CN114997218A - Recognition and detection method for polygonal abrasion of wheels of railway vehicle - Google Patents
Recognition and detection method for polygonal abrasion of wheels of railway vehicle Download PDFInfo
- Publication number
- CN114997218A CN114997218A CN202210556453.7A CN202210556453A CN114997218A CN 114997218 A CN114997218 A CN 114997218A CN 202210556453 A CN202210556453 A CN 202210556453A CN 114997218 A CN114997218 A CN 114997218A
- Authority
- CN
- China
- Prior art keywords
- polygonal
- abrasion
- model
- wave depth
- 1dcnn
- 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.)
- Pending
Links
- 238000005299 abrasion Methods 0.000 title claims abstract description 62
- 238000001514 detection method Methods 0.000 title abstract description 29
- 230000001133 acceleration Effects 0.000 claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000005457 optimization Methods 0.000 claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 10
- 239000002245 particle Substances 0.000 claims abstract description 9
- 238000012706 support-vector machine Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 14
- 230000004044 response Effects 0.000 claims description 12
- 230000008878 coupling Effects 0.000 claims description 8
- 238000010168 coupling process Methods 0.000 claims description 8
- 238000005859 coupling reaction Methods 0.000 claims description 8
- 238000011176 pooling Methods 0.000 claims description 8
- 238000012360 testing method Methods 0.000 claims description 7
- 238000013507 mapping Methods 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 7
- 238000012544 monitoring process Methods 0.000 abstract description 6
- 238000000605 extraction Methods 0.000 abstract description 5
- 238000010801 machine learning Methods 0.000 abstract description 4
- 238000013145 classification model Methods 0.000 abstract description 3
- 238000003672 processing method Methods 0.000 abstract description 3
- 238000004088 simulation Methods 0.000 description 6
- 229910000831 Steel Inorganic materials 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 239000010959 steel Substances 0.000 description 3
- 238000012800 visualization Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000003050 experimental design method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- 230000003068 static effect Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000013016 damping Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000008846 dynamic interplay Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61K—AUXILIARY EQUIPMENT SPECIALLY ADAPTED FOR RAILWAYS, NOT OTHERWISE PROVIDED FOR
- B61K9/00—Railway vehicle profile gauges; Detecting or indicating overheating of components; Apparatus on locomotives or cars to indicate bad track sections; General design of track recording vehicles
- B61K9/12—Measuring or surveying wheel-rims
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Mechanical Engineering (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
The invention discloses a recognition and detection method for polygonal abrasion of wheels of a railway vehicle, which comprises the following steps: s1, constructing a polygonal wear classification data set; s2, constructing and training a 1DCNN-SVM model; s3, polygonal abrasion classification and identification; s4, establishing a polygonal abrasion wave depth identification sample set; s5, constructing a KSM-PSO model and identifying the polygonal abrasion wave depth. The method omits the complex characteristic extraction process of the traditional signal processing and machine learning method, adopts the one-dimensional convolutional neural network to adaptively extract the characteristics of the axle box vertical acceleration signal, adopts the 1DCNN-SVM classification model to achieve the recognition rate of 99.82%, utilizes the Kergin agent model KSM and the particle swarm optimization algorithm PSO to quantitatively recognize the polygonal abrasion wave depth, has the error of not more than 2 percent, has the average relative error of only 0.50 percent, has the average time consumption of only 0.11s for recognizing single wave depth samples, meets the timeliness of online monitoring, and provides a new solution for recognizing the polygonal abrasion of the wheels of the rail vehicles/metro vehicles.
Description
Technical Field
The invention relates to the technical field of wheel wear identification and detection, in particular to an identification and detection method for polygonal wear of a wheel of a railway vehicle.
Background
The wheel polygon is characterized in that the adjacent wave crests are approximately equally spaced, and the abrasion of the wheel circumference is obviously wavy. The dynamic interaction of wheel and rail is more severe due to the polygonal shape of the wheels, the vibration of vehicle track parts such as wheel sets, axle boxes, steel rails and the like is intensified, and the long-term periodic wheel and rail vibration can cause fatigue failure (such as bolt loosening, bearing damage, steel rail fastener fracture and the like) and larger noise of the vehicle track parts, threaten the safe operation of trains and influence the riding comfort of passengers. As a result, subway operators often use wheel turnarounds to slow and eliminate polygon wear to reduce the adverse effects it brings. However, due to the lack of efficient wheel polygon detection means, the state of the wheel cannot be detected timely, accurately and quantitatively, the wheel maintenance is difficult to a certain degree, and the wheel maintenance cost is high.
At present, a detection method for polygonal abrasion of a wheel mainly comprises static detection and dynamic detection, wherein the static detection is to measure the wheel when a vehicle is not operated, the detection result is accurate but consumes much time, and the detection precision is also influenced by an operator; the dynamic detection method is used for detecting out-of-round of wheels by utilizing information such as vibration acceleration, displacement, wheel-rail interaction force and the like generated by running vehicles. The method can realize the online monitoring of the vehicle operation, and has the advantages of high detection speed and high practicability. The wheel rail force detection method and the steel rail vibration detection method in dynamic detection are characterized in that a detection device is arranged on a rail line, when a vehicle passes through, corresponding data can be obtained for judgment, and the defect is that the whole-process monitoring cannot be realized. The detection device can realize uninterrupted monitoring when being a vehicle-mounted sensor, the axle box vibration acceleration detection method is a common means for vehicle-mounted detection, and the detection device and the means are mature, low in cost and easy to maintain. In addition, there are also sound detection methods and optical detection techniques, but the application is less because of the technical difficulty or poor detection effect. For example, acoustic detection methods are susceptible to noise signal interference from adjacent wheel pairs; the optical detection method has high technical difficulty and high cost. Vibration signals generated in the running process of the vehicle usually adopt a time-frequency analysis method to extract corresponding characteristics of the vibration signals, and then training and classification are carried out through a machine learning method, so that the purpose of identifying wheel out-of-roundness is achieved. The traditional detection method has the problems of more time consumption, low universality, influence on train operation, high technical difficulty, complex operation, incapability of realizing the whole-process monitoring of vehicle operation and the like. The extraction of the signal features requires higher signal processing experience and needs rich prior knowledge, the process from signal processing to recognition is complex, and the recognition accuracy needs to be improved.
Therefore, it is desirable to provide a method for identifying and detecting polygonal wear of a wheel of a railway vehicle, so as to provide a technical means for wheel maintenance better.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a method for identifying and detecting the polygonal abrasion of the wheels of the railway vehicle, a one-dimensional convolution neural network is adopted to adaptively extract the characteristics of vertical acceleration signals of an axle box, when various wheel out-of-round forms exist, the classification identification precision can reach 99.82%, the average time consumption of identification of a single polygonal abrasion wave depth sample is only 0.11s, and the problems in the background art are solved.
In order to achieve the purpose, the invention provides the following technical scheme: a recognition and detection method for polygonal abrasion of a wheel of a railway vehicle comprises the following steps:
s1, constructing a polygonal abrasion classification data set;
s2, constructing and training a 1DCNN-SVM model;
s3, polygonal abrasion classification and identification;
s4, establishing a polygonal abrasion wave depth identification sample set;
s5, constructing a KSM-PSO model and identifying the polygonal abrasion wave depth.
Preferably, the constructing of the polygonal wear classification data set in step S1 specifically includes: a vertical vibration acceleration signal of an axle box is obtained by adopting a vehicle-mounted axle box acceleration sensor, the vertical acceleration of the axle box at the same speed is divided into M x N data set matrixes according to the data length to serve as polygonal abrasion classification data sets, and the polygonal abrasion classification data sets are divided into training sets and testing sets.
Preferably, the 1DCNN-SVM model of step S2 adopts a one-dimensional convolutional neural network 1DCNN to extract features of the input sample, and adopts a support vector machine SVM as a classifier;
the one-dimensional convolutional neural network 1DCNN comprises an input layer, 3 convolutional layers, 3 pooling layers and a full-connection layer; the number of convolution kernels of the three convolution layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 multiplied by 3, and the traversal step length is also all 1 multiplied by 2; the number of convolution kernels of the three pooling layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 multiplied by 2, and the step lengths are all 1 multiplied by 2.
Preferably, the training of the 1DCNN-SVM model specifically comprises: inputting training set data into 1DCNN to start training, introducing a Dropout layer into the full-connection layer, setting Dropout to be 0.5, selecting an Adam optimizer by the optimizer, wherein the learning rate is 0.001, the loss function is a cross entropy function, the size of a batch processing sample is 128, extracting output characteristics of the first full-connection layer as input characteristic vectors of the SVM after the model is trained, and finally obtaining a classification result.
Preferably, the support vector machine SVM is used as a classifier, when a classification task is performed, for m types of samples, one classifier is constructed for every two types of samples, m (m-1)/2 sub-classifiers are constructed together, when the class to which the sample belongs is predicted, each sub-classifier judges the sample and votes for the corresponding class, and when a decision is made, the class with a large number of votes is used as the class of the predicted sample.
Preferably, the polygonal wear classification identification is to label a normal wheel as "0", a low-order polygonal wheel as "1", and a high-order polygonal wheel as "2"; when the wheel polygon appears, the recognition result of the 1DCNN-SVM model is "1" or "2".
Preferably, the creating of the polygonal abrasion wave depth identification sample set comprises: and obtaining vertical acceleration RMS values of the axle box under different polygonal abrasion wave depths and vehicle speeds through tests or obtaining the vertical acceleration RMS values of the axle box under different working conditions by utilizing simulation of a vehicle-track coupling dynamic model to form a sample set for identifying the polygonal abrasion wave depths.
Preferably, the constructing of the KSM-PSO model and the polygonal abrasion wave depth recognition specifically include the steps of:
s51, according to the polygonal abrasion wave depth identification sample set established in the step S4, establishing a mapping relation among the polygonal abrasion wave depth, the vehicle speed and the axle box vertical acceleration RMS value by using a Krigin proxy model KSM to form a KSM response surface;
s52, substituting the vehicle speed and the axle box vertical acceleration RMS as a group of coordinates into the KSM response surface;
s53, performing wave depth solving by adopting a Particle Swarm Optimization (PSO);
and S54, completing the polygon abrasion wave depth recognition.
Preferably, the objective function of the wave depth solution in step S53 is as follows:
wherein H j In order to be a weighting parameter, the weighting parameter,
the jth component of the corresponding predicted value in the KSM for the sample point; y is j For the actual response component, n is the number of components.
The invention has the beneficial effects that: the method omits the complex characteristic extraction process of the traditional signal processing and machine learning method, adopts the one-dimensional convolutional neural network to adaptively extract the characteristics of the axle box vertical acceleration signal, the 1DCNN-SVM classification model can reach 99.82% of recognition rate, and utilizes a kriging proxy model and a particle swarm optimization algorithm to quantitatively recognize the polygonal abrasion wave depth, the error is not more than 2%, the average relative error is only 0.50%, the average time consumption of single wave depth sample recognition is only 0.11s, the timeliness of online monitoring is met, and a new solution is provided for the polygonal abrasion recognition of the wheels of the rail vehicles/metro vehicles.
Drawings
FIG. 1 is a flow chart of a rail vehicle wheel polygon identification;
FIG. 2 is a schematic diagram of axle box vertical acceleration signals for non-round wheels according to an embodiment, where FIG. 2(a) is normal, FIG. 2(b) is a low-order polygon, FIG. 2(c) is a high-order polygon, and FIG. 2(d) is random non-round;
FIG. 3 is a schematic diagram of a 1DCNN-SVM model;
FIG. 4 is a feature visualization and a 1DCNN-SVM confusion matrix map, FIG. 4(a) is a 1DCNN fully-connected layer feature visualization, and FIG. 4(b) is a 1DCNN-SVM confusion matrix map;
FIG. 5 is a schematic view of an example KSM response surface;
FIG. 6 is a graph showing the time consumption of an example PSO.
Detailed Description
The technical solutions in the embodiments of the present invention will be 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.
Convolutional Neural Network (CNN) was first applied to recognition of handwritten numbers, and is widely used in the fields of image processing, computer vision, fault diagnosis, and the like due to its "end-to-end" characteristic and strong adaptive feature extraction capability. The support vector machine is a classical binary classification algorithm based on a statistical theory, has a good classification effect under the condition of a small sample, and has good performance on the aspect of nonlinear classification.
The Kriging Surrogate Model (KSM) is an unbiased estimation Model that takes into account the spatial correlation of variables, and interpolates or regresses unknown sample information with an approximation function based on the correlation of known sample information based on the minimum variance criterion.
The Particle Swarm Optimization (PSO) has the advantages of less parameter adjustment, simple structure, high convergence speed, easiness in engineering realization and the like, and shows good performance when solving some classical function optimization problems. In view of this, a wheel polygon wear classification recognition model (1DCNN-SVM) based on a one-Dimensional Convolutional Neural Network (1-DCNN) and a Support Vector Machine (SVM), and a wheel polygon wear wave depth recognition model (KSM-PSO) based on a kriging agent model and a particle swarm optimization algorithm are proposed.
Example 1
Referring to fig. 1-6, the present invention provides a technical solution: a method for recognizing polygonal abrasion of wheels of a railway vehicle based on 1DCNN-SVM and KSM-PSO is disclosed, wherein the overall recognition process is shown in figure 1 and comprises the following steps: (1) constructing a polygonal abrasion classification data set; (2)1, constructing and training a DCNN-SVM model; (3) classifying and identifying polygonal abrasion; (4) establishing a polygonal abrasion wave depth identification sample set; (5) and constructing a KSM-PSO model and identifying the polygonal abrasion wave depth.
(1) Constructing a polygonal wear classification dataset
The axle box is directly connected with the wheel pair through the bearing, vibration transmitted from the wheel pair does not directly pass through a vibration damping system, vibration acceleration of the axle box can relatively directly reflect the characteristics of impact vibration of a wheel rail, and a vehicle-mounted axle box acceleration sensor can be used for obtaining a vertical vibration acceleration signal of the axle box. Fig. 2 is a typical wheel out-of-round form and its axle box vibration acceleration signal. The axle box vertical acceleration under a certain speed is divided into M x N data set matrixes according to a certain data length, such as 2048 data points, and the M x N data set matrixes are used as input sample sets, so that in order to facilitate the training and optimization of models, the input data sets are usually divided into training sets and testing sets in deep learning, and if p% in the sample sets is the training sets, the rest (1-p)% is used as the testing sets.
(2)1DCNN-SVM model construction and training
As shown in fig. 3, the 1DCNN-SVM framework of the present invention adopts 1DCNN to extract features of an input sample, and an SVM is used as a classifier. The one-dimensional convolutional neural network part comprises an input layer, three convolutional layers, three pooling layers and a full-connection layer. The reason that the number of convolution kernels of the first layer of convolution layers is large is to obtain the characteristics of signals at different angles, and the ReLU activation function is adopted as the activation function, so that the gradient disappearance phenomenon can be effectively avoided. The number of convolution kernels of the three layers of convolution layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 x 3, and the traversal step lengths are all 1 x 2. The number of convolution kernels of the three pooling layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 multiplied by 2, and the step lengths are all 1 multiplied by 2. The number of neurons in the first fully-connected layer was 100. The maximum pooling is selected by the pooling function, an SVM is adopted by the classification layer as a classifier, and the complete 1DCNN network structure parameters are shown in the table 1.
TABLE 11 DCNN model parameters
In order to accelerate the convergence of the model, a Dropout layer is introduced into the full-connection layer, the Dropout is set to be 0.5, an Adam optimizer is selected as the optimizer, the learning rate is 0.001, the loss function is a cross entropy function, and the size of a batch sample is 128. After the 1DCNN network model is trained, extracting output characteristics of a first full-connection layer as input characteristic vectors of the SVM, then training the SVM model, and finally obtaining a classification result.
When the SVM is used for multi-classification tasks, a one-to-one method is adopted, namely for m types of samples, a classifier is constructed for every two types of samples, m (m-1)/2 sub-classifiers are constructed together, when the types to which the samples belong are predicted, each sub-classifier judges the samples and votes for the corresponding types, and finally the types with more votes are used as the types of the predicted samples in decision making. The penalty factor of the support vector machine is determined by K-fold cross validation.
In addition, the 1DCNN-SVM model provided by the invention is established in a Keras deep learning library based on Python language. The PC hardware is configured as an i5-9300H processor 16GB memory Windows 10 system.
(3) Polygonal wear classification and identification
For the classification and identification of the polygonal wheel wear, a normal wheel label is defined as '0', a low-order polygonal wheel label is defined as '1', a high-order polygonal wheel label is defined as '2', and if other wheel out-of-round forms exist, the labels can be sequentially numbered according to the sequence numbers. Therefore, when the wheel polygon appears, the recognition result of the 1DCNN-SVM is "1" or "2".
(4) Establishing a polygonal abrasion wave depth identification sample set
After the polygon abrasion category is identified, the dominant polygon abrasion order can be determined according to the peak frequency by performing frequency spectrum analysis, envelope spectrum analysis and the like on the vertical acceleration signal of the shaft box. In order to realize the identification of the polygonal abrasion wave depth, a mapping relation among the polygonal abrasion wave depth, the vehicle speed and the axle box vertical acceleration RMS value needs to be established. In order to ensure that the sample points are uniformly filled in the design space, a uniform experimental design method is adopted to design and identify working conditions with different speeds and wave depths under the polygon orders, and under the condition of permission, axle box vertical acceleration RMS values under different polygon abrasion wave depths and vehicle speeds can be obtained through experiments. Considering that a certain cost exists in obtaining a sample set through a test method, a subway vehicle-track coupling dynamic model considering wheel set and track flexibility can be established by means of subway vehicle and track parameters. And (3) utilizing a subway vehicle-track coupling dynamic model to simulate to obtain axle box vertical acceleration RMS values under different working conditions, thereby forming a sample set for polygonal abrasion wave depth identification.
(5) Construction of KSM-PSO model and polygonal abrasion wave depth recognition
According to the constructed sample set for identifying the polygonal abrasion wave depth, a mapping relation among the polygonal abrasion wave depth, the vehicle speed and the axle box vertical acceleration RMS value is established by utilizing a Kergin proxy model KSM, so that a simulation model which is complex and time-consuming in calculation is replaced. The correlation function is chosen to be a commonly used gaussian model because it results in a relatively smooth and infinitely differentiable surface, and the KSM model regression part is set to a constant of 1. The RMS value and the vehicle speed of the vertical acceleration of the axle box are often known during wave depth identification, however, the KSM proxy model cannot directly determine the wave depth at the moment through the RMS value and the vehicle speed, so that the particle swarm optimization algorithm is introduced to solve the wave depth in the mapping relation constructed by the KSM. The wave depth solution optimization objective function is as follows:
wherein H j In order to weight the parameters of the system,
the jth component of the corresponding predicted value in the KSM for the sample point; y is j For the actual response component, n is the number of components.
And substituting the speed and the axle box vertical acceleration RMS value as a group of coordinates into a response surface constructed by the KSM, and solving the wave depth by adopting a particle swarm optimization algorithm according to the formula so as to identify the current polygonal abrasion wave depth.
Example 2
1. Description of the data
And establishing a subway vehicle-track rigid-flexible coupling dynamic model based on certain B-type subway vehicle parameters and track parameters, and verifying the simulation model by using the measured data. And acquiring axle box vertical acceleration data by using the vehicle-track coupling dynamic model.
2. Data set creation
The measured out-of-roundness of the metro wheels is used as the input of a metro vehicle-track coupling dynamic model, the track excitation is a U.S. quintuple spectrum, the sampling frequency is 5000Hz, and four types of samples, namely a normal wheel, a low-order polygonal wheel, a high-order polygonal wheel and a random non-circular wheel, are obtained. Typical wheel out-of-round forms and axle box accelerations are shown in FIG. 2, each type of sample has 960 pieces, each sample has length of 2048, a specific classification data set is shown in Table 2, 3840 samples are included in the data set, and a training set and a test set are split according to a ratio of 7: 3.
TABLE 2 Classification of data set constituents
The polygon wave depth data set takes a 7-order polygon as an example, in order to ensure that sample points uniformly fill a design space, a uniform experimental design method is adopted, 11 vehicle speeds (30, 35, …, 80km/h) and 10 wave depths (0.05, 0.1, … and 0.5mm) are subjected to simulation calculation by using an established subway vehicle-track coupling dynamic model, the track excitation is an American quintuple spectrum during simulation, and the sample condition obtained by the simulation calculation is shown in a table 3.
RMS value (m/s) of vertical acceleration of axle box at 37-step polygonal wear 2 )
3. Analysis of results
The punishment factor of the support vector machine is determined to be 0.75 through K-fold cross validation, and the classification accuracy of the 1DCNN-SVM is 99.82%. Feature visualization of the 1DCNN first layer fully-connected layer and the 1DCNN-SVM confusion matrix are shown in FIG. 4. It can be seen that the wrongly-divided samples are mainly classified into random noncircular classes by mistake, and the partial low-order and high-order components in the noncircular sequence of the low-order polygon are closer to the partial low-order and high-order components in the noncircular sequence of the random noncircular sequence by analyzing the order graphs of the noncircular sequence of the low-order polygon and the noncircular sequence of the random noncircular sequence, so that a small amount of low-order polygon abrasion samples are clustered into random noncircular classes.
The KSM response surface established according to table 3 is shown in fig. 5, and it can be seen from the graph that the response surface established by KSM is smoother, the sample points are all on the response surface, and the fitting degree is better; and randomly selecting 9 working conditions to verify the KSM model precision, wherein the verification results are shown in Table 4.
TABLE 4KSM model accuracy verification
The maximum relative error of the KSM prediction is 1.23%, the minimum relative error is 0.03%, and the average relative error is 0.49%.
The PSO algorithm is used to solve the wave depth, and the accuracy and time consumption of wave depth recognition are shown in table 5 and fig. 6, respectively.
TABLE 5 wave depth recognition accuracy
The maximum relative error of the PSO prediction is 1.33%, the minimum relative error is only 0.06%, and the average relative error is 0.50%. From the statistical results of fig. 6, the average time consumed for solving the PSO is 0.11s, and the accuracy and rapidity of wave depth identification are guaranteed when a large amount of data is faced.
The invention omits the complicated characteristic extraction process of the traditional signal processing and machine learning method, and the 1DCNN-SVM classification model can reach the recognition rate of 99.82%. And quantitatively identifying the polygonal abrasion wave depth by using a Kriging technology and a particle swarm optimization algorithm, wherein the error is not more than 2%, the average relative error is only 0.50%, and the average time consumption of single wave depth sample identification is only 0.11 s. And a new solution idea can be provided for recognizing the polygonal abrasion of the subway wheels.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes in the embodiments and/or modifications of the invention can be made, and equivalents and modifications of some features of the invention can be made without departing from the spirit and scope of the invention.
Claims (9)
1. A method for identifying and detecting polygonal abrasion of a wheel of a railway vehicle is characterized by comprising the following steps:
s1, constructing a polygonal wear classification data set;
s2, constructing and training a 1DCNN-SVM model;
s3, polygonal abrasion classification and identification;
s4, establishing a polygonal abrasion wave depth recognition sample set;
s5, constructing a KSM-PSO model and identifying the polygonal abrasion wave depth.
2. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: the step S1 of constructing the polygonal wear classification dataset specifically includes: a vertical vibration acceleration signal of an axle box is obtained by adopting a vehicle-mounted axle box acceleration sensor, the vertical acceleration of the axle box at the same speed is divided into M x N data set matrixes according to the data length to serve as polygonal abrasion classification data sets, and the polygonal abrasion classification data sets are divided into training sets and testing sets.
3. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: the 1DCNN-SVM model of step S2 adopts a one-dimensional convolutional neural network 1DCNN to extract the features of the input sample, and adopts a support vector machine SVM as a classifier;
the one-dimensional convolutional neural network 1DCNN comprises an input layer, 3 convolutional layers, 3 pooling layers and a full-connection layer; the number of convolution kernels of the three convolution layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 multiplied by 3, and the traversal step length is also all 1 multiplied by 2; the number of convolution kernels of the three pooling layers is 16, 8 and 4 respectively, the sizes of the convolution kernels are all 1 multiplied by 2, and the step lengths are all 1 multiplied by 2.
4. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: the 1DCNN-SVM model training specifically comprises the following steps: inputting training set data into 1DCNN to start training, introducing a Dropout layer into the full-connection layer, setting Dropout to be 0.5, selecting an Adam optimizer by the optimizer, wherein the learning rate is 0.001, the loss function is a cross entropy function, the size of a batch processing sample is 128, extracting output characteristics of the first full-connection layer as input characteristic vectors of the SVM after the model is trained, and finally obtaining a classification result.
5. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 3 or 4, wherein: the support vector machine SVM is used as a classifier, when classification tasks are carried out, for m types of samples, one classifier is constructed for every two types of samples, m (m-1)/2 sub-classifiers are constructed together, when the types to which the samples belong are predicted, each sub-classifier judges the samples and votes for the corresponding types, and finally the types with a large number of votes are used as the types of the predicted samples during decision making.
6. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: the polygonal abrasion classification identification is to label a normal wheel as '0', a low-order polygonal wheel as '1' and a high-order polygonal wheel as '2'; when the wheel polygon appears, the recognition result of the 1DCNN-SVM model is "1" or "2".
7. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: establishing a polygonal abrasion wave depth identification sample set: and obtaining vertical acceleration RMS values of the axle box under different polygonal abrasion wave depths and vehicle speeds through tests or obtaining the vertical acceleration RMS values of the axle box under different working conditions by utilizing a vehicle-track coupling dynamic model to simulate, thereby forming a sample set for identifying the polygonal abrasion wave depths.
8. The method for recognizing and detecting polygonal abrasion of the wheel of the railway vehicle according to claim 1, wherein: the construction of the KSM-PSO model and the polygonal abrasion wave depth identification specifically comprise the following steps:
s51, according to the polygonal abrasion wave depth identification sample set established in the step S4, a mapping relation among the polygonal abrasion wave depth, the vehicle speed and the axle box vertical acceleration RMS value is established by using a Kerriin surrogate model KSM, and a KSM response surface is formed;
s52, substituting the vehicle speed and the axle box vertical acceleration RMS value into a KSM response surface as a group of coordinates;
s53, performing wave depth solving by adopting a Particle Swarm Optimization (PSO);
and S54, completing the polygon abrasion wave depth recognition.
9. The method for identifying and detecting polygonal wear of a railway vehicle wheel according to claim 1, wherein the method comprises the following steps: the objective function of the wave depth solution performed in step S53 is as follows:
wherein H j In order to be a weighting parameter, the weighting parameter,
the jth component of the corresponding predicted value in the KSM for the sample point; y is j For the actual response component, n is the number of components.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210556453.7A CN114997218A (en) | 2022-05-20 | 2022-05-20 | Recognition and detection method for polygonal abrasion of wheels of railway vehicle |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210556453.7A CN114997218A (en) | 2022-05-20 | 2022-05-20 | Recognition and detection method for polygonal abrasion of wheels of railway vehicle |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114997218A true CN114997218A (en) | 2022-09-02 |
Family
ID=83027676
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210556453.7A Pending CN114997218A (en) | 2022-05-20 | 2022-05-20 | Recognition and detection method for polygonal abrasion of wheels of railway vehicle |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114997218A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115600086A (en) * | 2022-11-15 | 2023-01-13 | 西南交通大学（Cn） | Vehicle-mounted quantitative detection method for rail corrugation roughness based on convolution regression |
| CN115758289A (en) * | 2023-01-10 | 2023-03-07 | 西南交通大学 | Rail corrugation identification method based on multitask learning neural network |
| CN116952765A (en) * | 2023-09-19 | 2023-10-27 | 西南交通大学 | Method for inhibiting and quantitatively evaluating polygonal abrasion of wagon wheels |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006026446A2 (en) * | 2004-08-26 | 2006-03-09 | Symstream Technology Holdings Pty Ltd | Wave matrix mechanics method & apparatus |
| US20130288003A1 (en) * | 2011-01-13 | 2013-10-31 | Miba Frictec Gmbh | Friction material |
| CN105292177A (en) * | 2015-11-26 | 2016-02-03 | 唐智科技湖南发展有限公司 | Method for measuring track corrugation by utilizing axle box vibration and impact information |
| CN108593315A (en) * | 2018-04-25 | 2018-09-28 | 石家庄铁道大学 | The wheel polygon detecting method and terminal device of frequency domain character are vibrated based on axle box |
| WO2019185873A1 (en) * | 2018-03-29 | 2019-10-03 | Konux Gmbh | System and method for detecting and associating railway related data |
| CN111272550A (en) * | 2020-02-10 | 2020-06-12 | 西南交通大学 | Test method, test device, electronic equipment and storage medium |
| CN111428406A (en) * | 2020-03-23 | 2020-07-17 | 北京交通大学 | Method for estimating corrugation wavelength and corrugation depth of inner rail of small-radius curve of steel rail |
| CN112381027A (en) * | 2020-11-23 | 2021-02-19 | 西南交通大学 | Wheel polygon wave depth estimation method based on train axle box vertical acceleration signal |
| CN113276905A (en) * | 2021-04-30 | 2021-08-20 | 中铁二院工程集团有限责任公司 | Identification method and measurement method for distinguishing track corrugation and wheel polygon abrasion |
| CN113947130A (en) * | 2021-08-03 | 2022-01-18 | 西南交通大学 | Training and using method and device for regression prediction AI model of polygonal wear waveform of wheel |
-
2022
- 2022-05-20 CN CN202210556453.7A patent/CN114997218A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006026446A2 (en) * | 2004-08-26 | 2006-03-09 | Symstream Technology Holdings Pty Ltd | Wave matrix mechanics method & apparatus |
| US20130288003A1 (en) * | 2011-01-13 | 2013-10-31 | Miba Frictec Gmbh | Friction material |
| CN105292177A (en) * | 2015-11-26 | 2016-02-03 | 唐智科技湖南发展有限公司 | Method for measuring track corrugation by utilizing axle box vibration and impact information |
| WO2019185873A1 (en) * | 2018-03-29 | 2019-10-03 | Konux Gmbh | System and method for detecting and associating railway related data |
| CN108593315A (en) * | 2018-04-25 | 2018-09-28 | 石家庄铁道大学 | The wheel polygon detecting method and terminal device of frequency domain character are vibrated based on axle box |
| CN111272550A (en) * | 2020-02-10 | 2020-06-12 | 西南交通大学 | Test method, test device, electronic equipment and storage medium |
| CN111428406A (en) * | 2020-03-23 | 2020-07-17 | 北京交通大学 | Method for estimating corrugation wavelength and corrugation depth of inner rail of small-radius curve of steel rail |
| CN112381027A (en) * | 2020-11-23 | 2021-02-19 | 西南交通大学 | Wheel polygon wave depth estimation method based on train axle box vertical acceleration signal |
| CN113276905A (en) * | 2021-04-30 | 2021-08-20 | 中铁二院工程集团有限责任公司 | Identification method and measurement method for distinguishing track corrugation and wheel polygon abrasion |
| CN113947130A (en) * | 2021-08-03 | 2022-01-18 | 西南交通大学 | Training and using method and device for regression prediction AI model of polygonal wear waveform of wheel |
Non-Patent Citations (9)
| Title |
|---|
| YING SONG 等: "Railway Polygonized Wheel Detection Based on Numerical Time-Frequency Analysis of Axle-Box Acceleration", 《APPLIED SCIENCES》, vol. 10, no. 5, 28 February 2020 (2020-02-28), pages 1 - 18 * |
| 刘奇锋: "基于轴箱振动加速度的地铁车轮多边形辨识方法研究", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》, no. 2024, 15 February 2024 (2024-02-15), pages 033 - 497 * |
| 敖银辉 等: "基于非平衡数据的车辆轮对状态集成分类方法", 《西南交通大学学报》, vol. 52, no. 05, 15 September 2017 (2017-09-15), pages 852 - 858 * |
| 李欣: "基于支持向量机与最小二乘支持向量机的闸片识别及磨耗趋势预测", 《铁道技术监督》, vol. 46, no. 10, 15 October 2018 (2018-10-15), pages 29 - 32 * |
| 梁红琴 等: "基于轴箱垂向振动加速度的地铁车轮失圆状态诊断方法", 《中南大学学报(自然科学版)》, vol. 55, no. 01, 26 January 2024 (2024-01-26), pages 431 - 443 * |
| 蒋忠辉 等: "车辆轨道关键参数对高速铁路钢轨波磨发展的影响", 《机械工程学报》, vol. 54, no. 04, 20 February 2018 (2018-02-20), pages 57 - 63 * |
| 谢清林 等: "基于一维卷积神经网络的地铁钢轨波磨识别方法", 《中南大学学报(自然科学版)》, vol. 52, no. 04, 26 April 2021 (2021-04-26), pages 1371 - 1379 * |
| 金学松 等: "车轮非圆化磨耗问题研究进展", 《西南交通大学学报》, vol. 53, no. 01, 15 January 2018 (2018-01-15), pages 1 - 14 * |
| 陈伟 等: "高速列车车轮高阶多边形对车辆动力学性能的影响", 《铁道车辆》, vol. 52, no. 12, 10 December 2014 (2014-12-10), pages 4 - 9 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115600086A (en) * | 2022-11-15 | 2023-01-13 | 西南交通大学（Cn） | Vehicle-mounted quantitative detection method for rail corrugation roughness based on convolution regression |
| CN115758289A (en) * | 2023-01-10 | 2023-03-07 | 西南交通大学 | Rail corrugation identification method based on multitask learning neural network |
| CN115758289B (en) * | 2023-01-10 | 2023-04-25 | 西南交通大学 | Rail wave mill identification method based on multitask learning neural network |
| CN116952765A (en) * | 2023-09-19 | 2023-10-27 | 西南交通大学 | Method for inhibiting and quantitatively evaluating polygonal abrasion of wagon wheels |
| CN116952765B (en) * | 2023-09-19 | 2023-12-12 | 西南交通大学 | Method for inhibiting and quantitatively evaluating polygonal abrasion of wagon wheels |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pan et al. | LiftingNet: A novel deep learning network with layerwise feature learning from noisy mechanical data for fault classification | |
| CN114997218A (en) | Recognition and detection method for polygonal abrasion of wheels of railway vehicle | |
| Ravikumar et al. | Gearbox fault diagnosis based on Multi-Scale deep residual learning and stacked LSTM model | |
| CN110567720B (en) | Method for diagnosing depth confrontation of fault of fan bearing under unbalanced small sample scene | |
| CN110175434B (en) | Railway fastener system damage detection method based on convolutional neural network | |
| Yang et al. | Bearing fault automatic classification based on deep learning | |
| CN111037365B (en) | Cutter state monitoring data set enhancing method based on generative countermeasure network | |
| CN111623868B (en) | Convolutional neural network construction method for rail corrugation identification | |
| CN108984893A (en) | A kind of trend forecasting method based on gradient method for improving | |
| CN110297479B (en) | Hydroelectric generating set fault diagnosis method based on convolutional neural network information fusion | |
| CN106250613A (en) | A kind of wheel service state security domain is estimated and method for diagnosing faults | |
| CN106596149A (en) | Method for monitoring and diagnosing flat wheel state of rail vehicle based on VMD | |
| CN106021789A (en) | Fuzzy-intelligence-based rail car suspension system fault classification method and system | |
| CN105678343A (en) | Adaptive-weighted-group-sparse-representation-based diagnosis method for noise abnormity of hydroelectric generating set | |
| CN112329520B (en) | Truck bearing fault identification method based on generation countermeasure learning | |
| Cai et al. | EMD and GNN-AdaBoost fault diagnosis for urban rail train rolling bearings | |
| CN115358259A (en) | Self-learning-based unsupervised cross-working-condition bearing fault diagnosis method | |
| Chen et al. | Vibration-based damage detection of rail fastener using fully convolutional networks | |
| Li et al. | Fault diagnosis for rolling bearings of a freight train under limited fault data: Few-shot learning method | |
| Zhang et al. | High-accuracy and adaptive fault diagnosis of high-speed train bogie using dense-squeeze network | |
| Wu et al. | Quality estimation method for gear hobbing based on attention and adversarial transfer learning | |
| Liang et al. | Generalized composite multiscale diversity entropy and its application for fault diagnosis of rolling bearing in automotive production line | |
| CN116353660B (en) | High-speed railway wheel polygon fault detection method and system based on BWO-VMD | |
| CN116644278A (en) | Sample screening and feature extraction-based dual-model bearing anomaly detection algorithm | |
| CN110555457A (en) | Engine lubricating oil wear signal characteristic processing system and method |
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 |