CN111914320A - No-sample turnout fault diagnosis method based on deep learning - Google Patents
No-sample turnout fault diagnosis method based on deep learning Download PDFInfo
- Publication number
- CN111914320A CN111914320A CN202010508456.4A CN202010508456A CN111914320A CN 111914320 A CN111914320 A CN 111914320A CN 202010508456 A CN202010508456 A CN 202010508456A CN 111914320 A CN111914320 A CN 111914320A
- Authority
- CN
- China
- Prior art keywords
- turnout
- fault
- phase
- switch
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000003745 diagnosis Methods 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000013135 deep learning Methods 0.000 title claims abstract description 24
- 230000009471 action Effects 0.000 claims abstract description 34
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000004088 simulation Methods 0.000 claims abstract description 22
- 230000007246 mechanism Effects 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 33
- 238000011176 pooling Methods 0.000 claims description 20
- 238000013527 convolutional neural network Methods 0.000 claims description 16
- 230000004913 activation Effects 0.000 claims description 14
- 238000006243 chemical reaction Methods 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 11
- 230000008569 process Effects 0.000 claims description 10
- 238000010276 construction Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 230000006399 behavior Effects 0.000 claims description 5
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 4
- 238000001514 detection method Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 239000003550 marker Substances 0.000 claims description 2
- 230000001629 suppression Effects 0.000 claims description 2
- 238000012544 monitoring process Methods 0.000 description 12
- 238000002405 diagnostic procedure Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 3
- 238000013024 troubleshooting Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000011664 signaling Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Architecture (AREA)
- Civil Engineering (AREA)
- Structural Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Supply And Distribution Of Alternating Current (AREA)
- Protection Of Generators And Motors (AREA)
Abstract
A turnout fault diagnosis method based on deep learning under no-field fault samples comprises the following steps: the method comprises the following steps of establishing a simulation model: constructing a turnout action simulation model by using the action mechanism of a turnout switch machine; simulation to obtain a data sample: analyzing common faults of the turnout, obtaining turnout action current curves in different fault states based on a turnout action simulation model, and establishing a turnout fault sample set; establishing a fault diagnosis model: constructing a turnout fault diagnosis model by adopting a deep learning method, and training the turnout fault model by utilizing the turnout fault sample set; and inputting the image to be detected to the turnout fault diagnosis model to obtain a diagnosis result. The turnout fault diagnosis method can be used for diagnosing turnout faults with high recognition rate under the condition of no field fault sample, so that the turnout fault recognition efficiency is improved, and the driving safety is guaranteed.
Description
Technical Field
The invention relates to the field of rail transit key equipment fault diagnosis.
Background
With the rapid development of railways in China, higher requirements are put forward on the safety and reliability of railway signal equipment. Switches are key components of railway signaling equipment, and the state of the switches directly influences the safety and efficiency of railway transportation. The turnout for switching the train from one track to another is a key device for arranging the train route and realizing the route switching. The point switch is the switching equipment of the switch, used for realizing switching the switch, locking the switch and reflecting the position of the switch tongue. The failure is one of the main reasons causing major accidents of railways.
At present, the turnout is overhauled by combining a microcomputer monitoring system to monitor the action state of the turnout and a manual regular troubleshooting mode in China. The microcomputer monitoring system collects the data of the switch such as the action voltage, the action current and the like and displays the action current curve and the power curve of the switch in the system. If a curve different from the standard curve is found, the microcomputer monitoring system sends an alarm signal, and the experienced staff judges the type of the turnout fault according to the curve and arranges the field staff to go to overhaul. However, the fault prediction and diagnosis method based on the regular troubleshooting and the analysis of the microcomputer monitoring system information by the staff has low efficiency, large workload and low reliability, and can not meet the requirements of the existing high-speed railway on operation safety and efficiency.
To address the above-mentioned deficiencies, many researchers have conducted much research around the diagnosis of turnout failures. The method is mainly applied to diagnosing the turnout fault type based on a mathematical model, signal processing and knowledge. However, most of the existing researches need each kind of turnout fault samples, but the field cannot provide such many kinds of turnout fault samples; and the research requires raw data in a microcomputer monitoring system, but the raw data is difficult to obtain due to the privacy of the data.
Disclosure of Invention
The invention provides a turnout fault diagnosis method based on deep learning under a no-field fault sample, which can carry out turnout fault diagnosis with high recognition rate under the no-field fault sample, improve turnout fault recognition efficiency and ensure driving safety.
In order to achieve the above purpose, the invention provides the following technical scheme
A turnout fault diagnosis method based on deep learning under no-field fault samples comprises the following steps:
(1): establishing a simulation model: constructing a turnout action simulation model by using the action mechanism of a turnout switch machine;
(2): simulation to obtain a data sample: analyzing common faults of the turnout, obtaining turnout action current curves in different fault states based on the turnout action simulation model, and establishing a turnout fault sample set;
(3): establishing a fault diagnosis model: constructing a turnout fault diagnosis model by adopting a deep learning method, and training the turnout fault model by utilizing the turnout fault sample set;
(4) inputting an image to be detected to a turnout fault diagnosis model to obtain a diagnosis result;
wherein the step (1) further comprises:
(1a) the point switch is a switch device of a turnout, analyzes main parameters of the point switch, and constructs six modules of a turnout action simulation model: the device comprises a three-phase voltage source, a three-phase circuit breaker, a three-phase asynchronous motor, an RMS module, a gain module and a From workbench module;
(1b) connecting the first five modules in sequence, and adding three oscilloscopes into the model; the effective value of the three-phase stator current calculated by the RMS module is observed in the oscilloscope 1, the load torque curve of the three-phase asynchronous motor is checked in the oscilloscope 2, and the curves of the three-phase rotor current, the rotating speed and the mechanical torque are observed in the oscilloscope 3.
(1c) The From Workplace module is connected to the three-phase asynchronous motor;
in the step (1a), six module parameters are as follows:
(1a1) the voltage in the three-phase voltage source is 380V; the phase is 0 degree; the frequency is 50 Hz; the voltage does not change with time; no harmonic wave is generated; the generator type is swing;
(1a2) the initial state in the three-phase circuit breaker is open; switching the A phase, the B phase and the C phase; normally, the switching on time is 1/60 s; the three-phase voltage source is connected with the three-phase asynchronous motor, and the switching-off time is 396/60 s; the internal resistance is 0.01 Ω; the buffer resistance is 105 Ω; the buffer capacitance is infF;
(1a3) the rotor of the three-phase asynchronous motor is in a squirrel-cage shape; the mechanical input is torque; the reference frame is a rotor; no name is used to identify the bus tag; the rated power is 400W; the line voltage is 380V; the frequency is 50 Hz; the stator resistance is 0.5968 Ω; the stator inductance is 0.0003495H; the rotor resistance is 0.6258 Ω; the rotor inductance is 0.005473H; the mutual inductance is 0.0354H; an inertia constant of 0.05(kg.m 2); coefficient of friction 0.005879 (n.m.s); the number of pole pairs is 2; the initial slippage, the electrical angle, the stator current amplitude and the phase angle are all 0; saturation is not simulated; the sampling time is-1; the discrete solver model is a Tracpeoidal non iterative; mechanical power 1.492 × 106W;
(1a4) the RMS module fundamental frequency is 50 Hz; the initial root mean square value is 0; the sampling time is 0;
(1a5) the gain module setting is 0.03;
(1a6) in the From Workplace module, acquiring data of time and values, wherein the time is a time value when the torque changes, and the values are loaded torque values; the type of the output data is double; the sampling time is 0; interpolation and zero crossing detection are used;
wherein, the construction of the turnout fault sample set in the step (2) comprises the following steps:
(2a) acquiring a resistance curve of a turnout as a load torque curve of a motor;
(2b) the From Workplace module reads the load torque curve From the MATLAB working area and outputs the load torque curve to the three-phase asynchronous motor, and the process of load conversion during turnout conversion is simulated;
(2c) setting the closing time of the three-phase breaker under different fault categories according to the selected common turnout fault categories;
(2d) starting the simulation, starting a three-phase open-phase device, supplying power by a three-phase voltage source, and rotating a three-phase asynchronous motor after obtaining voltage; after the specified time is reached, the three-phase open-phase device is closed, the three-phase asynchronous motor cuts off the power supply, and the three-phase asynchronous motor stops working;
(2e) obtaining a turnout action current curve in an oscilloscope 1;
(2f) and (3) acquiring resistance curves of the turnout under different fault states, and repeating the processes (2a), (2b), (2c) and (2d) to obtain a turnout fault sample set.
In step (2c), the selectable common turnout fault categories are: sudden current increase, conversion time increase, excessive numerical value of 'small tail', two-phase small current, 'small tail' overlong, no 'small tail' fault, and no fault curve also including; the closing time of the three-phase breaker switch is determined by the switch fault category, and the corresponding relationship between the switch fault category and the closing time of the three-phase breaker switch is as follows:
(2c1) current sudden increase: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(2c2) the conversion time increases: the switch off time of the three-phase breaker is respectively set to be 6s and 7 s;
(2c3) the numerical value of the small tail exceeds the standard: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(2c4) two-phase small current: the opening time of the three-phase breaker is set to be 0s, and the closing time is set to be 3.4 s;
(2c5) too long "small tail": the switch off time of the three-phase breaker is respectively set to be 5.4s and 13 s;
(2c6) no "small tail": the switch off time of the three-phase breaker is respectively set to 5.4 s;
(2c7) normal curve: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
the construction of the turnout fault diagnosis model in the step (3) comprises the following steps:
(3a) dividing a turnout fault sample set into a training set and a testing set;
(3b) constructing a convolutional neural network structure based on the LENET;
(3c) leading the training set image into the convolutional neural network structure established in the step (3b) for training; training to obtain a high-precision convolutional neural network which is a turnout fault diagnosis model;
(3d) after training is finished, the test set is used for checking the performance of the finally selected optimal model;
the convolutional neural network structure of step (3b) is shown as follows:
(3b1) based on the LENET structure, the network comprises two convolution layers, a pooling layer and two full-connection layers;
(3b2) the convolution layer extracts basic characteristics of an original image, such as color, texture, shape and the like through convolution kernels of 7 x 7 and 3 x 3; the pooling layer adopts a maximum pooling method to carry out maximum sampling on the previous layer, and the size of a sliding window is 2 multiplied by 2; the fully-connected layer maps the feature representations of the convolutional and pooling layers to marker layers of the data sample; the network calculates the inner product of the output vector of the previous layer and the connection weight vector, then adds the offset, and obtains an output state of the whole network after the operation of the activation function; the activation function is formulated as follows:
where J × I is the size dimensions of the width and height of the convolution kernel, M × N is the size of the width and height of the input image, xm,nRepresenting the pixel value, y, at the (m, n) position in the input picturem′,n′Representing the corresponding calculation result; w is a weight and represents the influence of the corresponding feature x; f is an activation function, and nonlinear transformation is carried out on the inner product; b is an offset which serves to select the dividing line;
(3b3) the first layer and the second layer are convolution layers with 32 convolution kernels, and the sizes of the convolution kernels are 7 multiplied by 7 and 3 multiplied by 3 respectively; the third layer is the maximum pooling layer of the sliding window size 2 × 2, and a Dropout value of 0.25 is set thereto; the fourth layer and the fifth layer are full connection layers, 903-dimensional information is output, and a classification probability result is obtained; a Dropout value of 0.5 is set behind the fourth full-connection layer;
in the step (3c), the training method of the convolutional neural network comprises the following steps:
(3c1) inputting a training set of a turnout fault sample set into a convolutional neural network, wherein the size of a training set sample is 150 multiplied by 150; in the convolutional neural network, firstly, a convolutional layer performs convolutional operation and activation operation on an input layer so as to extract behavior space characteristics of the input layer; the activation function adopts a RELU function, and the formula is as follows:
f(x)＝max(0,x)
wherein x is an input vector; the RELU function performs unilateral suppression on the input vector;
(3c2) performing pooling operation on the behavior space characteristics of the convolutional layer through a pooling layer, wherein the pooling operation is used for compressing the quantity of data and parameters;
(3c3) the convolutional neural network passes through two convolutional layers and a pooling layer until reaching a full connection layer, the full connection layer performs full connection operation on input data with the size of 32 x 71, and an output vector with the dimension of 903 is generated through a RELU activation function; calculating the output vector of the full connection layer by a softmax function to obtain a final predicted value, calculating a loss function value of the predicted value and a real value by using a cross entropy loss function, and minimizing the loss function value;
(3c4) and continuously adjusting the network weight and bias by a random gradient descent method, and recalculating the loss function value until the loss function value tends to be stable or reaches a set iteration number, thereby obtaining the classified picture characteristics.
The verification process of step (3d) shown is as follows:
(3d1) inputting the verification set image into the turnout fault diagnosis model;
(3d2) the output of the model is a probability array, and each element value in the probability array respectively represents the probability of the corresponding turnout fault; the calculation formula of the recognition result is as follows:
Result＝argmax(Ri),i＝1,2,3,…
where i is the number of switch fault samples, and RiIs the probability array of each image, and the argmax function is to find the label corresponding to the maximum probability in the probability array and take the label as the switch fault identification result.
Wherein, the step (4) of inputting the image to be detected to obtain the diagnosis result comprises the following steps:
(4a) fitting microcomputer detection data into a curve picture by using MATLAB software;
(4b) inputting the curve picture into a diagnosis model, and correctly identifying the curve picture generated by the original data by the turnout diagnosis model;
(4c) the report of the paper or paper edition is cut out to contain the part of the turnout action current curve and is stored as a picture;
(4d) and inputting the curve picture into a turnout diagnosis model, and correctly identifying the curve picture obtained in a paper or paper report by the turnout diagnosis model.
The invention provides a turnout fault diagnosis method based on deep learning without field fault samples, which can be used for carrying out turnout fault diagnosis with high recognition rate without field fault samples, improving turnout fault recognition efficiency and ensuring driving safety.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a deep learning-based turnout fault diagnosis method without field fault samples according to an embodiment of the invention;
FIG. 2 is a turnout action model constructed by using a three-phase asynchronous motor according to an embodiment of the invention;
FIG. 3 is a flow chart of construction of a switch fault sample set in an embodiment of the present invention;
fig. 4 is a sample set of turnout faults constructed by the embodiment of the present invention, and the pictures are sequentially from top to bottom and from left to right: sudden current increase, conversion time increase, excessive numerical value of 'small tail', two-phase small current, 'small tail' overlong, no 'small tail' and no fault curve;
FIG. 5 is a schematic diagram of the construction of a turnout fault identification model according to an embodiment of the invention;
FIG. 6 is a diagram of a turnout fault identification model according to an embodiment of the present invention;
FIG. 7 is a graph of the accuracy trend of a convolutional neural model in accordance with an embodiment of the present invention;
FIG. 8 is a diagnostic process of data from the microcomputer monitoring system according to an embodiment of the present invention;
fig. 9 is a diagnostic process of switch action current curve pictures in paper or paper version of the report of the embodiment of the present invention.
Detailed Description
The technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings, and it should be understood that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
The invention relates to a turnout fault diagnosis method based on deep learning under a no-site fault sample.
Fig. 1 is a flowchart of a switch fault diagnosis method based on deep learning without field fault samples according to an embodiment of the present invention, including the following steps:
step S1: establishing a simulation model: constructing a turnout action simulation model by using the action mechanism of a turnout switch machine;
step S2: simulation to obtain a data sample: analyzing common faults of the turnout, obtaining turnout action current curves in different fault states based on the turnout action simulation model, and establishing a turnout fault sample set;
step S3: establishing a fault diagnosis model: constructing a turnout fault diagnosis model by adopting a deep learning method, and training the turnout fault model by utilizing the turnout fault sample set;
and step S4, inputting the image to be detected to the turnout fault diagnosis model to obtain a diagnosis result.
Step S1
Fig. 2 is a switch operation model constructed by using three-phase asynchronous motors. Taking the S700K switch as an example, the following parameters are mainly used:
(1) the rated voltage of the three-phase alternating-current asynchronous motor is 380V, and the rated power is 0.4 KW;
(2) the conversion force was 6000N;
(3) the action time is less than or equal to 6.6 s;
(4) the operating current is less than or equal to 2A;
(5) the stroke is 220 mm.
The turnout action model mainly comprises a three-phase voltage source, a three-phase circuit breaker, a three-phase asynchronous motor, an RMS module, a gain module and a From Workplace module; the six module parameters are set as follows:
(1) the voltage in the three-phase voltage source is 380V; the phase is 0 degree; the frequency is 50 Hz; the voltage does not change with time; no harmonic wave is generated; the generator type is swing;
(2) the initial state in the three-phase circuit breaker is open; switching the A phase, the B phase and the C phase; normally, the switching on time is 1/60 s; the three-phase voltage source is connected with the three-phase asynchronous motor, and the switching-off time is 396/60 s; the internal resistance is 0.01 Ω; the buffer resistance is 105 Ω; the buffer capacitance is infF;
(3) the rotor of the three-phase asynchronous motor is in a squirrel-cage shape; the mechanical input is torque; the reference frame is a rotor; no name is used to identify the bus tag; the rated power is 400W; the line voltage is 380V; the frequency is 50 Hz; the stator resistance is 0.5968 Ω; the stator inductance is 0.0003495H; the rotor resistance is 0.6258 Ω; the rotor inductance is 0.005473H; the mutual inductance is 0.0354H; an inertia constant of 0.05(k non g.m 2); coefficient of friction 0.005879 (n.m.s); the number of pole pairs is 2; the initial slippage, the electrical angle, the stator current amplitude and the phase angle are all 0; saturation is not simulated; the sampling time is-1; the discrete solver model is a Tracpezoidal iterative model; mechanical power 1.492 × 106W;
(4) the RMS module fundamental frequency is 50 Hz; the initial root mean square value is 0; the sampling time is 0;
(5) the gain module setting is 0.03;
(6) in the From Workplace module, acquiring data of time and values, wherein the time is a time value when the torque changes, and the values are loaded torque values; the type of the output data is double; the sampling time is 0; interpolation and zero crossing detection are used.
Connecting the first five modules in sequence, and adding three oscilloscopes into the model; observing the three-phase stator current effective value calculated by the RMS module in the oscilloscope 1, examining the load torque curve of the three-phase asynchronous motor in the oscilloscope 2, and observing the three-phase rotor current, rotating speed and mechanical torque curve in the oscilloscope 3; the From Workplace module is connected to a three-phase asynchronous motor.
Step S2
FIG. 3 is a flow chart of construction of a sample set of turnout faults;
firstly, acquiring a resistance curve of a turnout to obtain a load torque curve of a motor; the From Workplace module reads the load torque curve From the MATLAB working area and outputs the load torque curve to the three-phase asynchronous motor, and the process of load conversion during turnout conversion is simulated;
according to the selected turnout fault category, setting the closing time of the three-phase breaker under different fault categories; in this embodiment, six common turnout fault categories are selected, which are current sudden increase, conversion time increase, "small tail" value exceeds standard, two-phase small current, "small tail" overlong, no "small tail", and three-phase breaker closing time under different fault categories and normal conditions is set:
(1) current sudden increase: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(2) the conversion time increases: the switch off time of the three-phase breaker is respectively set to be 6s and 7 s;
(3) the numerical value of the small tail exceeds the standard: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(4) two-phase small current: the opening time of the three-phase breaker is set to be 0s, and the closing time is set to be 3.4 s;
(5) too long "small tail": the switch off time of the three-phase breaker is respectively set to be 5.4s and 13 s;
(6) no "small tail": the switch off time of the three-phase breaker is respectively set to 5.4 s;
(7) normal curve: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
starting the simulation, starting a three-phase open-phase device, supplying power by a three-phase voltage source, and rotating a three-phase asynchronous motor after obtaining voltage; after the specified time is reached, the three-phase open-phase device is closed, the three-phase asynchronous motor cuts off the power supply, and the three-phase asynchronous motor stops working; obtaining a turnout action current curve in an oscilloscope 1;
the resistance curves of the turnout under six fault states and normal conditions are collected, the process is repeated, and the obtained turnout fault sample set is shown in figure 4.
Step S3
Dividing a turnout fault sample set (including a non-fault sample) into a training set and a testing set for constructing a turnout fault diagnosis model, wherein fig. 5 is a training process of the model, performing iterative training on images of the sample set, and continuously adjusting parameters of a network until the training precision reaches an optimal model. The invention adopts a convolutional network model of the LENET to diagnose the turnout fault. For better explanation of the switch fault diagnosis method of the present embodiment, the following description is made with reference to the model parameters shown in fig. 6.
1. The training network comprises two convolution layers, a pooling layer and two full-connection layers; the first layer is a convolutional layer and receives the input of 3 × 150 × 150 images, where 150 × 150 refers to the width and height of the input image, and 3 refers to the RGB channels. The first layer of convolution kernels has the size of 7 multiplied by 7, the number of the convolution kernels is 30, the output shape is 32 multiplied by 144 characteristic diagram, the second layer of convolution kernels has the size of 3 multiplied by 3, and the output shape is 32 multiplied by 142 characteristic diagram; the third layer is the largest pooling layer with kernel size 2 × 2, after which Dropout value is set to 0.25 to prevent over-fitting; the fourth layer and the fifth layer are full connection layers, 903-dimensional information is output, and a classification probability result is obtained; and a Dropout value of 0.5 is set behind the fourth fully connected layer.
And then, inputting the images of the test set into a LENET model, and performing convolution operation and activation operation on the output of the previous layer by the next layer in the LENET model to extract the behavior space characteristics of the previous layer. In addition, through a random gradient descent method SGD optimizer, a Relu activation function performs nonlinear transformation, network weight and bias are continuously adjusted, a cross entropy function is used for recalculating a loss function value until the loss function value tends to be stable or reaches a set iteration number, and classified picture characteristics are obtained. This example sets the number of batches to 16 and the number of iterations to 10, based on the keras2.1.5 frame training.
And finally, solidifying the structure and parameters of the deep learning network model after training to obtain a turnout fault diagnosis model, wherein the accuracy change in the model training process is shown in FIG. 7, and the final accuracy of the model reaches 99.80%.
2. Inputting the test set image into the turnout fault diagnosis model; the output of the model is a probability array, and each element value in the probability array respectively represents the probability of the corresponding turnout fault; the calculation formula of the recognition result is as follows:
Result＝argmax(Ri),i＝1,2,3,…
where i is the number of switch fault samples, and RiIs the probability array of each image, and the argmax function is to find the label corresponding to the maximum probability in the probability array and take the label as the switch fault identification result.
Step S4
This step is an optional step.
The input image to be detected is the monitoring data of a microcomputer monitoring system and a turnout action current curve picture in a report of a paper or paper edition;
FIG. 8 is a diagnostic process of the monitoring data of the microcomputer monitoring system, wherein the monitoring data obtained from the microcomputer monitoring system is fitted into a curve picture by using MATLAB software; the picture comprises a normal curve and a curve without a small tail, the curve picture is input into a diagnosis model, and a turnout diagnosis model correctly identifies the two types of curves;
FIG. 9 is a diagnostic process of a switch action current curve picture in a paper or paper version of a report; cutting a part of a normal turnout action current curve in a file, and storing the part as a picture; and inputting the curve picture into a turnout diagnosis model, and correctly identifying the curve picture obtained in the report of the paper or paper edition as a normal curve by the turnout diagnosis model.
Claims (10)
1. A turnout fault diagnosis method based on deep learning under no-field fault samples is characterized by comprising the following steps:
(1): establishing a simulation model: constructing a turnout action simulation model by using the action mechanism of a turnout switch machine;
(2): simulation to obtain a data sample: analyzing common faults of the turnout, obtaining turnout action current curves in different fault states based on the turnout action simulation model, and establishing a turnout fault sample set;
(3): establishing a fault diagnosis model: constructing a turnout fault diagnosis model by adopting a deep learning method, and training the turnout fault model by utilizing the turnout fault sample set;
(4): and inputting the image to be detected to the turnout fault diagnosis model to obtain a diagnosis result.
2. The switch fault diagnosis method based on deep learning without field fault samples according to claim 1, wherein in the step (1), the switch action simulation model comprises the following steps:
(1a) the point switch is a switch device of a turnout, analyzes main parameters of the point switch, and constructs six modules of a turnout action simulation model: the device comprises a three-phase voltage source, a three-phase circuit breaker, a three-phase asynchronous motor, an RMS module, a gain module and a From workbench module;
(1b) the first five modules were connected in sequence and three oscilloscopes were added to the model: observing the three-phase stator current effective value calculated by the RMS module in the oscilloscope 1, examining the load torque curve of the three-phase asynchronous motor in the oscilloscope 2, and observing the three-phase rotor current, rotating speed and mechanical torque curve in the oscilloscope 3;
(1c) the From Workplace module is connected to a three-phase asynchronous motor.
3. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 2, wherein the six module parameters in the step (1a) are:
(1a1) the voltage in the three-phase voltage source is 380V; the phase is 0 degree; the frequency is 50 Hz; the voltage does not change with time; no harmonic wave is generated; the generator type is swing;
(1a2) the initial state in the three-phase circuit breaker is open; switching the A phase, the B phase and the C phase; normally, the switching on time is 1/60 s; the three-phase voltage source is connected with the three-phase asynchronous motor, and the switching-off time is 396/60 s; the internal resistance is 0.01 Ω; the buffer resistance is 105 Ω; the buffer capacitance is infF;
(1a3) the rotor of the three-phase asynchronous motor is in a squirrel-cage shape; the mechanical input is torque; the reference frame is a rotor; no name is used to identify the bus tag; the rated power is 400W; the line voltage is 380V; the frequency is 50 Hz; the stator resistance is 0.5968 Ω; the stator inductance is 0.0003495H; the rotor resistance is 0.6258 Ω; the rotor inductance is 0.005473H; the mutual inductance is 0.0354H; an inertia constant of 0.05(kg.m 2); coefficient of friction 0.005879 (n.m.s); the number of pole pairs is 2; the initial slippage, the electrical angle, the stator current amplitude and the phase angle are all 0; saturation is not simulated; the sampling time is-1; the discrete solver model is a Tracpeoidal non iterative; mechanical power 1.492 × 106W;
(1a4) the RMS module fundamental frequency is 50 Hz; the initial root mean square value is 0; the sampling time is 0;
(1a5) the gain module setting is 0.03;
(1a6) in the From Workplace module, acquiring data of time and values, wherein the time is a time value when the torque changes, and the values are loaded torque values; the type of the output data is double; the sampling time is 0; interpolation and zero crossing detection are used.
4. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 1, wherein the construction of the switch fault sample set in the step (2) comprises:
(2a) acquiring a resistance curve of a turnout as a load torque curve of a motor;
(2b) the From Workplace module reads the load torque curve From the MATLAB working area and outputs the load torque curve to the three-phase asynchronous motor, and the process of load conversion during turnout conversion is simulated;
(2c) setting the closing time of the three-phase breaker under different fault categories according to the selected common turnout fault categories;
(2d) starting the simulation, starting a three-phase open-phase device, supplying power by a three-phase voltage source, and rotating a three-phase asynchronous motor after obtaining voltage; after the specified time is reached, the three-phase open-phase device is closed, the three-phase asynchronous motor cuts off the power supply, and the three-phase asynchronous motor stops working;
(2e) obtaining a turnout action current curve in an oscilloscope 1;
(2f) and (3) acquiring resistance curves of the turnout in different fault states, and repeating the processes (2a), (2b), (2c), (2d) and (2e) to obtain a turnout fault sample set.
5. The switch fault diagnosis method based on deep learning without field fault samples according to claim 4, wherein the common switch fault categories selectable in the step (2c) are: the current suddenly increases, the conversion time increases, the numerical value of the small tail exceeds the standard, the two-phase small current, the small tail is overlong, the small tail fault does not exist, and in addition, the fault-free curve also needs to be included.
6. The construction method for diagnosing the turnout fault based on the deep learning without the field fault sample is characterized in that in the step (2c), the closing time of the three-phase breaker switch is determined by the turnout fault category, and the corresponding relationship between the turnout fault category and the closing time of the three-phase breaker switch is as follows:
(2c1) current sudden increase: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(2c2) the conversion time increases: the switch off time of the three-phase breaker is respectively set to be 6s and 7 s;
(2c3) the numerical value of the small tail exceeds the standard: the switch off time of the three-phase breaker is respectively set to be 5.4s and 6.6 s;
(2c4) two-phase small current: the opening time of the three-phase breaker is set to be 0s, and the closing time is set to be 3.4 s;
(2c5) too long "small tail": the switch off time of the three-phase breaker is respectively set to be 5.4s and 13 s;
(2c6) no "small tail": the switch off time of the three-phase breaker is respectively set to 5.4 s;
(2c7) no fault curve: the three-phase breaker switch off times were set to 5.4s and 6.6s, respectively.
7. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 1, wherein the step (3) of establishing the fault diagnosis model comprises the following steps:
(3a) dividing a turnout fault sample set into a training set and a testing set;
(3b) constructing a convolutional neural network structure based on LeNet;
(3c) leading the training set image into the convolutional neural network structure established in the step (3b) for training; training to obtain a high-precision convolutional neural network which is a turnout fault diagnosis model;
(3d) and after the training is finished, the test set is used for checking the performance of the finally selected optimal model.
8. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 7, wherein in step (3b), the step of constructing the convolutional neural network structure is as follows:
(3b1) based on a LeNet structure, the network comprises two convolution layers, a pooling layer and two full-connection layers;
(3b2) the convolution layer extracts basic characteristics of an original image, such as color, texture, shape and the like through convolution kernels of 7 x 7 and 3 x 3; the pooling layer adopts a maximum pooling method to carry out maximum sampling on the previous layer, and the size of a sliding window is 2 multiplied by 2; the fully-connected layer maps the feature representations of the convolutional and pooling layers to marker layers of the data sample; the network calculates the inner product of the output vector of the previous layer and the connection weight vector, then adds the offset, and obtains an output state of the whole network after the operation of the activation function; the activation function is formulated as follows:
where J × I is the size dimensions of the width and height of the convolution kernel, M × N is the size of the width and height of the input image, xm,nRepresenting the pixel value, y, at the (m, n) position in the input picturem′,n′Representing the corresponding calculation result; w is a weight and represents the influence of the corresponding feature x; f is an activation function, and nonlinear transformation is carried out on the inner product; b is an offset which serves to select the dividing line;
(3b3) the first layer and the second layer are convolution layers with 32 convolution kernels, and the sizes of the convolution kernels are 7 multiplied by 7 and 3 multiplied by 3 respectively; the third layer is the maximum pooling layer of the sliding window size 2 × 2, and a Dropout value of 0.25 is set thereto; the fourth layer and the fifth layer are full connection layers, 903-dimensional information is output, and a classification probability result is obtained; and a Dropout value of 0.5 is set behind the fourth fully connected layer.
9. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 7, wherein in step (3c), the training method of the convolutional neural network is as follows:
(3c1) inputting a training set of a turnout fault sample set into a convolutional neural network, wherein the size of a training set sample is 150 multiplied by 150; in the convolutional neural network, firstly, a convolutional layer performs convolutional operation and activation operation on an input layer so as to extract behavior space characteristics of the input layer; the activation function adopts a RELU function, and the formula is as follows:
f(x)＝max(0,x)
wherein x is an input vector; the RELU function performs unilateral suppression on the input vector;
(3c2) performing pooling operation on the behavior space characteristics of the convolutional layer through a pooling layer, wherein the pooling operation is used for compressing the quantity of data and parameters;
(3c3) the convolutional neural network passes through two convolutional layers and a pooling layer until reaching a full connection layer, the full connection layer performs full connection operation on input data with the size of 32 x 71, and an output vector with the dimension of 903 is generated through a RELU activation function; calculating the output vector of the full connection layer by a softmax function to obtain a final predicted value, calculating a loss function value of the predicted value and a real value by using a cross entropy loss function, and minimizing the loss function value;
(3c4) and continuously adjusting the network weight and bias by a random gradient descent method, and recalculating the loss function value until the loss function value tends to be stable or reaches a set iteration number, thereby obtaining the classified picture characteristics.
10. The switch fault diagnosis method based on deep learning without field fault samples as claimed in claim 7, wherein in step (3d), the verification process is as follows:
(3d1) inputting the verification set image into a turnout fault diagnosis model;
(3d2) the output of the model is a probability array, and each element value in the probability array respectively represents the probability of the corresponding turnout fault; the calculation formula of the recognition result is as follows:
Result＝argmax(Ri),i＝1,2,3,…
where i is the number of switch fault samples, and RiIs the probability array of each image, and the argmax function is to find the label corresponding to the maximum probability in the probability array and take the label as the switch fault identification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010508456.4A CN111914320B (en) | 2020-06-06 | 2020-06-06 | Sample-free turnout fault diagnosis method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010508456.4A CN111914320B (en) | 2020-06-06 | 2020-06-06 | Sample-free turnout fault diagnosis method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111914320A true CN111914320A (en) | 2020-11-10 |
| CN111914320B CN111914320B (en) | 2024-02-02 |
Family
ID=73237886
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010508456.4A Active CN111914320B (en) | 2020-06-06 | 2020-06-06 | Sample-free turnout fault diagnosis method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111914320B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112668715A (en) * | 2020-12-28 | 2021-04-16 | 卡斯柯信号有限公司 | Turnout switch machine abnormity diagnosis method and system based on machine learning |
| CN113139601A (en) * | 2021-04-25 | 2021-07-20 | 杭州电子科技大学 | Turnout fault detection method based on LSTM and attention mechanism |
| CN114137915A (en) * | 2021-11-18 | 2022-03-04 | 成都航天科工大数据研究院有限公司 | Fault diagnosis method for industrial equipment |
| CN114630352A (en) * | 2020-12-11 | 2022-06-14 | ***通信集团湖南有限公司 | Fault monitoring method and device of access equipment |
| CN114692969A (en) * | 2022-03-29 | 2022-07-01 | 西门子交通技术(北京)有限公司 | Point switch failure prediction method, device, electronic equipment and storage medium |
| CN114936583A (en) * | 2022-06-14 | 2022-08-23 | 中国科学技术大学 | Teacher-student model-based two-step field self-adaptive cross-user electromyogram pattern recognition method |
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 |
| CN106976468A (en) * | 2017-03-09 | 2017-07-25 | 南京理工大学 | A kind of switch breakdown diagnostic method based on DWT and C SVM |
| GB201902157D0 (en) * | 2019-02-15 | 2019-04-03 | Thales Holdings Uk Plc | Diognostic system and a method of diagnosing faults |
| CN109934335A (en) * | 2019-03-05 | 2019-06-25 | 清华大学 | High-speed railway track switch method for diagnosing faults based on interacting depth study |
| CN110596490A (en) * | 2019-09-11 | 2019-12-20 | 天津科技大学 | Intelligent detection method for railway turnout fault |
-
2020
- 2020-06-06 CN CN202010508456.4A patent/CN111914320B/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 |
| CN106976468A (en) * | 2017-03-09 | 2017-07-25 | 南京理工大学 | A kind of switch breakdown diagnostic method based on DWT and C SVM |
| GB201902157D0 (en) * | 2019-02-15 | 2019-04-03 | Thales Holdings Uk Plc | Diognostic system and a method of diagnosing faults |
| CN109934335A (en) * | 2019-03-05 | 2019-06-25 | 清华大学 | High-speed railway track switch method for diagnosing faults based on interacting depth study |
| CN110596490A (en) * | 2019-09-11 | 2019-12-20 | 天津科技大学 | Intelligent detection method for railway turnout fault |
Non-Patent Citations (1)
| Title |
|---|
| 许庆阳;刘中田;赵会兵;: "基于隐马尔科夫模型的道岔故障诊断方法", 铁道学报, no. 08 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114630352A (en) * | 2020-12-11 | 2022-06-14 | ***通信集团湖南有限公司 | Fault monitoring method and device of access equipment |
| CN114630352B (en) * | 2020-12-11 | 2023-08-15 | ***通信集团湖南有限公司 | Fault monitoring method and device for access equipment |
| CN112668715A (en) * | 2020-12-28 | 2021-04-16 | 卡斯柯信号有限公司 | Turnout switch machine abnormity diagnosis method and system based on machine learning |
| CN113139601A (en) * | 2021-04-25 | 2021-07-20 | 杭州电子科技大学 | Turnout fault detection method based on LSTM and attention mechanism |
| CN114137915A (en) * | 2021-11-18 | 2022-03-04 | 成都航天科工大数据研究院有限公司 | Fault diagnosis method for industrial equipment |
| CN114692969A (en) * | 2022-03-29 | 2022-07-01 | 西门子交通技术(北京)有限公司 | Point switch failure prediction method, device, electronic equipment and storage medium |
| CN114692969B (en) * | 2022-03-29 | 2023-05-12 | 西门子交通技术(北京)有限公司 | Switch machine fault prediction method, device, electronic equipment and storage medium |
| CN114936583A (en) * | 2022-06-14 | 2022-08-23 | 中国科学技术大学 | Teacher-student model-based two-step field self-adaptive cross-user electromyogram pattern recognition method |
| CN114936583B (en) * | 2022-06-14 | 2024-03-29 | 中国科学技术大学 | Dual-step field self-adaptive cross-user myoelectricity mode identification method based on teacher-student model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111914320B (en) | 2024-02-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111914320A (en) | No-sample turnout fault diagnosis method based on deep learning | |
| US11402438B2 (en) | Intelligent on-line diagnosis and positioning method for winding deformation of power transformers | |
| CN110376522B (en) | Motor fault diagnosis method of data fusion deep learning network | |
| Bessam et al. | Wavelet transform and neural network techniques for inter-turn short circuit diagnosis and location in induction motor | |
| CN112307950B (en) | Detail feature quantity extraction and intelligent analysis method for GIS vibration defect identification | |
| CN109444667B (en) | Power distribution network early fault classification method and device based on convolutional neural network | |
| CN107632258A (en) | A kind of fan converter method for diagnosing faults based on wavelet transformation and DBN | |
| CN113486868B (en) | Motor fault diagnosis method and system | |
| CN103760464A (en) | Small current grounding system fault line selecting method based on analytic graph solving and SVM | |
| CN112069930A (en) | Vibration signal processing method and device for improving GIS equipment fault diagnosis accuracy | |
| CN116342952B (en) | Transformer bushing abnormality identification method and system | |
| Júnior et al. | The design of multiple linear regression models using a genetic algorithm to diagnose initial short-circuit faults in 3-phase induction motors | |
| CN112816881B (en) | Battery pressure difference abnormality detection method, device and computer storage medium | |
| US11817809B2 (en) | Motor driving system converter fault diagnosis method based on adaptive sparse filtering | |
| Husari et al. | Stator turn fault diagnosis and severity assessment in converter-fed induction motor using flat diagnosis structure based on deep learning approach | |
| CN110045207A (en) | A kind of complex fault diagnostic method based on power grid architecture and multisource data fusion | |
| CN113093058A (en) | NPC three-level inverter open-circuit fault diagnosis method | |
| CN107037278A (en) | A kind of substandard intelligent substation method for diagnosing faults of IEC61850 | |
| Hussain et al. | Fault Detection and Identification Using Deep Learning Algorithms in InductionMotors. | |
| CN106019173A (en) | Real-time fault detection method applied to voltage source converter | |
| CN113610119B (en) | Method for identifying power transmission line development faults based on convolutional neural network | |
| Azeem et al. | Real-time harmonics analysis of digital substation equipment based on IEC-61850 using hybrid intelligent approach | |
| CN111060844A (en) | Interturn short-circuit fault diagnosis method and device for high-speed train traction transmission system | |
| CN109030976B (en) | Power electronic equipment monitoring and diagnosing system and method | |
| CN115311509A (en) | Power system transient stability evaluation method and system based on imaging data driving |
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 |