Metro rail corrugation detection method and system based on automatic feature extraction

Abstract

The application provides a subway rail corrugation detection method and system based on automatic feature extraction, belongs to the track defect detection technical field based on deep learning and time series, obtains subway train in-vehicle vibration data and in-vehicle noise data to be detected; a pre-trained detection model is used to process the obtained in-vehicle vibration data and in-vehicle noise data, and the subway rail corrugation state is obtained, wherein the corrugation state includes a corrugation characteristic wavelength and a corrugation wave depth amplitude. The application can realize accurate identification of the corrugation characteristic wavelength and the corrugation wave depth amplitude at the same time, can be arranged in the carriages of different trains, does not affect the normal operation of the vehicle, has the advantages of portability, high efficiency, accuracy, saving of operation cost and the like, is beneficial to promoting the change of the subway operation department from 'planned repair' to 'condition repair', and helps the green development of rail transit.

Description

Technical Field

[0001] This invention relates to the field of track defect detection technology based on deep learning and time series, specifically to a method and system for detecting rail corrugation in subways based on automatic feature extraction. Background Technology

[0002] Rail corrugation (or simply rail wear) refers to a periodic, wavy, uneven defect that appears on the longitudinal surface of the rail. Subway lines, due to their diverse track types, small curve radii, frequent starting and braking, and high operating density, experience more severe rail corrugation, leading to a series of problems such as fastener breakage, vehicle suspension fatigue, in-car vibration and noise, and secondary noise from buildings in nearby residential areas. Currently, rail grinding remains the most economical and effective measure for controlling rail corrugation. However, when scheduling rail grinding on-site, the selection of operational parameters such as grinding mileage, grinding depth, and number of grinding passes is often challenging. Previously, maintenance personnel would go down to the track during maintenance windows and use static measurement methods such as corrugation trolleys and steel rulers to obtain the rail corrugation status, combining this with vibration and noise complaint sections to manually select operational parameters such as grinding mileage. 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, there are various intelligent identification methods for rail corrugation, but they can only identify sections with or without rail corrugation. They cannot accurately identify key information such as the characteristic wavelength and wave depth amplitude of rail corrugation. Moreover, most of them identify rail corrugation by axle box acceleration, which has limitations in field application such as complex installation conditions or restrictions on installation, inability to be portable for measurement, and random installation in different carriages of a train. Summary of the Invention

[0004] The purpose of this invention is to provide a high-precision and high-efficiency method and system for detecting rail corrugation in subways based on automatic feature extraction, which can simultaneously identify key information such as characteristic wavelength and wave depth amplitude of rail corrugation, in order to solve at least one of the technical problems existing in the background art.

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

[0006] On the one hand, this invention provides a method for detecting rail corrugation in subways based on automatic feature extraction, comprising:

[0007] Acquire vibration and noise data inside the subway train to be tested;

[0008] Using a pre-trained detection model, the acquired in-vehicle vibration data and in-vehicle noise data are processed to obtain the corrugation state of the subway rails. The corrugation state includes the characteristic wavelength of corrugation and the amplitude of corrugation depth. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network is used to segment the vibration data and noise data into vibration spatial domain information and noise spatial domain information through displacement spatial transformation. The vibration signal recognition network is used to process the vibration spatial domain information to obtain the amplitude of corrugation depth. The noise signal recognition network is used to process the noise spatial domain information to obtain the characteristic wavelength of corrugation.

[0009] Optionally, the vibration data and noise data are segmented into vibration spatial domain information and noise spatial domain information through displacement spatial transformation, including:

[0010] Integrating the actual operating speed curve of the subway train, we obtain the curve of the train displacement changing with time, which is then calculated and determined by formula (1).

[0011]

[0012] By determining the spatial intercept length ω, the vibration and noise time domains of length T are segmented to obtain N segments of vibration time domain information and noise time domain information (0~T1, T1~T2, …, T). n-1 ~T n ), determined by formulas (2) to (3);

[0013]

[0014]

[0015] In the formula, T is the total test time, L is the total train mileage, ω is the spatial intercept length, and v represents the train speed.

[0016] Optionally, when training the detection model, the spatial domain information of vibration, noise, and rail corrugation in each segment is preprocessed to extract the time series vector of vibration signal, the normalized energy feature vector of noise signal, the spatial domain feature statistics and frequency domain feature statistics of rail corrugation signal in each segment. This includes: preprocessing the spatial domain information of vibration in each segment, bandpass filtering to eliminate other interference, setting the sampling window length, randomly sampling time series of the same time length from the spatial domain information of vibration in each segment to form a time series vector; using wavelet packet transform to decompose and reconstruct the spatial domain information of noise in each segment to eliminate signal noise and extract the normalized energy feature vector of different frequency bands; using 1 / 3 octave band to perform frequency domain transformation on the spatial domain information of rail corrugation in each segment and extracting the frequency domain feature statistics of the spatial domain information of rail corrugation in each segment.

[0017] Optionally, the time series with the same random sampling length specifically includes:

[0018] Obtain the time-domain information of each segment of vibration after displacement space transformation (0~T1, T1~T2, …, T n-1 ~T n Due to the different train speeds, the lengths of the vibration time-domain information for each segment are different.

[0019] Determine the minimum length L of the vibration time domain information for each segment. min The sampling window length l is determined by randomly sampling time series of the same length from the vibration time domain information of each segment, and is determined by formulas (4) to (5).

[0020] L min =min(L(0~T1),L(T1~T2),...,L(T n-1 ~T n (4)

[0021]

[0022] In the formula, α is the number of each segmented sampling window;

[0023] The frequency domain characteristic statistics include the characteristic wavelength of rail corrugation, and the spatial domain characteristic statistics include the peak-to-peak average value of the moving wave depth amplitude of rail corrugation and the percentage of peak-to-peak average value exceeding the limit, which are determined by formulas (6) to (10).

[0024] PPR(x,L)=(a1+a2+...+a n ) / n (6)

[0025]

[0026]

[0027]

[0028]

[0029] In the formula, PPR(x,L) is the peak-to-peak average amplitude of the moving wave depth of the rail corrugation, and a i Let L be the wave depth amplitude, L be the analysis window length, and n be the number of wave depth amplitude values ​​within the analysis window length. i This represents the percentage of peak-to-peak average values ​​exceeding the limit in different intervals.

[0030] Optionally, the input feature vector of the detection model can be represented as {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}.

[0031] Optionally, the noise signal recognition network is a linearly constrained wavelet neural network; the linearly constrained wavelet neural network uses the relationship between vehicle speed, wavelength and frequency to obtain a linear mapping matrix G, and applies the linear mapping matrix G to the output of the neural network so that the output satisfies specific constraints, which are determined by formulas (11) to (12).

[0032]

[0033] CGp = 0 (12)

[0034] In the formula, λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

[0035] Secondly, the present invention provides a subway rail corrugation detection system based on automatic feature extraction, comprising:

[0036] The acquisition module is used to acquire the vibration data and noise data inside the subway train to be detected.

[0037] The detection module is used to process the acquired in-vehicle vibration data and in-vehicle noise data using a pre-trained detection model to obtain the corrugation state of the subway rails. The corrugation state includes the characteristic wavelength of corrugation and the amplitude of corrugation depth. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network is used to segment the vibration data and noise data into vibration spatial domain information and noise spatial domain information through displacement spatial transformation. The vibration signal recognition network is used to process the vibration spatial domain information to obtain the amplitude of corrugation depth. The noise signal recognition network is used to process the noise spatial domain information to obtain the characteristic wavelength of corrugation.

[0038] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the subway rail corrugation detection method based on automatic feature extraction as described above.

[0039] Fourthly, the present invention provides a computer program product, including a computer program that, when run on one or more processors, is used to implement the subway rail corrugation detection method based on automatic feature extraction as described above.

[0040] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the subway rail corrugation detection method based on automatic feature extraction as described above.

[0041] The beneficial effects of this invention are: it can achieve accurate identification of both the characteristic wavelength and wave depth amplitude of rail corrugation at the same time, and it can be deployed in the carriages of different trains without affecting the normal operation of the vehicles. It has the advantages of being portable, efficient, accurate, and saving operating costs, which is conducive to promoting the transformation of subway operation departments from "planned maintenance" to "condition-based maintenance" and contributing to the green development of rail transit.

[0042] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0043] 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.

[0044] Figure 1 This is a schematic diagram illustrating the corrugation of subway rails in this invention.

[0045] Figure 2 This is a flowchart of the subway rail corrugation detection method based on automatic feature extraction as described in an embodiment of the present invention.

[0046] Figure 3 This is a structural diagram of the subway rail corrugation detection model constructed based on automatic feature extraction, as described in an embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram of time-domain information segmentation of in-vehicle vibration and noise according to an embodiment of the present invention.

[0048] Figure 5 This is a schematic diagram illustrating the acquisition of the time series vector of in-vehicle vibration signal according to an embodiment of the present invention.

[0049] Figure 6 This is a schematic diagram illustrating the acquisition of the normalized energy feature vector of in-vehicle noise signal according to an embodiment of the present invention.

[0050] Figure 7 This is a schematic diagram of the frequency domain feature statistics for obtaining spatial domain information of rail corrugation as described in an embodiment of the present invention.

[0051] Figure 8 This is a schematic diagram illustrating the acquisition of spatial domain characteristic statistics of rail corrugation as described in an embodiment of the present invention.

[0052] Figure 9 This is a schematic diagram of the sound pressure loss curve according to an embodiment of the present invention.

[0053] Figure 10 This is a flowchart of the subway rail corrugation detection device based on automatic feature extraction as described in an embodiment of the present invention.

[0054] Figure 11 This is a schematic diagram of a subway rail corrugation detection device based on automatic feature extraction, as described in an embodiment of the present invention. Detailed Implementation

[0055] 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.

[0056] 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.

[0057] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0058] 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, and / or groups thereof.

[0059] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0060] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0061] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0062] Example 1

[0063] In this embodiment 1, a subway rail corrugation detection system based on automatic feature extraction is first provided, including:

[0064] The acquisition module is used to acquire the vibration data and noise data inside the subway train to be detected.

[0065] The detection module is used to process the acquired in-vehicle vibration data and in-vehicle noise data using a pre-trained detection model to obtain the corrugation state of the subway rails. The corrugation state includes the characteristic wavelength of corrugation and the amplitude of corrugation depth. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network is used to segment the vibration data and noise data into vibration spatial domain information and noise spatial domain information through displacement spatial transformation. The vibration signal recognition network is used to process the vibration spatial domain information to obtain the amplitude of corrugation depth. The noise signal recognition network is used to process the noise spatial domain information to obtain the characteristic wavelength of corrugation.

[0066] In this embodiment 1, the above-described system is used to implement a method for detecting rail corrugation in subway trains based on automatic feature extraction, including:

[0067] Acquire vibration and noise data inside the subway train to be tested;

[0068] Using a pre-trained detection model, the acquired in-vehicle vibration data and in-vehicle noise data are processed to obtain the corrugation state of the subway rails. The corrugation state includes the characteristic wavelength of corrugation and the amplitude of corrugation depth. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network is used to segment the vibration data and noise data into vibration spatial domain information and noise spatial domain information through displacement spatial transformation. The vibration signal recognition network is used to process the vibration spatial domain information to obtain the amplitude of corrugation depth. The noise signal recognition network is used to process the noise spatial domain information to obtain the characteristic wavelength of corrugation.

[0069] Specifically, vibration data and noise data are segmented into vibration spatial domain information and noise spatial domain information through displacement space transformation, including:

[0070] Integrating the actual operating speed curve of the subway train, we obtain the curve of the train displacement changing with time, which is then calculated and determined by formula (1).

[0071]

[0072] By determining the spatial intercept length ω, the vibration and noise time domains of length T are segmented to obtain N segments of vibration time domain information and noise time domain information (0~T1, T1~T2, …, T). n-1 ~T n ), determined by formulas (2) to (3);

[0073]

[0074]

[0075] In the formula, T is the total test time, L is the total train mileage, ω is the spatial intercept length, and v represents the train speed.

[0076] When training the detection model, the spatial domain information of vibration, noise, and rail corrugation in each segment is preprocessed to extract the time series vector of vibration signal, the normalized energy feature vector of noise signal, the spatial domain feature statistics and frequency domain feature statistics of rail corrugation signal in each segment. This includes: preprocessing the spatial domain information of vibration in each segment, bandpass filtering to eliminate other interference, setting the sampling window length, randomly sampling time series of the same time length from the spatial domain information of vibration in each segment to form a time series vector; using wavelet packet transform to decompose and reconstruct the spatial domain information of noise in each segment to eliminate signal noise and extract the normalized energy feature vector of different frequency bands; and using 1 / 3 octave band to perform frequency domain transformation on the spatial domain information of rail corrugation in each segment to extract the frequency domain feature statistics of the spatial domain information of rail corrugation in each segment.

[0077] The aforementioned time series with the same random sampling length specifically include:

[0078] Obtain the time-domain information of each segment of vibration after displacement space transformation (0~T1, T1~T2, …, T n-1 ~T n Due to the different train speeds, the lengths of the vibration time-domain information for each segment are different.

[0079] Determine the minimum length L of the vibration time domain information for each segment. min The sampling window length l is determined by randomly sampling time series of the same length from the vibration time domain information of each segment, and is determined by formulas (4) to (5).

[0080] L min =min(L(0~T1),L(T1~T2),...,L(T n-1 ~T n (4)

[0081]

[0082] In the formula, α is the number of each segmented sampling window;

[0083] The frequency domain characteristic statistics include the characteristic wavelength of rail corrugation, and the spatial domain characteristic statistics include the peak-to-peak average value of the moving wave depth amplitude of rail corrugation and the percentage of peak-to-peak average value exceeding the limit, which are determined by formulas (6) to (10).

[0084] PPR(x,L)=(a1+a2+...+a n ) / n (6)

[0085]

[0086]

[0087]

[0088]

[0089] In the formula, PPR(x,L) is the peak-to-peak average amplitude of the moving wave depth of the rail corrugation, and a i Let L be the wave depth amplitude, L be the analysis window length, and n be the number of wave depth amplitude values ​​within the analysis window length. i This represents the percentage of peak-to-peak average values ​​exceeding the limit in different intervals.

[0090] The input feature vectors of the detection model are represented as {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}.

[0091] The noise signal recognition network is a linearly constrained wavelet neural network; the linearly constrained wavelet neural network uses the relationship between vehicle speed, wavelength and frequency to obtain a linear mapping matrix G, and applies the linear mapping matrix G to the output of the neural network so that the output satisfies specific constraints, which are determined by formulas (11) to (12).

[0092]

[0093] CGp = 0 (12)

[0094] In the formula, λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

[0095] Example 2

[0096] To address the shortcomings of existing intelligent rail corrugation detection algorithms, such as their inability to simultaneously identify key information like characteristic wavelengths and wave depth amplitudes, and their low efficiency in on-site deployment, this embodiment 2 provides a metro rail corrugation detection method based on automatic feature extraction. It utilizes in-vehicle vibration and noise information to extract feature vectors and trains a metro rail corrugation detection model based on automatic feature extraction, achieving accurate identification of characteristic wavelengths and wave depth amplitudes of rail corrugation. The metro rail corrugation detection model trained in this embodiment 2 can achieve real-time online detection of metro rail corrugation. Furthermore, it can be deployed in different train carriages without affecting normal train operation, offering advantages such as portability, efficiency, accuracy, and cost savings. This facilitates the transformation of metro operations from "planned maintenance" to "condition-based maintenance," contributing to the green development of rail transit.

[0097] In this embodiment 2, the subway rail corrugation detection method based on automatic feature extraction includes:

[0098] Obtain vibration and noise signals inside a subway train at different time periods, and obtain time-domain information of in-train vibration and noise based on the vibration and noise signals;

[0099] Acquire the spatial distribution of subway rail corrugation signals along the operating mileage, and obtain rail corrugation spatial domain information based on the rail corrugation signals;

[0100] The vibration and noise signals are divided into segmented vibration and noise spatial domain information of different lengths through displacement space transformation. The length of each segment of vibration and noise spatial domain information is equal to that of rail corrugation spatial domain information.

[0101] The vibration, noise and rail corrugation spatial domain information of each segment are preprocessed to extract the vibration signal time series vector, the noise signal normalized energy feature vector, the spatial domain feature statistics and frequency domain feature statistics of the rail corrugation signal of each segment.

[0102] Based on the time series vector and the normalized energy feature vector, a multi-source, multi-dimensional model input feature vector is obtained;

[0103] Based on the multi-source, multi-dimensional model input feature vector, a subway rail corrugation detection model based on automatic feature extraction is constructed, and the subway rail corrugation detection model based on automatic feature extraction is trained.

[0104] Input the in-vehicle vibration and noise data to be detected into the subway rail corrugation detection model based on automatic feature extraction to obtain the rail corrugation state.

[0105] In this embodiment 2, considering the significant differences in the time-frequency characteristics of in-vehicle vibration and noise signals generated when urban rail transit vehicles travel in corrugated sections and normal sections, and the relatively constant speed variation within a certain spatial domain, the in-vehicle noise signal caused by rail corrugation exhibits specific frequency band energy concentrations according to the fixed wavelength theorem. Furthermore, the in-vehicle vibration caused by the wave depth amplitude of rail corrugation is a highly regular time-series signal. Therefore, the in-vehicle noise signal caused by rail corrugation is processed using wavelet packet transformation to extract the normalized energy feature vector, which is then input into a wavelet neural network to identify the characteristic wavelength of rail corrugation. The vibration signal time-series vector is input into an LSTM neural network to identify the wave depth amplitude of rail corrugation. This constructs a metro rail corrugation detection model based on the aforementioned feature vectors. By combining in-vehicle vibration and noise signals, accurate identification of both the characteristic wavelength and wave depth amplitude of rail corrugation can be achieved simultaneously.

[0106] Compared to existing intelligent identification methods for rail corrugation, in this embodiment 2, after training with wavelet neural networks and LSTM neural networks, the collected in-vehicle vibration and noise signals are input into the trained subway rail corrugation detection model, which can quickly and efficiently achieve accurate identification of the characteristic wavelength and wave depth amplitude of rail corrugation.

[0107] The process of segmenting vibration and noise signals into different length segments of spatial domain information after displacement space transformation specifically includes:

[0108] Integrating the actual operating speed curve of the subway train, we obtain the curve of the train displacement changing with time, which is then calculated and determined by formula (1).

[0109]

[0110] By determining the spatial intercept length ω, the vibration and noise time domain of length T is segmented to obtain N segments of vibration and noise time domain information (0~T1, T1~T2, …, T). n-1 ~T n ), determined by formulas (2) to (3);

[0111]

[0112]

[0113] In the formula, T is the total test time, L is the total train mileage, and ω is the spatial intercept length.

[0114] The aforementioned preprocessing of the spatial domain information of vibration, noise, and rail corrugation in each segment, extracting the time series vector of the vibration signal, the normalized energy feature vector of the noise signal, and the spatial domain feature statistics and frequency domain feature statistics of the rail corrugation signal, specifically includes:

[0115] The vibration spatial domain information of each segment is preprocessed, bandpass filtering is used to remove other interference, the sampling window length is set, and time series with the same time length are randomly sampled from the vibration spatial domain information of each segment to form a time series vector.

[0116] Wavelet packet transform is used to decompose and reconstruct the spatial domain information of each segment of noise in order to eliminate signal noise and extract normalized energy feature vectors of different frequency bands.

[0117] Frequency domain transformation was performed on the spatial domain information of rail corrugation in each segment using a 1 / 3 octave band, and frequency domain feature statistics of the spatial domain information of rail corrugation in each segment were extracted.

[0118] The aforementioned time series with the same random sampling length specifically include:

[0119] Obtain the time-domain information of each segment of vibration after displacement space transformation (0~T1, T1~T2, …, T n-1 ~T n Due to the different train speeds, the lengths of the vibration time-domain information for each segment are different.

[0120] Determine the minimum length L of the vibration time domain information for each segment. min The sampling window length l is determined by randomly sampling time series of the same length from the vibration time domain information of each segment, and is determined by formulas (4) to (5).

[0121] L min =min(L(0~T1),L(T1~T2),...,L(T n-1 ~T n (4)

[0122]

[0123] In the formula, α is the number of each segmented sampling window;

[0124] The frequency domain characteristic statistics include the characteristic wavelength of rail corrugation, and the spatial domain characteristic statistics include the peak-to-peak average value of the moving wave depth amplitude of rail corrugation and the percentage of peak-to-peak average value exceeding the limit, which are determined by formulas (6) to (10).

[0125] PPR(x,L)=(a1+a2+...+a n ) / n (6)

[0126]

[0127]

[0128]

[0129]

[0130] In the formula, PPR(x,L) is the peak-to-peak average amplitude of the moving wave depth of the rail corrugation, and a i Let L be the wave depth amplitude, L be the analysis window length, and n be the number of wave depth amplitude values ​​within the analysis window length. i This represents the percentage of peak-to-peak average values ​​exceeding the limit in different intervals.

[0131] The multi-source, multi-dimensional model input feature vector is a model input feature vector composed of two dimensions generated by two sources (vibration and noise);

[0132] The input feature vector of the multi-source multidimensional model is represented as {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}.

[0133] Specifically, for multi-classification problems, hotspot coding is used to represent rail corrugation categories. The characteristic wavelength label set for rail corrugation is {no corrugation, 30-60mm, 60-90mm, 90-120mm, 120-150mm, 150-180mm, 180-210mm, 210-240mm, 240-270mm, 270-300mm}, and the wave depth amplitude label set for rail corrugation is {0-0.1mm, 0.1-0.2mm, 0.2-0.3mm, and above 0.3mm}.

[0134] The aforementioned subway rail corrugation detection model based on automatic feature extraction consists of two parts: an LSTM neural network and a linearly constrained wavelet neural network. Vibration is input into the LSTM neural network to obtain the rail corrugation wave depth amplitude category, and noise is input into the linearly constrained wavelet neural network to obtain the rail corrugation characteristic wavelength category.

[0135] The linearly constrained wavelet neural network described above obtains a linear mapping matrix G by utilizing the relationship between vehicle speed, wavelength, and frequency, and applies the linear mapping matrix G to the output of the neural network so that the output satisfies specific constraints, which are determined by formulas (11) to (12).

[0136]

[0137] CGp = 0 (12)

[0138] In the formula, λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

[0139] The training of the subway rail corrugation detection model based on automatic feature extraction includes:

[0140] The sample sets composed of vibration and noise input feature vectors are divided into training set, validation set and test set respectively; the ratio of training set, validation set and test set is 6:2:2.

[0141] The vibration training set and noise training set are input into the subway rail corrugation detection model based on automatic feature extraction, and the optimal parameters of the model are found by K-fold cross-validation and Bayesian search algorithm.

[0142] The validation set is input into a subway rail corrugation detection model based on automatic feature extraction, which uses the model's optimal parameters, to validate the model.

[0143] The test set is input into the trained subway rail corrugation detection model based on automatic feature extraction to detect rail corrugation and obtain the rail corrugation status.

[0144] In this method, the K-fold cross-validation divides the training set into K parts, with one part serving as the validation set and the remaining K-1 parts serving as the training set. The training is performed K times in total, with different parts being selected as the validation set and the remaining K-1 parts serving as the training set. After the traversal is complete, the best performing set is selected as the optimal parameters.

[0145] In this embodiment 2, a portable device for detecting rail corrugation in subways based on automatic feature extraction is provided based on the above method. This portable device supports the aforementioned method for detecting rail corrugation in subways based on automatic feature extraction. The portable device includes:

[0146] The sensor unit is used to acquire in-vehicle vibration and noise signals at different time periods. The vibration and noise signals are non-steady-state signals generated when the urban rail transit vehicle is in motion.

[0147] The acquisition unit is used to acquire vibration and noise signals to obtain vibration and noise time-domain information.

[0148] The data preprocessing unit is used to segment and denoise the vibration and noise time-domain information, extract the time series vectors of vibration signals in each segment, and the normalized energy feature vectors of noise signals to form the input feature vectors of the multi-source multi-dimensional model.

[0149] The rail corrugation detection unit deploys a trained subway rail corrugation detection model based on automatic feature extraction onto the device to detect the input vibration and noise dataset and obtain the rail corrugation status.

[0150] The visualization unit prints out the identified rail corrugation status (corrugation mileage, characteristic wavelength, and corrugation depth amplitude);

[0151] The fixed unit firmly attaches the portable device to the floor surface inside the subway car to ensure accurate vibration information of the vehicle is obtained.

[0152] The sensor unit integrates an acceleration sensor and a sound pressure sensor into a portable device, and the difference between the sound pressure signal in the portable device and the sound pressure signal in the actual sound field environment is represented by the sound pressure loss curve, which is specifically determined by formula (13).

[0153] S ri =S oi +L i (13)

[0154] In the formula, S ri S represents the sound pressure level at a point in the actual sound field at the center frequency i of 1 / 3 octave band. oi L represents the sound pressure level at a point on a portable device at the center frequency i of 1 / 3 octave band. i The sound pressure loss at the center frequency i of 1 / 3 octave band is determined experimentally;

[0155] The fixing unit consists of an adhesive material and a pressing device. Air pressure is used to push the adhesive material into a specific groove at the bottom of the portable device, thereby firmly adhering the device to the floor surface. When removing the device, air is extracted to release pressure and separate the device from the floor.

[0156] Example 3

[0157] like Figure 2 and Figure 3 As shown, this embodiment 3 provides a method for detecting rail corrugation in subways based on automatic feature extraction. The method includes:

[0158] Obtain vibration and noise signals inside a subway train at different time periods, and obtain time-domain information of in-train vibration and noise based on the vibration and noise signals;

[0159] Acquire the spatial distribution of subway rail corrugation signals along the operating mileage, and obtain rail corrugation spatial domain information based on the rail corrugation signals;

[0160] The vibration and noise signals are divided into segmented vibration and noise spatial domain information of different lengths through displacement space transformation. The length of each segment of vibration and noise spatial domain information is equal to that of rail corrugation spatial domain information.

[0161] The vibration, noise and rail corrugation spatial domain information of each segment are preprocessed to extract the vibration signal time series vector, the noise signal normalized energy feature vector, the spatial domain feature statistics and frequency domain feature statistics of the rail corrugation signal of each segment.

[0162] Based on the time series vector and the normalized energy feature vector, a multi-source, multi-dimensional model input feature vector is obtained;

[0163] Based on the multi-source, multi-dimensional model input feature vector, a subway rail corrugation detection model based on automatic feature extraction is constructed, and the subway rail corrugation detection model based on automatic feature extraction is trained.

[0164] Input the in-vehicle vibration and noise data to be detected into the subway rail corrugation detection model based on automatic feature extraction to obtain the rail corrugation state.

[0165] In this example, the process of segmenting vibration and noise signals into different lengths of segmented vibration and noise spatial domain information through displacement space transformation specifically includes:

[0166] Integrating the actual operating speed curve of the subway train, we obtain the curve of the train displacement changing with time, which is then calculated and determined by formula (1).

[0167]

[0168] By determining the spatial intercept length ω, the vibration and noise time domain of length T is segmented to obtain N segments of vibration and noise time domain information (0~T1, T1~T2, …, T). n-1 ~T n ), determined by formulas (2) to (3);

[0169]

[0170]

[0171] In the formula, T is the total test time, L is the total train mileage, and ω is the spatial intercept length.

[0172] In this example, the preprocessing of the vibration, noise, and rail corrugation spatial domain information of each segment, and the extraction of the vibration signal time series vector, the noise signal normalized energy feature vector, and the spatial domain feature statistics and frequency domain feature statistics of the rail corrugation signal for each segment, specifically includes:

[0173] The vibration spatial domain information of each segment is preprocessed, bandpass filtering is used to remove other interference, the sampling window length is set, and time series with the same time length are randomly sampled from the vibration spatial domain information of each segment to form a time series vector.

[0174] Wavelet packet transform is used to decompose and reconstruct the spatial domain information of each segment of noise in order to eliminate signal noise and extract normalized energy feature vectors of different frequency bands.

[0175] Frequency domain transformation was performed on the spatial domain information of rail corrugation in each segment using a 1 / 3 octave band, and frequency domain feature statistics of the spatial domain information of rail corrugation in each segment were extracted.

[0176] In this example, the time series with the same random sampling length specifically includes:

[0177] Obtain the time-domain information of each segment of vibration after displacement space transformation (0~T1, T1~T2, …, T n-1 ~T n Due to the different train speeds, the lengths of the vibration time-domain information for each segment are different.

[0178] Determine the minimum length L of the vibration time domain information for each segment. min The sampling window length l is determined by randomly sampling time series of the same length from the vibration time domain information of each segment, and is determined by formulas (4) to (5).

[0179] L min =min(L(0~T1),L(T1~T2),...,L(T n-1 ~T n (4)

[0180]

[0181] In the formula, α is the number of each segmented sampling window;

[0182] The frequency domain characteristic statistics include the characteristic wavelength of rail corrugation, and the spatial domain characteristic statistics include the peak-to-peak average value of the moving wave depth amplitude of rail corrugation and the percentage of peak-to-peak average value exceeding the limit, which are determined by formulas (6) to (10).

[0183] PPR(x,L)=(a1+a2+...+a n ) / n (6)

[0184]

[0185]

[0186]

[0187]

[0188] In the formula, PPR(x,L) is the peak-to-peak average amplitude of the moving wave depth of the rail corrugation, and a i Let L be the wave depth amplitude, L be the analysis window length, and n be the number of wave depth amplitude values ​​within the analysis window length. i This represents the percentage of peak-to-peak average values ​​exceeding the limit in different intervals.

[0189] In this example, the multi-source, multi-dimensional model input feature vector is a model input feature vector composed of two dimensions generated by two sources (vibration and noise);

[0190] The input feature vector of the multi-source multidimensional model is represented as {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}.

[0191] Specifically, for multi-classification problems, hotspot coding is used to represent rail corrugation categories. The characteristic wavelength label set for rail corrugation is {no corrugation, 30-60mm, 60-90mm, 90-120mm, 120-150mm, 150-180mm, 180-210mm, 210-240mm, 240-270mm, 270-300mm}, and the wave depth amplitude label set for rail corrugation is {0-0.1mm, 0.1-0.2mm, 0.2-0.3mm, and above 0.3mm}.

[0192] In this example, the rail corrugation detection model based on automatic feature extraction consists of two parts: an LSTM neural network and a linearly constrained wavelet neural network. Vibration is input into the LSTM neural network to obtain the rail corrugation depth category, and noise is input into the linearly constrained wavelet neural network to obtain the rail corrugation characteristic wavelength category.

[0193] The linearly constrained wavelet neural network described above obtains a linear mapping matrix G by utilizing the relationship between vehicle speed, wavelength, and frequency, and applies the linear mapping matrix G to the output of the neural network so that the output satisfies specific constraints, which are determined by formulas (11) to (12).

[0194]

[0195] CGp = 0 (12)

[0196] In the formula, λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

[0197] In this example, the training of the subway rail corrugation detection model based on automatic feature extraction includes:

[0198] The sample sets composed of vibration and noise input feature vectors are divided into training set, validation set and test set respectively; the ratio of training set, validation set and test set is 6:2:2.

[0199] The vibration training set and noise training set are input into the subway rail corrugation detection model based on automatic feature extraction, and the optimal parameters of the model are found by K-fold cross-validation and Bayesian search algorithm.

[0200] The validation set is input into a subway rail corrugation detection model based on automatic feature extraction, which uses the model's optimal parameters, to validate the model.

[0201] The test set is input into the trained subway rail corrugation detection model based on automatic feature extraction to detect rail corrugation and obtain the rail corrugation status.

[0202] In this method, the K-fold cross-validation divides the training set into K parts, with one part serving as the validation set and the remaining K-1 parts serving as the training set. The training is performed K times in total, with different parts being selected as the validation set and the remaining K-1 parts serving as the training set. After the traversal is complete, the best performing set is selected as the optimal parameters.

[0203] In this embodiment, considering the significant differences in the time-frequency characteristics of in-vehicle vibration and noise signals generated when urban rail transit vehicles travel in corrugated sections and normal sections, and the relatively constant speed of the train within a certain spatial domain, the in-vehicle noise signal caused by rail corrugation exhibits specific frequency band energy concentrations according to the fixed wavelength theorem. Furthermore, the in-vehicle vibration caused by the wave depth amplitude of rail corrugation is a highly regular time-series signal. Therefore, wavelet packet transform is performed on the in-vehicle noise signal caused by rail corrugation to extract the normalized energy feature vector of the noise signal, which is then input into a wavelet neural network to identify the characteristic wavelength of rail corrugation. The time-series vector of the vibration signal is input into an LSTM neural network to identify the wave depth amplitude of rail corrugation. A metro rail corrugation detection model based on these feature vectors is constructed. By combining in-vehicle vibration and noise signals, accurate identification of both the characteristic wavelength and wave depth amplitude of rail corrugation can be achieved simultaneously. After training with wavelet neural networks and LSTM neural networks, the collected vibration and noise signals inside the vehicle can be input into the trained subway rail corrugation detection model to quickly and efficiently achieve accurate identification of the characteristic wavelength and wave depth amplitude of rail corrugation.

[0204] Example 4

[0205] like Figures 1 to 8 As shown, the difference between this embodiment and embodiment 3 is that this embodiment is based on the automatic feature extraction-based subway rail corrugation detection method of embodiment 1, and is implemented as follows:

[0206] The process consists of seven steps: 1) Measure vibration and noise signals inside the subway train at different time periods to obtain vibration and noise time-domain information; 2) Perform displacement space transformation and segmentation on the vibration and noise time-domain information to obtain vibration and noise spatial domain information corresponding to each segment; 3) Randomly sample time series of the same duration for each segment of vibration spatial domain information to form a time series vector; 4) Extract normalized energy feature vectors by performing wavelet packet transform on each segment of noise spatial domain information; 5) Construct the model input feature vectors {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}; 6) Divide the model input feature vectors composed of vibration and noise into training set, validation set, and test set, and train and evaluate the subway rail corrugation detection model based on automatic feature extraction; 7) Identify the vibration and noise data in the test set to obtain the rail corrugation state.

[0207] The equipment used includes:

[0208] An accelerometer sensor is installed on the floor inside a subway train to measure vibration signals of the floor.

[0209] A sound pressure sensor, which is installed inside a subway train car, is used to measure the noise signal inside the car;

[0210] The data acquisition module is connected to the accelerometer and the sound pressure sensor to collect vibration and noise signals and obtain vibration and noise time-domain information.

[0211] 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.

[0212] 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 noise signals generated by rail corrugation are obtained, so as to achieve the best training effect.

[0213] The following are examples of implementation methods:

[0214] (1) Data description: On-site test was conducted on a section of a subway line in a certain city in China. The vibration and noise signals inside the vehicle were obtained using the multi-channel 24AD high-precision data acquisition system of Dongfang Institute.

[0215] (2) Data Segmentation: The time-domain information of vibration and noise from different time periods is transformed by displacement space and segmented to obtain segmented vibration and noise spatial domain information of equal spatial length, such as... Figure 4 As shown.

[0216] (3) Input vector: A time series vector is constructed by randomly sampling time series of the same length from each segment of the vibration spatial domain information, such as... Figure 5 As shown; wavelet packet transform is performed on the spatial domain information of each segment of noise to extract normalized energy feature vectors for different frequency bands; such as Figure 6 As shown; the spatial domain information of rail corrugation in each segment is transformed into the frequency domain using a 1 / 3 octave band, and the frequency domain feature statistics of the spatial domain information of rail corrugation in each segment are extracted, such as... Figure 7 As shown; the spatial domain information of rail corrugation in each segment is processed to calculate the peak-to-peak average value and the percentage of peak-to-peak average value exceeding the limit for the moving wave depth amplitude of rail corrugation, such as... Figure 8 As shown.

[0217] (4) Constructing a sample dataset: The input vectors processed above are used to construct multi-source, multi-dimensional model input feature vectors such as {noise normalized energy feature vector; rail corrugation feature wavelength} and {vibration time series vector; rail corrugation wave depth amplitude} to form a sample dataset.

[0218] (5) Divide the sample dataset: Randomly divide the sample dataset into training set, validation set and test set, with the ratio of training set, validation set and test set being 6:2:2.

[0219] (6) Model training: Input the training set into the model, perform K-fold cross-validation and Bayesian search to obtain the optimal parameters of the model, and use the optimal parameters of the model to obtain the optimal model for detecting rail corrugation in subways.

[0220] (7) Model validation: Input the test set into the optimal model for detecting rail corrugation in subways and evaluate the model's accuracy and generalization ability.

[0221] (8) Metro rail corrugation detection based on automatic feature extraction: Input the in-vehicle vibration and noise samples to be detected into the metro rail corrugation detection model based on automatic feature extraction to obtain the rail corrugation state.

[0222] Example 5

[0223] like Figures 9-11 As shown, the difference between this embodiment and Embodiment 3 is that this embodiment provides a portable device for detecting subway rail corrugation based on automatic feature extraction. This portable device supports the aforementioned subway rail corrugation detection method based on automatic feature extraction. The portable device includes:

[0224] The sensor unit is used to acquire in-vehicle vibration and noise signals at different time periods. The vibration and noise signals are non-steady-state signals generated when the urban rail transit vehicle is in motion.

[0225] The acquisition unit is used to acquire vibration and noise signals to obtain vibration and noise time-domain information.

[0226] The data preprocessing unit is used to segment and denoise the vibration and noise time-domain information, extract the time series vectors of vibration signals in each segment, and the normalized energy feature vectors of noise signals to form the input feature vectors of the multi-source multi-dimensional model.

[0227] The rail corrugation detection unit deploys a trained subway rail corrugation detection model based on automatic feature extraction onto the device to detect the input vibration and noise dataset and obtain the rail corrugation status.

[0228] The visualization unit prints out the identified rail corrugation status (corrugation mileage, characteristic wavelength, and corrugation depth amplitude);

[0229] The fixed unit firmly attaches the portable device to the floor surface inside the subway car to ensure accurate vibration information of the vehicle is obtained.

[0230] The sensor unit integrates an acceleration sensor and a sound pressure sensor into a portable device, and the difference between the sound pressure signal in the portable device and the sound pressure signal in the actual sound field environment is represented by the sound pressure loss curve, which is specifically determined by formula (13).

[0231] S ri =S oi +L i (13)

[0232] In the formula, Sri is the sound pressure value at a point in the actual sound field at the center frequency i of the 1 / 3 octave band, Soi is the sound pressure value at a point in the actual sound field at the center frequency i of the 1 / 3 octave band as measured by a portable device, and Li is the sound pressure loss value at the center frequency i of the 1 / 3 octave band, which is determined by experiments.

[0233] The fixing unit consists of an adhesive material and a pressing device. Air pressure is used to push the adhesive material into a specific groove at the bottom of the portable device, thereby firmly adhering the device to the floor surface. When removing the device, air is extracted to release pressure and separate the device from the floor.

[0234] The execution process of each unit can be carried out according to the steps of the subway rail corrugation detection method based on automatic feature extraction described in Example 3. These steps will not be repeated in this example.

[0235] Example 6

[0236] This embodiment 6 provides a non-transitory computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they implement the subway rail corrugation detection method based on automatic feature extraction as described above.

[0237] Example 7

[0238] This embodiment 7 provides a computer program product, including a computer program that, when run on one or more processors, is used to implement the subway rail corrugation detection method based on automatic feature extraction as described above.

[0239] Example 8

[0240] This embodiment 8 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes instructions to implement the subway rail corrugation detection method based on automatic feature extraction as described above.

[0241] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0242] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0243] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0244] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0245] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. A method for detecting rail corrugation in subway trains based on automatic feature extraction, characterized in that, include: Acquire vibration and noise data inside the subway train to be tested; Using a pre-trained detection model, the acquired in-vehicle vibration data and in-vehicle noise data are processed to obtain the rail corrugation state of the subway. The corrugation state includes the characteristic wavelength and depth amplitude of the corrugation. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network segments the vibration and noise data into vibration spatial domain information and noise spatial domain information through displacement space transformation. The vibration signal recognition network processes the vibration spatial domain information to obtain the depth amplitude of the corrugation. The noise signal recognition network processes the noise spatial domain information to obtain the characteristic wavelength of the corrugation. The noise signal recognition network is a linearly constrained wavelet neural network. This linearly constrained wavelet neural network uses the relationship between vehicle speed, wavelength, and frequency to obtain a linear mapping matrix G, which is then applied to the neural network output to satisfy specific constraints. ; Where λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

2. The method for detecting rail corrugation in subways based on automatic feature extraction according to claim 1, characterized in that, Vibration and noise data are segmented into vibration spatial domain information and noise spatial domain information through displacement spatial transformation, including: integrating the actual operating speed curve of the subway train to obtain the curve of train displacement changing with time. Determine the spatial intercept length ω, and cut the vibration and noise time domain of length T to obtain N segments of vibration time domain information and noise time domain information (0~T1, T1~T2, ..., T). n-1 ~T n ): , In the formula, T is the total test time, L is the total train mileage, and ω is the spatial intercept length. This indicates the train's speed.

3. The method for detecting rail corrugation in subways based on automatic feature extraction according to claim 1, characterized in that, When training the detection model, the spatial domain information of vibration, noise, and rail corrugation in each segment is preprocessed to extract the time series vector of vibration signal, the normalized energy feature vector of noise signal, and the spatial domain feature statistics and frequency domain feature statistics of rail corrugation signal for each segment. This includes: preprocessing the spatial domain information of vibration in each segment, bandpass filtering to eliminate other interference, setting the sampling window length, randomly sampling time series of the same time length from the spatial domain information of vibration in each segment to form a time series vector; using wavelet packet transform to decompose and reconstruct the spatial domain information of noise in each segment to eliminate signal noise, and extracting the normalized energy feature vector of different frequency bands; and using 1 / 3 octave band to perform frequency domain transformation on the spatial domain information of rail corrugation in each segment, and extracting the frequency domain feature statistics of the spatial domain information of rail corrugation in each segment.

4. The method for detecting rail corrugation in subways based on automatic feature extraction according to claim 3, characterized in that, The aforementioned time series with the same random sampling length specifically include: Obtain the time-domain information of each segment of vibration after displacement space transformation (0~T1, T1~T2, ..., T). n-1 ~T n Due to the different train speeds, the time-domain information lengths of vibration in each segment are different; Determine the minimum length L of the vibration time domain information for each segment. min The sampling window length is l, and time series of the same length are randomly sampled from the vibration time domain information of each segment: ； ； In the formula, α is the number of sampling windows for each segment; The frequency domain characteristic statistics include the characteristic wavelength of rail corrugation, and the spatial domain characteristic statistics include the peak-to-peak average amplitude of the moving wave depth of rail corrugation and the percentage of peak-to-peak average exceeding the limit. ； ； ； ； ； In the formula, PPR(x, L) is the peak-to-peak average amplitude of the moving wave depth of rail corrugation, and a i Let L be the wave depth amplitude, L be the analysis window length, and n be the number of wave depth amplitude values ​​within the analysis window length. i This represents the percentage of peak-to-peak average values ​​exceeding the limit in different intervals.

5. The method for detecting rail corrugation in subways based on automatic feature extraction according to claim 4, characterized in that, The input feature vectors of the detection model are represented as {noise normalized energy feature vector; rail corrugation characteristic wavelength} and {vibration time series vector; rail corrugation wave depth amplitude}.

6. A subway rail corrugation detection system based on automatic feature extraction, characterized in that, include: The acquisition module is used to acquire the vibration data and noise data inside the subway train to be detected. The detection module processes the acquired in-vehicle vibration and noise data using a pre-trained detection model to obtain the rail corrugation state of the subway. The corrugation state includes the characteristic wavelength and depth amplitude of the corrugation. The detection model includes a data segmentation network, a vibration signal recognition network, and a noise signal recognition network. The data segmentation network segments the vibration and noise data into vibration spatial domain information and noise spatial domain information through displacement space transformation. The vibration signal recognition network processes the vibration spatial domain information to obtain the depth amplitude of the corrugation. The noise signal recognition network processes the noise spatial domain information to obtain the characteristic wavelength of the corrugation. The noise signal recognition network is a linearly constrained wavelet neural network. This linearly constrained wavelet neural network uses the relationship between vehicle speed, wavelength, and frequency to obtain a linear mapping matrix G, which is then applied to the neural network output to satisfy specific constraints. ; Where λ is the rail corrugation wavelength, v is the train speed, f is the vehicle passing frequency, C is the constraint vector, G is the mapping matrix, and p is the neural network output.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the subway rail corrugation detection method based on automatic feature extraction as described in any one of claims 1-5.

8. A computer program product, characterized in that, The method includes a computer program that, when run on one or more processors, implements the subway rail corrugation detection method based on automatic feature extraction as described in any one of claims 1-5.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions that implement the subway rail corrugation detection method based on automatic feature extraction as described in any one of claims 1-5.

Images