Metro rail corrugation measurement method based on seq2seq model and vibration data fusion

Abstract

The application provides a subway rail corrugation measurement method based on a Seq2Seq model and vibration and noise data fusion. The method comprises the following steps: constructing a time series data set corresponding to the subway train floor longitudinal acceleration and train speed based on subway train historical data; constructing a subway train mileage matching prediction model based on a Seq2Seq model, training the subway train mileage matching prediction model by using the time series data set, inputting the in-train floor longitudinal acceleration of the subway train to be detected into the trained subway train mileage matching prediction model, and obtaining the running speed of the subway train to be detected; calculating a rail corrugation vibration and noise comprehensive index according to the in-train vibration and noise data of the subway train to be detected, and determining the wavelength and amplitude information corresponding to the corrugation. The model accurately predicts the speed and mileage of the subway train through the vehicle body longitudinal acceleration, and accurately identifies the characteristic wavelength and wave depth amplitude of the rail corrugation by using the in-train floor vertical vibration acceleration and in-train sound pressure signal fusion.

Description

Technical Field

[0001] This invention relates to the field of rail corrugation measurement technology, and in particular to a method for measuring rail corrugation in subways based on the fusion of a Seq2Seq model and vibration and noise data. Background Technology

[0002] Rail corrugation refers to a periodic, wavy, uneven wear pattern that appears along the longitudinal surface of the rail. In recent years, with the rapid development of urban rail transit, rail corrugation has become more severe in subway lines due to factors such as diverse track types, small curve radii, frequent starting and braking, and high operating density. This leads to a series of problems, including fastener breakage, vehicle suspension fatigue, in-vehicle vibration and noise, and secondary noise from buildings in residential areas along the line. Currently, rail grinding remains the most economical and effective measure for controlling rail corrugation. However, when scheduling rail grinding on-site, it is often necessary to select operational parameters such as grinding mileage, grinding depth, and number of grinding passes. In the past, the rail corrugation status was mostly obtained by maintenance personnel during maintenance windows using static measurement methods such as corrugation trolleys and steel rulers. The grinding mileage and other operational parameters were then manually selected based on vibration and noise complaint sections. This approach not only occupies maintenance personnel and has low inspection efficiency, but also often only schedules grinding after passenger complaints, severely impacting the passenger experience and hindering the green development of rail transit. In addition, due to the limited number of rail grinding vehicles, how to schedule grinding times according to the rail corrugation condition and make reasonable use of rail grinding vehicles to achieve the transformation from "planned maintenance" to "condition-based maintenance" is also an important issue faced by the on-site operations department.

[0003] Currently, several intelligent identification methods for rail corrugation exist in existing technologies. However, due to the weak GPS positioning signal when the subway is running in tunnels, and the difficulty in easily obtaining train speed measurement data for wheelsets, the existing intelligent identification algorithms for rail corrugation can only roughly identify whether a certain section has corrugation. Moreover, they mostly rely on axle box acceleration or acoustic signals for identification, which also presents problems such as difficulty in obtaining raw data and low testing efficiency in field applications. Therefore, there is an urgent need for a high-precision and high-efficiency rapid measurement method for subway rail corrugation that can simultaneously identify key information such as the characteristic wavelength and depth amplitude of rail corrugation, as well as the location of corrugation. Summary of the Invention

[0004] The embodiments of the present invention provide a method for measuring rail corrugation in subways based on the fusion of a Seq2Seq model and vibration and noise data, so as to achieve accurate identification of the characteristic wavelength and wave depth amplitude of rail corrugation.

[0005] To achieve the above objectives, the present invention adopts the following technical solution.

[0006] A method for measuring rail corrugation in subways based on the fusion of a Seq2Seq model and vibration and noise data includes:

[0007] A time-series dataset of floor longitudinal acceleration and train speed was constructed based on historical subway train data.

[0008] A metro train mileage matching prediction model based on the Seq2Seq model is constructed. The metro train mileage matching prediction model based on the Seq2Seq model is trained using the time series dataset corresponding to the longitudinal acceleration and train speed to obtain the trained metro train mileage matching prediction model.

[0009] The longitudinal acceleration of the floor inside the subway train to be detected is input into the trained subway train mileage matching prediction model based on the Seq2Seq model to obtain the running speed of the subway train to be detected.

[0010] The rail corrugation vibration and noise comprehensive index of the subway train to be tested is calculated based on the preset distribution range of rail corrugation wavelength and the running speed, floor vertical vibration acceleration and in-vehicle sound pressure signal of the subway train to be tested. The specific location of corrugation and the corresponding characteristic wavelength and wave depth amplitude are determined based on the corrugation vibration and noise comprehensive index.

[0011] Preferably, the construction of the time series dataset corresponding to floor longitudinal acceleration and train speed based on historical subway train data includes:

[0012] Based on historical data of subway trains, vertical and longitudinal vibration acceleration signals of the floor surface at the location of the bogie inside the subway train are obtained, and time-domain information of vertical and longitudinal vibration of the floor inside the train is obtained based on the vertical and longitudinal vibration acceleration signals.

[0013] Acquire the measured sound pressure signal inside the subway train at the location of the bogie, and obtain the time-domain information of the sound pressure inside the train based on the sound pressure signal;

[0014] Acquire the measured speed signal from the speed measuring wheelset of the subway train, and obtain the train speed signal based on the speed signal;

[0015] Based on the preset window length, the longitudinal vibration acceleration signal and train speed signal of the floor surface are segmented into segmented floor longitudinal acceleration and train speed signals, generating a time series dataset corresponding to the longitudinal vibration acceleration signal and train speed.

[0016] Preferably, the step of segmenting the longitudinal vibration acceleration signal and train speed signal of the floor surface into segmented floor longitudinal acceleration and train speed signals according to a preset window length, and generating a time series dataset corresponding to the longitudinal vibration acceleration signal and train speed, includes:

[0017] The longitudinal vibration acceleration signal and the train speed signal are resampled to ensure that the longitudinal acceleration time series and the train speed time series have the same length, denoted as L. s ;

[0018] The longitudinal acceleration time series is x i (i = 0, 1, ..., x) n The train speed time series is y. j (j = 0, 1, ..., y) n If the time step is set to K, the number of samples in the constructed dataset is determined by equation (1).

[0019] N = L s -K+1 (1)

[0020] The time series dataset corresponding to the longitudinal vibration acceleration signal and the train speed is determined by equation (2);

[0021]

[0022] Preferably, the construction of the metro train mileage matching prediction model based on the Seq2Seq model involves training the metro train mileage matching prediction model based on the Seq2Seq model using the time series dataset corresponding to the longitudinal acceleration and train speed to obtain a trained metro train mileage matching prediction model, including:

[0023] A metro train mileage matching prediction model based on the Seq2Seq model is constructed. The metro train mileage matching prediction model belongs to the Many to Many type of recurrent neural network. The encoder part uses a recurrent neural network model to encode the longitudinal acceleration time series and outputs the hidden state of the last step. The decoder is also composed of a recurrent neural network and uses the last hidden state of the encoder and the train speed time series as input.

[0024] The time series datasets corresponding to the longitudinal acceleration and train speed are divided into training set, validation set and test set. The training set is input into the subway train mileage matching prediction model, and the optimal parameters of the model are found through K-fold cross-validation.

[0025] The validation set is input into the metro train mileage matching prediction model using the optimal parameters of the model to validate the metro train mileage matching prediction model. After successful validation, a trained metro train mileage matching prediction model based on the Seq2Seq model is obtained. The input data of the trained metro train mileage matching prediction model based on the Seq2Seq model is the time series of longitudinal acceleration of the metro train floor, and the output data is the time series of the corresponding metro train running speed.

[0026] Preferably, the step of calculating the comprehensive index of rail corrugation vibration and noise of the subway train under test based on the preset distribution range of rail corrugation wavelength and the running speed, floor vertical vibration acceleration, and in-vehicle sound pressure signal of the subway train under test includes:

[0027] The system acquires the operating speed, floor vertical vibration acceleration signal, and in-vehicle sound pressure signal of the subway train to be tested.

[0028] Based on the preset distribution range of the corrugated wavelength of the subway rails and the train speed of the subway train to be tested, the vertical vibration acceleration of the floor and the sound pressure signal inside the train to be tested are segmented and bandpass filtered.

[0029] The vibration index and noise index corresponding to the filtered floor vertical vibration acceleration signal and the in-vehicle sound pressure signal are calculated. Floor vertical vibration acceleration signals and in-vehicle sound pressure signals with vibration index and noise index greater than preset thresholds are retained. The combined wave and noise index is further calculated on the retained floor vertical vibration acceleration signals and in-vehicle sound pressure signals.

[0030] Preferably, the step of performing segmented bandpass filtering on the vertical vibration acceleration of the floor and the sound pressure signal inside the metro train to be tested, based on a preset distribution range of metro rail corrugation wavelengths and the train's operating speed, includes:

[0031] Assume that the vertical vibration acceleration signal of the subway train floor and the sound pressure signal inside the train have the same length, denoted as L. v The segment length is set to W, and the overlap rate is set to 3 / 4. If the remaining signal length is greater than or equal to W, zeros are padded to make the length equal to W; otherwise, the remaining signal is discarded.

[0032] According to the preset distribution range of subway rail corrugation wavelength [λ] L ,λ H Find the corresponding upper and lower cutoff frequencies [f]. L ,f H A bandpass filter based on the window function method is designed with an order of n. This filter is used to perform bandpass filtering on the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal of the subway train to be detected, obtaining the filtered segmented floor vertical vibration acceleration signal sig. vb and the in-vehicle sound pressure signal sig no ;

[0033] The upper and lower cutoff frequencies are determined by equation (3);

[0034]

[0035] In the formula, The average speed of segmented trains is determined based on the metro train mileage matching prediction model.

[0036] Preferably, the calculation of the vibration index and noise index corresponding to the filtered floor vertical vibration acceleration signal and the in-vehicle sound pressure signal includes:

[0037] Calculate the vertical vibration acceleration signal sig of the segmented floor. vb The self-power spectral density, in [f L ,f H Finding the peak frequency f in the frequency band vmax According to [f vmax -D,f vmax +D] Continue to analyze the segmented vertical vibration acceleration signal sig vb With bandpass filtering, the vibration index is determined by equation (4);

[0038]

[0039] In the formula, VI(i) is the vibration index of the i-th frame of the floor vertical vibration acceleration signal, and sig vb This is the vibration signal after the first filtering. `filter` is a bandpass filter, `D` is half the passband width, and `f`... vmax S is the peak frequency in the power spectral density of the vibration signal in the i-th frame. Vtotal S represents the effective value of the vibration signal after the first filtering. Vi The effective value of the vibration signal after the second filtering is determined by equation (5);

[0040]

[0041] In the formula, W V The length of each frame of the vibration signal;

[0042] Calculate the segmented in-vehicle sound pressure signal sig no The self-power spectral density, in [f L ,f H Finding the peak frequency f in the frequency band nmax According to [f nmax -D,f nmax +D] Continue to analyze the segmented in-vehicle sound pressure signal sig no For bandpass filtering, the noise figure is determined by equation (6);

[0043]

[0044] In the formula, NI(i) is the noise index of the i-th frame of the noise signal, and sig no This is the noise signal after the first filtering. `filter` is a window function bandpass filter, `D` is half the passband width, and `f`... nmaxS is the peak frequency of the noise signal in the i-th frame from its power spectral density. Ntotal S represents the effective value of the noise signal after the first filtering. Ni The effective value of the noise signal after the second filtering is determined by equation (7);

[0045]

[0046] In the formula, W N The length of each frame of noise signal.

[0047] Preferably, the retained floor vertical vibration acceleration signals and in-vehicle sound pressure signals with vibration and noise indices greater than preset thresholds are further used to calculate a comprehensive wave-slip noise index, including:

[0048] The vibration index threshold is determined to be VI. m The noise figure threshold is NI. m If the vibration index VI(i) corresponding to the vertical vibration acceleration signal of the segmented floor is greater than VI m Furthermore, the noise index NI(i) corresponding to the segmented in-vehicle sound pressure signal is greater than NI. m Then calculate the cross power spectrum of the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal;

[0049] In [f L ,f H Finding the peak frequency f in the frequency band vnmax According to [f vnmax -D,f vnmax +D] For bandpass filtering of the acoustic-vibration composite signal, the wave-slip noise composite index is determined by equation (9);

[0050]

[0051] In the formula, CVNI(i) is the combined wave-scraping noise index of the i-th frame, and sig vn The signal is a composite acoustic-vibration signal, filter is a window function bandpass filter, D is half the passband width, and f vnmax S is the peak frequency in the cross-power spectrum of the i-th frame acoustic vibration signal. VNtotal S represents the effective value of the acoustic-vibration composite signal after the first filtering. VNi The effective value of the acoustic-vibration composite signal after the second filtering is determined by equation (10);

[0052]

[0053] Preferably, determining the specific location of wave wear and the corresponding characteristic wavelength and wave depth amplitude based on the wave wear noise composite index includes:

[0054] The train speed v(t) is obtained from the metro train mileage matching prediction model based on the Seq2Seq model, and the train position is determined by equation (12).

[0055]

[0056] Based on the peak frequency f corresponding to the segmented acoustic-vibration composite signal vnmax Based on the average vehicle speed V, the wavelength of rail corrugation can be identified.

[0057] Determined by equation (13);

[0058]

[0059] The severity of rail corrugation, i.e., the range of corrugation depth amplitude, is identified based on the corrugation noise comprehensive index corresponding to the segmented acoustic-vibration composite signal, and is determined by equation (14):

[0060]

[0061] As can be seen from the technical solutions provided by the embodiments of the present invention described above, the present invention uses longitudinal acceleration of the in-vehicle floor and time series data of train speed to train a metro train mileage matching prediction model based on the Seq2Seq model. The longitudinal acceleration of the vehicle body is used to accurately predict the speed and mileage of the metro train. At the same time, the vertical vibration acceleration of the in-vehicle floor and the sound pressure signal inside the vehicle are fused to achieve accurate identification of the characteristic wavelength and wave depth amplitude of rail corrugation.

[0062] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of the invention. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram illustrating the corrugation of subway rails, provided as an embodiment of the present invention.

[0065] Figure 2 The flowchart illustrates a rapid measurement method for subway rail corrugation based on the fusion of a Seq2Seq model and vibration and noise data, provided as an embodiment of the present invention.

[0066] Figure 3This is a structural diagram of a subway train mileage matching prediction model based on a Seq2Seq model, provided in an embodiment of the present invention.

[0067] Figure 4 This is a schematic diagram of the basic components of an ATP system for a subway train, provided as an embodiment of the present invention.

[0068] Figure 5 This is a schematic diagram of a vehicle floor vertical and longitudinal vibration acceleration and vehicle sound pressure test provided in an embodiment of the present invention.

[0069] Figure 6 This is a schematic diagram of floor longitudinal acceleration and train speed data segmentation provided in an embodiment of the present invention.

[0070] Figure 7 This is a schematic diagram of bandpass filtering for vertical vibration acceleration signals of a floor, provided as an embodiment of the present invention.

[0071] Figure 8 This is a schematic diagram of a vibration index calculation result provided in an embodiment of the present invention.

[0072] Figure 9 This is a schematic diagram of a noise index calculation result provided in an embodiment of the present invention.

[0073] Figure 10 This is a schematic diagram illustrating the calculation results of a comprehensive wave-slip noise index provided in an embodiment of the present invention. Detailed Implementation

[0074] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0075] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or couplings. The term “and / or” as used herein includes any and all combinations of one or more of the associated listed items.

[0076] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as herein.

[0077] To facilitate understanding of the embodiments of the present invention, the following will provide further explanation and description with reference to the accompanying drawings and several specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0078] Example 1

[0079] An embodiment of the present invention provides a schematic diagram illustrating the corrugation of subway rails, as shown below. Figure 1 As shown in the figure, a flowchart of a method for measuring rail corrugation in subways based on the fusion of a Seq2Seq model and vibration and noise data provided in this embodiment of the invention is as follows: Figure 2 As shown, the processing steps include the following:

[0080] Step S1: Based on historical data of subway trains, obtain the vertical and longitudinal vibration acceleration signals of the floor surface at the location of the bogie inside the subway train, and obtain the time domain information of the vertical and longitudinal vibration of the floor inside the train based on the vertical and longitudinal vibration acceleration signals.

[0081] Step S2: Obtain the measured sound pressure signal inside the subway train at the location of the bogie, and obtain the time-domain information of the sound pressure inside the train based on the sound pressure signal;

[0082] Step S3: Obtain the measured speed signal from the speed measuring wheelset of the subway train, and obtain the train speed signal based on the speed signal;

[0083] Step S4: According to the preset window length, the longitudinal vibration acceleration signal and train speed signal of the floor surface are divided into segmented floor longitudinal acceleration and train speed signals to generate a time series dataset corresponding to the longitudinal vibration acceleration signal and train speed.

[0084] Step S5: Construct a subway train mileage matching prediction model based on the Seq2Seq model. Input the time series dataset of {floor longitudinal acceleration, train speed} and train the subway train mileage matching prediction model based on the Seq2Seq model. Figure 3 This is a structural diagram of a subway train mileage matching prediction model based on a Seq2Seq model, provided in an embodiment of the present invention.

[0085] Step S6: Input the longitudinal acceleration of the subway train floor to be detected into the trained subway train mileage matching prediction model based on the Seq2Seq model to obtain the subway train running speed and mileage.

[0086] Step S7: Based on the preset distribution range of subway rail corrugation wavelength and train speed, the vertical vibration acceleration of the floor and the sound pressure signal inside the train are segmented and bandpass filtered.

[0087] Step S8: Calculate the vibration index and noise index corresponding to the filtered floor vertical vibration acceleration and the in-vehicle sound pressure signal;

[0088] Step S9: Further calculate the comprehensive wave-slip vibration-noise index for floor vertical vibration acceleration signals and sound pressure signals whose vibration index and noise index are greater than the preset threshold;

[0089] Step S10: Determine the wave wavelength and wave depth amplitude based on the wave noise comprehensive index, and determine the specific location where the wave occurs based on the metro train mileage matching prediction model.

[0090] Specifically, step S4 above, which involves segmenting the floor longitudinal acceleration and train speed signals into segmented floor longitudinal acceleration and train speed signals according to a preset window length, and generating a time series dataset with the shape {floor longitudinal acceleration, train speed}, specifically includes:

[0091] The longitudinal vibration acceleration signal and the train speed signal are resampled to ensure that the longitudinal acceleration time series and the train speed time series have the same length, denoted as L. s ;

[0092] The longitudinal acceleration time series is x i (i = 0, 1, ..., x) n The train speed time series is y. j (j = 0, 1, ..., y) n If the time step is set to K, the number of samples in the constructed dataset is determined by equation (1).

[0093] N = L s -K+1 (1)

[0094] The format of the time series dataset {floor longitudinal acceleration, train speed} is determined by equation (2).

[0095]

[0096] Specifically, the metro train mileage matching prediction model based on the Seq2Seq model in step S5 above belongs to the Many to Many type of recurrent neural network. The encoder part uses GRU (Gate Recurrent Unit, recurrent neural network) to encode the longitudinal acceleration time series and output the hidden state of the last step. The decoder is also composed of GRU recurrent neural network and uses the last hidden state of the encoder and the train speed time series as input.

[0097] The time series datasets corresponding to the longitudinal acceleration and train speed are divided into a training set, a validation set, and a test set; wherein the ratio of the training set, validation set, and test set is 6:2:2.

[0098] The training of the metro train mileage matching prediction model based on the Seq2Seq model includes:

[0099] The training set is input into a metro train mileage matching prediction model based on the Seq2Seq model, and the optimal parameters of the model are found through K-fold cross-validation.

[0100] The validation set is input into the metro train mileage matching prediction model based on the Seq2Seq model using the optimal parameters of the model to validate the model. After successful validation, the trained metro train mileage matching prediction model based on the Seq2Seq model is obtained. The trained metro train mileage matching prediction model based on the Seq2Seq model can output the corresponding metro train running speed time series by inputting the longitudinal acceleration time series of the metro train's interior floor.

[0101] The test set is input into the trained metro train mileage matching prediction model based on the Seq2Seq model to perform train mileage matching prediction and obtain the train running speed and mileage.

[0102] Specifically, in step S7 above, the vertical vibration acceleration of the floor and the sound pressure signal inside the train are segmented and bandpass filtered according to the preset distribution range of the corrugated wavelength of the subway rail and the train running speed.

[0103] The vertical vibration acceleration signal of the floor is equal in length to the sound pressure signal inside the vehicle, denoted as L. v The segment length is set to W, and the overlap rate is set to 3 / 4. If the remaining signal length is greater than or equal to W, zeros are padded to make the length equal to W; otherwise, the remaining signal is discarded.

[0104] According to the preset distribution range of subway rail corrugation wavelength [λ] L ,λ H Find the corresponding upper and lower cutoff frequencies [f]. L ,f HA bandpass filter based on the window function method is designed with an order of n. This filter is used to perform bandpass filtering on the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal, obtaining the filtered segmented floor vertical vibration acceleration signal sig. vb and the in-vehicle sound pressure signal sig no ;

[0105] The upper and lower cutoff frequencies are determined by equation (3);

[0106]

[0107] In the formula, The average speed of segmented trains is determined based on the metro train mileage matching prediction model.

[0108] Specifically, the calculation of the filtered floor vertical vibration acceleration and the vibration index and noise index corresponding to the in-vehicle sound pressure signal in step S8 above includes:

[0109] Calculate the vertical vibration acceleration signal sig of the segmented floor. vb The self-power spectral density, in [f L ,f H Finding the peak frequency f in the frequency band vmax According to [f vmax -D,f vmax +D] Continue to analyze the segmented vertical vibration acceleration signal sig vb With bandpass filtering, the vibration index is determined by equation (4);

[0110]

[0111] In the formula, VI(i) is the vibration index of the i-th frame of the vibration signal, and sig vb This is the vibration signal after the first filtering. `filter` is a bandpass filter, `D` is half the passband width, and `f`... vmax S is the peak frequency in the power spectral density of the vibration signal in the i-th frame. Vtotal S represents the effective value of the vibration signal after the first filtering. Vi The effective value of the vibration signal after the second filtering is determined by equation (5);

[0112]

[0113] In the formula, W V The length of each frame of the vibration signal;

[0114] Calculate the segmented in-vehicle sound pressure signal sig no The self-power spectral density, in [f L ,f H Finding the peak frequency f in the frequency band nmaxAccording to [f nmax -D,f nmax +D] Continue to analyze the segmented in-vehicle sound pressure signal sig no For bandpass filtering, the noise figure is determined by equation (6);

[0115]

[0116] In the formula, NI(i) is the noise index of the i-th frame of the noise signal, and sig no This is the noise signal after the first filtering. `filter` is a window function bandpass filter, `D` is half the passband width, and `f`... nmax S is the peak frequency of the noise signal in the i-th frame from its power spectral density. Ntotal S represents the effective value of the noise signal after the first filtering. Ni The effective value of the noise signal after the second filtering is given by equation [equation missing].

[0117] (7) Determine;

[0118]

[0119] In the formula, W N The length of each frame of noise signal.

[0120] Specifically, in this example of step S8 above, the statement based on [f] vmax -D,f vmax +D] Continue to analyze the segmented vertical vibration acceleration signal sig vb Bandpass filtering, according to [f nmax -D,f nmax +D] Continue to analyze the segmented in-vehicle sound pressure signal sig no Bandpass filtering, specifically includes:

[0121] Based on the peak frequency f in the power spectral density of the segmented vertical vibration acceleration signal and the segmented in-vehicle sound pressure signal. vmax and f nmax The passband width 2D is determined by equation (8):

[0122]

[0123] In the formula, f imax (i = v or n) represents the peak frequency f vmax or f nmax , A(f imax f represents the power spectral density amplitude corresponding to the peak frequency. h () represents the larger frequency value corresponding to the amplitude, f l () represents the smaller frequency value corresponding to the amplitude;

[0124] According to fimax The passband width and 2D determine the upper and lower cutoff frequencies [f] imax -D,f imax +D], design a window function bandpass filter to bandpass filter the segmented floor vertical vibration acceleration signal and the segmented in-vehicle sound pressure signal respectively, and obtain the wave wear characterization signal C. Vi and C Ni .

[0125] Specifically, in step S9 above, the vertical vibration acceleration signal and sound pressure signal of the floor with vibration index and noise index greater than the preset threshold are further calculated to obtain the comprehensive wave-slip vibration-noise index, which includes:

[0126] Based on statistical analysis of a large amount of measured data, the vibration index threshold was determined to be VI. m The noise figure threshold is NI. m If the vibration index VI(i) corresponding to the vertical vibration acceleration signal of the segmented floor is greater than VI m Furthermore, the noise index NI(i) corresponding to the segmented in-vehicle sound pressure signal is greater than NI. m Then calculate the cross power spectrum of the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal;

[0127] In [f L ,f H Finding the peak frequency f in the frequency band vnmax According to [f vnmax -D,f vnmax +D] For bandpass filtering of the acoustic-vibration composite signal, the wave-slip noise composite index is determined by equation (9);

[0128]

[0129] In the formula, CVNI(i) is the combined wave-scraping noise index of the i-th frame, and sig vn The signal is a composite acoustic-vibration signal, filter is a window function bandpass filter, D is half the passband width, and f vnmax S is the peak frequency in the cross-power spectrum of the i-th frame acoustic vibration signal. VNtotal S represents the effective value of the acoustic-vibration composite signal after the first filtering. VNi The effective value of the acoustic-vibration composite signal after the second filtering is given by equation [equation missing].

[0130] (10) Confirmed;

[0131]

[0132] In this example, the calculation of the wave-slip noise composite index of the acoustic-vibration composite signal is based on [f vnmax -D,f vnmax +D] Bandpass filtering of acoustic-vibration composite signals specifically includes:

[0133] The acoustic-vibration composite signal is calculated based on the vertical vibration acceleration signal of the floor and the sound pressure signal inside the vehicle, and is determined by equation (11);

[0134] sig vn =sig vb ×sig no (11)

[0135] In the formula, sig vb The signal is the vertical vibration acceleration of the floor after the first filtering, sig no This is the sound pressure signal inside the vehicle after the first filtering.

[0136] According to f vnmax The passband width and 2D determine the upper and lower cutoff frequencies [f] vnmax -D,f vnmax +D], design a window function bandpass filter for the acoustic-vibration composite signal sig vn Bandpass filtering yields the erosion characterization signal C. VNi .

[0137] Specifically, step S10 above, which involves determining the wave wavelength and wave depth amplitude based on the wave noise comprehensive index and determining the specific location of wave generation based on the metro train mileage matching prediction model, includes:

[0138] The train speed v(t) is obtained from the metro train mileage matching prediction model based on the Seq2Seq model, and the train position is determined by equation (12).

[0139]

[0140] Based on the peak frequency f corresponding to the segmented acoustic-vibration composite signal vnmax Based on the average vehicle speed V, the wavelength of rail corrugation can be identified.

[0141] Determined by equation (13);

[0142]

[0143] The severity of rail corrugation, i.e. the range of the wave depth amplitude, is identified by the corrugation noise comprehensive index corresponding to the segmented acoustic and vibration composite signal, and is determined by equation (14).

[0144]

[0145] The working principle of this invention is as follows: Considering that GPS positioning cannot be effectively used to measure vehicle speed when urban rail transit vehicles are traveling in tunnels, and that the train's own operational data is not easily obtained, this invention utilizes the relationship between the longitudinal acceleration of the vehicle body and its speed to determine the train's position. Furthermore, the time-frequency characteristics of the vertical vibration acceleration and sound pressure signal inside the vehicle generated in corrugated and normal rail sections differ significantly. By fusing acoustic and vibration data, the corrugation component in the signal can be effectively extracted. Therefore, this invention uses time-series data of longitudinal acceleration of the vehicle floor and train speed to train a metro train mileage matching and prediction model based on a Seq2Seq model. It accurately predicts the speed and mileage of the metro train through longitudinal acceleration of the vehicle body, and simultaneously uses the fusion of vertical vibration acceleration of the vehicle floor and sound pressure signal inside the vehicle to accurately identify the characteristic wavelength and depth amplitude of rail corrugation.

[0146] Example 2

[0147] The difference between this embodiment and Embodiment 1 is that this embodiment is based on the rapid measurement method for subway rail corrugation based on the fusion of Seq2Seq model and vibration and noise data in Embodiment 1, and is implemented as follows:

[0148] The process consists of the following ten steps: 1) Measure the vertical and longitudinal vibration acceleration signals of the subway train floor and the sound pressure signal inside the train; 2) Obtain the actual train speed signal based on the subway train's onboard ATP system; 3) Resample the subway train floor longitudinal acceleration and train speed signals at the same frequency to construct a {floor longitudinal acceleration, train speed} dataset; 4) Divide the {floor longitudinal acceleration, train speed} dataset into a training set, a validation set, and a test set; 5) Construct a subway train mileage matching prediction model based on the Seq2Seq model, train it on the training set, evaluate it on the validation set, and finally obtain the result. 6) Train a metro train mileage matching prediction model based on the Seq2Seq model and input it into the test set for testing; 7) Perform bandpass filtering on the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal according to the preset metro rail corrugation wavelength distribution range; 8) Calculate the vibration index and noise index corresponding to the filtered floor vertical vibration acceleration signal and the in-vehicle sound pressure signal respectively; 9) If both the vibration index and the noise index are greater than the preset threshold, calculate the acoustic-vibration composite signal; 10) Calculate the corrugation vibration-noise comprehensive index corresponding to the acoustic-vibration composite signal; 11) Output key information such as the characteristic wavelength and wave depth amplitude of rail corrugation based on the corrugation vibration-noise comprehensive index.

[0149] The equipment used in the method of this invention includes:

[0150] An accelerometer, which is installed on the floor inside a subway train, is used to measure the vertical and longitudinal vibration acceleration signals of the floor inside the train.

[0151] A sound pressure sensor, which is installed inside a subway train car, is used to measure the sound pressure signal inside the car;

[0152] The pulse speed sensor, which is mounted on the train axle, is part of the onboard ATP system and is used to collect the train's running speed.

[0153] The data acquisition module is connected to the accelerometer and the sound pressure sensor to collect vibration and sound pressure signals and obtain vibration and noise time-domain information.

[0154] The data processing module is connected to the data acquisition module and is used to process the data acquired by the data acquisition module into a format that can be used by a computer to identify the wave pattern.

[0155] The acceleration sensor and sound pressure sensor are arranged on the floor above the bogie of the subway train, which ensures that the relevant vibration and sound pressure signals generated by rail corrugation are obtained to achieve the best recognition effect.

[0156] The following are examples of implementation methods:

[0157] (1) Data Description

[0158] A field test was conducted on a section of a subway line in a certain city in China. Train speed signals were obtained using train speed signals recorded by the onboard ATP system. Figure 4 As shown.

[0159] The vibration and sound pressure signals inside the vehicle are acquired using a multi-channel 24AD high-precision data acquisition system, such as... Figure 5 As shown.

[0160] (2) Data partitioning

[0161] The floor longitudinal acceleration and train speed signals are resampled to ensure that the time series lengths of floor longitudinal acceleration and train speed are equal. The signals are then segmented using a preset window length to obtain a one-to-one corresponding {floor longitudinal acceleration, train speed} dataset, such as... Figure 6 As shown.

[0162] (3) Divide the sample dataset

[0163] The sample dataset is randomly divided into a training set, a validation set, and a test set, with the ratio of the training set, validation set, and test set being 6:2:2.

[0164] (4) Model Training

[0165] The training set is input into the model, and K-fold cross-validation is performed to obtain the optimal parameters of the model. Using the optimal parameters, the optimal model of the subway train mileage matching prediction model is obtained.

[0166] (5) Signal filtering

[0167] The obtained vertical vibration acceleration signal of the floor and the sound pressure signal inside the vehicle were bandpass filtered according to the preset distribution range of subway rail corrugation. The filter was a window function filter with an order of n = 30. Figure 7 As shown.

[0168] (6) Calculate the vibration index

[0169] The filtered vertical vibration acceleration signal of the floor is segmented according to a preset window length, and the vibration index corresponding to each segment is calculated, such as... Figure 8 As shown.

[0170] (7) Calculate the noise figure

[0171] The filtered in-vehicle sound pressure signal is segmented according to a preset window length, and the noise index corresponding to each segment of the in-vehicle sound pressure signal is calculated, such as... Figure 9 As shown.

[0172] (8) Calculate the comprehensive index of corrugation noise

[0173] If the vibration index and noise index are greater than preset thresholds, the acoustic-vibration composite signal is calculated and segmented according to a preset window length. The wave-slip vibration-noise composite index corresponding to each segment is then calculated. Figure 10 As shown.

[0174] (9) Output of test results

[0175] Based on the peak frequency f corresponding to the cross-power spectral density of the floor vertical vibration acceleration signal and the in-vehicle sound pressure signal. vnmax The characteristic wavelength of rail corrugation is obtained from the train speed output by the metro train mileage matching prediction model; the corrugation depth amplitude information is obtained from the corrugation vibration and noise comprehensive index, such as... Figure 10 As shown.

[0176] In summary, compared with existing intelligent detection methods for rail corrugation, the present invention utilizes vertical and longitudinal vibration acceleration of the vehicle floor and sound pressure signals inside the vehicle to accurately identify the mileage caused by rail corrugation, and can also identify key information such as the characteristic wavelength and wave depth amplitude of the corresponding rail corrugation.

[0177] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of one embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention.

[0178] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0179] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for apparatus or system embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. The apparatus and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0180] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for measuring rail corrugation in subway trains based on the fusion of a Seq2Seq model and vibration and noise data, characterized in that, include: A time-series dataset of floor longitudinal acceleration and train speed was constructed based on historical subway train data. A metro train mileage matching prediction model based on the Seq2Seq model is constructed. The metro train mileage matching prediction model based on the Seq2Seq model is trained using the time series dataset corresponding to the longitudinal acceleration and train speed to obtain the trained metro train mileage matching prediction model. The longitudinal acceleration of the floor inside the subway train to be detected is input into the trained subway train mileage matching prediction model based on the Seq2Seq model to obtain the running speed of the subway train to be detected. The rail corrugation vibration and noise comprehensive index of the subway train to be tested is calculated based on the preset distribution range of rail corrugation wavelength and the running speed, floor vertical vibration acceleration and in-vehicle sound pressure signal of the subway train to be tested. The specific location of corrugation and the corresponding characteristic wavelength and wave depth amplitude are determined based on the corrugation vibration and noise comprehensive index. The aforementioned construction of a time-series dataset corresponding to floor longitudinal acceleration and train speed based on historical subway train data includes: Based on historical data of subway trains, vertical and longitudinal vibration acceleration signals of the floor surface at the location of the bogie inside the subway train are obtained, and time-domain information of vertical and longitudinal vibration of the floor inside the train is obtained based on the vertical and longitudinal vibration acceleration signals. Acquire the sound pressure signal inside the subway train at the location of the bogie, and obtain the time-domain information of the sound pressure inside the train based on the sound pressure signal; Acquire the speed signal from the speed measuring wheelset of the subway train, and obtain the train speed signal based on the speed signal; According to the preset window length, the longitudinal vibration acceleration signal and train speed signal of the floor surface are segmented into segmented floor longitudinal acceleration and train speed signals, and a time series dataset corresponding to the longitudinal vibration acceleration signal and train speed is generated. The method of calculating the comprehensive index of rail corrugation vibration and noise of the subway train under test based on the preset wavelength distribution range of subway rail corrugation and the running speed, floor vertical vibration acceleration, and in-vehicle sound pressure signal of the subway train under test includes: The system acquires the operating speed, floor vertical vibration acceleration signal, and in-vehicle sound pressure signal of the subway train to be tested. Based on the preset distribution range of the corrugated wavelength of the subway rails and the train speed of the subway train to be tested, the vertical vibration acceleration of the floor and the sound pressure signal inside the train to be tested are segmented and bandpass filtered. Calculate the vibration index and noise index corresponding to the filtered floor vertical vibration acceleration signal and the in-vehicle sound pressure signal. Retain the floor vertical vibration acceleration signal and the in-vehicle sound pressure signal whose vibration index and noise index are greater than the preset threshold. Further calculate the wave-slip vibration-noise comprehensive index for the retained floor vertical vibration acceleration signal and the in-vehicle sound pressure signal. The wave wavelength and wave depth amplitude are determined based on the wave noise comprehensive index, and the specific location of the wave occurrence is determined based on the metro train mileage matching prediction model.

2. The method according to claim 1, characterized in that, The process of segmenting the longitudinal vibration acceleration signal and train speed signal of the floor surface into segmented floor longitudinal acceleration and train speed signals according to a preset window length, and generating a time series dataset corresponding to the longitudinal vibration acceleration signal and train speed, includes: The longitudinal vibration acceleration signal and the train speed signal are resampled to ensure that the longitudinal acceleration time series and the train speed time series have the same length, denoted as L. s ; The longitudinal acceleration time series is x i (i = 0, 1,…, x n The train speed time series is y. j (j = 0,1,…, y n If the time step is set to K, then the number of samples in the constructed dataset is determined by equation (1); (1) The time series dataset corresponding to the longitudinal vibration acceleration signal and the train speed is determined by equation (2); (2)。 3. The method according to claim 1, characterized in that, The construction of the metro train mileage matching prediction model based on the Seq2Seq model involves training the model using the time series dataset corresponding to the longitudinal acceleration and train speed to obtain a trained metro train mileage matching prediction model, including: A metro train mileage matching prediction model based on the Seq2Seq model is constructed. The metro train mileage matching prediction model belongs to the Many to Many type of recurrent neural network. The encoder part uses a recurrent neural network model to encode the longitudinal acceleration time series and outputs the hidden state of the last step. The decoder is also composed of a recurrent neural network and uses the last hidden state of the encoder and the train speed time series as input. The time series datasets corresponding to the longitudinal acceleration and train speed are divided into training set, validation set and test set. The training set is input into the subway train mileage matching prediction model, and the optimal parameters of the model are found through K-fold cross-validation. The validation set is input into the metro train mileage matching prediction model using the optimal parameters of the model to validate the metro train mileage matching prediction model. After successful validation, a trained metro train mileage matching prediction model based on the Seq2Seq model is obtained. The input data of the trained metro train mileage matching prediction model based on the Seq2Seq model is the time series of longitudinal acceleration of the metro train floor, and the output data is the time series of the corresponding metro train running speed.

4. The method according to claim 1, characterized in that, The method of performing segmented bandpass filtering on the vertical vibration acceleration of the floor and the sound pressure signal inside the subway train, based on a preset wavelength distribution range for subway rail corrugation and the train's operating speed, includes: Assume that the vertical vibration acceleration signal of the subway train floor and the sound pressure signal inside the train have the same length, denoted as L. v The segment length is set to W, and the overlap rate is set to 3 / 4. If the remaining signal length is greater than or equal to W, zeros are padded to make the length equal to W; otherwise, the remaining signal is discarded. According to the preset distribution range of subway rail corrugation wavelength [λ] L , λ H Find the corresponding upper and lower cutoff frequencies [f]. L ,f H A bandpass filter based on the window function method is designed with an order of n. This filter is used to perform bandpass filtering on the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal of the subway train to be detected, obtaining the filtered segmented floor vertical vibration acceleration signal sig. vb and the in-vehicle sound pressure signal sig no ; The upper and lower cutoff frequencies are determined by equation (3); (3) In the formula, The average speed of segmented trains is determined based on the metro train mileage matching prediction model.

5. The method according to claim 4, characterized in that, The calculation of the vibration index and noise index corresponding to the filtered floor vertical vibration acceleration signal and the in-vehicle sound pressure signal includes: Calculate the vertical vibration acceleration signal sig of the segmented floor. vb The self-power spectral density, in [f L , f H Finding the peak frequency f in the frequency band vmax According to [f vmax -D, f vmax +D] Continue to analyze the segmented vertical vibration acceleration signal sig vb With bandpass filtering, the vibration index is determined by equation (4); (4) In the formula, VI(i) is the vibration index of the i-th frame of the floor vertical vibration acceleration signal, and sig vb This is the vibration signal after the first filtering. `filter` is a bandpass filter, `D` is half the passband width, and `f`... vmax S is the peak frequency in the power spectral density of the vibration signal in the i-th frame. Vtotal S represents the effective value of the vibration signal after the first filtering. Vi The effective value of the vibration signal after the second filtering is determined by equation (5); (5) In the formula, W V The length of each frame of the vibration signal; Calculate the segmented in-vehicle sound pressure signal sig no The self-power spectral density, in [f L , f H Finding the peak frequency f in the frequency band nmax According to [f nmax -D, f nmax +D] Continue to analyze the segmented in-vehicle sound pressure signal sig no For bandpass filtering, the noise figure is determined by equation (6); (6) In the formula, NI(i) is the noise index of the i-th frame of the noise signal, and sig no This is the noise signal after the first filtering. `filter` is a window function bandpass filter, `D` is half the passband width, and `f`... nmax S is the peak frequency of the noise signal in the i-th frame from its power spectral density. Ntotal S represents the effective value of the noise signal after the first filtering. Ni The effective value of the noise signal after the second filtering is determined by equation (7); (7) In the formula, W N The length of each frame of noise signal.

6. The method according to claim 1, characterized in that, The retained floor vertical vibration acceleration signals and in-vehicle sound pressure signals with vibration and noise indices greater than preset thresholds are further used to calculate a comprehensive wave-slip noise index, including: The vibration index threshold is determined to be VI. m The noise figure threshold is NI. m If the vibration index VI(i) corresponding to the vertical vibration acceleration signal of the segmented floor is greater than VI m Furthermore, the noise index NI(i) corresponding to the segmented in-vehicle sound pressure signal is greater than NI. m Then calculate the cross power spectrum of the segmented floor vertical vibration acceleration signal and the in-vehicle sound pressure signal; In [f L , f H Finding the peak frequency f in the frequency band vnmax According to [f vnmax -D, f vnmax +D] For bandpass filtering of the acoustic-vibration composite signal, the wave-slip noise composite index is determined by equation (9); (9) In the formula, CVNI(i) is the combined wave-scraping noise index of the i-th frame, and sig vn The signal is a composite acoustic-vibration signal, filter is a window function bandpass filter, D is half the passband width, and f vnmax S is the peak frequency in the cross-power spectrum of the i-th frame acoustic vibration signal. VNtotal S represents the effective value of the acoustic-vibration composite signal after the first filtering. VNi The effective value of the acoustic-vibration composite signal after the second filtering is determined by equation (10); (10)。 7. The method according to claim 1, characterized in that, The method of determining the specific location of wave wear and the corresponding characteristic wavelength and wave depth amplitude based on the wave wear noise comprehensive index includes: The train speed v(t) is obtained from the metro train mileage matching prediction model based on the Seq2Seq model, and the train position is determined by equation (12). (12) Based on the peak frequency f corresponding to the segmented acoustic-vibration composite signal vnmax The wavelength of rail corrugation is identified by the average vehicle speed V and determined by equation (13); (13) The severity of rail corrugation, i.e., the range of corrugation depth amplitude, is identified based on the corrugation noise comprehensive index corresponding to the segmented acoustic-vibration composite signal, and is determined by equation (14): (14)。

Images