, and the normal curve data volume acquired through a microcomputer monitoring system is far greater than the fault curve data volume, so that serious unbalance exists in the data. And secondly, performing feature extraction and neural network training based on the existing current curve data aiming at turnout fault diagnosis of a neural network algorithm, wherein the problems of complex model calculation and the like caused by overlong training time under the condition of insufficient accuracy of a neural network model under the condition of small data volume and large data volume exist.
The SMOTE is a technology for synthesizing minority class oversampling, and the algorithm mainly synthesizes new minority class samples by partial minority class samples, and controls the quantity and distribution of the new minority samples, thereby realizing the balance of the multi-class samples and the minority class samples. GAN is a model of the generation of a challenge network, through the mutual challenge and game between the two networks of generator G and arbiter D, to the extent that the samples generated by the generator have the ability to be spurious. The fault data set can be expanded by SMOTE and GAN methods to achieve the balance of the number of fault samples and normal samples.
Cloud computing is an effective way to efficiently process large amounts of data, and can be used as a neural network switch fault diagnosis network architecture to process large amounts of data. However, the cloud server is usually deployed at a fixed location, sometimes far from the terminal, and as the railway coverage further increases, the cloud computing network bandwidth load will have a serious influence on the stability of the whole operation and maintenance system, so that the requirement of the increasing railway operation and maintenance system is difficult to be met by a pure cloud computing network architecture.
Edge computing is to process the service of edge terminals by deploying servers with computing and storage capabilities at the network edge. The most important characteristics are that the edge nodes are physically close to the edge terminals, and compared with the traditional cloud computing mode, the characteristics can effectively reduce delay, save energy efficiency, be beneficial to privacy protection, reduce bandwidth occupation, improve system instantaneity and the like. For the railway operation and maintenance system, a large number of business processing processes can be completed at a local edge layer by edge calculation without the need of cloud processing, so that the business processing efficiency of the whole operation and maintenance system is greatly improved, and the cloud load is reduced. And as the mobile terminal is closer to the train and the like, the quick response can be provided, and the occurrence of faults is reduced so as to ensure the safe operation of the train.
Therefore, based on an edge computing network architecture, the data enhancement technologies such as SMOTE and GAN are combined, an artificial intelligent neural network is adopted, a large number of data processing models are trained and configured to a central node, tasks with small calculation amount such as data preprocessing are configured to edge nodes, the problems of wide turnout equipment distribution, unbalanced data, occupied cloud computing bandwidth, insufficient computing capacity of the edge nodes and the like can be effectively solved, and the stability and the high efficiency of the existing intelligent operation and maintenance system are improved.
Disclosure of Invention
The invention aims to provide a turnout intelligent fault diagnosis method based on an edge computing network architecture, which solves the problems of unbalanced data, complex model calculation aiming at an artificial intelligent neural network fault diagnosis algorithm, occupied cloud computing bandwidth, insufficient computing capacity of edge nodes and the like in the prior art.
The technical scheme adopted by the invention is as follows: a turnout intelligent fault diagnosis method based on an edge computing network architecture comprises the following specific operation steps:
step 1, the method generally divides a turnout fault diagnosis network architecture into three layers, namely an access layer, an edge cloud layer and a center cloud layer;
step 2, the central cloud layer uses the existing real data to carry out data enhancement through SMOTE and GAN methods, is used for training an initial neural network model, and transmits the trained model to the edge cloud layer;
step 3, the access layer collects the starting current curve data of the switch machine through a microcomputer monitoring system and sends the starting current curve data to the edge cloud layer for data storage, data preprocessing and fault diagnosis;
step 4, after preprocessing the data, the edge cloud layer performs fault diagnosis on one hand, and transmits the data to the central cloud layer for data storage and neural network model training on the other hand, and transmits the data which cannot be diagnosed to the central cloud layer for recording processing;
step 5, on the basis of the step 4, when the real data amount in the central cloud layer reaches more than 1 ten thousand, the real data from the edge cloud layer is not received any more, and only the fault data and unidentified data determined after the intelligent diagnosis of the edge cloud layer are received; the central cloud layer carries out neural network training according to a time period T, and the obtained neural network model is transmitted to the edge cloud layer.
The present invention is also characterized in that,
the access layer comprises a switch machine, a sensor, a microcomputer monitoring system and the like; the edge cloud layer comprises a server and a database of edge nodes; the central cloud layer comprises a large server and a database; the central cloud layer has extremely strong data computing power and data storage capacity compared with the servers of the edge cloud layer.
The edge cloud layer is used for data preprocessing and fault diagnosis based on the neural network model, and the calculation of the neural network model is performed on the center cloud layer.
The step 2 is specifically as follows:
the central cloud layer stores more than 400 pieces of historical real same type turnout action current curve data, and the data are classified according to normal data and 6 common fault types, wherein more than 50 pieces of 6 similar fault current curve data are respectively stored, and more than 100 pieces of normal data are stored; the central cloud layer uses a synthetic minority oversampling technology SMOTE and an anti-network GAN generation method to enhance data through the data; the 6 similar fault current curve data are respectively increased to more than 500 pieces, the normal class data are increased to more than 1000 pieces, and the total data is more than 4000 pieces;
wherein, the normal data type is marked as f0;
the type of the fault state of the foreign matter existing between the rails is marked as f1;
the type of the fluctuation fault state in the transition stage is marked as f2;
the type of the fluctuation fault state in the locking stage is marked as f3;
the type of the fluctuation fault state in the starting stage is marked as f4;
the difficult fault state type in the locking stage is marked as f5;
the type of the stator and rotor mixed line fault state is marked as f6.
When current curve data enhancement is carried out, a GAN method is used for four data types f0, f1, f4 and f6, and an SMOTE method is used for three data types f2, f3 and f5;
the data generated by the SMOTE and GAN methods are used for CNN neural network training, and an initial neural network model is generated for diagnosing turnout faults; the model can carry out turnout fault diagnosis through the data preprocessed by the given current curve, and the diagnosis result is divided into a fault-free state, 6 fault states and an unidentified state; the central cloud layer transmits the trained neural network model to each edge cloud layer node;
and if the normal data and various fault data in the real data reach more than 1 ten thousand, canceling sample data generated by the SMOTE and GAN methods, and training a CNN model by adopting the real data from the access layer.
Step 5, period T is defined according to the requirements of the user and defaults to 1 week.
The step 3 is specifically as follows:
the access layer transmits the current curve data received by the microcomputer monitoring system to the edge cloud layer node nearby, and performs data storage, data preprocessing and fault diagnosis;
the data preprocessing is to perform data length compensation, data segmentation and feature extraction on the received current curve data. Then, performing fault diagnosis on the preprocessed data by using a neural network diagnosis model of the transmission of the central cloud layer;
the data length compensation means that the data length of the current curve is zero-compensated to 576;
the data segmentation refers to dividing the compensated data into three sections according to a sliding window method, namely unlocking, converting and locking stages respectively;
the feature extraction means that 11 widely used feature extraction equations are used for describing geometric features of the curve, including import and export differences, average values, root mean square values, variances, checksums, kurtosis values, peak factors, waveform factors, pulse factors and margin factors.
The step 4 is specifically as follows:
step 4.1, each node of the edge cloud layer transmits the preprocessed data stored in the database to the center cloud layer in real time and stores the preprocessed data in the database; and setting a time period of T days, and combining the stored data by the central cloud layer to train the neural network model. After training the neural network model, transmitting the neural network model to each edge cloud layer node;
step 4.2, after the edge cloud layer performs data preprocessing on the current curve data from the access layer, fault diagnosis is performed on one hand, and the current curve data is transmitted to the center cloud layer on the other hand;
the fault diagnosis result is divided into a normal state, 6 common fault states and an unidentified state; if the state is normal, no treatment is carried out; if the fault is 6, giving an alarm to a switch fault maintainer to process the fault type; if the curve data is in an unidentified state, transmitting the curve data to a central cloud layer for recording;
the recording processing refers to storing the curve data of the diagnosed unidentified state as a new unidentified state type, and appointing an maintainer to overhaul a turnout section where the unidentified state occurs, so as to determine whether the turnout has faults, fault reasons and state types. Finally, storing the analysis result in a server;
when the unidentified current curve data of the same class reaches more than 50, the data set is expanded by SMOTE and GAN methods, and the method is used for generating a CNN model of a new fault class.
The step 5 is specifically as follows:
the normal current curve data quantity is far greater than the fault current curve data, when the normal current curve data quantity reaches more than 1 ten thousand, the normal current data is not received any more, and only the fault current curve data and unidentified current curve data are received, so that the accuracy of the neural network model is improved; in order to save resources and improve efficiency, the time period can be prolonged, and the transmission frequency can be reduced.
The beneficial effects of the invention are as follows: the invention provides a turnout fault diagnosis method based on an edge computing network architecture, which aims at solving the problem of uneven distribution of turnout curve data samples and provides a method for enhancing data of a few fault samples by using an SMOTE and GAN network method. Aiming at the problems of complex model calculation, occupied cloud calculation bandwidth, insufficient computing capacity of edge nodes and the like of a turnout fault diagnosis method using an artificial intelligent neural network algorithm, the patent adopts a method based on an edge calculation network architecture, and the fault diagnosis network architecture is layered into three layers, namely an access layer, an edge cloud layer and a central cloud layer. The preprocessing calculation and fault diagnosis of a large amount of data are placed on the edge cloud layer, and the CNN model training with complex calculation is only carried out on the central cloud layer, so that the network time delay and bandwidth of fault diagnosis are greatly reduced, the problems of wide distribution of turnout equipment, occupied cloud calculation bandwidth, insufficient calculation capacity of edge nodes and the like can be effectively solved, and the stability and the high efficiency of the existing intelligent operation and maintenance system are improved.
Drawings
FIG. 1 is a schematic diagram of a three-layer network architecture of a switch fault diagnosis method based on an edge computing network architecture of the present invention;
FIG. 2 is a schematic diagram of a three-layer network physical structure of a switch fault diagnosis method based on an edge computing network architecture;
FIG. 3 is a general flow diagram of a switch fault diagnosis method based on an edge computing network architecture of the present invention;
FIG. 4 is a flow chart of a primary switch fault diagnosis of a switch fault diagnosis method based on an edge computing network architecture of the present invention;
fig. 5 is a schematic diagram of SMOTE method of the switch fault diagnosis method based on the edge computing network architecture of the present invention.
Detailed Description
As shown in fig. 1, the switch fault diagnosis method based on the edge computing network architecture is implemented according to the following steps:
step 1, the turnout fault diagnosis method based on the edge computing network architecture generally divides a turnout fault diagnosis network into three layers of an access layer, an edge cloud layer and a center cloud layer.
And 2, carrying out data enhancement on the central cloud layer by using a small amount of existing real data through an SMOTE and GAN method, training an initial neural network model, and transmitting the trained model to the edge cloud layer.
And 3, the access layer collects current curve data of the switch machine through the sensor and sends the current curve data to the edge cloud layer for data storage, data preprocessing and fault diagnosis.
And 4, transmitting the data after the data preprocessing to the central cloud layer by the edge cloud layer for data storage and neural network training, and transmitting the data which cannot be diagnosed to the central cloud layer by the edge cloud layer for recording processing.
And 5, on the basis of the step 4, when the real data amount in the central cloud layer reaches a certain amount (more than 1 ten thousand are recommended), the real data from the edge cloud layer is not received any more, and only the fault data and unidentified data determined after the intelligent diagnosis of the edge cloud layer are received. The central cloud layer carries out neural network training according to a time period T, the obtained neural network model is transmitted to the edge cloud layer, and T can be defaulted to 1 week according to user definition.
As shown in fig. 2, step 1 specifically includes the following steps:
the method divides the switch fault diagnosis network architecture into three layers, namely an access layer, an edge cloud layer and a center cloud layer. The access layer comprises a switch machine, a sensor, a microcomputer monitoring system and the like; the edge cloud layer comprises servers, databases and the like of edge nodes; the central cloud layer comprises a large server and a database; the central cloud layer has extremely strong data computing power and data storage capacity compared with the servers of the edge cloud layer.
As shown in fig. 3, step 2 is specifically as follows:
the central cloud layer stores a small amount (more than 400 recommended) of historical real same-model turnout action current curve data, and the data can be classified according to normal data and 6 common fault types, wherein the 6 similar fault current curve data are respectively more than 50 and more than 100. The central cloud layer uses a synthetic minority class oversampling technique (SMOTE) and a Generation Antagonism Network (GAN) method to enhance data through these small amounts of data. The 6 kinds of similar fault current curve data are respectively increased to more than 500 pieces, and the normal class data are increased to more than 1000 pieces, and the total number of the data is more than 4000 pieces.
The SMOTE algorithm flow is as follows:
(1) For each sample in the minority class, calculating the distance from the sample to each other sample by taking Euclidean distance, mahalanobis distance and other calculation methods as standards;
(2) Setting a value of 5, and selecting the nearest neighbor sample; secondly, setting a value representing the sampling multiplying power, and determining the number of finally generated samples;
(3) Randomly selecting a neighbor in the neighbor samples;
(4) Finally, a number from 0 to 1 is randomly set to represent the position proportion of the newly synthesized sample on the connecting line of the original minority sample and the randomly selected neighbor sample.
The steps of the SMOTE method are schematically shown in figure 5. In fig. 5, circles represent normal majority samples, five-pointed star in the middle represents minority samples, five-pointed star around represents neighbor samples of minority samples, and black rectangle is new minority samples.
The GAN algorithm flow is as follows:
(1) Initializing a generator G and a discriminator D;
(2) During each iteration:
(1) a fixed generator G which only updates the parameters of the discriminator D;
(2) fixing the discriminator D, and only updating parameters of the generator G;
GAN objective optimization function:
the data generated by the SMOTE and GAN methods are used for CNN neural network training, and an initial neural network model is generated for switch fault diagnosis. The model can carry out turnout fault diagnosis through the data preprocessed by the given current curve, and the diagnosis result is divided into a normal state, 6 fault states and an unidentified state. The central cloud layer transmits the trained neural network model to each edge cloud layer node.
As shown in fig. 3 and 4, step 3 is specifically as follows:
and 3.1, as shown in fig. 3, the access layer transmits the current curve data received by the sensor to the edge cloud layer node nearby, and performs data storage, data preprocessing and fault diagnosis.
The data preprocessing is to perform data length compensation, data segmentation and eigenvalue calculation on the received current curve data. And then, performing fault diagnosis on the preprocessed data by using a neural network diagnosis model for sending the central cloud layer.
Step 3.2, if the diagnosis result according to the CNN model is normal, not processing; if the fault exists, different fault treatments are respectively carried out according to different fault types. The specific six common faults and maintenance proposals are as follows:
fault state f1: foreign matter exists between the rails.
And (3) fault treatment: and overhauling the switch tongue rail and the stock rail to see whether foreign matters exist between the switch tongue rail and the stock rail.
Fault state f2: the transition phase fluctuates.
And (3) fault treatment: and overhauling the switch machine and the friction belt.
Fault state f3: the locking phase fluctuates.
And (3) fault treatment: and overhauling the cable box.
Fault state f4: the start-up phase fluctuates.
And (3) fault treatment: an automatic shutter for maintenance.
Fault state f5: the locking phase is difficult.
And (3) fault treatment: and (5) overhauling the turnout.
Fault state f6: and (5) fixing and rotor mixing lines.
And (3) fault treatment: and overhauling the switch machine.
As shown in fig. 3, step 4 is specifically as follows:
and 4.1, setting a certain time period T, transmitting the preprocessed data stored in the database to the central cloud layer by the edge cloud layer, storing the preprocessed data received by the central cloud layer from each edge cloud layer node into the database, and combining the preprocessed data with the stored data to train the neural network model. And after training, the neural network model is still sent to each edge cloud layer node.
And 4.2, when the unidentified current curve data is diagnosed by the edge cloud layer, timely alarming to switch fault overhauling personnel and recording the record at a central cloud layer server. When the same kind of unidentified current curve data reaches more than 50, the data set can be expanded by the SMOTE method and the GAN method, and the method is used for generating a CNN model of a new fault kind. The new fault class can be added into the conventional fault class by means of manually identifying the fault type, the fault maintenance mode and the generation of the fault diagnosis model.
The step 5 is specifically as follows:
the amount of normal current curve data collected by the microcomputer monitoring system in the access layer is often much greater than the fault current curve data, so the normal curve data set required for the central cloud layer training neural network model is much simpler than the acquisition of the fault curve data set. When the normal current curve data amount reaches a certain amount (more than 1 ten thousand are recommended), the normal current curve data from the access layer is not needed any more, so that only fault current curve data and unidentified current curve data are needed to be received at the moment, and the accuracy of the neural network model is improved. In order to save resources and improve efficiency, the time period can be prolonged, and the transmission frequency can be reduced.
And after the time period is up, training the CNN model through the stored preprocessed data set, and transmitting the training model to each edge node.
An example of the identification method of the present invention is to take part of the history data as an input data set. Data enhancement is firstly carried out on a central cloud layer through SMOTE and GAN methods. And then performing data preprocessing and model training based on a sample set consisting of the generated data samples and the real samples as a sample set of the CNN model. And then transmitting the trained CNN model to each node of the edge cloud layer for fault diagnosis, wherein a fault diagnosis flow chart is shown in fig. 4. And after the edge cloud layer carries out data preprocessing on the turnout current curve data received from the access layer, fault diagnosis is carried out on one hand, data storage is carried out on the other hand, and the data are transmitted to the center cloud layer for model training until a certain time period is reached. Because the data volume of normal current is far more than the data volume of fault current, the data volume distribution is uneven and can lead to poor training effect, so that the normal action current data of the same type turnout of the central cloud layer is collected to more than 1 ten thousand, the normal data from the edge cloud layer is not received any more, and only the fault data is received. And if the normal data and various fault data in the real data reach more than 1 ten thousand, canceling sample data generated by the SMOTE and GAN methods, and training a CNN model by adopting the real data from the access layer.
Claims (7)
1. The intelligent switch fault diagnosis method based on the edge computing network architecture is characterized by comprising the following specific operation steps:
step 1, the method generally divides a turnout fault diagnosis network architecture into three layers, namely an access layer, an edge cloud layer and a center cloud layer;
step 2, the central cloud layer uses the existing real data to carry out data enhancement through SMOTE and GAN methods, is used for training an initial neural network model, and transmits the trained model to the edge cloud layer;
step 3, the access layer collects the starting current curve data of the switch machine through a microcomputer monitoring system and sends the starting current curve data to the edge cloud layer for data storage, data preprocessing and fault diagnosis;
step 4, after preprocessing the data, the edge cloud layer performs fault diagnosis on one hand, and transmits the data to the central cloud layer for data storage and neural network model training on the other hand, and transmits the data which cannot be diagnosed to the central cloud layer for recording processing;
step 5, on the basis of the step 4, when the real data amount in the central cloud layer reaches more than 1 ten thousand, the real data from the edge cloud layer is not received any more, and only the fault data and unidentified data determined after the intelligent diagnosis of the edge cloud layer are received; the central cloud layer carries out neural network training according to a time period T, and the obtained neural network model is transmitted to the edge cloud layer.
2. The intelligent fault diagnosis method for the turnout based on the edge computing network architecture as claimed in claim 1, wherein the access layer comprises a point machine, a sensor, a microcomputer monitoring system and the like; the edge cloud layer comprises a server and a database of edge nodes; the central cloud layer comprises a large server and a database; the central cloud layer has extremely strong data computing capacity and data storage capacity compared with the servers of the edge cloud layer;
the edge cloud layer is used for data preprocessing and fault diagnosis based on the neural network model, and the calculation of the neural network model is performed on the center cloud layer.
3. The intelligent switch fault diagnosis method based on the edge computing network architecture as claimed in claim 2, wherein the step 2 is specifically as follows:
the central cloud layer stores more than 400 pieces of historical real same type turnout action current curve data, and the data are classified according to normal data and 6 common fault types, wherein more than 50 pieces of 6 similar fault current curve data are respectively stored, and more than 100 pieces of normal data are stored; the central cloud layer uses a synthetic minority oversampling technology SMOTE and an anti-network GAN generation method to enhance data through the data; the 6 similar fault current curve data are respectively increased to more than 500 pieces, the normal class data are increased to more than 1000 pieces, and the total data is more than 4000 pieces;
wherein, the normal data type is marked as f0;
the type of the fault state of the foreign matter existing between the rails is marked as f1;
the type of the fluctuation fault state in the transition stage is marked as f2;
the type of the fluctuation fault state in the locking stage is marked as f3;
the type of the fluctuation fault state in the starting stage is marked as f4;
the difficult fault state type in the locking stage is marked as f5;
the type of the stator and rotor mixed line fault state is marked as f6;
when current curve data enhancement is carried out, a GAN method is used for four data types f0, f1, f4 and f6, and an SMOTE method is used for three data types f2, f3 and f5;
the data generated by the SMOTE and GAN methods are used for CNN neural network training, and an initial neural network model is generated for diagnosing turnout faults; the model can carry out turnout fault diagnosis through the data preprocessed by the given current curve, and the diagnosis result is divided into a fault-free state, 6 fault states and an unidentified state; the central cloud layer transmits the trained neural network model to each edge cloud layer node;
and if the normal data and various fault data in the real data reach more than 1 ten thousand, canceling sample data generated by the SMOTE and GAN methods, and training a CNN model by adopting the real data from the access layer.
4. The intelligent switch fault diagnosis method based on the edge computing network architecture according to claim 1, wherein the period T in the step 5 is defined according to user requirements, and defaults to 1 week.
5. The intelligent switch fault diagnosis method based on the edge computing network architecture according to claim 3, wherein the step 3 is as follows:
the access layer transmits the current curve data received by the microcomputer monitoring system to the edge cloud layer node nearby, and performs data storage, data preprocessing and fault diagnosis;
the data preprocessing is to perform data length supplementing, data segmentation and feature extraction on the received current curve data; then, performing fault diagnosis on the preprocessed data by using a neural network diagnosis model of the transmission of the central cloud layer;
the data length compensation means that the data length of the current curve is zero-compensated to 576;
the data segmentation refers to dividing the compensated data into three sections according to a sliding window method, namely unlocking, converting and locking stages respectively;
the feature extraction means that 11 widely used feature extraction equations are used for describing geometric features of the curve, including import and export differences, average values, root mean square values, variances, checksums, kurtosis values, peak factors, waveform factors, pulse factors and margin factors.
6. The intelligent switch fault diagnosis method based on the edge computing network architecture as claimed in claim 3, wherein the step 4 is specifically as follows:
step 4.1, each node of the edge cloud layer transmits the preprocessed data stored in the database to the center cloud layer in real time and stores the preprocessed data in the database; setting a time period of T days, and combining the stored data by the central cloud layer to train a neural network model; after training the neural network model, transmitting the neural network model to each edge cloud layer node;
step 4.2, after the edge cloud layer performs data preprocessing on the current curve data from the access layer, fault diagnosis is performed on one hand, and the current curve data is transmitted to the center cloud layer on the other hand;
the fault diagnosis result is divided into a normal state, 6 common fault states and an unidentified state; if the state is normal, no treatment is carried out; if the fault is 6, giving an alarm to a switch fault maintainer to process the fault type; if the curve data is in an unidentified state, transmitting the curve data to a central cloud layer for recording;
the recording processing refers to the steps of storing the curve data of the diagnosed unidentified state as a new unidentified state type, appointing an maintainer to overhaul a turnout section where the unidentified state occurs, and determining whether the turnout has faults, fault reasons and state types; finally, storing the analysis result in a server;
when the unidentified current curve data of the same class reaches more than 50, the data set is expanded by SMOTE and GAN methods, and the method is used for generating a CNN model of a new fault class.
7. The intelligent switch fault diagnosis method based on the edge computing network architecture as claimed in claim 3, wherein the step 5 is specifically as follows:
the normal current curve data quantity is far greater than the fault current curve data, when the normal current curve data quantity reaches more than 1 ten thousand, the normal current data is not received any more, and only the fault current curve data and unidentified current curve data are received, so that the accuracy of the neural network model is improved; in order to save resources and improve efficiency, the time period can be prolonged, and the transmission frequency can be reduced.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110757298.0A CN113641486B (en) | 2021-07-05 | 2021-07-05 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110757298.0A CN113641486B (en) | 2021-07-05 | 2021-07-05 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113641486A CN113641486A (en) | 2021-11-12 |
| CN113641486B true CN113641486B (en) | 2024-03-01 |
Family
ID=78416724
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110757298.0A Active CN113641486B (en) | 2021-07-05 | 2021-07-05 | Intelligent turnout fault diagnosis method based on edge computing network architecture |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113641486B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114415641A (en) * | 2022-01-29 | 2022-04-29 | 重庆长安汽车股份有限公司 | Privacy protection method and system for remote fault diagnosis |
| CN117081502B (en) * | 2023-10-16 | 2024-02-06 | 锦浪科技股份有限公司 | Photovoltaic direct current arc monitoring platform and control method thereof |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN106709567A (en) * | 2016-12-14 | 2017-05-24 | 河北省科学院应用数学研究所 | Method for diagnosing switch faults based on deep learning model |
| WO2020244134A1 (en) * | 2019-06-05 | 2020-12-10 | 华南理工大学 | Multi-task feature sharing neural network-based intelligent fault diagnosis method |
| CN112348170A (en) * | 2020-11-10 | 2021-02-09 | 交控科技股份有限公司 | Fault diagnosis method and system for turnout switch machine |
| CN112738092A (en) * | 2020-12-29 | 2021-04-30 | 北京天融信网络安全技术有限公司 | Log data enhancement method, classification detection method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11416748B2 (en) * | 2019-12-18 | 2022-08-16 | Sap Se | Generic workflow for classification of highly imbalanced datasets using deep learning |
-
2021
- 2021-07-05 CN CN202110757298.0A patent/CN113641486B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2015158198A1 (en) * | 2014-04-17 | 2015-10-22 | 北京泰乐德信息技术有限公司 | Fault recognition method and system based on neural network self-learning |
| CN106709567A (en) * | 2016-12-14 | 2017-05-24 | 河北省科学院应用数学研究所 | Method for diagnosing switch faults based on deep learning model |
| WO2020244134A1 (en) * | 2019-06-05 | 2020-12-10 | 华南理工大学 | Multi-task feature sharing neural network-based intelligent fault diagnosis method |
| CN112348170A (en) * | 2020-11-10 | 2021-02-09 | 交控科技股份有限公司 | Fault diagnosis method and system for turnout switch machine |
| CN112738092A (en) * | 2020-12-29 | 2021-04-30 | 北京天融信网络安全技术有限公司 | Log data enhancement method, classification detection method and system |
Non-Patent Citations (2)
| Title |
|---|
| 云/边缘协同的轴承故障诊断方法;张文龙;胡天亮;王艳洁;魏永利;;计算机集成制造系统;20201231(第03期);全文 * |
| 基于条件式Wasserstein生成对抗网络的电力变压器故障样本增强技术;刘云鹏;许自强;和家慧;王权;高树国;赵军;;电网技术;20201231(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113641486A (en) | 2021-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107941537B (en) | A kind of mechanical equipment health state evaluation method | |
| CN110929918B (en) | 10kV feeder fault prediction method based on CNN and LightGBM | |
| CN113641486B (en) | Intelligent turnout fault diagnosis method based on edge computing network architecture | |
| CN110441629B (en) | Fault diagnosis method and device based on turnout switch machine operating current curve | |
| CN107179503A (en) | The method of Wind turbines intelligent fault diagnosis early warning based on random forest | |
| CN106443447B (en) | A kind of aerogenerator fault signature extracting method based on iSDAE | |
| CN107909118A (en) | A kind of power distribution network operating mode recording sorting technique based on deep neural network | |
| CN104200396B (en) | A kind of wind turbine component fault early warning method | |
| CN108802525A (en) | Equipment fault intelligent Forecasting based on small sample | |
| CN107340766B (en) | Power scheduling alarm signal text based on similarity sorts out and method for diagnosing faults | |
| CN109597401A (en) | A kind of equipment fault diagnosis method based on data-driven | |
| CN111199257A (en) | Fault diagnosis method and device for high-speed rail driving equipment | |
| CN103176128A (en) | Method and system for forcasting state of wind generating set and diagnosing intelligent fault | |
| CN105894177A (en) | Decision-making-tree-algorithm-based analysis and evaluation method for operation risk of power equipment | |
| CN108537259A (en) | Train control on board equipment failure modes and recognition methods based on Rough Sets Neural Networks model | |
| CN111046583A (en) | Switch machine fault diagnosis method based on DTW algorithm and ResNet network | |
| Yang et al. | Fault detection of wind turbine generator bearing using attention-based neural networks and voting-based strategy | |
| CN110309884A (en) | Electricity consumption data anomalous identification system based on ubiquitous electric power Internet of Things net system | |
| CN108876163A (en) | The transient rotor angle stability fast evaluation method of comprehensive causality analysis and machine learning | |
| CN106897945A (en) | The clustering method and equipment of wind power generating set | |
| CN112200263B (en) | Self-organizing federal clustering method applied to power distribution internet of things | |
| CN111539381B (en) | Construction method of wind turbine bearing fault classification diagnosis model | |
| CN116070140B (en) | Power distribution substation safe operation state monitoring system and method | |
| CN114167837B (en) | Intelligent fault diagnosis method and system for railway signal system | |
| CN110458405A (en) | It is a kind of based on electric power-Characteristics of micrometeorology data power system security method for early warning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |