Steel rail corrugation feature extraction method and device, terminal equipment and storage medium
By synchronously collecting and analyzing vibration and noise signals on passenger trains, forming vibration and noise fusion characteristics, and inputting them into a neural network model, the problem of limited representation ability caused by single signals in existing technologies is solved, and more accurate rail corrugation detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN TECH UNIV
- Filing Date
- 2025-12-22
- Publication Date
- 2026-04-17
AI Technical Summary
Existing rail corrugation detection methods rely on a single signal source, which limits their characterization capabilities. They are unable to achieve coordinated analysis and feature-level fusion of vibration and noise signals, and cannot comprehensively and robustly characterize the complete physical state of corrugation, resulting in information blind spots.
By integrating the sensor terminal of the passenger train to synchronously collect vibration and noise signals, performing time-domain, frequency-domain, and time-frequency-domain analysis, and after standardization processing, splicing them together to form vibration and noise fusion features, and inputting them into a convolutional neural network model to identify the wavelength and depth information of rail corrugation.
It enhances information complementarity, provides richer criteria for judgment, improves the accuracy of extracting rail corrugation wavelength and depth information, reduces operation and maintenance costs, and supports online monitoring of the entire line in all weather conditions.
Smart Images
Figure CN121880877A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of rail inspection technology, and in particular to a method, apparatus, terminal equipment and storage medium for extracting rail corrugation features. Background Technology
[0002] Rail corrugation (wavy wear), as one of the main forms of rail damage, requires timely detection and maintenance. Current rail corrugation measurement and identification technologies, based on their technical approaches and equipment, suffer from limitations in their characterization capabilities due to a single signal source and incomplete information dimensions. Most existing methods rely solely on either vibration or noise as a single signal source. However, vibration signals are severely filtered out during transmission through the train's suspension system, resulting in insensitivity to key high-frequency characteristics of corrugation excitation. While noise signals retain high-frequency information, they are easily interfered with by environmental background noise such as passenger conversations and announcements. This inherent limitation of a single signal source prevents the extracted feature set from comprehensively and robustly characterizing the complete physical state of corrugation, creating an information blind spot. Furthermore, the lack of an effective vibration-noise fusion mechanism makes it difficult to handle complex operating conditions: current technologies have failed to achieve collaborative analysis and feature-level fusion of vibration and noise signals. Summary of the Invention
[0003] In view of this, embodiments of this application provide a method, apparatus, terminal equipment, and storage medium for extracting rail corrugation features, which can effectively solve the problem of limited characterization capabilities.
[0004] In a first aspect, embodiments of this application provide a method for extracting rail corrugation features, including: Vibration and noise signals inside the train body are collected synchronously through a sensor terminal integrated into the passenger train. The vibration signal is subjected to time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis to obtain vibration signal characteristics; the noise signal is subjected to time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis to obtain noise signal characteristics. The vibration signal features and the noise signal features are standardized separately, and the standardized vibration signal features and the noise signal features are then concatenated to form a vibration-noise fusion feature. The vibration and noise fusion features are input into a trained convolutional neural network model to identify and output the rail corrugation wavelength and depth information at the corresponding mileage location.
[0005] In some embodiments, the method further includes: The severity of rail corrugation is determined based on the identified wave depth information, and maintenance recommendations are generated. When the wave depth exceeds a preset threshold, an early warning mechanism is triggered, and an early warning message is pushed to the track maintenance management system.
[0006] In some embodiments, performing time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the vibration signal to obtain vibration signal characteristics includes: The vibration acceleration signal is analyzed in the time domain and frequency domain, and vibration features are extracted from the vibration acceleration signal by combining the physical mapping relationship between the wave-slip wavelength and the wave depth. The vibration signal features include time domain features and frequency domain features, wherein the frequency domain features include the energy ratio of multiple wavelength critical frequencies and the mean and median of the wave depth index of each frequency band. The step of performing time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics includes: The noise signal is analyzed in the time domain, frequency domain, and time-frequency domain, and combined with the wave-slip depth calculation model to extract noise features and form noise signal features; the noise signal features include time domain features, frequency domain energy features, and wavelet energy values based on wavelet transform.
[0007] In some embodiments, the vibration signal is a triaxial acceleration signal with a sampling frequency not lower than a preset first frequency; the noise signal is a full-band audio signal with a sampling frequency not lower than a preset second frequency.
[0008] In some embodiments, the vibration-noise fusion feature is a 64-dimensional feature vector; The step of concatenating the standardized vibration signal features with the noise signal features to form vibration-noise fusion features includes: The 30-dimensional features from the vibration signal features and the 34-dimensional features from the noise signal features are directly spliced together in sequence after standardization.
[0009] In some embodiments, the output layer of the convolutional neural network model is divided into a first branch and a second branch; The first branch is used to classify the wave-polishing wavelengths, dividing the wave-polishing wavelengths into four categories: a first wavelength range, a second wavelength range, a third wavelength range, and a composite wavelength. The second branch is used to classify wave depths, dividing the wave depths into three categories: a first wave depth interval, a second wave depth interval, and a third wave depth interval.
[0010] In some embodiments, the wave depth index is a dimensionless parameter calculated from a mapping function between wave depth and signal energy established based on measured wave wear data.
[0011] Secondly, this application also provides a rail corrugation feature extraction device, comprising: The data acquisition module is used to synchronously collect vibration and noise signals inside the passenger train through a sensor terminal integrated into the passenger train. The data analysis module is used to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the vibration signal to obtain vibration signal characteristics, and to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics. The data fusion module is used to standardize the vibration signal features and the noise signal features respectively, and then concatenate the standardized vibration signal features with the noise signal features to form vibration-noise fusion features; The data recognition module is used to input the vibration and noise fusion features into the trained convolutional neural network model and identify and output the rail corrugation wavelength and depth information at the corresponding mileage position.
[0012] Thirdly, this application also provides a terminal device, which includes a processor and a memory, the memory storing a computer program, and the processor being used to execute the rail corrugation feature extraction method described above.
[0013] Fourthly, this application also provides a readable storage medium storing a computer program that, when executed on a processor, implements the rail corrugation feature extraction method described above.
[0014] The embodiments of this application have the following beneficial effects: The solution in this embodiment synchronously collects vibration and noise signals inside the passenger train by integrating a sensor terminal. Then, the model features of the vibration and noise signals are extracted and fused, and the neural network model is used for identification. By fusing the vibration and noise signals, the complementarity of information is enhanced, so as to obtain more accurate and effective identification results and overcome the problem of limited representation ability. Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 A schematic flowchart of a method for extracting rail corrugation features according to an embodiment of this application is shown; Figure 2 This invention illustrates a schematic diagram of a neural network model structure according to an embodiment of this application; Figure 3 A schematic diagram of a rail corrugation feature extraction device according to an embodiment of this application is shown. Detailed Implementation
[0017] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.
[0018] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.
[0019] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.
[0020] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.
[0021] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features described herein can be combined with each other.
[0022] To address the problems of existing technologies, this application provides a method for extracting rail corrugation features. The method involves synchronously collecting vibration and noise signals from within the passenger train using a microelectromechanical system (MEMS) sensor terminal integrated into the train. The vibration signals are analyzed in the time domain, frequency domain, and time-frequency domain to obtain vibration signal features. Similarly, the noise signals are analyzed in the time domain, frequency domain, and time-frequency domain to obtain noise signal features. Both the vibration and noise signal features are standardized, and then concatenated with the standardized vibration and noise signal features to form a vibration-noise fusion feature. This fusion feature is then input into a trained convolutional neural network model to identify and output the rail corrugation wavelength and depth information at the corresponding mileage location. By fusing the vibration and noise signals, the complementary nature of the information is enhanced, providing richer criteria for judgment and enabling the extraction of rail corrugation wavelength and depth information, thus making a solid contribution to subsequent judgment.
[0023] The following examples illustrate the method for extracting the corrugation features of rails.
[0024] Figure 1 A flowchart of a rail corrugation feature extraction method according to an embodiment of this application is shown. Exemplarily, the rail corrugation feature extraction method includes the following steps: In step S100, vibration and noise signals inside the train body are collected synchronously through a sensor terminal integrated into the passenger train.
[0025] The method in this embodiment is applied to the detection of rail corrugation, which refers to the periodic, wavy wear on the rail surface along the longitudinal direction, manifesting as regular or irregular undulations and depressions. It is mainly caused by rolling contact fatigue, vibration resonance, and friction between the wheel and rail. It can lead to increased train noise, intensified vibration, and decreased passenger comfort; in severe cases, it can cause damage to the track structure and even the risk of derailment.
[0026] The microelectromechanical system (MEMS) sensing terminal is a device that integrates an onboard accelerometer and a microphone, which can collect vibration and noise signals from inside the vehicle. The vibration signal is the mechanical vibration caused by the interaction between the wheels and rails during train operation, which is transmitted to the interior of the carriage through the bogies and car body.
[0027] The noise signal is an acoustic emission signal generated in the wheel-rail contact area, which propagates through the air into the carriage.
[0028] It primarily collects triaxial acceleration vibration signals and full-band audio signals simultaneously. This terminal features a high sampling rate, low power consumption, and wireless transmission capabilities, allowing for long-term deployment on routinely operating trains to achieve all-weather, full-line data acquisition.
[0029] The triaxial acceleration signal has a sampling frequency of not less than 500Hz; the noise signal is a full-band audio signal with a sampling frequency of not less than 44.1kHz.
[0030] Step S200: Perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the vibration signal to obtain vibration signal characteristics; perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics.
[0031] Next, the extracted vibration and noise signals will be processed separately. Time-domain and frequency-domain analyses will be performed on the vibration acceleration signals. Based on the physical mapping relationship between wave-worn wavelength and wave depth, vibration features will be extracted from the vibration acceleration signals. These vibration signal features include time-domain and frequency-domain features. The frequency-domain features include the energy ratios of multiple wavelength critical frequencies and the mean and median of the wave depth index for each frequency band. The wave depth index is an empirical or semi-empirical indicator constructed based on measured wave depth data, used to quantify the correlation between signals in different frequency bands and the actual wave-worn depth.
[0032] Time-domain analysis refers to directly analyzing the statistical characteristics of signal amplitude changes over time. Frequency-domain analysis refers to converting the original time-domain signal to frequency space for analysis, revealing the energy distribution patterns of the signal. Time-frequency domain analysis refers to methods for joint time-frequency analysis of non-stationary signals (such as signals whose frequency varies with time).
[0033] Similarly, the noise signal is analyzed in the time domain, frequency domain, and time-frequency domain, and combined with the wave-slip depth calculation model to extract noise features and form noise signal features; the noise signal features include time domain features, frequency domain energy features, and wavelet energy values based on wavelet transform.
[0034] Taking vibration signals as an example, time-domain analysis can extract features such as mean, variance, peak value, and kurtosis. Then, frequency-domain analysis can be performed to extract spectral features such as center frequency, bandwidth, and harmonic components. Finally, through time-frequency domain analysis, wavelet transform can be used to obtain time-frequency graph features and capture the characteristics of non-stationary signals.
[0035] For example, time-domain feature extraction can capture transient shocks and overall fluctuation trends. It can extract 15 typical time-domain statistics, covering dimensions such as amplitude distribution, dispersion, and non-Gaussianity, including: maximum value, minimum value, peak-to-peak value, absolute mean, median, variance, standard deviation, root mean square (RMS), and skewness, which reflects asymmetry.
[0036] Kurtosis characterizes impulse intensity, while interquartile range (IQR) and mean absolute deviation (MAD) are used to characterize noise immunity.
[0037] The mean and median of the absolute difference reflect the degree of drastic change between adjacent sampling points.
[0038] By counting the points past the mean, the frequency at which the signal crosses the baseline is measured, which indirectly relates to periodic damage.
[0039] These characteristics collectively characterize the abnormal vibration energy concentration and repetitive impact behavior caused by the train passing through the corrugated section, and are an important basis for identifying initial damage.
[0040] Frequency domain feature extraction is used to reveal wavelength-related resonant frequency components. For example, FFT transformation is performed on the original vibration signal and the power spectral density (PSD) is calculated to extract the energy proportion of key frequency bands. Five frequency bands are divided into 0–100Hz, 100–200Hz, 200–300Hz, 300–400Hz, and 400–500Hz, covering the main vibration response range of the wheel-rail system.
[0041] A wavelength-critical frequency mapping model is introduced: based on the relationship between vehicle speed and wavelength, common wave milling wavelengths are divided into (e.g., 30 mm to 40 mm, 40 mm to 50 mm, 50 mm to 100 mm) and converted into corresponding excitation frequency ranges.
[0042] For example, at an operating speed of 80 km / h (≈22.2 m / s), the critical frequency corresponding to a 30 mm wavelength is approximately 740 Hz. The critical frequency corresponding to a 50 mm wavelength is approximately 444 Hz.
[0043] Therefore, in practical applications, the frequency band allocation can be dynamically adjusted to focus the analysis on the theoretically expected frequency band.
[0044] In addition, the ratio of spectral energy to the energy of the adjacent background noise within each target wavelength range can be calculated to form a critical frequency energy ratio characteristic, which significantly enhances the sensitivity to specific wavelength bands.
[0045] When performing time-frequency domain feature extraction, the primary goal is to capture the temporal localization capability of non-stationary impact events. For example, Discrete Wavelet Transform (DWT) can be used to perform multi-resolution analysis on noise signals, decomposing to the fourth level using db4 or sym5 wavelet basis functions to extract the energy of four key sub-bands: 125–250Hz: corresponding to the subharmonic components excited by short-to-medium wavelength wave wear; 250–500Hz: the wheel-rail resonance region, where energy is concentrated; 500–1000Hz: the typical impact noise frequency band; and 1000–2000Hz: high-frequency friction and microcrack propagation acoustic emission. These four wavelet energy features can effectively capture transient, localized wheel-rail interaction events, and are particularly suitable for detecting early-stage weak wave wear.
[0046] Finally, the vibration and noise characteristics are obtained. For ease of subsequent explanation, the vibration characteristics are represented as follows: It includes 15 time-domain features and 15 frequency-domain features.
[0047] The noise characteristics are represented as follows: It includes 15 time-domain features, 15 frequency-domain features, and 4 time-frequency-domain features.
[0048] Step S300: Standardize the vibration signal features and the noise signal features respectively, and then splice the standardized vibration signal features and the noise signal features to form a vibration-noise fusion feature. During standardization, Z-score standardization can be used to obtain the standardized vibration signal characteristics and noise signal characteristics.
[0049] Next, the standardized vibration and noise feature vectors are directly concatenated, and the resulting fused feature expression is: .
[0050] After concatenation, a unified 64-dimensional vibration-noise fusion feature vector is formed, which serves as the input for subsequent classification models. The concatenation order is fixed and traceable, facilitating later feature importance analysis. The standardized 30-dimensional vibration feature vector and the 34-dimensional noise feature vector are directly concatenated along their feature dimensions to form a 64-dimensional vibration-noise fusion feature vector. This process achieves deep fusion from the physical channel level to the feature representation level, preserving the unique information of each signal while enhancing the overall discriminative ability.
[0051] Step S400: Input the vibration and noise fusion features into the trained convolutional neural network model to identify and output the rail corrugation wavelength and depth information at the corresponding mileage position.
[0052] A convolutional neural network model is constructed, taking the fused 64-dimensional feature vector as input and outputting two branches. The convolutional neural network model is as follows: Figure 2 As shown, the output layer of the convolutional neural network model is divided into a first branch and a second branch.
[0053] The first branch is used to classify the wave-polishing wavelengths, dividing them into four categories: a first wavelength range, a second wavelength range, a third wavelength range, and a composite wavelength. The second branch is used to classify the wave depth, dividing it into three categories: a first wave depth range, a second wave depth range, and a third wave depth range.
[0054] As an example, wavelengths can be classified into four categories: 30 mm–40 mm, 40 mm–50 mm, 50 mm–100 mm, and composite wavelengths. Wave depth can be classified into three categories: less than or equal to 0.01 mm (healthy), 0.01–0.02 mm (mild), and greater than or equal to 0.02 mm (severe).
[0055] The aforementioned model can be trained using supervised learning, with labels provided by actual measurement data from a wave mill, ensuring the output results have practical engineering significance. Ultimately, it can automatically generate wave mill distribution maps and maintenance priority suggestions based on the identification results.
[0056] After determining the severity of rail corrugation based on the identified wave depth, maintenance recommendations can be generated. Furthermore, when the identified wave depth exceeds a preset threshold, an early warning mechanism is triggered, and an early warning message is pushed to the track maintenance management system.
[0057] The aforementioned preset threshold can be 0.02 mm. That is, when the wave depth is classified as severe wave wear, an early warning mechanism can be triggered and an early warning message can be pushed to the track maintenance management system. The track maintenance management system is a system that monitors the safety status of the entire track. The method in this embodiment can be an independent function running within the track maintenance management system or a function terminal attached to the outside of the system.
[0058] The method in this embodiment effectively compensates for the information blind spots of a single signal by fusing two heterogeneous signals: vibration and noise. Experiments show that under complex operating conditions (such as different vehicle speeds, track curvature, and environmental noise), the recognition accuracy of this method is about 18%-25% higher than that of a single vibration or noise method, especially performing excellently in short wavelength (<50mm) and shallow wavelength (<0.02mm) scenarios. Furthermore, it enhances feature representation capabilities and expands information dimensions. This embodiment constructs a 64-dimensional fused feature space including time domain, frequency domain, time-frequency domain, and physically guided features (wavelength-to-energy ratio, wavelength-depth index), increasing the number of features by more than 110% compared to a single signal, greatly enriching the discrimination criteria that can be used for classification and improving the model's generalization ability. It achieves low-cost, high-frequency, and full-coverage online monitoring. Based on passenger train deployment, it eliminates the need for dedicated inspection vehicles or additional maintenance windows, allowing continuous data collection during daily operation, truly realizing an intelligent monitoring mode of "detection as operation," significantly reducing maintenance costs. The integration of physical mechanisms and data-driven approaches enhances model interpretability. This embodiment introduces the "critical frequency-energy ratio" and "wave depth index" based on wheel-rail dynamics into feature engineering, enabling the model to not only rely on black-box learning but also better reflect the real physical process, thus enhancing the reliability of the results and facilitating understanding and adoption by engineers. In summary, this embodiment provides an efficient, accurate, and robust wave wear feature extraction method, solving the performance bottleneck problem caused by the single signal in existing onboard detection technologies, and promoting the development of intelligent operation and maintenance of rail transit infrastructure.
[0059] Figure 3 A schematic diagram of a rail corrugation feature extraction device according to an embodiment of this application is shown. Exemplarily, the rail corrugation feature extraction device includes: Data acquisition module 10 is used to synchronously acquire vibration and noise signals inside the vehicle body through a microelectromechanical system sensing terminal integrated into the passenger train. Data analysis module 20 is used to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the vibration signal to obtain vibration signal characteristics, and to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics; The data fusion module 30 is used to standardize the vibration signal features and the noise signal features respectively, and then splice the standardized vibration signal features and the noise signal features to form vibration-noise fusion features; The data recognition module 40 is used to input the vibration and noise fusion features into the trained convolutional neural network model and identify and output the rail corrugation wavelength and depth information at the corresponding mileage position.
[0060] It is understood that the apparatus in this embodiment corresponds to the xx method in the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.
[0061] This application also provides a terminal device, exemplary of which includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to enable the terminal device to perform the functions of the various modules in the above-described rail corrugation feature extraction method or the above-described rail corrugation feature extraction device.
[0062] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.
[0063] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.
[0064] This application also provides a readable storage medium for storing the computer program used in the aforementioned terminal device.
[0065] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0066] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.
[0067] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0068] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A method of rail corrugation feature extraction, characterized in that, include: Vibration and noise signals inside the train body are collected synchronously through a sensor terminal integrated into the passenger train. The vibration signal is subjected to time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis to obtain vibration signal characteristics; the noise signal is subjected to time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis to obtain noise signal characteristics. The vibration signal features and the noise signal features are standardized separately, and the standardized vibration signal features and the noise signal features are then concatenated to form a vibration-noise fusion feature. The vibration and noise fusion features are input into a trained convolutional neural network model to identify and output the rail corrugation wavelength and depth information at the corresponding mileage location.
2. The steel rail roughness feature extraction method of claim 1, wherein, Also includes: The severity of rail corrugation is determined based on the identified wave depth information, and maintenance recommendations are generated. When the wave depth exceeds a preset threshold, an early warning mechanism is triggered, and an early warning message is pushed to the track maintenance management system.
3. The method for extracting rail corrugation features according to claim 1, characterized in that, The vibration signal is subjected to time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis to obtain vibration signal characteristics, including: The vibration acceleration signal is analyzed in the time domain and frequency domain, and vibration features are extracted from the vibration acceleration signal by combining the physical mapping relationship between the wave-slip wavelength and the wave depth. The vibration signal features include time domain features and frequency domain features, wherein the frequency domain features include the energy ratio of multiple wavelength critical frequencies and the mean and median of the wave depth index of each frequency band. The step of performing time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics includes: The noise signal is analyzed in the time domain, frequency domain, and time-frequency domain, and combined with the wave-slip depth calculation model to extract noise features, which constitute noise signal features. The noise signal features include time domain features, frequency domain energy features, and wavelet energy values based on wavelet transform.
4. The method for extracting rail corrugation features according to claim 1, characterized in that, The vibration signal is a triaxial acceleration signal with a sampling frequency not lower than a preset first frequency; the noise signal is a full-band audio signal with a sampling frequency not lower than a preset second frequency.
5. The method for extracting rail corrugation features according to claim 1, characterized in that, The vibration-noise fusion feature is a 64-dimensional feature vector; The step of concatenating the standardized vibration signal features with the noise signal features to form vibration-noise fusion features includes: The 30-dimensional features from the vibration signal features and the 34-dimensional features from the noise signal features are directly spliced together in sequence after standardization.
6. The method for extracting rail corrugation features according to claim 3, characterized in that, The output layer of the convolutional neural network model is divided into a first branch and a second branch; The first branch is used to classify the wave-polishing wavelengths, dividing the wave-polishing wavelengths into four categories: a first wavelength range, a second wavelength range, a third wavelength range, and a composite wavelength. The second branch is used to classify wave depths, dividing the wave depths into three categories: a first wave depth interval, a second wave depth interval, and a third wave depth interval.
7. The method for extracting rail corrugation features according to claim 3, characterized in that, The wave depth index is a dimensionless parameter calculated based on the mapping relationship function between wave depth and signal energy established from measured wave wear data.
8. A device for extracting rail corrugation features, characterized in that, include: The data acquisition module is used to synchronously collect vibration and noise signals inside the passenger train through a sensor terminal integrated into the passenger train. The data analysis module is used to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the vibration signal to obtain vibration signal characteristics, and to perform time-domain analysis, frequency-domain analysis, and time-frequency-domain analysis on the noise signal to obtain noise signal characteristics. The data fusion module is used to standardize the vibration signal features and the noise signal features respectively, and then concatenate the standardized vibration signal features with the noise signal features to form vibration-noise fusion features; The data recognition module is used to input the vibration and noise fusion features into the trained convolutional neural network model and identify and output the rail corrugation wavelength and depth information at the corresponding mileage position.
9. A terminal device, characterized in that, The terminal device includes a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the rail corrugation feature extraction method according to any one of claims 1-7.
10. A readable storage medium, characterized in that, It stores a computer program, which, when executed on a processor, implements the rail corrugation feature extraction method according to any one of claims 1-7.