Machine learning-based wheel wear recognition method, device, equipment and medium
By acquiring the axle box vibration acceleration signal and vehicle speed signal of the vehicle for feature extraction and time-frequency analysis, using the coefficient of variation to identify the polygonal features of the wheel, and using a machine learning model for wear identification, the problem of low accuracy and reliance on manual labor in wheel polygonal wear detection is solved, and accurate quantitative wear identification is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHUZHOU ELECTRIC LOCOMOTIVE CO LTD
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-02
AI Technical Summary
Existing methods for detecting polygonal wear on wheels suffer from low accuracy in polygon order recognition and reliance on manual intervention.
By acquiring the axle box vibration acceleration signal and vehicle speed signal of the target vehicle, feature extraction and time-frequency analysis are performed. The coefficient of variation is used to identify the polygonal features of the wheel, and a machine learning model is used to identify wear, determine the dominant order and wear degree.
It achieves accurate existence determination of polygonal features of wheels and automatic quantitative identification of wear depth, overcoming the shortcomings of traditional methods that rely on manual labor and have low accuracy, and improving the identification accuracy.
Smart Images
Figure CN122132920A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wheel polygon fault recognition technology, and in particular to wheel wear recognition methods, devices, equipment and media based on machine learning. Background Technology
[0002] The polygonal shape of a rail vehicle wheel, where the wheel tread is circumferential, causes significant vibration and noise during vehicle operation, increases wheel-rail impact, and accelerates wear on the rails and wheels. This uneven wear not only reduces vehicle comfort and stability but can also damage rails or vehicle components, increasing maintenance costs.
[0003] Current methods for detecting polygonal wear on wheels involve simple vibration analysis of the vehicle. However, this method suffers from low accuracy in polygon order recognition and reliance on manual intervention. Summary of the Invention
[0004] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for wheel wear identification based on machine learning. This method can identify vehicle wear by judging stable frequency components based on the coefficient of variation and utilizing a machine learning model, overcoming the shortcomings of reliance on manual labor and low identification accuracy, and achieving quantitative wear identification. The specific solution is as follows:
[0005] Firstly, this application provides a machine learning-based method for identifying wheel wear, including:
[0006] The target axle box vibration acceleration signal and the target vehicle speed signal are acquired during the operation of the target vehicle, and feature extraction is performed on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data;
[0007] Peak retrieval is performed on the time-frequency feature data to obtain the corresponding frequency peaks, the amplitude data corresponding to each frequency peak along the time axis is determined, and the coefficient of variation of each set of amplitude data corresponding to each frequency peak is calculated.
[0008] Each of the aforementioned coefficients of variation is compared with a preset coefficient of variation threshold, and the presence of wheel polygon features in the time-frequency feature data is determined based on the comparison results. If wheel polygon features are present in the time-frequency feature data, the corresponding target polygon order is determined based on the vehicle speed signal corresponding to the wheel polygon features, and the dominant order in the target polygon order is determined. The wheel polygon features represent the stable frequency components in the time-frequency feature data caused by wheel polygon wear.
[0009] The target wear analysis model is used to analyze the dominant order and the corresponding target time-frequency characteristic data to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
[0010] Optionally, acquiring the target axle box vibration acceleration signal and the target vehicle speed signal during vehicle operation includes:
[0011] Acquire the initial axle box vibration acceleration signal and the initial vehicle speed signal during the operation of the target vehicle, and align the initial axle box vibration acceleration signal and the initial vehicle speed signal on the time axis;
[0012] The aligned initial axle box vibration acceleration signal and the initial vehicle speed signal are supplemented with missing values and cleaned of outliers to obtain the target axle box vibration acceleration signal and the target vehicle speed signal.
[0013] Optionally, the step of extracting features from the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data includes:
[0014] The target axle box vibration acceleration signal and the target vehicle speed signal are denoised to obtain corresponding denoised data. The denoised data is then decomposed using fully integrated empirical mode decomposition and adaptive noise technology to obtain several IMF components.
[0015] Extract the time-domain and frequency-domain feature parameters of each IMF component to obtain an initial feature set, and perform time-frequency transformation on the initial feature set to obtain the time-frequency feature data.
[0016] Optionally, determining the dominant order in the order of the target polygon includes:
[0017] Determine the amplitude data corresponding to each of the target polygon orders, and determine the polygon order with the largest amplitude data among the target polygon orders as the dominant order.
[0018] Optionally, before analyzing the dominant order and corresponding target time-frequency characteristic data using the target wear analysis model, the method further includes:
[0019] Using a pre-set simulation platform, simulated fault samples covering different polygon orders, wear depths, and operating speeds, along with corresponding axle box vibration acceleration signals, are generated to construct a full-condition fault sample set.
[0020] The target wear analysis model is determined from the initial wear analysis models based on the training effect of each initial wear analysis model using the full-condition fault sample set.
[0021] Optionally, after analyzing the dominant order and corresponding target time-frequency characteristic data using the target wear analysis model, the method further includes:
[0022] The wear warning level of the target vehicle is determined according to the preset wheel wear classification and the wear degree of the target vehicle's wheels, and the target vehicle is then subjected to corresponding maintenance treatment based on the wear warning level.
[0023] Optionally, the machine learning-based wheel wear recognition method further includes: if there are no wheel polygon features in the time-frequency feature data, then skip the process of wheel wear recognition for the target vehicle.
[0024] Secondly, this application provides a wheel wear recognition device based on machine learning, comprising:
[0025] The feature extraction module is used to acquire the target axle box vibration acceleration signal and the target vehicle speed signal when the target vehicle is running, and to extract features from the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data;
[0026] The coefficient of variation calculation module is used to perform peak retrieval on the time-frequency feature data to obtain the corresponding frequency peaks, determine the amplitude data corresponding to each frequency peak along the time axis, and calculate the coefficient of variation of each set of amplitude data corresponding to each frequency peak.
[0027] The dominant order determination module is used to compare each of the coefficients of variation with a preset coefficient of variation threshold, and determine whether there is a wheel polygon feature in the time-frequency feature data based on the comparison results. If there is a wheel polygon feature in the time-frequency feature data, the corresponding target polygon order is determined based on the vehicle speed signal corresponding to the wheel polygon feature, and the dominant order in the target polygon order is determined. The wheel polygon feature represents the stable frequency component in the time-frequency feature data caused by wheel polygon wear.
[0028] The data analysis module is used to analyze the dominant order and the corresponding target time-frequency characteristic data using a target wear analysis model to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
[0029] Thirdly, this application provides an electronic device, comprising:
[0030] Memory, used to store computer programs;
[0031] A processor is used to execute the computer program to implement the aforementioned machine learning-based wheel wear recognition method.
[0032] Fourthly, this application provides a computer-readable storage medium for storing a computer program that, when executed by a processor, implements the aforementioned machine learning-based wheel wear recognition method.
[0033] This application first acquires the target axle box vibration acceleration signal and the target vehicle speed signal during the operation of the target vehicle, and performs feature extraction on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data. Then, it performs peak retrieval on the time-frequency feature data to obtain the corresponding frequency peaks, determines the amplitude data corresponding to each frequency peak along the time axis, and calculates the coefficient of variation of each set of amplitude data corresponding to each frequency peak. Then, it compares each coefficient of variation with a preset coefficient of variation threshold, and determines whether there is a wheel polygon feature in the time-frequency feature data based on the comparison results. If there is a wheel polygon feature in the time-frequency feature data, it determines the corresponding target polygon order based on the vehicle speed signal corresponding to the wheel polygon feature, and determines the dominant order in the target polygon order. The wheel polygon feature represents the stable frequency component in the time-frequency feature data caused by wheel polygon wear. Finally, it uses a target wear analysis model to analyze the dominant order and the corresponding target time-frequency feature data to obtain the wheel wear degree of the target vehicle. The target wear analysis model is an artificial intelligence model obtained based on machine learning technology. Therefore, this application achieves accurate existence judgment and dominance order locking of wheel polygon features by using a stable frequency component discrimination method based on the coefficient of variation, thus solving the problem of ambiguous order identification in traditional methods. By inputting the dominant order and corresponding features into a machine learning model for analysis, it achieves automatic quantitative identification of wheel wear depth, overcoming the shortcomings of traditional methods that rely on manual labor and have low accuracy. Attached Figure Description
[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0035] Figure 1This is a schematic diagram of the process of a machine learning-based wheel wear recognition method disclosed in this application;
[0036] Figure 2 This application discloses a flowchart for wheel wear identification.
[0037] Figure 3 This is a schematic diagram of a vehicle speed signal disclosed in this application;
[0038] Figure 4 This is a schematic diagram of a shaft box vibration acceleration signal disclosed in this application;
[0039] Figure 5 This is a data preprocessing flowchart disclosed in this application;
[0040] Figure 6 This application discloses a data time-frequency transformation flowchart;
[0041] Figure 7 This is a schematic diagram of an IMF correlation coefficient disclosed in this application;
[0042] Figure 8 This is a schematic diagram of a frequency retrieval method disclosed in this application;
[0043] Figure 9 This application discloses a flowchart for order discrimination driven by the coefficient of variation.
[0044] Figure 10 This is a schematic diagram of a polygon order discrimination result disclosed in this application;
[0045] Figure 11 This application discloses a flowchart of a method for establishing a dynamic simulation platform.
[0046] Figure 12 This application discloses a model screening flowchart;
[0047] Figure 13 This is a schematic diagram of a model training result disclosed in this application;
[0048] Figure 14 This is a schematic diagram illustrating the construction of a fault sample as disclosed in this application;
[0049] Figure 15 This is a schematic diagram of a wear prediction result disclosed in this application;
[0050] Figure 16 This is a schematic diagram of a hardware system disclosed in this application;
[0051] Figure 17 This is a schematic diagram of the structure of a machine learning-based wheel wear recognition device disclosed in this application;
[0052] Figure 18 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation
[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0054] Current methods for detecting polygonal wear on wheels suffer from low accuracy in polygon order recognition and reliance on manual intervention. To address this, this application presents a machine learning-based wheel wear recognition method. This method identifies vehicle wear by using stable frequency components based on the coefficient of variation and employing a machine learning model, thus overcoming the drawbacks of manual intervention and low accuracy.
[0055] See Figure 1 As shown, this embodiment of the invention discloses a wheel wear recognition method based on machine learning, including:
[0056] Step S11: Obtain the target axle box vibration acceleration signal and the target vehicle speed signal when the target vehicle is running, and perform feature extraction on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain the corresponding time-frequency feature data.
[0057] The general process for wheel wear identification in this embodiment is as follows: Figure 2 As shown, the process includes steps such as data acquisition and transmission, data preprocessing, feature extraction, machine learning model optimization, and wear identification.
[0058] In this embodiment, the process of acquiring the target axle box vibration acceleration signal and the target vehicle speed signal during the operation of the target vehicle includes: acquiring the initial axle box vibration acceleration signal and the initial vehicle speed signal during the operation of the target vehicle, and aligning the initial axle box vibration acceleration signal and the initial vehicle speed signal with the time axis; and supplementing missing values and cleaning outliers in the aligned initial axle box vibration acceleration signal and the initial vehicle speed signal to obtain the target axle box vibration acceleration signal and the target vehicle speed signal.
[0059] Specifically, in this embodiment, a vibration acceleration sensor with a sampling frequency ≥1kHz is used to collect axle box vibration acceleration signal (i.e., initial axle box vibration acceleration signal), and a vehicle speed sensor is used to collect vehicle speed signal (i.e., initial vehicle speed signal). After processing by a pre-processor, the signals are synchronously transmitted to the onboard host for real-time identification; simultaneously, they are stored in the onboard TCMS (Train Control and Management System) and BeiDou positioning system via the vehicle network bus, and packaged and sent to the ground platform system via a 5G wireless transmission channel, where they are stored in a standard database in time-series format. During system deployment, the sensor installation positions and hardware configurations need to be adjusted according to the structural characteristics of different rail transit equipment to ensure stable system operation. The collected vehicle speed signal and vibration acceleration signal are respectively as follows: Figure 3 and Figure 4 As shown.
[0060] The process of data preprocessing is as follows Figure 5 As shown, the axle box vibration acceleration signal and vehicle speed signal are time-series aligned, and missing values are filled using linear or polynomial interpolation to maintain the original data characteristics and avoid distortion. Data cleaning is performed through duplicate data detection, outlier data detection, and logical error verification to remove outliers and duplicates. Focusing on the target analysis speed range, vibration data within the corresponding timestamp range is selected to ensure the relevance of the analysis. Specifically, firstly, the vibration acceleration signal and vehicle speed signal are time-series aligned to ensure accurate timestamp matching; missing values are filled using linear interpolation, and abnormal jump values are smoothed; duplicate and erroneous data are removed through duplicate data detection and logical error verification; based on the target analysis speed range (e.g., 60~160km / h), data segments with relatively flat speeds are selected, and vibration acceleration data at the corresponding timestamps are extracted to ensure the data duration is at least 10 seconds.
[0061] The missing value imputation mentioned above can be achieved by using polynomial interpolation instead of linear interpolation, and the noise reduction methods can be achieved by using low-pass filtering or wavelet denoising instead of band-pass filtering. During data preprocessing, the filtering parameters and feature selection criteria need to be adjusted according to the actual operating conditions to ensure that the preprocessing effect is suitable for different scenarios.
[0062] In addition, in this embodiment, feature extraction is performed on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data. This includes: denoising the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding denoised data, and decomposing the denoised data using fully integrated empirical mode decomposition and adaptive noise technology to obtain several IMF (Intrinsic Mode Function) components; extracting the time-domain feature parameters and frequency-domain feature parameters of each IMF component to obtain an initial feature set, and performing time-frequency transformation on the initial feature set to obtain time-frequency feature data.
[0063] That is, such as Figure 6 As shown, this embodiment uses low-pass or band-pass filtering to denoise the filtered data and remove external noise interference. Empirical Mode Decomposition (IMF) is used to decompose the denoised data into multiple IMF components. The correlation coefficient, energy entropy, fuzzy entropy, kurtosis, and root mean square value (i.e., time-domain and frequency-domain feature parameters) of each IMF component are extracted as feature parameters (i.e., time-frequency feature data). Principal components are selected using single signal features or composite feature values of multiple signal features. Short-time Fourier transform or wavelet transform is performed on the principal components to enhance effective signal features. Specifically, firstly, bandpass filtering is used to denoise the filtered vibration acceleration signal, preserving the frequency range where polygonal faults might occur; then, CEEMDAN empirical mode decomposition is performed on the filtered data to obtain multiple IMF components; the correlation coefficient between each IMF component and the original signal, as well as characteristic parameters such as energy entropy, fuzzy entropy, kurtosis, and root mean square value, are calculated; principal components are selected based on the correlation coefficient and the importance of the characteristic parameters; finally, short-time Fourier transform is performed on the principal components to obtain the time-frequency feature map. (Appendix) Figure 7 The graph shows the correlation coefficients and characteristic parameters extracted from each IMF component. Figure 8 The image shows the time-frequency transformation result of STFT (Short-Time Fourier Transform). The specific steps of CEEMDAN empirical mode decomposition are as follows:
[0064] (1) Adding white noise: Adding a segment of white noise n(t) to the original signal x(t) yields a new signal:
[0065] ;
[0066] (2) Perform EMD: For each noisy signal Perform EMD (Empirical Mode Decomposition) to extract the IMF:
[0067] ;
[0068] Where k is the number of IMFs extracted. It is a residual signal.
[0069] (3) Repeat the process: Repeat the above steps for signals with different noises to obtain multiple IMFs.
[0070] 4) Average IMF: The final IMF is obtained by averaging the IMFs obtained from all repeated experiments.
[0071] ;
[0072] The formulas for calculating the correlation coefficient and root mean square are as follows:
[0073] ;
[0074] ;
[0075] in: Indicates the overall covariance; and Let x and y be the standard deviations, respectively. By substituting each order of the IMF component into x and the original signal into y, the correlation coefficients between each order of the IMF component and the original signal can be obtained.
[0076] Given a continuous-time signal x(t), the calculation process of the Short-Time Fourier Transform (STFT) can be expressed as follows:
[0077] ;
[0078] in: This is the result of STFT, representing the spectral information at time t and frequency f; It is the original signal; These are window functions used to window signals; options include Hanning window, Hamming window, and other window functions. It is a complex exponential function, representing the frequency component in the Fourier transform.
[0079] The aforementioned empirical mode decomposition can be replaced by EMD or CEEMD (Complete Ensemble Empirical Mode Decomposition), and the time-frequency transformation can be replaced by wavelet transform or WVD (Wignerr-Ville distribution) transform. Furthermore, the order, wear depth, and operating speed coverage of the fault sample set in this embodiment can be adjusted according to the actual application scenario.
[0080] Step S12: Perform peak retrieval on the time-frequency feature data to obtain the corresponding frequency peaks, determine the amplitude data corresponding to each frequency peak along the time axis, and calculate the coefficient of variation of each set of amplitude data corresponding to each frequency peak.
[0081] This embodiment requires obtaining the STFT time-frequency plot of the principal components, retrieving global frequency peaks and ensuring that the peaks are staggered; extracting the amplitude data of each frequency peak along the time axis, and calculating the coefficient of variation (CV) of each amplitude group.
[0082] Step S13: Compare each of the coefficients of variation with a preset coefficient of variation threshold, and determine whether there is a wheel polygon feature in the time-frequency feature data based on the comparison results. If there is a wheel polygon feature in the time-frequency feature data, determine the corresponding target polygon order based on the vehicle speed signal corresponding to the wheel polygon feature, and determine the dominant order in the target polygon order. The wheel polygon feature represents the stable frequency component in the time-frequency feature data caused by wheel polygon wear.
[0083] In this embodiment, the process of determining the dominant order in the target polygon order includes: determining the amplitude data corresponding to each target polygon order, and determining the polygon order with the largest amplitude data in each target polygon order as the dominant order.
[0084] Specifically, the coefficient of variation (CV) of the amplitude data of each frequency peak along the time axis after time-frequency transformation is calculated. When there is a CV value < for example 0.3, it is determined to contain polygonal features of the wheel; when all CV values are > 0.3, it is determined not to contain polygonal features. For frequency bands containing polygonal features, the polygon order is calculated by combining frequency and vehicle speed, and the dominant order with the largest amplitude is identified.
[0085] It is understandable that if the time-frequency feature data does not contain wheel polygon features, the process of identifying wheel wear on the target vehicle will be skipped.
[0086] Specifically, according to the appendix Figure 9 The process shown involves obtaining the STFT time-frequency plot of the principal components, retrieving global frequency peaks and ensuring that the peaks are staggered; extracting the amplitude data of each frequency peak along the time axis, and calculating the coefficient of variation (CV) of each amplitude group; if there is a CV value < 0.3, it is determined that there are polygonal features, and the polygon order is calculated by combining frequency and vehicle speed to lock the dominant order. The specific steps are as follows: (1) First, obtain the STFT time-frequency diagram of the principal component of vibration acceleration; (2) Perform global frequency peak retrieval on the time-frequency diagram according to the magnitude of the time-frequency amplitude, and ensure that the frequency peaks are staggered; (3) Extract all amplitude data of each frequency peak along the time axis; (4) Calculate the coefficient of variation (CV) value of each group of amplitudes, and determine whether there is a continuous frequency band with a large amplitude based on the coefficient of variation (CV) value, thereby determining whether it contains polygonal features. If the coefficient of variation (CV) value of each group is greater than 0.3, it is considered that there are no polygonal features. If there is a coefficient of variation (CV) value less than 0.3, it is considered that it contains polygonal features; (5) For the frequency band containing polygonal features, calculate the polygonal order based on the frequency and vehicle speed; (6) Extract time-domain and frequency-domain features from the vibration data containing polygonal features, including the root mean square (RMS) value in the time domain, energy entropy, and the main frequency amplitude in the frequency domain. Input these feature values and the calculated order into the polygonal quantitative identification model for quantitative identification of polygonal wear. Appendix Figure 10 The order discrimination result diagram identifies the first three main orders of the two polygonal wheels and locks the dominant order with the largest amplitude. In quantitative identification, the dominant order of the wheel polygon with the largest amplitude is mainly considered.
[0087] The formulas for calculating the main characteristic parameters are as follows:
[0088] ;
[0089] ;
[0090] ;
[0091] ;
[0092] ;
[0093] ;
[0094] in, It is the signal mean. CV is the standard deviation of the signal, and CV is the coefficient of variation. It is the frequency of the nth-order polygon, and v is the vehicle speed. R is the wavelength of the nth-order polygon, R is the radius of the wheel, and n is the order of the polygon.
[0095] Step S14: Analyze the dominant order and corresponding target time-frequency feature data using the target wear analysis model to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
[0096] In this embodiment, before analyzing the dominant order and corresponding target time-frequency characteristic data using the target wear analysis model, the method further includes: generating simulated fault samples covering different polygon orders, wear depths, and operating speeds, as well as corresponding axle box vibration acceleration signals, using a preset simulation platform to construct a full-condition fault sample set; training different initial wear analysis models using the full-condition fault sample set, and determining the target wear analysis model from each initial wear analysis model based on the training effect of each initial wear analysis model.
[0097] Specifically, such as Figure 11 As shown in the process, a vehicle dynamics model is built based on the target dynamics simulation platform, and its accuracy is verified through real vehicle tests. Wheel polygon fault samples covering common fault orders (e.g., 5-15), wear wave depths of 0-0.2mm, and operating speeds of 60-160km / h are generated. The axle box vibration acceleration corresponding to each fault sample is simulated and calculated, forming a full-condition fault sample set to compensate for the lack of measured samples. Based on the fault sample set, time-domain and frequency-domain features are extracted as input, and wear wave depth is used as output. Nonlinear mapping models are constructed using Kernel Extreme Learning Machine (KELM), Radial Basis Function Network (RBFNN), and Ensemble Learning Method (XGBoost), respectively. The coefficient of determination is used as the evaluation index for model training accuracy. The training speed and recognition accuracy of each model are compared, and the optimal quantitative recognition model with a coefficient of determination ≥0.95 is selected and packaged. The above model selection process is as follows: Figure 12 As shown, the training accuracy of different machine learning algorithms is as follows: Figure 13 As shown, the process of generating a fault sample set is as follows: Figure 14 As shown.
[0098] The coefficient of determination is used as a metric for the accuracy of iterative training in machine learning, and the formula is as follows:
[0099] ;
[0100] in: To verify the number of sample points, Let be the actual value of the target response at the i-th sample point, and be the predicted value of the surrogate model. This is the approximate mean of all sample points.
[0101] In addition, after analyzing the dominant order and the corresponding target time-frequency characteristic data using the target wear analysis model, the process also includes: determining the wear warning level of the target vehicle according to the preset wheel wear classification and the wheel wear degree of the target vehicle, and performing corresponding maintenance on the target vehicle according to the wear warning level.
[0102] Specifically, the dominant order and its corresponding time-domain and frequency-domain features are input into the optimal quantitative identification model, and the wear wave depth is output. Based on the wear wave depth, four warning levels are defined: 0~0.05mm is Level 1 alarm (monthly tracking), 0.05~0.1mm is Level 2 alarm (weekly tracking), 0.1~0.15mm is Level 3 alarm (repair within one week), and >0.15mm is Level 4 alarm (repair upon same-day entry into inventory). Corresponding maintenance strategies are output, and the final wear identification result is as follows: Figure 15 As shown.
[0103] In addition, this embodiment requires the deployment of system hardware and software modules, such as system hardware... Figure 16 As shown. The hardware modules include a data acquisition module (2kHz vibration acceleration sensor, vehicle speed sensor), a preprocessing module, a storage module, a 5G transmission module, an onboard host module, a ground analysis platform module, and a warning output module; the software modules adopt a modular design, with data preprocessing, order discrimination, and quantitative identification modules deployed on the onboard host, and sample set construction and model training optimization modules deployed on the ground analysis platform; the onboard host achieves real-time identification with a response time ≤0.1s, and the ground platform achieves offline deep analysis and model iterative optimization.
[0104] Therefore, this application achieves accurate existence judgment and dominance order locking of wheel polygon features by using a stable frequency component discrimination method based on the coefficient of variation, thus solving the problem of ambiguous order identification in traditional methods. By inputting the dominant order and corresponding features into a machine learning model for analysis, it achieves automatic quantitative identification of wheel wear depth, overcoming the shortcomings of traditional methods that rely on manual labor and have low accuracy.
[0105] See Figure 17 As shown, this embodiment of the invention discloses a wheel wear recognition device based on machine learning, comprising:
[0106] The feature extraction module 11 is used to acquire the target axle box vibration acceleration signal and the target vehicle speed signal when the target vehicle is running, and to extract features from the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data;
[0107] The coefficient of variation calculation module 12 is used to perform peak retrieval on the time-frequency feature data to obtain the corresponding frequency peaks, determine the amplitude data corresponding to each frequency peak along the time axis, and calculate the coefficient of variation of each set of amplitude data corresponding to each frequency peak.
[0108] The dominant order determination module 13 is used to compare each of the coefficients of variation with a preset coefficient of variation threshold, and determine whether there is a wheel polygon feature in the time-frequency feature data according to the comparison results. If there is a wheel polygon feature in the time-frequency feature data, the corresponding target polygon order is determined according to the vehicle speed signal corresponding to the wheel polygon feature, and the dominant order in the target polygon order is determined. The wheel polygon feature represents the stable frequency component in the time-frequency feature data caused by wheel polygon wear.
[0109] The data analysis module 14 is used to analyze the dominant order and the corresponding target time-frequency characteristic data using the target wear analysis model to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
[0110] In some specific embodiments, the feature extraction module 11 may specifically include:
[0111] The signal alignment unit is used to acquire the initial axle box vibration acceleration signal and the initial vehicle speed signal when the target vehicle is running, and to align the initial axle box vibration acceleration signal and the initial vehicle speed signal on the time axis.
[0112] An outlier cleaning unit is used to supplement missing values and clean outliers in the aligned initial axle box vibration acceleration signal and initial vehicle speed signal to obtain the target axle box vibration acceleration signal and the target vehicle speed signal.
[0113] In some specific embodiments, the feature extraction module 11 may specifically include:
[0114] The data decomposition unit is used to perform noise reduction processing on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding noise-reduced data, and to decompose the noise-reduced data using fully integrated empirical mode decomposition and adaptive noise technology to obtain several IMF components.
[0115] The time-frequency transformation unit is used to extract the time-domain feature parameters and frequency-domain feature parameters of each IMF component to obtain an initial feature set, and to perform time-frequency transformation on the initial feature set to obtain the time-frequency feature data.
[0116] In some specific embodiments, the dominant order determination module 13 may specifically include:
[0117] The dominant order determination unit is used to determine the amplitude data corresponding to each of the target polygon orders, and to determine the polygon order with the largest amplitude data among the target polygon orders as the dominant order.
[0118] In some specific embodiments, the data analysis module 14 further includes:
[0119] The sample set construction unit is used to generate simulated fault samples covering different polygon orders, wear depths and operating speeds, as well as corresponding axle box vibration acceleration signals, using a preset simulation platform to construct a full-condition fault sample set.
[0120] The model training unit is used to train different initial wear analysis models using the full-condition fault sample set, and to determine the target wear analysis model from each initial wear analysis model based on the training effect of each initial wear analysis model.
[0121] In some specific embodiments, the data analysis module 14 further includes:
[0122] The vehicle maintenance unit is used to determine the wear warning level of the target vehicle according to the preset wheel wear classification and the wear degree of the target vehicle's wheels, and to perform corresponding maintenance on the target vehicle according to the wear warning level.
[0123] In some specific embodiments, the machine learning-based wheel wear recognition device further includes:
[0124] The process skip unit is used to skip the process of identifying wheel wear on the target vehicle if there is no wheel polygon feature in the time-frequency feature data.
[0125] Furthermore, embodiments of this application also disclose an electronic device, Figure 18 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.
[0126] Figure 18 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the machine learning-based wheel wear recognition method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.
[0127] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.
[0128] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.
[0129] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the machine learning-based wheel wear recognition method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.
[0130] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned machine learning-based wheel wear recognition method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.
[0131] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.
[0132] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0133] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.
[0134] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0135] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for identifying wheel wear based on machine learning, characterized in that, include: The target axle box vibration acceleration signal and the target vehicle speed signal are acquired during the operation of the target vehicle, and feature extraction is performed on the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data; Peak retrieval is performed on the time-frequency feature data to obtain the corresponding frequency peaks, the amplitude data corresponding to each frequency peak along the time axis is determined, and the coefficient of variation of each set of amplitude data corresponding to each frequency peak is calculated. Each of the aforementioned coefficients of variation is compared with a preset coefficient of variation threshold, and the presence of wheel polygon features in the time-frequency feature data is determined based on the comparison results. If wheel polygon features are present in the time-frequency feature data, the corresponding target polygon order is determined based on the vehicle speed signal corresponding to the wheel polygon features, and the dominant order in the target polygon order is determined. The wheel polygon features represent the stable frequency components in the time-frequency feature data caused by wheel polygon wear. The target wear analysis model is used to analyze the dominant order and the corresponding target time-frequency characteristic data to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
2. The wheel wear recognition method based on machine learning according to claim 1, characterized in that, The acquisition of the target axle box vibration acceleration signal and the target vehicle speed signal during vehicle operation includes: Acquire the initial axle box vibration acceleration signal and the initial vehicle speed signal during the operation of the target vehicle, and align the initial axle box vibration acceleration signal and the initial vehicle speed signal on the time axis; The aligned initial axle box vibration acceleration signal and the initial vehicle speed signal are supplemented with missing values and cleaned of outliers to obtain the target axle box vibration acceleration signal and the target vehicle speed signal.
3. The wheel wear recognition method based on machine learning according to claim 1, characterized in that, The step of extracting features from the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data includes: The target axle box vibration acceleration signal and the target vehicle speed signal are denoised to obtain corresponding denoised data. The denoised data is then decomposed using fully integrated empirical mode decomposition and adaptive noise technology to obtain several IMF components. Extract the time-domain and frequency-domain feature parameters of each IMF component to obtain an initial feature set, and perform time-frequency transformation on the initial feature set to obtain the time-frequency feature data.
4. The wheel wear identification method based on machine learning according to claim 1, characterized in that, Determining the dominant order in the order of the target polygon includes: Determine the amplitude data corresponding to each of the target polygon orders, and determine the polygon order with the largest amplitude data among the target polygon orders as the dominant order.
5. The wheel wear recognition method based on machine learning according to claim 1, characterized in that, Before analyzing the dominant order and corresponding target time-frequency characteristic data using the target wear analysis model, the method further includes: Using a pre-set simulation platform, simulated fault samples covering different polygon orders, wear depths, and operating speeds, along with corresponding axle box vibration acceleration signals, are generated to construct a full-condition fault sample set. The target wear analysis model is determined from the initial wear analysis models based on the training effect of each initial wear analysis model using the full-condition fault sample set.
6. The wheel wear recognition method based on machine learning according to claim 1, characterized in that, After analyzing the dominant order and corresponding target time-frequency characteristic data using the target wear analysis model, the method further includes: The wear warning level of the target vehicle is determined according to the preset wheel wear classification and the wear degree of the target vehicle's wheels, and the target vehicle is then subjected to corresponding maintenance treatment based on the wear warning level.
7. The wheel wear identification method based on machine learning according to any one of claims 1 to 6, characterized in that, Also includes: If the time-frequency feature data does not contain wheel polygon features, then the process of identifying wheel wear on the target vehicle is skipped.
8. A wheel wear recognition device based on machine learning, characterized in that, include: The feature extraction module is used to acquire the target axle box vibration acceleration signal and the target vehicle speed signal when the target vehicle is running, and to extract features from the target axle box vibration acceleration signal and the target vehicle speed signal to obtain corresponding time-frequency feature data; The coefficient of variation calculation module is used to perform peak retrieval on the time-frequency feature data to obtain the corresponding frequency peaks, determine the amplitude data corresponding to each frequency peak along the time axis, and calculate the coefficient of variation of each set of amplitude data corresponding to each frequency peak. The dominant order determination module is used to compare each of the coefficients of variation with a preset coefficient of variation threshold, and determine whether there is a wheel polygon feature in the time-frequency feature data based on the comparison results. If there is a wheel polygon feature in the time-frequency feature data, the corresponding target polygon order is determined based on the vehicle speed signal corresponding to the wheel polygon feature, and the dominant order in the target polygon order is determined. The wheel polygon feature represents the stable frequency component in the time-frequency feature data caused by wheel polygon wear. The data analysis module is used to analyze the dominant order and the corresponding target time-frequency characteristic data using a target wear analysis model to obtain the wheel wear degree of the target vehicle; wherein, the target wear analysis model is an artificial intelligence model obtained based on machine learning technology.
9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the machine learning-based wheel wear recognition method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Used to store a computer program, which, when executed by a processor, implements the machine learning-based wheel wear recognition method as described in any one of claims 1 to 7.