Railway roadbed disease rapid diagnosis method based on spectrogram analysis

Abstract

The invention discloses a railway roadbed disease rapid diagnosis method based on spectrogram analysis, and relates to the technical field of signal processing and track engineering. The method comprises the following steps: firstly, acquiring acceleration signal samples influenced by typical roadbed diseases by using acceleration sensors mounted at different positions of a vehicle and a roadbed; secondly, performing multi-source information fusion by adopting DS data fusion; combining different frequency scale features to construct a disease feature vector, and using a spectrogram theory to form a railway roadbed disease spectrogram; and then, carrying out spectral analysis on the acceleration signal monitored in real time, and comparing with a disease spectrum library to judge the roadbed disease. The method has the beneficial effects that the health condition of the railroad bed can be reflected from different angles and levels through the fused signal source, the limitation of a single detection means is overcome, and the comprehensiveness of disease detection is improved; the disease degree and position can be accurately evaluated by using a spectrogram theory and a positioning system, the railway roadbed disease diagnosis efficiency is remarkably improved, and the monitoring cost is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the fields of signal processing and track engineering technology and the field of bridge structure passive control technology, and in particular to a method for rapid diagnosis of railway roadbed diseases based on spectrum analysis. Background Art

[0002] Railway subgrade is the foundation of railway tracks, and its health is crucial to railway transportation safety. However, due to the influence of various factors such as natural environment, traffic load and material aging, railway subgrade may suffer from various diseases, such as uneven settlement, mud oozing and slipping, etc. If these diseases are not discovered and treated in time, they may have serious consequences.

[0003] At present, the detection of railway subgrade diseases mainly relies on regular manual inspections or single detection technology. Such methods have the following limitations. First, the efficiency of manual re-inspection is low and it is greatly affected by subjective factors, making it difficult to achieve comprehensive coverage. In addition, the detection equipment is complex, requiring high technical quality of the detection personnel, and the detection efficiency is low; secondly, a single detection technology can often only reflect one aspect of the characteristics of the subgrade disease and cannot comprehensively evaluate the health of the subgrade. In addition, the existing technology also has shortcomings in real-time monitoring and early warning, making it difficult to achieve rapid response and processing of diseases. With the development of information technology and signal analysis technology, multi-source information fusion technology and signal analysis methods based on spectral theory have provided new ideas for the detection of railway subgrade diseases. Multi-source information fusion technology can integrate multiple signal sources, and through data level integration and feature level extraction, it can achieve comprehensive, real-time analysis and accurate diagnosis of subgrade diseases. However, how to effectively integrate different information sources and how to quickly and accurately diagnose diseases from real-time detection signals are still technical problems that need to be solved urgently.

[0004] Aiming at the problems of low detection efficiency, insufficient accuracy and poor real-time performance in the existing railway subgrade disease monitoring technology, the present invention proposes a rapid diagnosis method for railway subgrade diseases in the time-frequency domain based on multi-source information fusion and spectral graph theory. The method uses time-frequency domain analysis technology to convert time-domain detection signals into signal statistical parameters in the time-frequency domain, and forms a disease spectrum library based on confidence assignment function, DS data fusion and spectral graph theory. On this basis, the real-time detection signal is diagnosed based on the correlation of characteristic vectors, aiming to improve the efficiency, accuracy and real-time performance of railway subgrade disease detection and provide strong guarantee for railway transportation safety. Summary of the invention

[0005] The purpose of the present invention is to provide a rapid diagnosis method for railway roadbed defects based on spectrum analysis. The method first collects acceleration time-history defect signal samples at different positions of vehicles and roadbeds during train travel using multi-source signal fusion technology; then, based on spectrum theory, the fused defect feature vectors are grouped into defect structures at different frequency scales to form a defect spectrum library; finally, the real-time detection of multi-source acceleration signals, satellite positioning systems, and defect spectrum libraries, as well as the correlation of characteristic vectors based on the spectrum, is combined to achieve real-time diagnosis and detection of defect types, degrees, locations, etc.

[0006] The technical solution adopted by the present invention is: a rapid diagnosis method for railway subgrade diseases based on spectrum analysis, the method steps are as follows:

[0007] Step 1: Obtain typical railway subgrade disease signals;

[0008] Acceleration sensors are installed on the train body, bogie, wheelset, track plate, mortar layer, base, and subgrade surface to obtain the vibration acceleration typical disease signals (x i (t));

[0009] Step 2: Time-frequency domain transformation of typical disease signals and construction of disease feature vectors;

[0010] The typical disease signal is cleaned based on autocorrelation, and the mean of the upper and lower envelopes of the signal is constructed from the cleaned typical disease signal. The eigenfunction is obtained based on the bipolarity and symmetry of the signal, and then its frequency characteristics are analyzed to determine its stability. The stationary and non-stationary eigenfunctions are added separately to obtain the stationary and non-stationary components of the typical disease sample signal (Y S (t) and Y NS (t));

[0011] The obtained signal stationary and non-stationary components are subjected to Fourier and wavelet transforms to obtain the spectrum of the stationary and non-stationary sub-signals (Y S (ω) and Y NS (ω)), construct the defect spectrum feature vector δY(ω) based on the energy difference of frequency octave, harmonic and wavelet based on the time-frequency spectrum of stationary and non-stationary sub-signals;

[0012] Step 3: Disease sample signal fusion;

[0013] According to the disease characteristic equation, trust assignment function, probability distribution trust, and DS synthesis rule, the obtained disease spectrum feature vector δY(ω) is used to construct fused disease information;

[0014] Step 4: Disease signal grouping;

[0015] After signal fusion, the probability distribution confidence of typical diseases is generated (m N (A i )), based on the spectrum theory, the fused disease feature vector δY(ω) is grouped into disease structures at different frequency scales to form multiple typical amplitude-frequency characteristic spectra of railway subgrade diseases, and the amplitude-frequency characteristic spectra and typical railway subgrade disease types are numbered;

[0016] Step 5: Real-time disease identification of monitoring signals;

[0017] Collect the vehicle and roadbed acceleration signals during the operation of the train in real time, repeat steps one to three, and determine the defect type of the vibration signal to be tested based on the correlation of the characteristic vector of the spectrum, that is, perform correlation analysis on the frequency spectrum feature vector of the vibration data to be tested and the common feature vector of the spectrum of each type of typical railway roadbed disease, and determine whether the frequency spectrum feature vector is within the spectrum family corresponding to the common disease. If so, it is considered that such typical disease has occurred in the roadbed at the corresponding position of the signal. At this time, the location where this type of disease occurs is given according to the location information recorded by the satellite positioning system; if not, it is considered that the roadbed at the corresponding position of the monitoring signal is in normal condition. Then, the remaining characteristic spectra of other types of diseases are continuously judged;

[0018] Step 6: Manual re-inspection and repair of roadbed defects;

[0019] The test results are fed back to the visualization module and manual re-inspection system in real time. The inspectors conduct manual re-inspection of the disease alarm location, locate the disease in the 3D model according to the location of the disease, and record the structural part, disease type and degree of the disease in text and pictures.

[0020] The advantages of the present invention compared with the prior art are:

[0021] 1. The multi-source information fusion technology is used to fuse the acceleration time history signals from different positions of the vehicle and the roadbed. Different from the traditional single data source detection method, it effectively enhances the spatial diversity of the data and provides a comprehensive signal analysis, increasing the accuracy and robustness of the diagnosis;

[0022] 2. By constructing a disease spectrum feature vector based on frequency doubling, harmonics and wavelet energy difference, the present invention can more accurately characterize the characteristics of the disease;

[0023] 3. The spectral graph theory is used to structurally group the disease feature vectors to form a spectrum family of typical railway roadbed diseases. Different from the existing detection method based on time-course signals, the present invention uses Fourier transform and wavelet transform to convert the signal into frequency domain and time-frequency domain, which can quickly capture the time-varying characteristics of the signal and realize rapid diagnosis of railway roadbed diseases;

[0024] 4. By combining the generated spectrum of typical railway subgrade defects with a machine vector machine to determine the defects of monitoring signals, and locating the defects based on satellite positioning information, the present invention realizes automated real-time diagnosis and detection of defect types, degrees, and locations, eliminating tedious and complex manual inspection work and significantly improving the efficiency of railway subgrade defect diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of the implementation of a method for rapid diagnosis of railway subgrade diseases based on spectrum analysis of the present invention;

[0026] Figure 2 The acceleration time history signals of vehicles and embankment at different positions with some typical railway embankment diseases - including multiple figures ( Figure 2 (a) Figure 2 (b) Figure 2 (c) Figure 2 (d) Figure 2 (e) Figure 2 (f) Figure 2 (g) acceleration signals corresponding to the wheelset, frame, body of the vehicle and the track slab, mortar layer, base and surface layer of the subgrade on the roadbed respectively);

[0027] Figure 3 The signal regenerated based on autocorrelation processing for typical disease signals - including multiple figures ( Figure 3 (a) Figure 3 (b) and Figure 3 (c) The signal after autocorrelation cleaning, and the stationary and non-stationary signal components after decomposition, respectively);

[0028] Figure 4 is the time spectrum of the non-stationary signal component represented by the time spectrum; -- including multiple figures ( Figure 4 (a) and (b) correspond to the signal time-frequency diagrams with and without disease respectively)

[0029] Figure 5 is the characteristic vector of the typical disease spectrum;

[0030] Figure 6 This is a schematic diagram of disease diagnosis results. DETAILED DESCRIPTION

[0031] The present invention provides a method for rapid diagnosis of railway subgrade diseases based on spectrum analysis. The specific implementation process of the method is as follows: Figure 1 As shown, the specific process can be summarized into the following five steps:

[0032] Step 1: Obtain typical railway subgrade disease signals;

[0033] Use a vibration information collection system including acceleration sensors to obtain acceleration signals x of vehicles and different locations on the embankment (vehicles: wheelsets, bogies, car bodies; embankments: track plates, mortar cushions, bases, and embankment surface layers) with different typical railway embankment diseases (uneven embankment settlement, arching, sliding, mud bubbling, etc.). i ,i=1,2,3,...,7(corresponding to Figure 2 (a) Figure 2 (b) Figure 2 (c) Figure 2 (d) Figure 2 (e) Figure 2 (f) Figure 2 (g));

[0034] Step 2: Time-frequency domain transformation of typical disease signals and construction of disease feature vectors;

[0035] right Figure 2 The typical disease signal in the image is cleaned based on autocorrelation according to formula (1). Figure 2 (g) is taken as an example, the result after the autocorrelation is clear is as follows Figure 3 The autocorrelation signal in a is shown.

[0036]

[0037] Among them, x i (t) is the acceleration signal time history of the acceleration sensor at the i-th location; N is the number of signal detection positions; t is the time of detecting the signal.

[0038] Generate a white noise signal:

[0039] W(t)~N(0,1)(2)where N represents the white noise generation operator;

[0040] Construct a Toeplitz matrix:

[0041]

[0042] Where n represents the number of time points; Rx is the same as formula (1).

[0043] Filter the white noise to obtain the regenerated signal

[0044]

[0045] The regenerated signal The empirical mode decomposition is used to decompose the stationary and non-stationary signals regenerated by the autocorrelation to extract the characteristics of the diseased signal; the reconstructed signal Y(t) is subjected to empirical mode decomposition to obtain the intrinsic mode functions (IMFs) of the stationary and non-stationary parts (IMFsS (t) and IMF NS (t)):

[0046]

[0047] Among them, r n (t) is the final residual;

[0048] The stationary and non-stationary eigenfunctions are added together to obtain the stationary and non-stationary components of the typical disease sample signal (Y S (t) and Y NS (t), corresponding to Figure 3 b and Figure 3 c):

[0049]

[0050] Among them, S represents the index set of stationary I MF, and NS represents the index set of non-stationary I MF.

[0051] The obtained signal stationary and non-stationary components are constructed based on the spectrum analysis of the stable signal, the time-frequency domain analysis of the non-stationary signal components, and the reference deviation to construct the defect feature vector of the frequency multiple, harmonic and wavelet energy difference.

[0052]

[0053] Where k is the number of harmonics and harmonics, M is the number of selected scales; the harmonic energy E k ; Harmonic energy H k ; Wavelet energy W(a); The constructed spectrum feature vector is normalized by formula 9 to obtain the spectrum of the stationary and non-stationary sub-signals under disease and non-disease conditions ( Figure 4 ):

[0054]

[0055] Step 3: Disease sample signal fusion;

[0056] The obtained disease spectrum feature vector δY(ω) is constructed into fusion disease information according to the disease characteristic equation (Equation 10), trust assignment function (Equation 11), probability distribution trust (Equation 12), and DS synthesis rule (Equation 14);

[0057] The Signal The fusion is performed according to the following disease characteristic equation:

[0058]

[0059] Among them, δY(ω) and δZ are the disease characteristic vector and disease pattern vector respectively, δy i and δz i Its components, i = 1,...7, correspond to the vehicle body, bogie, wheelset, and the track plate, mortar layer, base and subgrade surface of the roadbed; H is the system matrix, which is given by a large amount of field measured data;

[0060] The disease pattern vector (δZ) needs to be solved in combination with the obtained disease characteristic vector (δY) and formula (1). Since the characteristic vector δY obtained from each monitoring sample of a specific roadbed disease has a certain degree of error and random changes, the disease pattern vector (δZ) obtained each time deviates from the true state to varying degrees. Therefore, it is necessary to calculate the trust function of multi-source disease diagnosis results and self-iterative DS signal fusion to obtain the signal z. i and signal z j For example, the basic confidence assignment function m between the two signals is:

[0061]

[0062] Among them, m i and m j is x i , x j The corresponding basic confidence assignment function; the similarity coefficient a of the two signal sources ij for

[0063]

[0064] When the similarity coefficient a between the evidences ij The closer it is to 1, the i With z j The more similar; when a ij =1, z i and z j Completely consistent; when a ij =1, z i and z j Complete Conflict:

[0065]

[0066] K(m i )∈(0,1), the larger the value, the more likely the disease mode z is. i On the contrary, the smaller the value, the lower the trust of the disease pattern xi.

[0067] According to the new probability distribution trust, the frequency domain statistical parameters of n information sources are synthesized n-1 times using the DS synthesis rule to weighted average the single set of evidence, so as to construct the fused disease information;

[0068]

[0069] Among them, A i For diseases, such as uneven settlement of roadbed; m i (i=1,2,…,n) is the disease A i The probability of assigning trust; K(m i ) is disease A i The weight of m N (A i ) is the confidence level of the new probability distribution generated after signal fusion;

[0070] Step 4: Disease signal grouping;

[0071] After signal fusion, the probability distribution trust of typical diseases is generated. Based on the spectral graph theory, the probability distribution trust (m N (A i )) and the fused defect feature vector δY(ω) are used to group the defect structures at different frequency scales to form multiple typical amplitude-frequency feature vector spectra Z of railway subgrade defects. In order to ensure the comparability of features at different scales, all features need to be standardized. Assume that the feature set is Z = {z1,z2,...,z n}, the standardized feature is Z′={z′1,z′2,...,z′ n The first step in using spectral graph theory is to construct the disease spectrum characteristics and statistical feature sets into a graph structure

[0072] G=(Z',E) (15)

[0073] Among them, Z' is the feature set, representing different feature vectors, and E is the edge set, representing the similarity between features;

[0074] Calculate the similarity between feature vectors, assuming that the feature vector z i and z j The similarity measure between them is w ij

[0075]

[0076] Among them, σ is an adjustable parameter used to control the scale of the similarity measure;

[0077] Construct an adjacency matrix A, where A ij Represents the similarity measure between node i and node j:

[0078] A ij =w ij (17)

[0079] Construct the degree matrix D, whose diagonal elements Dii Represents the degree of node i, that is, the sum of the weights of the edges connected to node i:

[0080] D ii =∑ j A ij (18)

[0081] Construct the Laplacian matrix L of the graph and perform eigenvalue decomposition:

[0082] L=DA (19)

[0083] Perform eigenvalue decomposition on the Laplace matrix L to obtain its eigenvalues ​​and eigenvectors:

[0084] Lv=λv (20)

[0085] Among them, λ is the eigenvalue and v is the eigenvector.

[0086] The eigenvectors of the Laplace matrix are used to group the disease features, and the first k eigenvectors of the Laplace matrix L are selected to form a matrix V = [v1, v2, …, v k ], map each node i to the feature vector space and obtain a new embedding representation:

[0087] y i =[v i1 ,v i2 ,…,v ik ] (twenty one)

[0088] Cluster analysis is performed on the embedded representation to divide the nodes into k spectral clusters, each of which represents an inherent structural group of diseases. According to the clustering results, the disease features of the same category are classified into one spectral cluster to construct the railway subgrade disease spectrum ( Figure 5 ).

[0089] Step 5: Real-time disease identification of monitoring signals;

[0090] Taking uneven roadbed settlement as an example, the vehicle and roadbed acceleration signals during train operation are collected in real time, and steps one to three are repeated to perform correlation analysis on the frequency spectrum feature vector of the vibration signal with uneven roadbed settlement to be tested and the common feature vector of the spectrum of each typical railway roadbed disease. At the same time, the location where this type of disease occurs is given according to the position information recorded by the satellite positioning system; if not, it is considered that the roadbed state at the corresponding position of the monitoring signal is normal. Then, the remaining characteristic spectra of other types of diseases are continuously judged. The frequency spectrum feature vector of the measured vibration data is correlated with the common feature vector of the spectrum of each typical railway roadbed disease, and finally the type and degree of the disease are reflected in the form of a rose diagram ( Figure 6), where diseases exceeding the threshold of 0.6 require manual re-inspection and maintenance.

[0091] Step 6: Manual re-inspection and repair of roadbed defects;

[0092] The inspectors will conduct manual re-inspection of the disease alarm location, and provide textual descriptions and pictures of the structural location, disease type, and disease severity of the disease.

[0093] The test results show that the method for rapid diagnosis of railway subgrade diseases in time and frequency domains described in the present invention overcomes the limitations of a single detection method after fusion of the signal source, improves the comprehensiveness of disease detection, and significantly improves the efficiency of railway subgrade disease diagnosis, which can effectively reduce monitoring costs.

[0094] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A rapid diagnosis method for railway subgrade diseases based on spectral analysis, which is a rapid diagnosis method for railway subgrade diseases in time and frequency domains based on multi-source information fusion and spectral theory, and is characterized by: Step 1: Obtain typical railway subgrade disease signals; Acceleration sensors are installed on the wheels, frame, and body of the vehicle, as well as on the track slab, mortar layer, base, and subgrade surface of the subgrade to collect vibration acceleration signals of the vehicle and subgrade affected by various typical subgrade diseases during train travel. Step 2: Time-frequency domain transformation of typical disease signals and construction of disease feature vectors; The acquired signal is processed by autocorrelation, and then the regenerated signal is decomposed into stationary and non-stationary signals; the decomposed stationary and non-stationary signals are transformed into frequency domain and time-frequency domain respectively; the transformed spectrum vector is used to construct the defect spectrum feature vector based on the frequency multiple, harmonic and wavelet energy difference; Step 3: Disease sample signal fusion; For the obtained disease spectrum feature vector, the fused disease information is constructed according to the disease characteristic equation, trust assignment function, probability distribution trust, and DS synthesis rule; Step 4: Disease signal grouping; Calculate the probability distribution trust of the fused signal, use the spectrum theory to calculate the corresponding disease feature vector and group the disease structure to generate the typical railway roadbed disease feature spectrum; Step 5: Real-time disease identification of monitoring signals; The acceleration signals of the vehicle and the roadbed during the train operation are collected in real time, and steps 1 to 3 are repeated to calculate the correlation between the spectral feature vector of the vibration data to be tested and the spectral feature vector of the typical railway roadbed disease, and determine whether the spectral feature vector contains typical diseases.

2. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 1 is characterized in that: The multi-source information fusion technology includes the following contents: autocorrelation regeneration of the collected disease signal samples, as shown in Formula 1, and removing the interference of random vibration signal components to highlight the disease signal components; Among them, x i (t) is the acceleration signal time history of the acceleration sensor at the i-th location; N is the number of signal detection locations; t is the time of signal detection; Generate a white noise signal: W(t)~N(0,1)(2)where N represents the white noise generation operator; Construct a Toeplitz matrix: Where n represents the number of time points; Rx is the same as formula (1); Filter the white noise to obtain the regenerated signal 3. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 1 or 2, characterized in that: The regenerated signal The empirical mode decomposition is used to perform stationary and non-stationary decomposition of the autocorrelation regenerated signal to extract the characteristics of the disease signal; Perform empirical mode decomposition to obtain the stationary and non-stationary intrinsic mode functions (IMFs): Among them, r n (t) is the final residual; Decompose the intrinsic mode function (IMF) into stationary and non-stationary signals Among them, S represents the index set of stationary I MF, and NS represents the index set of non-stationary I MF.

4. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 3 is characterized by: The stable signal Y S (t) is analyzed by Fourier transform, while the non-stationary signal Y NS (t) Part 1 uses wavelet transform to perform time-frequency spectrum analysis; based on the spectrum results of stable signal components, the spectrum time-domain statistics of unstable signal components, and the reference deviation, the defect feature vectors of frequency multiples, harmonics, and wavelet energy differences are constructed to construct the defect feature vector δY(ω): Where k is the number of harmonics and harmonics, M is the number of selected scales; the harmonic energy E k ; Harmonic energy H k ; Wavelet energy W(a); The constructed spectral feature vector is standardized to eliminate dimension and scale differences:

5. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 4 is characterized by: The method for acquiring the disease characteristic vector δY(ω) is: combining the basic confidence assignment function and the DS data fusion technology to fuse the multi-source information of the frequency doubling, harmonic and wavelet energy difference disease characteristic vector information.

6. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 5 is characterized by: The signals from different measuring points are fused according to the following disease characteristic equation: Among them, δY(ω) and δZ are the disease characteristic vector and disease pattern vector respectively, δy i and δz i Its components, i = 1,...7, correspond to the vehicle body, bogie, wheelset, and the track plate, mortar layer, base and subgrade surface of the roadbed; H is the system matrix, which is given by a large amount of field measured data; The disease pattern vector δZ needs to be solved in combination with the obtained disease characteristic vector δY and formula (1). Since the characteristic vector δY obtained from each monitoring sample of a specific roadbed disease has a certain degree of error and random changes, the disease pattern vector δZ obtained each time deviates from the true state to varying degrees. Therefore, it is necessary to calculate the trust function of multi-source disease diagnosis results and self-iterative DS signal fusion to obtain the signal z. i and signal z j For example, the basic confidence assignment function m between the two signals is: Among them, m i and m j is x i , x j The corresponding basic confidence assignment function; the similarity coefficient a of the two signal sources ij for When the similarity coefficient a between the evidences ij The closer it is to 1, the i With z j The more similar; when a ij =1, z i and z j Completely consistent; when a ij =1, z i and z j Complete Conflict: K(m i )∈(0,1), the larger the value, the more likely the disease mode z is. i On the contrary, the smaller the value, the lower the trust of the disease pattern xi.

7. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 6 is characterized by: According to the new probability distribution trust, the frequency domain statistical parameters of n information sources are synthesized n-1 times using the DS synthesis rule to weighted average the single set of evidence, so as to construct the fused disease information; Among them, A i For diseases, such as uneven settlement of roadbed; m i (i=1,2,…,n) is the disease A i The probability of assigning trust; K(m i ) is disease A i The weight of m N (A i ) is the confidence level of the new probability distribution generated after signal fusion; The disease fusion information combines the features of different time scales or frequency scales to construct the frequency multiple and harmonic disease feature vectors and the wavelet energy difference disease feature vectors to reveal the feature sets of the signal at different frequency scales.

8. The method for rapid diagnosis of railway subgrade diseases based on spectrum analysis according to claim 7 is characterized by: Using spectrum theory, the spectrum characteristics and statistical feature sets of the diseases are grouped into inherent structure of the diseases to form a railway subgrade disease database; Assume that the feature set is Z = {z1,z2,...,z n }, the standardized feature is Z′={z1′,z2′,...,z n The first step of using spectral graph theory is to construct the disease spectrum characteristics and statistical feature sets into a graph structure. G=(Z,E) (15) Among them, V is the feature set, representing different feature vectors, and E is the edge set, representing the similarity between features; Calculate the similarity between feature vectors, assuming that the feature vector z i and z j The similarity measure between them is w ij Among them, σ is an adjustable parameter used to control the scale of the similarity measure; Construct an adjacency matrix A, where A ij Represents the similarity measure between node i and node j: A ij =w ij (17) Construct the degree matrix D, whose diagonal elements D ii Represents the degree of node i, that is, the sum of the weights of the edges connected to node i: D ii =∑ j A ij (18) Construct the Laplacian matrix L of the graph and perform eigenvalue decomposition: L=DA (19) Perform eigenvalue decomposition on the Laplace matrix L to obtain its eigenvalues ​​and eigenvectors: Lv=λv (20) Among them, λ is the eigenvalue, v is the eigenvector; The eigenvectors of the Laplace matrix are used to group the disease features, and the first k eigenvectors of the Laplace matrix L are selected to form a matrix V = [v1, v2, …, v k ]; Map each node i to the feature vector space and obtain a new embedding representation: y i =[v i1 ,v i2 ,…,v ik ] (21) Cluster analysis is performed on the embedded representation to divide the nodes into k spectral clusters, each of which represents an inherent structural group of defects. According to the clustering results, the defect features of the same category are classified into a spectral cluster to construct a railway subgrade defect spectrum library. Each entry in the spectrum includes the defect category and spectral characteristics.

9. A rapid diagnosis system for railway subgrade diseases based on spectral analysis, the system being based on the rapid diagnosis method for railway subgrade diseases based on spectral analysis according to any one of claims 1 to 8, characterized in that: The characteristic vector correlation analysis of the spectrum characteristic parameters of the real-time acceleration signals of vehicles and roadbeds during train operation is carried out with the disease spectrum library, and a disease diagnosis conclusion is generated to determine whether there are roadbed diseases and the type of disease. At the same time, the Beidou satellite positioning system and visualization module are combined to display the type and location of the disease, and feedback is given to the manual inspection system in real time.

Images