CN113159179B - Subway and subway bogie running state identification method and system - Google Patents
Subway and subway bogie running state identification method and system Download PDFInfo
- Publication number
- CN113159179B CN113159179B CN202110436534.9A CN202110436534A CN113159179B CN 113159179 B CN113159179 B CN 113159179B CN 202110436534 A CN202110436534 A CN 202110436534A CN 113159179 B CN113159179 B CN 113159179B
- Authority
- CN
- China
- Prior art keywords
- layer
- neural network
- convolutional neural
- network model
- input
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/08—Railway vehicles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Probability & Statistics with Applications (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
The invention discloses a method and a system for identifying the running state of a subway and a subway bogie, which comprises the steps of collecting running data of the subway bogie by adopting a plurality of sensors, carrying out denoising processing on the running data, then carrying out multi-domain feature extraction on the running data, training a convolutional neural network model by using the extracted multi-domain features and the running state, and optimizing the parameters of the model by adopting a particle swarm optimization algorithm during training; the operation data is acquired by adopting multiple sensors, so that the data source is more reliable; the model is trained by the multi-domain characteristics, so that a more reliable recognition result can be obtained; and optimizing the model parameters and the classifier parameters by adopting a particle swarm optimization algorithm, so that the model is more reliable, and the identification result of the running state of the metro bogie has good reliability.
Description
Technical Field
The invention belongs to the technical field of subway bogie state identification, and particularly relates to a subway bogie running state identification method and system based on a multi-feature-domain convolutional neural network, and a subway.
Background
The subway is a beneficial supplement of urban rail transit. The bogie in the subway vehicle is positioned at the lowest part of the subway vehicle and between the vehicle body and the track, bears and transmits various loads from the vehicle body and the track, and simultaneously relieves the power action of the bogie. The metro bogie mainly comprises a framework (without a swing bolster and an H-shaped structure), a wheel set, a primary suspension, a secondary suspension, an anti-rolling torsion bar, a basic braking system, a central traction unit, a wheel rim lubricating system, an auxiliary device and other parts, is one of the most important parts of the metro vehicle, and has great significance for safe and reliable operation of the metro vehicle due to timely identification of the operation state.
Due to the complexity of the metro vehicle bogie structure and the uncertainty and nonlinearity of various influencing factors in the operation state, the conventional single characteristic domain state identification method is difficult to effectively identify the operation state of the metro bogie, and the reliability of the identification result is reduced.
The convolutional neural network is an algorithm comprising convolution calculation, and is a feedforward neural network with a deep structure, which is one of typical algorithms for deep learning. The convolutional neural network structure comprises an input layer, an output layer, a convolutional layer, a pooling layer and a full-connection layer. The convolutional neural network model realizes the updating of the weight parameters by minimizing a loss function based on a gradient descent method, and the gradient of the convolutional neural network model is reversely adjusted layer by layer from back to front until the first layer parameters of the network are updated. Compared with the traditional neural network framework, the convolutional neural network has the characteristics that the concepts of weight sharing and receptive field are introduced, so that the learning parameter quantity is greatly reduced, and the convolutional neural network has better learning capability; the convolutional neural network is more suitable for processing mass data, learning and extracting key information in the mass information, and realizing nonlinear complex feature extraction; local features of all data can be extracted, data dimensionality is reduced, feature integrity is guaranteed, and running state identification of equipment is achieved.
Disclosure of Invention
The invention aims to provide a method and a system for identifying the running state of a subway and a subway bogie, and aims to solve the problem that the running state of the subway bogie is difficult to effectively identify by using a traditional single characteristic domain state identification method, so that the reliability of an identification result is low.
The invention solves the technical problems through the following technical scheme: a method for identifying the running state of a metro bogie comprises the following steps:
acquiring operation data of a subway bogie;
denoising the operation data to obtain noiseless operation data;
sensitive feature extraction is carried out on the noiseless operation data, the sensitive features comprise time domain features, frequency domain features and time-frequency domain features, and sensitive feature vectors formed by the sensitive features are used as input feature vectors of a convolutional neural network model;
constructing a convolutional neural network model according to the sensitive characteristic vector and the running state of the metro bogie;
and training the convolutional neural network model by taking the input feature vector as an input training sample and the running state of the subway bogie as an output training sample, and optimizing parameters of the convolutional neural network model and parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
Furthermore, the operation data is acquired by four three-way acceleration sensors, and the four three-way acceleration sensors are respectively installed near wheel sets of the subway bogie.
And further, denoising the operating data by adopting a wavelet packet multi-threshold denoising method.
Further, the time domain features include peak-to-peak, mean, variance, slope, form factor, and peak factor values;
the frequency domain characteristic is a frequency variance.
Further, extracting the time-frequency domain features by adopting rapid complementary set empirical mode decomposition, wherein the specific implementation process is as follows:
performing fast complementary set empirical mode decomposition on the noiseless operation data to obtain a plurality of IMF components and a residual component;
and carrying out Fourier transform on each IMF component and calculating the total energy of a frequency domain, wherein the total energy is used as a time-frequency domain characteristic.
Further, the convolutional neural network model comprises an input layer, an output layer and at least one intermediate layer, wherein the intermediate layers are sequentially connected in series or the intermediate layers are connected in series and in parallel to form a combined layer;
the output end of the input layer is connected with the input end of the first intermediate layer or the combined layer, and the output end of the last intermediate layer or the combined layer is connected with the input end of the output layer.
The invention also provides a system for identifying the running state of the metro bogie, which comprises the following components:
the data acquisition unit is used for acquiring the operation data of the metro bogie;
the data preprocessing unit is used for denoising the operation data to obtain noiseless operation data;
the characteristic extraction unit is used for extracting sensitive characteristics of the noiseless operation data, wherein the sensitive characteristics comprise time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, and sensitive characteristic vectors formed by the sensitive characteristics are used as input characteristic vectors of a convolutional neural network model;
the model building unit is used for building a convolutional neural network model according to the sensitive characteristic vector and the running state of the subway bogie;
and the model training optimization unit is used for training the convolutional neural network model by taking the input characteristic vector as an input training sample and the running state of the subway bogie as an output training sample, and optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
Further, the data preprocessing unit is specifically configured to: and denoising the operating data by adopting a wavelet packet multi-threshold denoising method.
Further, the feature extraction unit is further specifically configured to:
and extracting the time-frequency domain characteristics by adopting rapid complementary set empirical mode decomposition, wherein the specific implementation process is as follows:
performing fast complementary set empirical mode decomposition on the noiseless operation data to obtain a plurality of IMF components and a residual component;
and carrying out Fourier transform on each IMF component and calculating the total energy of a frequency domain, wherein the total energy is used as a time-frequency domain characteristic.
The invention also provides a subway, which comprises the subway bogie running state identification system.
Advantageous effects
Compared with the prior art, the invention has the advantages that:
1. the operation data is acquired by adopting multiple sensors, so that the data source is more reliable;
2. training a convolutional neural network model by using a feature vector consisting of time domain features, frequency domain features and time-frequency domain features, so that a more reliable identification result can be obtained;
3. and optimizing the model parameters and the classifier parameters by adopting a particle swarm optimization algorithm, so that the model is more reliable, and the identification result of the running state of the subway bogie has good reliability.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only one embodiment of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
Fig. 1 is a flowchart of a method for identifying an operating state of a metro bogie according to an embodiment of the present invention;
FIG. 2 is a first embodiment of multiple intermediate layers connected in series and in parallel to form a composite layer in an example of the present invention;
FIG. 3 is a second embodiment of multiple intermediate layers connected in series and in parallel to form a composite layer in an example of the present invention;
FIG. 4 is a third embodiment of multiple intermediate layers connected in series and in parallel to form a composite layer in an example of the present invention;
FIG. 5 is a fourth embodiment of multiple intermediate layers connected in series and in parallel to form a composite layer in an example of the present invention;
FIG. 6 is a fifth embodiment of multiple intermediate layers connected in series and in parallel to form a composite layer in an example of the present invention.
Detailed Description
The technical solutions in the present invention are clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. 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.
The subway bogie mainly comprises a framework, a wheel pair assembly, a primary suspension, a secondary suspension, an anti-rolling torsion bar, a basic braking system, a central traction unit, a rim lubrication system assembly, an auxiliary device and other parts, and the conventional single-characteristic-domain state identification method is difficult to effectively identify the running state of the subway bogie due to the complexity of the structure of the subway bogie and the uncertainty and nonlinearity of various influencing factors in the running state, so that the reliability of an identification result is low.
Aiming at the structural complexity, nonlinearity and uncertainty in the operation state identification of the metro bogie, the invention provides a method and a system for identifying the operation state of the metro bogie based on a multi-feature-domain convolutional neural network.
The technical solution of the present application will be described in detail below with specific examples. The following several specific embodiments may be combined with each other, and details of the same or similar concepts or processes may not be repeated in some embodiments.
As shown in fig. 1, the method for identifying an operating state of a metro bogie provided in this embodiment includes the following steps:
1. and acquiring the operation data of the subway bogie.
According to the structure of the subway bogie, a subway bogie experiment platform is built, and the experiment platform comprises four three-way acceleration sensors, a data acquisition card and a PC (personal computer) besides the subway bogie; the method comprises the following steps that four three-way acceleration sensors are respectively arranged near wheel sets of a subway bogie, each three-way acceleration sensor corresponds to one wheel, each three-way acceleration sensor acquires operation data of the corresponding wheel and transmits the acquired operation data to a PC (personal computer) through a data acquisition card, and denoising processing, sensitive feature extraction, construction of a convolutional neural network model and training optimization of the model are all completed in the PC.
In this embodiment, three-way acceleration sensor chooses PCB triaxial acceleration vibration sensor 356A16 type for use, has the advantage of high accuracy, high sensitivity, and the sensor of high accuracy + multisensor makes the operation data source more reliable. The data acquisition card selects NI PXI-1042. The PC can also be replaced by a singlechip, a microprocessor, a DSP, a PLC and the like.
2. And denoising the operation data to obtain noiseless operation data.
In order to avoid the influence of noise on the operation data and further influence the recognition result, a wavelet packet multi-threshold denoising method is adopted to denoise the operation data. The wavelet packet multi-threshold denoising method is based on the fact that operation data and noise have different statistical characteristics after wavelet transformation, the energy of the operation data corresponds to a wavelet coefficient with a large amplitude and is mainly concentrated on high frequency, the noise energy corresponds to a wavelet coefficient with a small amplitude, a first threshold is set according to the characteristics, the main component of the wavelet coefficient larger than the first threshold is considered to be operation data, and the operation data is reserved after contraction; the main component of the wavelet coefficient smaller than the first threshold is considered as noise, and the noise is eliminated, so that the purpose of denoising is achieved. In denoising, the main components of the wavelet coefficients larger than the first threshold are generally considered to contain noise, and in order to improve the denoising effect, a second threshold is set, and the main components of the wavelet coefficients larger than the first threshold are subjected to wavelet decomposition and denoising again by adopting the second threshold. In this embodiment, a satisfactory denoising effect can be achieved by performing cubic wavelet decomposition and denoising.
3. And extracting sensitive features of the noiseless operation data, wherein the sensitive features comprise time domain features, frequency domain features and time-frequency domain features, and sensitive feature vectors formed by the sensitive features are used as input feature vectors of the convolutional neural network model.
And respectively drawing an acceleration time domain curve graph according to the running data in the X direction, the Y direction and the Z direction acquired by the four three-way acceleration sensors, and finding out the running data corresponding to the maximum amplitude in the acceleration time domain curve graph as a sensitive characteristic data set, thereby obtaining four sensitive characteristic data sets which are respectively marked as J1, J2, J3 and J4.
The time domain features include peak-to-peak, mean, variance, slope, form factor, and crest factor values, denoted as S1, S2, S3, S4, S5, S6, respectively. The frequency domain features are frequency variances, denoted as P1. The time domain features and the frequency features are extracted by the existing method.
In this embodiment, the time-frequency domain features are extracted by adopting a rapid complementary set empirical mode decomposition, and the specific implementation process is as follows:
performing fast complementary set empirical mode decomposition (FCEEMD) on the noiseless operation data to obtain a plurality of IMF components and a residual component;
and performing Fourier transform on each IMF component, and calculating the total energy of the frequency domain, wherein the total energy is marked as N, and the total energy N is used as a time-frequency domain characteristic.
Compared with the EMD, the FCEEMD has the advantages of inhibiting modal aliasing, improving the calculation efficiency and weakening the end point effect more effectively.
The sensitive feature vector composed of the time domain feature, the frequency domain feature and the time-frequency domain feature is marked as T, then
T = [ S1, S2, S3, S4, S5, S6, P1, N ], operation data acquired by each three-way acceleration sensor in each sampling period can obtain T through processing in steps 2 and 3, a plurality of T can be obtained by four three-way acceleration sensors in a plurality of sampling periods, the T form input samples in training samples, each T corresponds to one bogie operation state, the T correspond to a plurality of bogie operation states, and the bogie operation states form output samples in the training samples. When the convolutional neural network model is trained, a certain T is used as an input vector, the bogie running state corresponding to the T is used as output, and the model is trained by adopting a plurality of Ts and the bogie running state corresponding to the T.
In training, the operation state of the bogie is coded, the normal state is coded as (1, 0), and the fault state is coded as (0, 1).
4. And constructing a convolutional neural network model according to the sensitive characteristic vector and the running state of the subway bogie.
The sensitive characteristic vector T comprises 8 characteristic quantities, the running states of the bogie are two, namely normal and fault, so that the number of input layer nodes of the constructed convolutional neural network model is 8, and the number of output layer nodes is 2.
The convolutional neural network model comprises an input layer, an output layer and at least one intermediate layer, wherein the intermediate layers are sequentially connected in series or the intermediate layers are connected in series and in parallel to form a combined layer; the output end of the input layer is connected with the input end of the first intermediate layer or the combination layer, and the output end of the last intermediate layer or the combination layer is connected with the input end of the output layer.
Illustratively, assume that the intermediate layers include four, a first intermediate layer, a second intermediate layer, a third intermediate layer, and a fourth intermediate layer. The multiple intermediate layers are sequentially connected in series, namely the first intermediate layer, the second intermediate layer, the third intermediate layer and the fourth intermediate layer are sequentially connected in series, the output end of the input layer is connected with the input end of the first intermediate layer, and the output end of the fourth intermediate layer is connected with the input end of the output layer.
Specific embodiments of the multiple intermediate layers connected in series and in parallel to form the combined layer include, but are not limited to, the following:
the first embodiment: the first intermediate layer and the second intermediate layer are connected in parallel and then connected in series with the third intermediate layer and the fourth intermediate layer in sequence, as shown in fig. 2; the output end of the input layer is connected with the input ends of the first middle layer and the second middle layer, and the output end of the fourth middle layer is connected with the input end of the output layer.
Second embodiment: the second intermediate layer and the third intermediate layer are connected in parallel to form a combined layer, and the first intermediate layer, the combined layer and the fourth intermediate layer are sequentially connected in series as shown in fig. 3; the output end of the input layer is connected with the input end of the first middle layer, and the output end of the fourth middle layer is connected with the input end of the output layer.
The third embodiment: the third intermediate layer and the fourth intermediate layer are connected in parallel to form a combined layer, and the first intermediate layer, the second intermediate layer and the combined layer are sequentially connected in series as shown in fig. 4; the output end of the input layer is connected with the input end of the first middle layer, and the output end of the combined layer is connected with the input end of the output layer.
Fourth embodiment: the first middle layer and the second middle layer are connected in parallel to form a first combined layer, the third middle layer and the fourth middle layer are connected in parallel to form a second combined layer, and the first combined layer and the second combined layer are connected in series as shown in FIG. 5; the output of input layer is connected with the input of first combination layer, and the output of second combination layer is connected with the input of output layer.
Fifth embodiment: the first intermediate layer, the second intermediate layer, the third intermediate layer and the fourth intermediate layer are connected in parallel to form a combined layer, as shown in fig. 6; the output end of the input layer is connected with the input end of the combination layer, and the output end of the combination layer is connected with the input end of the output layer.
5. And training the convolutional neural network model by taking the input characteristic vector as an input training sample and the running state of the metro bogie as an output training sample, and optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
Inputting a certain T into a convolutional neural network model to obtain the output of the model, calculating a loss function and accuracy according to the output and the running state code of the subway bogie corresponding to the T, if the loss function and the accuracy do not meet the expected requirements, optimizing parameters of the convolutional neural network model and parameters of a softmax classifier by adopting a particle swarm optimization algorithm, and updating the weight and the offset of the model by adopting an Adam iterative update algorithm until the loss function and the accuracy meet the expected requirements; inputting the next T into the convolutional neural network model to obtain the output of the model, calculating a loss function and the accuracy according to the output and the running state code of the subway bogie corresponding to the T, if the loss function and the accuracy do not meet the expected requirements, optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm, and updating the weight and the offset of the model by adopting an Adam iterative update algorithm until the loss function and the accuracy meet the expected requirements; 823060, 8230; and by analogy, training of all the training samples on the convolutional neural network model is completed, and the trained convolutional neural network model is obtained.
The parameters of the model comprise a learning rate, an optimizer, iteration times, an activation function, an implicit neuron number, weight initialization, a Dropout method, regularization and the like. The particle swarm optimization algorithm is adopted to optimize the model parameters and the softmax classifier parameters, so that the training efficiency can be improved, and the recognition rate is improved when the training time is close; the algorithm accelerates error convergence, and can achieve or even exceed the recognition effect of other traditional methods by only needing a small number of iteration times.
During identification, the operation data acquired by the three-way acceleration sensor in real time is input into a trained convolutional neural network model after denoising and sensitive feature extraction, and then the operation state of the subway bogie can be identified.
The embodiment also provides a subway bogie running state identification system, includes:
and the data acquisition unit is used for acquiring the operation data of the subway bogie.
According to the structure of the subway bogie, a subway bogie experiment platform is built, and the experiment platform comprises four three-way acceleration sensors, a data acquisition card and a PC (personal computer) besides the subway bogie; the method comprises the steps that four three-way acceleration sensors are respectively arranged near wheel sets of a subway bogie, each three-way acceleration sensor corresponds to one wheel, each three-way acceleration sensor collects operation data of the corresponding wheel and transmits the collected operation data to a PC through a data collection card, denoising processing, sensitive feature extraction, convolutional neural network model construction and model training optimization are all completed in the PC, namely the PC comprises a data acquisition unit, a data preprocessing unit, a feature extraction unit, a model construction unit and a model training optimization unit.
In the embodiment, the three-direction acceleration sensor selects a PCB three-axis acceleration vibration sensor 356A16 type, so that the method has the advantages of high precision and high sensitivity, and the high-precision sensor and the multiple sensors make the source of the operation data more reliable. The data acquisition card selects NI PXI-1042. The PC can also be replaced by a singlechip, a microprocessor, a DSP, a PLC and the like.
And the data preprocessing unit is used for denoising the operation data to obtain noiseless operation data.
In order to avoid the influence of noise on the operation data and further influence the recognition result, a wavelet packet multi-threshold denoising method is adopted to denoise the operation data. The wavelet packet multi-threshold denoising method is based on that operation data and noise have different statistical characteristics after wavelet transformation, the energy of the operation data corresponds to a wavelet coefficient with a larger amplitude and is mainly concentrated at high frequency, the noise energy corresponds to a wavelet coefficient with a smaller amplitude, a first threshold is set according to the characteristic, the main component of the wavelet coefficient larger than the first threshold is considered as operation data, and the operation data is reserved after shrinkage; the main component of the wavelet coefficient smaller than the first threshold is considered as noise, and the noise is eliminated, so that the purpose of denoising is achieved. In denoising, the main components of the wavelet coefficients larger than the first threshold are generally considered to contain noise, and in order to improve the denoising effect, a second threshold is set, and the main components of the wavelet coefficients larger than the first threshold are subjected to wavelet decomposition and denoising again by adopting the second threshold. In this embodiment, a satisfactory denoising effect can be achieved by performing cubic wavelet decomposition and denoising.
And the characteristic extraction unit is used for extracting sensitive characteristics of the noiseless operation data, the sensitive characteristics comprise time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, and sensitive characteristic vectors formed by the sensitive characteristics are used as input characteristic vectors of a convolutional neural network model.
The method comprises the steps of respectively drawing an acceleration time domain curve graph according to running data in X, Y and Z directions acquired by four three-way acceleration sensors, finding out running data corresponding to the maximum amplitude in the acceleration time domain curve graph as a sensitive feature data set (for a single three-way acceleration sensor, the running data in the X, Y and Z directions can be acquired, and only the running data in one direction is taken as the sensitive feature data through the acceleration time domain curve graph), and accordingly acquiring four sensitive feature data sets which are respectively marked as J1, J2, J3 and J4.
The time domain features include peak-to-peak, mean, variance, slope, form factor, and crest factor values, denoted as S1, S2, S3, S4, S5, S6, respectively. The frequency domain features are frequency variances, denoted as P1.
In this embodiment, the time-frequency domain features are extracted by adopting a rapid complementary set empirical mode decomposition, and the specific implementation process is as follows:
performing fast complementary set empirical mode decomposition (FCEEMD) on the noiseless operation data to obtain a plurality of IMF components and a residual component;
and performing Fourier transform on each IMF component, and calculating the total energy of the frequency domain, wherein the total energy is marked as N, and the total energy N is used as a time-frequency domain characteristic.
Compared with the EMD, the FCEEMD has the advantages of inhibiting modal aliasing, improving the calculation efficiency and weakening the end point effect more effectively.
The sensitive feature vector composed of the time domain feature, the frequency domain feature and the time-frequency domain feature is marked as T, then
T = [ S1, S2, S3, S4, S5, S6, P1, N ], operation data acquired by each three-way acceleration sensor in each sampling period can obtain T through processing in steps 2 and 3, a plurality of T can be obtained by four three-way acceleration sensors in a plurality of sampling periods, the T form input samples in training samples, each T corresponds to one bogie operation state, the T correspond to a plurality of bogie operation states, and the bogie operation states form output samples in the training samples. When the convolutional neural network model is trained, a certain T is used as an input vector, the bogie running state corresponding to the T is used as output, and the model is trained by adopting a plurality of Ts and the bogie running state corresponding to the T.
In training, the operation state of the bogie is coded, the normal state is coded as (1, 0), and the fault state is coded as (0, 1).
And the model building unit is used for building a convolutional neural network model according to the sensitive characteristic vector and the running state of the subway bogie.
The sensitive characteristic vector T comprises 8 characteristic quantities, the running states of the bogie are two, namely normal and fault, so that the number of input layer nodes of the constructed convolutional neural network model is 8, and the number of output layer nodes is 2.
The convolutional neural network model comprises an input layer, an output layer and at least one intermediate layer, wherein the intermediate layers are sequentially connected in series or the intermediate layers are connected in series and in parallel to form a combined layer; the output end of the input layer is connected with the input end of the first intermediate layer or the combination layer, and the output end of the last intermediate layer or the combination layer is connected with the input end of the output layer.
Illustratively, assume that the intermediate layers include four, a first intermediate layer, a second intermediate layer, a third intermediate layer, and a fourth intermediate layer. The multiple intermediate layers are sequentially connected in series, namely the first intermediate layer, the second intermediate layer, the third intermediate layer and the fourth intermediate layer are sequentially connected in series, the output end of the input layer is connected with the input end of the first intermediate layer, and the output end of the fourth intermediate layer is connected with the input end of the output layer.
Specific embodiments of the multiple intermediate layers connected in series and in parallel to form the combined layer include, but are not limited to, the following:
the first embodiment: the first intermediate layer is connected with the second intermediate layer in parallel and then is sequentially connected with the third intermediate layer and the fourth intermediate layer in series, as shown in fig. 2; the output end of the input layer is connected with the input ends of the first middle layer and the second middle layer, and the output end of the fourth middle layer is connected with the input end of the output layer.
Second embodiment: the second intermediate layer and the third intermediate layer are connected in parallel to form a combined layer, and the first intermediate layer, the combined layer and the fourth intermediate layer are sequentially connected in series as shown in fig. 3; the output end of the input layer is connected with the input end of the first middle layer, and the output end of the fourth middle layer is connected with the input end of the output layer.
The third embodiment: the third intermediate layer and the fourth intermediate layer are connected in parallel to form a combined layer, and the first intermediate layer, the second intermediate layer and the combined layer are sequentially connected in series as shown in fig. 4; the output end of the input layer is connected with the input end of the first middle layer, and the output end of the combined layer is connected with the input end of the output layer.
Fourth embodiment: the first middle layer and the second middle layer are connected in parallel to form a first combined layer, the third middle layer and the fourth middle layer are connected in parallel to form a second combined layer, and the first combined layer and the second combined layer are connected in series as shown in FIG. 5; the output of input layer is connected with the input of first combination layer, and the output of second combination layer is connected with the input of output layer.
Fifth embodiment: the first intermediate layer, the second intermediate layer, the third intermediate layer and the fourth intermediate layer are connected in parallel to form a combined layer, as shown in fig. 6; the output end of the input layer is connected with the input end of the combination layer, and the output end of the combination layer is connected with the input end of the output layer.
And the model training optimization unit is used for training the convolutional neural network model by taking the input characteristic vector as an input training sample and taking the running state of the subway bogie as an output training sample, and optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
Inputting a certain T into a convolutional neural network model to obtain the output of the model, calculating a loss function and accuracy according to the output and the running state code of the subway bogie corresponding to the T, if the loss function and the accuracy do not meet the expected requirements, optimizing parameters of the convolutional neural network model and parameters of a softmax classifier by adopting a particle swarm optimization algorithm, and updating the weight and the offset of the model by adopting an Adam iterative update algorithm until the loss function and the accuracy meet the expected requirements; inputting the next T into the convolutional neural network model to obtain the output of the model, calculating a loss function and the accuracy according to the output and the running state code of the subway bogie corresponding to the T, if the loss function and the accuracy do not meet the expected requirements, optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm, and updating the weight and the offset of the model by adopting an Adam iterative update algorithm until the loss function and the accuracy meet the expected requirements; 823060, 8230; and analogizing in turn, completing the training of all the training samples on the convolutional neural network model, and obtaining the trained convolutional neural network model.
The parameters of the model include learning rate, optimizer, iteration number, activation function, hidden neuron number, weight initialization, dropout method, regularization and the like. The particle swarm optimization algorithm is adopted to optimize the model parameters and the softmax classifier parameters, so that the training efficiency can be improved, and the recognition rate is improved when the training time is close; the algorithm accelerates error convergence, and can achieve or even be superior to the recognition effect of other traditional methods only by a small number of iteration times.
During identification, the operation data acquired by the three-way acceleration sensor in real time is input into a trained convolutional neural network model after denoising and sensitive feature extraction, and then the operation state of the subway bogie can be identified.
The above disclosure is only for the specific embodiments of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive of changes or modifications within the technical scope of the present invention, and shall be covered by the scope of the present invention.
Claims (3)
1. A method for identifying the running state of a metro bogie is characterized by comprising the following steps:
acquiring operation data of a subway bogie; the running data is acquired by four three-way acceleration sensors which are respectively arranged near wheel sets of a subway bogie, and each three-way acceleration sensor corresponds to one wheel;
denoising the operation data by adopting a wavelet packet multi-threshold denoising method to obtain noiseless operation data;
sensitive feature extraction is carried out on the noiseless operation data, the sensitive features comprise time domain features, frequency domain features and time-frequency domain features, and sensitive feature vectors formed by the sensitive features are used as input feature vectors of a convolutional neural network model; and extracting the time-frequency domain characteristics by adopting rapid complementary set empirical mode decomposition, wherein the specific implementation process comprises the following steps: performing rapid complementary set empirical mode decomposition on the noise-free operation data to obtain a plurality of IMF components and a residual component; performing Fourier transform on each IMF component and calculating total energy of a frequency domain, wherein the total energy is used as a time-frequency domain characteristic; the time domain features include peak-to-peak, mean, variance, slope, form factor, and crest factor value; the frequency domain characteristic is a frequency variance;
constructing a convolutional neural network model according to the sensitive characteristic vector and the running state of the subway bogie; the convolutional neural network model comprises an input layer, an output layer and at least one intermediate layer, wherein the intermediate layers are connected in series and in parallel to form a combined layer; the output end of the input layer is connected with the input end of the combination layer, and the output end of the combination layer is connected with the input end of the output layer;
and training the convolutional neural network model by taking the input feature vector as an input training sample and the running state of the subway bogie as an output training sample, and optimizing parameters of the convolutional neural network model and parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
2. A subway bogie running state recognition system, comprising:
the data acquisition unit is used for acquiring the operation data of the subway bogie; the running data is acquired by four three-way acceleration sensors which are respectively arranged near wheel sets of a subway bogie, and each three-way acceleration sensor corresponds to one wheel;
the data preprocessing unit is used for denoising the operating data by adopting a wavelet packet multi-threshold denoising method to obtain noiseless operating data;
the characteristic extraction unit is used for extracting sensitive characteristics of the noiseless operation data, wherein the sensitive characteristics comprise time domain characteristics, frequency domain characteristics and time-frequency domain characteristics, and sensitive characteristic vectors formed by the sensitive characteristics are used as input characteristic vectors of a convolutional neural network model; and extracting the time-frequency domain characteristics by adopting rapid complementary set empirical mode decomposition, wherein the specific implementation process is as follows: performing fast complementary set empirical mode decomposition on the noiseless operation data to obtain a plurality of IMF components and a residual component; performing Fourier transform on each IMF component and calculating the total energy of a frequency domain, wherein the total energy is used as a time-frequency domain characteristic; the time domain features include peak-to-peak, mean, variance, slope, form factor, and crest factor value; the frequency domain characteristic is a frequency variance;
the model building unit is used for building a convolutional neural network model according to the sensitive characteristic vector and the running state of the subway bogie; the convolutional neural network model comprises an input layer, an output layer and at least one intermediate layer, wherein the intermediate layers are connected in series and in parallel to form a combined layer; the output end of the input layer is connected with the input end of the combination layer, and the output end of the combination layer is connected with the input end of the output layer;
and the model training optimization unit is used for training the convolutional neural network model by taking the input characteristic vector as an input training sample and the running state of the subway bogie as an output training sample, and optimizing the parameters of the convolutional neural network model and the parameters of the softmax classifier by adopting a particle swarm optimization algorithm during training to obtain the trained convolutional neural network model.
3. A subway comprising the subway bogie operation state recognition system as claimed in claim 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110436534.9A CN113159179B (en) | 2021-04-22 | 2021-04-22 | Subway and subway bogie running state identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110436534.9A CN113159179B (en) | 2021-04-22 | 2021-04-22 | Subway and subway bogie running state identification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113159179A CN113159179A (en) | 2021-07-23 |
| CN113159179B true CN113159179B (en) | 2023-04-18 |
Family
ID=76869543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110436534.9A Active CN113159179B (en) | 2021-04-22 | 2021-04-22 | Subway and subway bogie running state identification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113159179B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760839A (en) * | 2016-02-22 | 2016-07-13 | 重庆大学 | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine |
| CN110399816A (en) * | 2019-07-15 | 2019-11-01 | 广西大学 | A kind of high-speed train bottom foreign matter detecting method based on Faster R-CNN |
| WO2020156348A1 (en) * | 2019-01-31 | 2020-08-06 | 青岛理工大学 | Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002541448A (en) * | 1999-04-01 | 2002-12-03 | シーメンス シュヴァイツ アクチエンゲゼルシャフト | Method and apparatus for monitoring bogies of a multi-axle vehicle |
| CN105092467B (en) * | 2015-09-09 | 2018-05-18 | 南车工业研究院有限公司 | A kind of quick monitoring device of high-speed railway wheel tread flat and method |
| CN205373871U (en) * | 2016-01-22 | 2016-07-06 | 华东交通大学 | High speed train wheel space coupling vibration testing appearance |
| CN108600984B (en) * | 2018-04-18 | 2020-12-22 | 华东交通大学 | Remote monitoring device and method for running state of locomotive bogie |
| CN109597043B (en) * | 2018-11-16 | 2023-05-26 | 江苏科技大学 | Radar signal identification method based on quantum particle swarm convolutional neural network |
| CN110606105B (en) * | 2019-09-23 | 2020-08-25 | 西南交通大学 | Safety monitoring system for walking part of unmanned subway vehicle |
| CN111899905B (en) * | 2020-08-05 | 2022-11-01 | 哈尔滨工程大学 | Fault diagnosis method and system based on nuclear power device |
-
2021
- 2021-04-22 CN CN202110436534.9A patent/CN113159179B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760839A (en) * | 2016-02-22 | 2016-07-13 | 重庆大学 | Bearing fault diagnosis method based on multi-feature manifold learning and support vector machine |
| WO2020156348A1 (en) * | 2019-01-31 | 2020-08-06 | 青岛理工大学 | Structural damage identification method based on ensemble empirical mode decomposition and convolution neural network |
| CN110399816A (en) * | 2019-07-15 | 2019-11-01 | 广西大学 | A kind of high-speed train bottom foreign matter detecting method based on Faster R-CNN |
Non-Patent Citations (2)
| Title |
|---|
| 吴昀璞 ; 金炜东 ; 黄颖坤 ; .基于多域融合CNN的高速列车转向架故障检测.系统仿真学报.2018,(第11期),第4492-4497页. * |
| 翟国锐 ; 贺延芳 ; 魏秀琨 ; .城市轨道交通钢轨波浪形磨耗检测仿真研究.北京交通大学学报.2017,(第04期),第77-84、90页. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113159179A (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| Liu et al. | Multidimensional feature fusion and ensemble learning-based fault diagnosis for the braking system of heavy-haul train | |
| CN105973595A (en) | Diagnosis method of rolling bearing fault | |
| CN113687433B (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
| CN113704868B (en) | Fault diagnosis method based on multi-objective optimization flywheel energy storage system bearing | |
| CN111397901A (en) | Rolling bearing fault diagnosis method based on wavelet and improved PSO-RBF neural network | |
| CN113705396A (en) | Motor fault diagnosis method, system and equipment | |
| CN112381027B (en) | Wheel polygon wave depth estimation method based on train axle box vertical acceleration signal | |
| CN115758208A (en) | Traction converter fault diagnosis method and device, computer equipment and storage medium | |
| CN113203914A (en) | Underground cable early fault detection and identification method based on DAE-CNN | |
| CN112803398A (en) | Load prediction method and system based on empirical mode decomposition and deep neural network | |
| CN116933175A (en) | Electric automobile charging load prediction method and device | |
| Liang et al. | 1d convolutional neural networks for fault diagnosis of high-speed train bogie | |
| CN113159179B (en) | Subway and subway bogie running state identification method and system | |
| CN114386452A (en) | Method for detecting faults of sun wheel of nuclear power circulating water pump | |
| Wang et al. | Contrastive learning-based bayes-adaptive meta-reinforcement learning for active pantograph control in high-speed railways | |
| CN113392475A (en) | Intelligent fault identification method for speed reducer of industrial robot | |
| CN116467570B (en) | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving | |
| CN110569966B (en) | Data processing method and device and electronic equipment | |
| CN116205135A (en) | SO based on data decomposition and neural network 2 Emission prediction method and system | |
| Qi et al. | A new deep fusion network for automatic mechanical fault feature learning | |
| Zheng et al. | A novel deep learning architecture and its application in dynamic load monitoring of the vehicle system | |
| Qifeng et al. | Application of wavelet de-noising method in vibration signal analysis of elevator car | |
| CN112197983A (en) | Train service performance identification method | |
| Zhang et al. | Wavelet neural network-based short-term passenger flow forecasting on urban rail transit |
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 |