A rail wave depth prediction method, device, equipment and medium

Abstract

The application discloses a steel rail wave depth prediction method and device, equipment and medium, relates to the wheel-rail monitoring technology field of track traffic, and comprises the following steps: synchronously acquiring sound signals and vibration signals of a track vehicle running along a steel rail to be predicted; extracting sound features and vibration features related to rail wear from the sound signals and vibration signals; inputting a to-be-predicted feature vector constructed based on the sound features, the vibration features, and working condition information of the track vehicle running along the steel rail to be predicted into a preset steel rail wave depth prediction model, so that the to-be-predicted feature vector is assigned to multiple predictors, each predictor performs prediction processing on the to-be-predicted feature vector according to a preset decision rule, and corresponding prediction values are output; and determining and outputting a predicted wave depth value of the steel rail to be predicted based on the prediction values output by the multiple predictors. The high-precision wave depth value can be predicted and output.

Description

Technical Field

[0001] This invention relates to the field of rail transit wheel-rail monitoring technology, and in particular to a method, device, equipment and medium for predicting rail wave depth. Background Technology

[0002] Rail corrugation is a major form of damage in wheel-rail systems, occurring in urban rail transit, high-speed rail, and freight rail. Rail corrugation exacerbates the interaction between the wheel and rail, causing excessive vibration of wheel-rail system components, thus affecting the comfort, safety, and stability of operating vehicles. Severe rail corrugation can lead to the breakage or even detachment of wheel-rail system components. The development of rail corrugation not only poses safety risks but also accelerates rail damage, thereby increasing rail maintenance costs. Currently, rail corrugation detection methods can be divided into direct measurement and indirect measurement methods. Direct measurement involves directly contacting the actual geometry of the rail surface with a measuring instrument to calculate the wavelength and depth parameters of the corrugated wear. Based on rail grinding and maintenance standards, it is determined whether grinding and maintenance are necessary. Commonly used measuring instruments include ruler measurement, contact displacement sensor measurement, and inertial reference methods. This method offers high measurement accuracy and is unaffected by the wheel tread surface, but it requires experienced personnel to work at night, consuming significant manpower and costs. Indirect measurement involves measuring the dynamic response of the vehicle or track caused by corrugation, such as collecting noise signals from the wheel-rail system and vibration acceleration signals from the vehicle axle box, to calculate parameters such as the wavelength and depth of rail corrugation under the contact action between the wheel and the track. This method can achieve real-time and online monitoring of the track and can determine the location of rail corrugation based on abnormal vibration acceleration or noise. However, rail corrugation detection is greatly affected by parameters such as vehicle speed, wheel tread out-of-roundness, vehicle structure, and track conditions. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide a method, apparatus, equipment, and medium for predicting rail corrugation depth, which can improve the efficiency of rail corrugation detection and the quantitative accuracy of corrugation, so that the predicted rail corrugation depth results can guide track maintenance personnel to formulate reasonable grinding strategies. The specific solution is as follows:

[0004] In a first aspect, this application discloses a method for predicting rail wave depth, including:

[0005] Simultaneously acquire sound and vibration signals of the rail vehicle running along the rail to be predicted;

[0006] Extract the acoustic and vibration features related to rail corrugation from the acoustic and vibration signals;

[0007] The feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, is input into the preset rail wave depth prediction model so that the feature vector to be predicted is assigned to multiple predictors, so that each predictor performs prediction processing on the feature vector to be predicted according to the preset decision rules and outputs the corresponding prediction value.

[0008] The predicted wave depth value of the rail to be predicted is determined and output based on the predicted values ​​output by each of the predictors.

[0009] Optionally, extracting the sound and vibration features related to rail corrugation from the sound signal and the vibration signal includes:

[0010] The sound signal and the vibration signal are respectively subjected to DC removal processing to obtain the first sound signal and the first vibration signal, respectively;

[0011] The frequency components of the non-rail corrugation characteristic frequency band in the first sound signal and the first vibration signal are filtered to obtain the second sound signal and the second vibration signal.

[0012] Extract the acoustic features related to rail corrugation from the second acoustic signal;

[0013] Extract vibration features related to rail corrugation from the second vibration signal.

[0014] Optionally, the step of filtering the frequency components of the non-rail corrugation characteristic frequency bands in the first sound signal and the first vibration signal to obtain the second sound signal and the second vibration signal includes:

[0015] Configure the non-corrugating characteristic frequency band corresponding to the rail running process to be predicted;

[0016] The first sound signal and the first vibration signal are band-stop filtered using the non-wave-slip characteristic frequency band to obtain the second sound signal and the second vibration signal.

[0017] Accordingly, extracting sound features related to rail corrugation from the second sound signal and extracting vibration features related to rail corrugation from the second vibration signal includes:

[0018] The sound pressure level is extracted from the second sound signal to obtain sound characteristics;

[0019] The effective value of vibration acceleration is extracted from the second vibration signal to obtain the vibration characteristics.

[0020] Optionally, the step of inputting the feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model includes:

[0021] The sound features, vibration features, and information on the vehicle speed and curve radius of the rail vehicle as it runs along the rail to be predicted are scalarized and then combined in sequence to obtain the feature vector to be predicted.

[0022] The feature vector to be predicted is input into the preset rail wave depth prediction model.

[0023] Optionally, before inputting the feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, the method further includes:

[0024] The initial rail wave depth prediction model based on the random forest model was trained using the training sample set. The training model learned the nonlinear mapping relationship between historical sound features, historical vibration features, historical working condition information and the rail wave depth value after kilometer benchmark alignment, so as to obtain the preset rail wave depth prediction model.

[0025] Optionally, before training the initial rail wave depth prediction model based on the random forest model using the training sample set, the method further includes:

[0026] Simultaneously acquire the initial sound and vibration signals of the rail vehicle running along the rail under different historical operating conditions;

[0027] The initial sound signal and the initial vibration signal are preprocessed to obtain the preprocessed sound signal and the preprocessed vibration signal.

[0028] Based on the wheel speed and wheel diameter acquired synchronously with the initial sound signal and the initial vibration signal, the kilometer marker information corresponding to the preprocessed sound signal and the preprocessed vibration signal is calculated to obtain the preprocessed sound signal and the preprocessed vibration signal carrying the kilometer marker information.

[0029] Preprocessing, including kilometer mark calculation, is performed on the rail displacement waveform data collected by the corrugation detection vehicle during its operation along the rail under different historical working conditions to calculate the rail wave depth value carrying kilometer mark information.

[0030] The preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information are compared and processed to obtain the preprocessed sound signal, the preprocessed vibration signal, and the rail wave depth value after kilometer marker comparison.

[0031] Historical sound features and historical vibration features related to rail corrugation were extracted from the preprocessed sound signal after the kilometer marker alignment and the preprocessed vibration signal after the kilometer marker alignment, respectively.

[0032] Training samples are constructed based on the historical sound features, historical vibration features, historical working condition information, and rail wave depth values ​​after the kilometer marker is aligned, and all training samples under different working conditions are used as the training sample set.

[0033] Optionally, the preprocessing of the rail displacement waveform data collected by the corrugation detection vehicle during its operation along the rail under different historical working conditions, including kilometer marker calculation, to calculate the rail wave depth value carrying kilometer marker information, includes:

[0034] The rail displacement waveform data of the whole frequency band was collected by the corrugated testing vehicle during the operation of the rail under different historical working conditions.

[0035] The rail displacement waveform data of the entire frequency band is subjected to bandpass filtering to obtain rail displacement waveform data of each wavelength band.

[0036] According to the preset sampling window length of each wavelength band, the rail displacement waveform data of the corresponding wavelength band is processed by sliding non-overlapping slicing to obtain each waveform data slice, and the kilometer marker information of each waveform data slice is calculated to obtain the rail displacement waveform data carrying the kilometer marker information after slicing.

[0037] Calculate the peak-to-peak value of the rail displacement waveform data carrying kilometer marker information after each slice;

[0038] The arithmetic mean of all peak-to-peak values ​​in the sliced ​​rail displacement waveform data carrying kilometer marker information is taken as the rail wave depth value of the center point of the sliced ​​rail displacement waveform data carrying kilometer marker information.

[0039] Optionally, the calibration process for the preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information includes:

[0040] Using the sample kilometer range of the preprocessed sound signal and the preprocessed vibration signal as the reference interval, find and extract all rail wave depth values ​​located within the reference interval;

[0041] Calculate the preset quantile amplitude of the rail wave depth value, and use it as the rail wave depth value after being aligned with the sample kilometer range.

[0042] Optionally, the calibration process for the preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information includes:

[0043] Using the sample kilometer range of the preprocessed sound signal and the preprocessed vibration signal as the reference interval, find and extract all rail wave depth values ​​located within the reference interval;

[0044] Calculate the average value of the rail wave depth, and use the average value as the rail wave depth value after being aligned with the sample kilometer range.

[0045] Optionally, the step of training the initial rail wave depth prediction model based on the random forest model using the training sample set includes:

[0046] Multiple training subsets are randomly selected from the training sample set using a self-sampling method.

[0047] Select any one or more of the historical sound features, the historical vibration features, and the historical operating condition information from the training subset to construct a feature subset;

[0048] Based on the feature subset, recursively process according to the rule of maximizing the reduction of node impurity to achieve node splitting until the preset stopping growth condition is met, thus completing the construction of a predictor;

[0049] The constructed multiple predictors are used as the trained preset rail wave depth prediction model.

[0050] Optionally, the recursive processing based on the feature subset according to the rule of maximizing the reduction of node impurity to achieve node splitting includes:

[0051] For each node to be split in the decision tree of the random forest model, traverse every feature in the current feature subset and all candidate split points;

[0052] Calculate the impurity of the set of child nodes corresponding to each of the candidate split points;

[0053] Based on the features that maximize the reduction of the impurity of the parent node and the target candidate split point, a rule for maximizing the reduction of the node impurity of the node to be split is constructed, and recursive processing of the node to be split is performed according to the rule for maximizing the reduction of the node impurity to achieve node splitting.

[0054] Optionally, after determining and outputting the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors, the method further includes:

[0055] The predicted wave depth value is compared with the standard value of rail wave depth for the current wavelength band;

[0056] If the predicted wave depth value exceeds the standard value of the rail wave depth for the current wavelength band, a grinding and maintenance alarm message is generated for the corresponding mileage location of the rail to be predicted.

[0057] Secondly, this application discloses a rail wave depth prediction device, comprising:

[0058] The signal acquisition module is used to simultaneously acquire the sound and vibration signals of the rail vehicle running along the rail to be predicted.

[0059] The feature extraction module is used to extract sound features and vibration features related to rail corrugation from the sound signal and the vibration signal;

[0060] The prediction module is used to input the feature vector to be predicted, which is constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor can perform prediction processing on the feature vector to be predicted according to the preset decision rules and output the corresponding prediction value.

[0061] The result output module is used to determine and output the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors.

[0062] Thirdly, this application discloses an electronic device, including:

[0063] Memory, used to store computer programs;

[0064] A processor is used to execute the computer program to implement the steps of the aforementioned disclosed rail wave depth prediction method.

[0065] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed rail wave depth prediction method.

[0066] As can be seen, this application discloses a method for simultaneously acquiring sound and vibration signals of a rail vehicle running along a rail to be predicted; extracting sound and vibration features related to rail corrugation from the sound and vibration signals; inputting a feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into a preset rail corrugation prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor performs prediction processing on the feature vector to be predicted according to a preset decision rule and outputs a corresponding prediction value; and determining and outputting the predicted corrugation value of the rail to be predicted based on the prediction values ​​output by each predictor. Therefore, by simultaneously acquiring sound and vibration signals and extracting sound and vibration features related to rail corrugation, the synchronous acquisition means simultaneously obtaining the structural response and acoustic fingerprint of the corrugation excitation, forming a two-dimensional, complementary representation of the same corrugation. This provides more comprehensive feature information than relying solely on a single-dimensional vibration signal. Furthermore, combining acoustic and vibration features with operating condition information allows the model to understand the operating conditions under which the signal was generated. This enables the model to isolate or compensate for the influence of operating condition changes on the acoustic and vibration signals during the prediction process. Based on the learned real and stable physical relationship between acoustic and vibration features and wave depth under different speeds and curves, the model can predict and output high-precision wave depth values. Attached Figure Description

[0067] 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.

[0068] Figure 1 This is a flowchart of a rail wave depth prediction method disclosed in this application;

[0069] Figure 2 This application discloses a specific method for predicting rail wave depth.

[0070] Figure 3 This is a flowchart of an acoustic and vibration signal preprocessing method disclosed in this application;

[0071] Figure 4 This is a flowchart of a rail displacement waveform data preprocessing method disclosed in this application;

[0072] Figure 5 This is a schematic diagram of the displacement waveform data after segmentation disclosed in this application;

[0073] Figure 6This application discloses a flowchart of a method for calibrating acoustic vibration signals with rail wave depth data.

[0074] Figure 7 This is a flowchart of another method for aligning acoustic vibration signals with rail wave depth data disclosed in this application;

[0075] Figure 8 This is a flowchart of a pre-set rail wave depth prediction model training method disclosed in this application;

[0076] Figure 9(a) is a wave depth trend diagram before removing rail seams and weld seams as disclosed in this application;

[0077] Figure 9(b) is a wave depth trend diagram after removing rail seams and weld seams as disclosed in this application;

[0078] Figure 10 This application discloses a wave depth trend diagram before and after acoustic-vibration synergistic benchmarking;

[0079] Figure 11(a) is a trend diagram of the sound dB value and rail wave depth value after the kilometer marker is aligned with the marker disclosed in this application;

[0080] Figure 11(b) is a trend diagram of the effective vibration value and rail wave depth value after the kilometer marker is aligned with the marker disclosed in this application.

[0081] Figure 12(a) shows the prediction effect of a training set in a random forest model disclosed in this application;

[0082] Figure 12(b) shows the prediction effect of a test set disclosed in this application on a random forest model;

[0083] Figure 13 This is a schematic diagram of the structure of a rail wave depth prediction device disclosed in this application;

[0084] Figure 14 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0085] The technical solutions of 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 the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0086] Rail corrugation is a major form of damage in wheel-rail systems, occurring in urban rail transit, high-speed rail, and freight trains. Rail corrugation exacerbates the interaction between the wheel and rail, causing excessive vibration of wheel-rail system components. This impacts the comfort, safety, and stability of operating vehicles, and in severe cases, can lead to component breakage or even detachment. The development of rail corrugation not only poses safety risks but also accelerates rail damage, thereby increasing rail maintenance costs.

[0087] Rail corrugation detection methods can be divided into direct measurement and indirect measurement methods. Direct measurement involves using measuring instruments to directly contact the actual geometry of the rail surface to calculate the wavelength and depth parameters of the corrugated wear. Based on rail grinding and maintenance standards, it is determined whether grinding and maintenance are necessary. Commonly used measuring instruments include ruler measurement, contact displacement sensor measurement, and inertial reference methods. This method offers high measurement accuracy and is unaffected by the wheel tread surface. Indirect measurement measures the dynamic response of the vehicle or track caused by corrugation, such as collecting noise signals from the wheel-rail system and vibration acceleration signals from the vehicle axle box. This allows for the calculation of parameters such as the wavelength and depth of rail corrugation under wheel-rail contact. This method enables real-time and online monitoring of the track and can determine the location of rail corrugation based on abnormal vibration acceleration or noise.

[0088] Direct measurement methods for detecting rail corrugation require experienced personnel to work at night, consuming significant manpower and costs. Indirect measurement methods can install vibration acceleration sensors or sound sensors on operating trains to achieve online monitoring of rail corrugation across the entire line; however, rail corrugation detection is significantly affected by parameters such as vehicle speed, wheel tread out-of-roundness, vehicle structure, and track conditions.

[0089] Traditional direct measurement methods for detecting rail corrugation offer high accuracy, but with the increasing operational mileage of urban rail transit, their efficiency is declining, hindering timely rail grinding and maintenance by track maintenance personnel. Indirect measurement methods, such as vehicle-mounted rail corrugation detection, are more efficient and can monitor the rail condition across the entire line. However, the calculated rail corrugation depth deviates from the actual depth. This is primarily because the collected vibration acceleration signals include not only rail corrugation information but also information on the vibration of the vehicle's own structure and the track system, leading to inaccurate quantitative analysis of rail corrugation.

[0090] Therefore, this invention provides a rail corrugation depth prediction scheme, which can improve the efficiency of rail corrugation detection and the quantitative accuracy of corrugation, so that the predicted rail corrugation depth results can guide the maintenance personnel to formulate reasonable grinding strategies.

[0091] like Figure 1 As shown, the present invention provides a method for predicting rail wave depth, comprising:

[0092] Step S11: Synchronously acquire the sound and vibration signals of the rail vehicle running along the rail to be predicted.

[0093] In this embodiment, as Figure 2 As shown, during the operation of the rail vehicle along the rail to be predicted, vibration and sound signals are collected synchronously using vibration sensors installed in the axle box and microphones installed on the bogie.

[0094] Step S12: Extract the sound features and vibration features related to rail corrugation from the sound signal and the vibration signal.

[0095] In this embodiment, as Figure 2 As shown, the sound signal and the vibration signal are respectively subjected to DC removal processing to obtain a first sound signal and a first vibration signal; the frequency components of the non-rail corrugation characteristic frequency band in the first sound signal and the first vibration signal are filtered to obtain a second sound signal and a second vibration signal; it can be understood that the collected sound signal and vibration signal (acoustic-vibration signal) are preprocessed, specifically including DC removal processing and filtering processing, and the specific preprocessing process is as follows. Figure 3As shown, the synchronously acquired sound and vibration signals are first converted to units to obtain the converted sound and vibration signals. The arithmetic mean of all sampling points in each unit-converted sound and vibration segment is then calculated. This mean is subtracted from the value of each sampling point in the segment to obtain a zero-mean signal, effectively completing the DC removal process and yielding the first sound and vibration signals. Next, the first sound and vibration signals are filtered. Specifically, a non-corrugating characteristic frequency band corresponding to the rail running process to be predicted is configured. This non-corrugating characteristic frequency band is then used to perform band-stop filtering on the first sound and vibration signals to obtain the second sound and vibration signals. It is understood that filtering is applied to the frequency components within the non-corrugating frequency range of the first sound and vibration signals, such as wheel rotation frequency and higher-order frequencies, and the inherent frequencies of components like track fasteners and ballast, to prevent non-corrugating frequency information from affecting the corrugating feature extraction results. By analyzing samples from corrugated and non-corrugated sections of the line, the non-corrugated frequency range of the line can be determined (the non-corrugated characteristic frequency bands of different lines can be configured accordingly). It should be noted that after configuring the non-corrugated characteristic frequency bands, this embodiment uses band-stop filtering in time-domain filtering to filter the first sound signal and the first vibration signal. Furthermore, there are many filtering methods; for other time-domain filtering and frequency-domain filtering (inverse fast Fourier transform), when the non-corrugated characteristic frequency band is fixed, the filtering results of different filtering methods will not differ significantly.

[0096] In this embodiment, the sound pressure level is extracted from the second sound signal to obtain sound features; the effective value of vibration acceleration is extracted from the second vibration signal to obtain vibration features. It can be understood that the A-weighted sound pressure level (referred to as the sound dB value) is extracted from the second sound signal, and the effective value of vibration acceleration (RMS) is extracted from the second vibration signal to obtain sound features and vibration features respectively. It should be noted that the sound pressure level is only one feature of the sound signal related to rail corrugation; this embodiment uses it as the sound feature of the sound signal. Other features characterizing corrugation can also be selected, and no specific limitation is made.

[0097] Step S13: Input the feature vector to be predicted, which is constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor can perform prediction processing on the feature vector to be predicted according to the preset decision rules and output the corresponding prediction value.

[0098] In this embodiment, the sound features, vibration features, and the vehicle speed and curve radius information of the rail vehicle running along the rail to be predicted are scalarized and then combined sequentially to obtain the feature vector to be predicted. The feature vector to be predicted is then input into a preset rail corrugation prediction model. It can be understood that, to correlate vehicle operating conditions, the vehicle speed and curve radius information, along with the sound and vibration features, are scalarized and then combined in a certain order, for example, in the order of [sound features, vibration features, vehicle speed information, curve radius information] to form the feature vector to be predicted. The reason for selecting vehicle speed and curve radius information as the operating conditions to be combined is that the vehicle speed determines the frequency of the wheel impact on the rail. When the frequency at a certain vehicle speed is the same as the natural frequency of the rail, it will cause an impact (resonance) between the wheel and the rail. The impact energy will accelerate the development of rail corrugation. The definitions of vehicle speed and frequency are: ;in, For vehicle speed, For wavelength, The frequency is used for information about curve radius. Curve radius information is used because the lateral creep force on curved sections is huge, resulting in intense wheel-rail interaction, which is the main cause of rail corrugation. Generally, the smaller the curve radius, the higher the risk and severity of rail corrugation.

[0099] In this embodiment, the preset rail wave depth prediction model assigns the feature vector to be predicted to multiple predictors, so that each predictor can perform prediction processing on the feature vector to be predicted according to the preset decision rules and output the corresponding predicted value. Before obtaining the preset rail wave depth prediction model, an initial rail wave depth prediction model needs to be constructed and trained using a training sample set. The specific process of obtaining the training sample set is as follows:

[0100] The system synchronously acquires initial sound and vibration signals of the rail vehicle during its operation along the rail under different historical operating conditions. It preprocesses these signals to obtain preprocessed sound and vibration signals. Based on the wheel speed and wheel diameter acquired synchronously with the initial sound and vibration signals, it calculates the kilometer marker information corresponding to the preprocessed sound and vibration signals to obtain preprocessed sound and vibration signals carrying kilometer marker information. Finally, it preprocesses the rail displacement waveform data acquired by the corrugation detection vehicle during its operation along the rail under different historical operating conditions, including kilometer marker calculation, to obtain preprocessed sound and vibration signals carrying kilometer marker information. The text describes a process for obtaining rail corrugation depth values. It details how to perform calibration on the pre-processed sound and vibration signals carrying kilometer marker information, and the rail corrugation depth values. The text then describes extracting historical sound and vibration features related to rail corrugation from these signals. Finally, it mentions constructing training samples based on these historical features, vibration features, historical operating condition information, and the calibration-aligned rail corrugation depth values, using all training samples under different operating conditions as the training sample set. It notes that the historical operating condition information includes two types: vehicle speed and track curve radius. Vehicle speed may have slight errors at different times, making it dynamic information; the track curve radius is static information and does not change over time.

[0101] Understandably, the initial sound and vibration signals under different historical working conditions are acquired synchronously, and then the rail displacement waveform data is collected by a corrugated steel testing vehicle. The synchronously acquired initial sound and vibration signals are then processed as follows: Figure 3 The preprocessing flow shown will not be elaborated further. After filtering the two signals to obtain the preprocessed sound and vibration signals, kilometer marker calculation is required to spatially match the onboard sensor data (sound and vibration data) with the corrugation detection vehicle data (rail displacement waveform data). Since the sound and vibration sensors collect waveforms that change over time, but rail corrugation is a geometric damage distributed along the longitudinal direction of the rail, the final requirement is to know the location (kilometer marker) and the depth of the corrugation. Therefore, by multiplying the rotational speed by the wheel diameter by π, the distance traveled by the wheel can be calculated, thus mapping the sampling time point to the distance traveled by the train. Combining this with the starting kilometer marker, each sound and vibration data point can be assigned a spatial coordinate, i.e., a kilometer marker.

[0102] Specifically, a rail displacement waveform data across the entire frequency band is collected during rail operation under different historical operating conditions using a wave-pattern inspection vehicle. The full-frequency rail displacement waveform data is then bandpass filtered to obtain rail displacement waveform data for each wavelength band. Based on the preset sampling window length for each wavelength band, the rail displacement waveform data for that wavelength band is subjected to sliding non-overlapping slicing to obtain waveform data slices. The kilometer marker information for each waveform data slice is calculated to obtain rail displacement waveform data carrying the kilometer marker information after slicing. The peak-to-peak value of the rail displacement waveform data carrying the kilometer marker information after each slice is calculated. The arithmetic mean of all peak-to-peak values ​​within the rail displacement waveform data carrying the kilometer marker information after slicing is taken as the rail wave depth value at the center point of the rail displacement waveform data carrying the kilometer marker information after slicing.

[0103] Understandably, further, such as Figure 4 As shown, the rail displacement waveform data collected by the corrugated mill is preprocessed. Specifically, the corrugated mill collects rail displacement waveform data with a sampling accuracy of 1 mm / point. The collected full-frequency rail displacement waveform data is bandpass filtered to obtain rail displacement waveform data for each wavelength band. Based on the sampling window length and depth of light (PPR) values ​​for different wavelength bands in Table 1, the rail displacement waveform data for each wavelength band is sliced ​​into non-overlapping slices according to its sampling window length. The kilometer marker information after slicing each waveform data is calculated to obtain the rail displacement waveform data carrying the kilometer marker information after slicing.

[0104] Table 1. Sampling window length and wavelength depth standards for different wavelength bands

[0105]

[0106] Calculate the peak-to-peak average (PPR) of the segmented displacement waveform data, which is also the wave depth value, such as... Figure 5 As shown, the PPR value is defined as follows:

[0107] ;

[0108] in, , , Each represents the length of a moving window. L The first wave depth value, the second wave depth value, and the nth wave depth value.

[0109] Then, when the corrugation inspection vehicle collects rail displacement waveform data, it includes both information related to rail corrugation and information unrelated to rail corrugation. For example, when the corrugation inspection vehicle passes over rail joints, welds, and turnouts, it can cause the calculated PPR value to be abnormally high, so this data needs to be discarded. Therefore, the PPR data corresponding to the abnormally high PPR value caused by rail joints, welds, etc., is discarded to obtain the preprocessed corrugation depth (PPR) value.

[0110] In this embodiment, the rail wave depth value is determined as follows: In one specific implementation, using the sample kilometer range of the preprocessed sound signal and preprocessed vibration signal as a reference interval, all rail wave depth values ​​located within the reference interval are searched and extracted; the average value of the rail wave depth values ​​is calculated, and this average value is used as the rail wave depth value after calibration with the sample kilometer range. It can be understood that, due to errors between the vehicle-mounted sound and vibration coordination system and the corrugation detection vehicle measurement system, it is necessary to calibrate the sound and vibration data with the rail wave depth data using kilometer benchmarks. The kilometer benchmark calibration flowchart is as follows... Figure 6 As shown, based on the preprocessed sound signal and preprocessed vibration signal, the starting kilometer marker S and ending kilometer marker E of a single sample are obtained to obtain the sample kilometer marker range [S, E]. The kilometer markers corresponding to the preprocessed wave depth data are traversed. If the kilometer marker corresponding to the wave depth (PPR) value is within the range of the sound and vibration sample kilometer markers [S, E], then the wave depth (PPR) value and its corresponding kilometer marker are recorded. The average value of all wave depth (PPR) values ​​within the range of the sound and vibration single sample kilometer markers [S, E] is calculated, and the calculated average values ​​are used as the corresponding wave depth values ​​of the sound and vibration single samples, respectively, to achieve the alignment of sound and vibration data with rail wave depth data kilometer markers.

[0111] Furthermore, in another specific embodiment, using the sample kilometer range of the preprocessed sound signal and preprocessed vibration signal as a reference interval, all rail wave depth values ​​located within the reference interval are found and extracted; the preset quantile amplitude of the rail wave depth values ​​is calculated as the rail wave depth value after calibration with the sample kilometer range. It is understood that due to errors between the vehicle-mounted sound and vibration coordination system and the corrugation detection vehicle measurement system, it is necessary to calibrate the sound and vibration data with the rail wave depth data using kilometer benchmarks. The kilometer benchmark calibration flowchart is as follows... Figure 7As shown, based on the preprocessed sound signal and preprocessed vibration signal, the starting kilometer marker S and ending kilometer marker E of a single sample are obtained to obtain the sample kilometer marker range [S, E]. The kilometer markers corresponding to the preprocessed wave depth data are traversed. If the wave depth (PPR) value corresponding to the kilometer marker is within the range of the sound and vibration sample kilometer markers [S, E], then the wave depth (PPR) value and its corresponding kilometer marker are recorded. Since the distribution of corrugation on the track is uneven, there may be a section of track with good condition and another section with corrugation. Averaging the entire range, the wave depth value (larger) corresponding to corrugation will be averaged with the wave depth value (smaller) corresponding to normal track conditions. Therefore, sections with corrugation will be underestimated. Thus, by calculating the sound... The preset quantile amplitude (N%) is calculated for all wave depth (PPR) values ​​within the kilometer marker [S, E] of the vibration single sample. The preset quantile ranges from 85% to 99%. In this embodiment, any quantile within the preset quantile range is used as the preset quantile, and the wave depth value of that quantile is used as the preset quantile amplitude. Taking a preset quantile of 85% as an example, the corresponding preset quantile amplitude is the target wave depth value of 85%. If the wave depth amplitude of 85% of the measurement points is less than or equal to the target wave depth value, and the wave depth value of 5% of the measurement points is greater than or equal to the target wave depth value, then the target wave depth value is used as the corresponding wave depth value for the sound and vibration single samples, thereby achieving the alignment of sound and vibration data with the rail wave depth data kilometer marker.

[0112] In this embodiment, after acquiring the sound signal, vibration signal and rail corrugation depth data after the kilometer markers are aligned, the working condition information is further extracted and associated. Specifically, when establishing the rail corrugation depth prediction model, it is necessary to extract the relevant features that characterize rail corrugation. Commonly used features are time-domain features and frequency-domain features. For example, time-domain features mainly include features such as mean, peak value, effective value and kurtosis, while frequency-domain features mainly include frequency domain effective value, spectral energy value and so on.

[0113] Based on the preprocessed sound and vibration signals, sound pressure level and effective vibration values ​​are extracted, and coupled with operating condition information such as vehicle speed and track curve radius, these are used as input features for the prediction model to obtain training samples. Sound pressure level characterizes the overall noise level caused by rail corrugation; the more severe the rail corrugation, the higher the sound pressure level. The A-weighted sound pressure level over continuous time is defined as follows:

[0114] ;

[0115] in, Indicates A-weighted sound pressure. The reference sound pressure level is set to a value of [value missing]. .

[0116] Effective value of vibration acceleration: Characterizes the overall level of vibration energy; the more severe the rail corrugation, the greater the effective value of acceleration. The effective value of vibration acceleration over continuous time is defined as follows:

[0117] ;

[0118] in, This represents the amplitude of the vibration acceleration sample point. Indicates the length of the sample.

[0119] Vehicle speed: The vehicle's speed determines the frequency of the wheels impacting the rails. When the frequency at a certain vehicle speed matches the natural frequency of the rails, it will trigger an impact (resonance) between the wheels and the rails. The impact energy will accelerate the development of rail corrugation. The definitions of vehicle speed and frequency are as follows:

[0120] ;

[0121] in, For vehicle speed, For wavelength, For frequency.

[0122] Curve radius: Due to the huge lateral creep force in curved sections, the interaction between the wheel and rail is intense, which is the main cause of rail corrugation. Generally speaking, the smaller the curve radius, the higher the risk and the more severe the rail corrugation.

[0123] The model associates operating condition characteristics with sound and vibration characteristics. Based on historical sound features, historical vibration features, historical operating condition information, and rail wave depth values ​​after kilometer marker alignment, a training sample is constructed. This training sample is then used to further construct a training sample set. After obtaining the training sample set, the initial rail wave depth prediction model based on the random forest model is trained using this set. The training model learns the nonlinear mapping relationship between historical sound features, historical vibration features, historical operating condition information, and rail wave depth values ​​after kilometer marker alignment, thus obtaining the constructed preset rail wave depth prediction model. It can be understood that the random forest model improves overall prediction accuracy by constructing multiple decision trees (predictors) and combining the prediction results of each decision tree (predictor). A random forest consists of different types and numbers of decision trees (predictors), each of which is independent.

[0124] Specifically, multiple training subsets are randomly selected from the training sample set using a bootstrap sampling method; one or more of the historical sound features, historical vibration features, and historical working condition information are selected from the training subsets to construct a feature subset; based on the feature subset, recursively processing is performed according to the rule of maximizing node impurity reduction to achieve node splitting, until a preset stopping growth condition is met, thus completing the construction of a predictor; the constructed multiple predictors are used as the preset rail wave depth prediction model after training. Specifically, for each node to be split in the decision tree of the random forest model, each feature in the current feature subset and all candidate split points are traversed; the impurity of the child node set corresponding to each candidate split point is calculated; based on the feature that maximizes the reduction of the parent node's impurity and the target candidate split point, a rule for maximizing node impurity reduction for the node to be split is constructed, and recursive processing is performed on the node to be split according to the rule of maximizing node impurity reduction to achieve node splitting. It can be understood that the specific process of training the model to learn the nonlinear mapping relationship between historical sound features, historical vibration features, historical working condition information, and the rail wave depth value after kilometer benchmark alignment is as follows: Figure 8 As shown, a bootstrap sampling method is used in the training sample set to randomly select multiple training subsets (each subset is the same size as the original dataset, and repeated sampling is allowed) to train different decision trees (predictors). When splitting at each node of each decision tree, only a specific number of features are randomly selected instead of using all features, which increases the diversity between trees. A complete decision tree is built for each training subset and trained based on a split point of the selected features until a stopping condition is met, such as reaching the maximum depth, the number of samples in the node being less than a certain threshold, or being unable to continue splitting. Each decision tree makes predictions on new data to obtain the predicted values ​​output by each decision tree.

[0125] Step S14: Determine and output the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors.

[0126] In this embodiment, the average of the predicted values ​​of all decision trees is taken as the predicted wave depth value of the rail to be predicted in the output of the random forest model.

[0127] In this embodiment, the predicted wave depth value is compared with the standard value of rail wave depth for the current wavelength band; if the predicted wave depth value exceeds the standard value of rail wave depth for the current wavelength band, a grinding and maintenance alarm message is generated for the corresponding mileage location of the rail to be predicted.

[0128] As can be seen, this application discloses a method for simultaneously acquiring sound and vibration signals of a rail vehicle running along a rail to be predicted; extracting sound and vibration features related to rail corrugation from the sound and vibration signals; inputting a feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into a preset rail corrugation prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor performs prediction processing on the feature vector to be predicted according to a preset decision rule and outputs a corresponding prediction value; and determining and outputting the predicted corrugation value of the rail to be predicted based on the prediction values ​​output by each predictor. Therefore, by simultaneously acquiring sound and vibration signals and extracting sound and vibration features related to rail corrugation, the synchronous acquisition means simultaneously obtaining the structural response and acoustic fingerprint of the corrugation excitation, forming a two-dimensional, complementary representation of the same corrugation. This provides more comprehensive feature information than relying solely on a single-dimensional vibration signal. Furthermore, combining acoustic and vibration features with operating condition information allows the model to understand the operating conditions under which the signal was generated. This enables the model to isolate or compensate for the influence of operating condition changes on the acoustic and vibration signals during the prediction process. Based on the learned real and stable physical relationship between acoustic and vibration features and wave depth under different speeds and curves, the model can predict and output high-precision wave depth values.

[0129] In this embodiment, taking the collection of acoustic and vibration co-location data and rail displacement waveform data from a corrugation detection vehicle in section A of a certain line as an example, we illustrate the method for predicting rail wave depth based on a random forest regression model. Table 2 below shows the detailed information of the test data.

[0130] Table 2 Test Data

[0131]

[0132] Preprocessing was performed on the acoustic-vibration coordination data and rail displacement waveform data from the corrugation detection vehicle in section A of Table 2 above. The wavelength in this section is 30~100mm. The sliding step size and sliding window length for calculating the wave depth (PPR) value of the rail displacement waveform data were both 600mm. Figure 9(a) shows the wave depth (PPR) trend before removing rail gaps and welds, and Figure 9(b) shows the wave depth (PPR) trend before and after removing rail gaps and welds, which cause the wave depth (PPR) value to be larger. As can be seen from Figures 9(a) and 9(b), the wave depth values ​​of the two sections K34+800~K35+000 and K35+800~K35+900 are greater than the standard rail wave depth value for the 30~100mm wavelength range, therefore, rail corrugation exists in these sections.

[0133] The preprocessed acoustic-vibration co-processing data and rail wave depth data were compared using kilometer-scale benchmarks. The wave depth trends before and after the comparison with the acoustic-vibration co-processing data are as follows: Figure 10As shown in Figure 11(a), the trend of sound dB value and rail corrugation depth (PPR) value after kilometer marker alignment is shown in Figure 11(b), and the trend of vibration effective value and rail corrugation depth (PPR) value after kilometer marker alignment is shown in Figure 11(a) and Figure 11(b). It can be seen from Figures 11(a) and 11(b) that the trends of sound dB value, vibration effective value, and rail corrugation depth after kilometer marker alignment are generally consistent, thus indicating that sound dB value and vibration effective value can effectively reflect rail corrugation conditions.

[0134] The sound dB value, vibration effective value, vehicle speed, track curve radius, and rail wave depth (PPR) value of the single sample data after kilometer benchmark alignment were used as input features for the random forest model. There are a total of 85 samples in this interval. 90% of the data (77 samples) were used for training the random forest model, and 10% of the data (8 samples) were used for testing the random forest model.

[0135] Figures 12(a) and 12(b) show the prediction results of the constructed random forest model on the training and test sets. As can be seen from Figures 12(a) and 12(b), the predicted wave depth results of the model on both the training and test sets are basically consistent with the actual rail wave depth results, with only a slight deviation in amplitude. To further illustrate the accuracy of the predicted wave depth of the constructed random forest model, the root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (COP) of the model prediction results on both the training and test sets were calculated. The results are shown in Tables 3 and 4, respectively. As can be seen from Tables 3 and 4, the root mean square error and mean absolute error calculated for both the training and test sets are relatively small, and the coefficient of determination is relatively large (close to 1), indicating that the random forest model has high accuracy in predicting wave depth.

[0136] Table 3. Evaluation metrics for the model used to calculate predicted wave depth values ​​on the training set.

[0137]

[0138] Table 4. Evaluation metrics for the model used to calculate predicted wave depth values ​​on the test set.

[0139]

[0140] The root mean square error (RMSE) measures the deviation between the predicted and actual values. A smaller RMSE indicates a smaller deviation and higher accuracy of the prediction model. It is defined as follows:

[0141] ;

[0142] in, It is the quantity of actual values. It is the first One actual value, It is the first One predicted value.

[0143] Mean Absolute Error (MAE) represents the average of the absolute differences between predicted and actual values, and is defined as follows:

[0144] ;

[0145] in, It is the absolute value of the difference between the actual value and the predicted value.

[0146] Coefficient of determination ( A metric used to measure a model's ability to explain the variability of actual values. A larger value indicates that the model can better explain the variation of the target variable, and the model fit is better. It is defined as:

[0147] ;

[0148] in, This represents the average of the actual values.

[0149] Therefore, constructing a random forest-based rail corrugation depth prediction model can improve the detection efficiency and accuracy of rail corrugation, replacing measuring instruments and reducing the inspection and maintenance costs for track maintenance departments. It can predict the rail condition in corrugated sections, anticipate the subsequent development of rail corrugation, and formulate reasonable rail grinding strategies, such as grinding time and amount, to achieve precise rail grinding and extend rail service life. Analysis of test data shows that the constructed rail corrugation depth prediction model has high prediction accuracy and largely matches the actual rail corrugation depth data.

[0150] like Figure 13 As shown, the present invention also discloses a rail wave depth prediction device, comprising:

[0151] Signal acquisition module 11 is used to synchronously acquire sound and vibration signals of the rail vehicle running along the rail to be predicted;

[0152] Feature extraction module 12 is used to extract sound features and vibration features related to rail corrugation from the sound signal and the vibration signal;

[0153] Prediction module 13 is used to input the feature vector to be predicted, which is constructed based on the sound features, the vibration features and the working condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor performs prediction processing on the feature vector to be predicted according to the preset decision rules and outputs the corresponding prediction value.

[0154] The result output module 14 is used to determine and output the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors.

[0155] Therefore, compared to current methods that use measuring instruments to measure the actual geometry of the rail surface, rail corrugation detection offers higher accuracy but lower efficiency. Vehicle-mounted rail corrugation detection methods can effectively overcome the low efficiency of measuring instruments, but they still suffer from inaccuracies in quantitative analysis. By simultaneously acquiring vibration and sound signals, as well as rail displacement data collected by the corrugation detection vehicle, and starting from the fault mechanism, a machine learning-based rail corrugation depth prediction method is established by extracting sound and vibration characteristics related to rail corrugation and corroding them with operating condition information such as vehicle speed and track curve radius. This method can effectively improve the detection efficiency and quantitative diagnostic accuracy of rail corrugation.

[0156] Furthermore, embodiments of this application also disclose an electronic device, Figure 14 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.

[0157] Figure 14 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 rail wave depth prediction method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0158] 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.

[0159] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0160] 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.

[0161] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the rail wave depth prediction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0162] 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 rail wave depth prediction method. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0163] 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.

[0164] 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. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0165] 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.

[0166] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. 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 the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting rail wave depth, characterized in that, include: Simultaneously acquire sound and vibration signals of the rail vehicle running along the rail to be predicted; Extract the acoustic and vibration features related to rail corrugation from the acoustic and vibration signals; The feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, is input into the preset rail wave depth prediction model so that the feature vector to be predicted is assigned to multiple predictors, so that each predictor performs prediction processing on the feature vector to be predicted according to the preset decision rules and outputs the corresponding prediction value. The predicted wave depth value of the rail to be predicted is determined and output based on the predicted values ​​output by each of the predictors. The extraction of sound and vibration features related to rail corrugation from the sound and vibration signals includes: The sound signal and the vibration signal are respectively subjected to DC removal processing to obtain the first sound signal and the first vibration signal, respectively; The frequency components of the non-rail corrugation characteristic frequency band in the first sound signal and the first vibration signal are filtered to obtain the second sound signal and the second vibration signal. Extract the acoustic features related to rail corrugation from the second acoustic signal; Extract vibration features related to rail corrugation from the second vibration signal; The step of filtering the frequency components of the non-rail corrugation characteristic frequency bands in the first sound signal and the first vibration signal to obtain the second sound signal and the second vibration signal includes: Configure the non-corrugating characteristic frequency band corresponding to the rail running process to be predicted; The first sound signal and the first vibration signal are band-stop filtered using the non-wave-slip characteristic frequency band to obtain the second sound signal and the second vibration signal. Accordingly, extracting sound features related to rail corrugation from the second sound signal and extracting vibration features related to rail corrugation from the second vibration signal includes: The sound pressure level is extracted from the second sound signal to obtain sound characteristics; The effective value of vibration acceleration is extracted from the second vibration signal to obtain the vibration characteristics; The step of inputting the feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model includes: The sound features, vibration features, and information on the vehicle speed and curve radius of the rail vehicle as it runs along the rail to be predicted are scalarized and then combined in sequence to obtain the feature vector to be predicted. The feature vector to be predicted is input into the preset rail wave depth prediction model; Before inputting the feature vector to be predicted, constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, the method further includes: The initial rail wave depth prediction model based on the random forest model was trained using the training sample set. The training model learned the nonlinear mapping relationship between historical sound features, historical vibration features, historical working condition information and the rail wave depth value after kilometer benchmark alignment, so as to obtain the preset rail wave depth prediction model. The process of training the initial rail wave depth prediction model based on the random forest model using the training sample set includes: Multiple training subsets are randomly selected from the training sample set using a self-sampling method. Select any one or more of the historical sound features, the historical vibration features, and the historical operating condition information from the training subset to construct a feature subset; Based on the feature subset, recursively process according to the rule of maximizing the reduction of node impurity to achieve node splitting until the preset stopping growth condition is met, thus completing the construction of a predictor; The constructed multiple predictors are used as the trained preset rail wave depth prediction model.

2. The rail wave depth prediction method according to claim 1, characterized in that, Before training the initial rail wave depth prediction model based on the random forest model using the training sample set, the following steps are also included: Simultaneously acquire the initial sound and vibration signals of the rail vehicle running along the rail under different historical operating conditions; The initial sound signal and the initial vibration signal are preprocessed to obtain the preprocessed sound signal and the preprocessed vibration signal. Based on the wheel speed and wheel diameter acquired synchronously with the initial sound signal and the initial vibration signal, the kilometer marker information corresponding to the preprocessed sound signal and the preprocessed vibration signal is calculated to obtain the preprocessed sound signal and the preprocessed vibration signal carrying the kilometer marker information. Preprocessing, including kilometer mark calculation, is performed on the rail displacement waveform data collected by the corrugation detection vehicle during its operation along the rail under different historical working conditions to calculate the rail wave depth value carrying kilometer mark information. The preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information are compared and processed to obtain the preprocessed sound signal, the preprocessed vibration signal, and the rail wave depth value after kilometer marker comparison. Historical sound features and historical vibration features related to rail corrugation were extracted from the preprocessed sound signal after the kilometer marker alignment and the preprocessed vibration signal after the kilometer marker alignment, respectively. Training samples are constructed based on the historical sound features, historical vibration features, historical working condition information, and rail wave depth values ​​after the kilometer marker is aligned, and all training samples under different working conditions are used as the training sample set.

3. The rail wave depth prediction method according to claim 2, characterized in that, The preprocessing of rail displacement waveform data collected by the corrugation detection vehicle during its operation along the rail under different historical operating conditions, including kilometer marker calculation, yields the rail wave depth value carrying kilometer marker information, including: The rail displacement waveform data of the whole frequency band was collected by the corrugated testing vehicle during the operation of the rail under different historical working conditions. The rail displacement waveform data of the entire frequency band is subjected to bandpass filtering to obtain rail displacement waveform data of each wavelength band. According to the preset sampling window length of each wavelength band, the rail displacement waveform data of the corresponding wavelength band is processed by sliding non-overlapping slicing to obtain each waveform data slice, and the kilometer marker information of each waveform data slice is calculated to obtain the rail displacement waveform data carrying the kilometer marker information after slicing. Calculate the peak-to-peak value of the rail displacement waveform data carrying kilometer marker information after each slice; The arithmetic mean of all peak-to-peak values ​​in the sliced ​​rail displacement waveform data carrying kilometer marker information is taken as the rail wave depth value of the center point of the sliced ​​rail displacement waveform data carrying kilometer marker information.

4. The rail wave depth prediction method according to claim 2, characterized in that, The process of aligning the preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information includes: Using the sample kilometer range of the preprocessed sound signal and the preprocessed vibration signal as the reference interval, find and extract all rail wave depth values ​​located within the reference interval; Calculate the preset quantile amplitude of the rail wave depth value, and use it as the rail wave depth value after being aligned with the sample kilometer range.

5. The rail wave depth prediction method according to claim 2, characterized in that, The process of aligning the preprocessed sound signal carrying kilometer marker information, the preprocessed vibration signal carrying kilometer marker information, and the rail wave depth value carrying kilometer marker information includes: Using the sample kilometer range of the preprocessed sound signal and the preprocessed vibration signal as the reference interval, find and extract all rail wave depth values ​​located within the reference interval; Calculate the average value of the rail wave depth, and use the average value as the rail wave depth value after being aligned with the sample kilometer range.

6. The rail wave depth prediction method according to claim 1, characterized in that, The recursive processing based on the feature subset according to the rule of maximizing the reduction of node impurity to achieve node splitting includes: For each node to be split in the decision tree of the random forest model, traverse every feature in the current feature subset and all candidate split points; Calculate the impurity of the set of child nodes corresponding to each of the candidate split points; Based on the features that maximize the reduction of the impurity of the parent node and the target candidate split point, a rule for maximizing the reduction of the node impurity of the node to be split is constructed, and recursive processing of the node to be split is performed according to the rule for maximizing the reduction of the node impurity to achieve node splitting.

7. The rail wave depth prediction method according to claim 1, characterized in that, After determining and outputting the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors, the method further includes: The predicted wave depth value is compared with the standard value of rail wave depth for the current wavelength band; If the predicted wave depth value exceeds the standard value of the rail wave depth for the current wavelength band, a grinding and maintenance alarm message is generated for the corresponding mileage location of the rail to be predicted.

8. A rail wave depth prediction device, characterized in that, include: The signal acquisition module is used to simultaneously acquire the sound and vibration signals of the rail vehicle running along the rail to be predicted. The feature extraction module is used to extract sound features and vibration features related to rail corrugation from the sound signal and the vibration signal; The prediction module is used to input the feature vector to be predicted, which is constructed based on the sound features, the vibration features, and the operating condition information of the rail vehicle running along the rail to be predicted, into the preset rail wave depth prediction model, so as to distribute the feature vector to be predicted to multiple predictors, so that each predictor can perform prediction processing on the feature vector to be predicted according to the preset decision rules and output the corresponding prediction value. The result output module is used to determine and output the predicted wave depth value of the rail to be predicted based on the predicted values ​​output by each of the predictors. The feature extraction module is specifically used to perform DC removal processing on the sound signal and the vibration signal respectively to obtain a first sound signal and a first vibration signal; to filter the frequency components of the non-rail corrugation characteristic frequency band in the first sound signal and the first vibration signal to obtain a second sound signal and a second vibration signal; to extract the sound features related to rail corrugation from the second sound signal; and to extract the vibration features related to rail corrugation from the second vibration signal. The rail wave depth prediction device is further configured to configure a non-wavelength characteristic frequency band corresponding to the rail running process to be predicted; to perform band-stop filtering on the first sound signal and the first vibration signal using the non-wavelength characteristic frequency band to obtain a second sound signal and a second vibration signal; and to extract the sound pressure level from the second sound signal to obtain sound characteristics. The effective value of vibration acceleration is extracted from the second vibration signal to obtain the vibration characteristics; The prediction module is specifically used to scalarize and process the sound features, vibration features, and vehicle speed and curve radius information of the rail vehicle running along the rail to be predicted, and then combine them in sequence to obtain the feature vector to be predicted; and input the feature vector to be predicted into the preset rail wave depth prediction model. The rail wave depth prediction device is also used to train the initial rail wave depth prediction model based on the random forest model using the training sample set, so as to train the model to learn the nonlinear mapping relationship between historical sound features, historical vibration features and historical working condition information and the rail wave depth value after the kilometer marker is aligned, so as to obtain the preset rail wave depth prediction model. The rail wave depth prediction device is further configured to randomly extract multiple training subsets from the training sample set using a self-sampling method; select any one or more of the historical sound features, historical vibration features, and historical working condition information from the training subsets to construct a feature subset; recursively process the feature subsets according to the rule of maximizing node impurity reduction to achieve node splitting until a preset stopping growth condition is met, thereby completing the construction of a predictor; and use the constructed multiple predictors as the preset rail wave depth prediction model after training.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the rail wave depth prediction 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; wherein, when the computer program is executed by a processor, it implements the steps of the rail wave depth prediction method as described in any one of claims 1 to 7.

Images