Rail surface damage identification method, system and device based on improved discrete entropy

Abstract

The application discloses a rail surface damage identification method and system based on improved discrete entropy, and equipment. The method comprises the following steps: adopting a normal cumulative distribution function to perform standardization processing on an original rail profile data sequence; generating a vector group based on the standardized sequence; corresponding each vector in the vector group with each discrete mode, and calculating the weighted probability distribution of each discrete mode according to the complexity of each vector; calculating the improved discrete entropy entropy value by adopting Rényi entropy according to the weighted probability distribution; obtaining the entropy value distribution range of each type of damage; and identifying the rail surface damage based on the improved discrete entropy entropy value and the entropy value distribution range. The application can automatically and high-precisely realize the extraction of various damage characteristics, mine valuable information in the damage data, and provide decision-making help and information support for railway staff.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for heavy-haul railways, and in particular to a method, system and equipment for identifying rail surface damage based on improved discrete entropy. Background Technology

[0002] To date, there are two main methods for geometric measurement of rail profile damage: contact measurement and non-contact measurement. Contact measurement primarily involves mechanical contact, requiring the probe of the inspection equipment to contact the rail. This fault detection method has been widely used in the Chinese railway system, but it suffers from low measurement accuracy and high labor intensity, failing to meet the current needs of domestic railway development. A newer fault detection method is the non-contact rail profile inspection system equipped with a laser profile sensor. This system collects rail surface data using a laser sensor and forms the rail profile by establishing and transforming the coordinates of the data points. This new inspection technology boasts high automation, high accuracy, and fast inspection speed. Using this system and the track profile data it collects and processes, we can extract monitoring data features and identify surface damage on in-service rails.

[0003] Essentially, rail surface damage detection based on profile data is a pattern recognition process. For identifying rail surface damage types, the analysis object is the shape of the track profile, which requires extracting valuable features from the perspective of temporal amplitude. While various types of fluctuations in the detected object can reflect hidden features, simply comparing amplitude values ​​is insufficient; it's necessary to consider the various patterns of amplitude values ​​and the differences between them. In fact, these features are closely related to determining the specific fault category of the analyzed object. Therefore, effectively distinguishing different patterns becomes crucial for different fault types. However, existing information entropy-based identification methods for effectively extracting patterns from track profile data have the following shortcomings:

[0004] 1. Existing information entropy methods are often based on the most primitive Shannon entropy. However, Shannon entropy only considers the probability distribution of variables without taking into account the intrinsic meaning of the variables' values. Therefore, the calculated entropy values ​​often result in inaccurate uncertainty / complexity measurements.

[0005] 2. Existing methods often fail to consider the magnitude relationship of variable amplitudes when analyzing the probability distribution characteristics of variables. Therefore, the amplitude information of the analyzed data is often lost during the extraction process.

[0006] 3. Existing methods lack necessary preprocessing for the detection data to be analyzed, which prevents the effective extraction of hidden high-dimensional information in the data. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, and device for identifying rail surface damage based on improved discrete entropy, so as to automatically and accurately extract various damage features, mine valuable information from damage data, and provide decision-making assistance and information support for railway staff.

[0008] To achieve the above objectives, the present invention provides the following solution:

[0009] A method for identifying rail surface damage based on improved discrete entropy, the method comprising:

[0010] The original rail profile data sequence was standardized using the normal cumulative distribution function;

[0011] Generate vector sets based on standardized sequences;

[0012] Each vector in the vector group is associated with each discrete pattern, and the weighted probability distribution of each discrete pattern is calculated based on the complexity of each vector.

[0013] Based on the weighted probability distribution, the improved discrete entropy value is calculated using Rényi entropy;

[0014] Obtain the entropy distribution range of various types of damage;

[0015] Rail surface damage is identified based on the improved discrete entropy value and the entropy value distribution range.

[0016] Optionally, the original rail profile data sequence is standardized using a normal cumulative distribution function, specifically including:

[0017] The original rail profile data sequence is mapped to a normalized sequence with a value range of [0, 1] using the normal cumulative distribution function.

[0018] Optionally, a vector set is generated based on the standardized sequence, specifically including:

[0019] Quantize the standardized sequence;

[0020] The quantized sequence is reconstructed in phase space to generate a vector set.

[0021] Optionally, the formula for calculating the weighted probability distribution is as follows:

[0022]

[0023] Among them, P w (π v0v1...vm-1 ) represents each discrete mode The weighted probability distribution, w represents the indicator function of the discrete pattern u to be matched. i π represents the complexity weight of the i-th vector. i Represents the i-th discrete pattern. Let N represent the vector with quantization level c and length m, and let N represent the number of vectors.

[0024] Alternatively, the improved formula for calculating discrete entropy is as follows:

[0025]

[0026] Where RDE represents improved discrete entropy value, x represents the original profile data, m represents the dimension of the vector, c represents the quantization level, and q represents the adjustment parameter.

[0027] A rail surface damage identification system based on improved discrete entropy, the system comprising:

[0028] The standardization module is used to standardize the original rail profile data sequence using the normal cumulative distribution function;

[0029] The vector group generation module is used to generate vector groups based on standardized sequences.

[0030] The weighted probability distribution calculation module is used to map each vector in the vector group to each discrete pattern, and calculate the weighted probability distribution of each discrete pattern according to the complexity of each vector.

[0031] An improved discrete entropy value calculation module is used to calculate the improved discrete entropy value based on the weighted probability distribution using Rényi entropy.

[0032] The entropy distribution range acquisition module is used to obtain the entropy distribution range of various types of damage;

[0033] The damage identification module is used to identify rail surface damage based on the improved discrete entropy value and the entropy value distribution range.

[0034] Optionally, the vector group generation module specifically includes:

[0035] Quantization unit, used to quantize a standardized sequence;

[0036] The phase space reconstruction unit is used to reconstruct the phase space of the quantized sequence to generate a vector group.

[0037] An electronic device includes a memory and a processor, the memory storing a computer program, and the processor running the computer program to enable the electronic device to perform the above-described rail surface damage identification method based on improved discrete entropy.

[0038] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for identifying rail surface damage based on improved discrete entropy.

[0039] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0040] (1) The present invention uses an improved discrete mode that can simultaneously extract amplitude magnitude information and amplitude change information from the profile data;

[0041] (2) The improved discrete mode of this invention can extract the amplitude information of the original data more accurately;

[0042] (3) The present invention is not sensitive to the influence of noise, and compared with other fault identification methods, the accuracy of this method is higher. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in 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 The flowchart is a method for identifying rail surface damage based on improved discrete entropy provided in Embodiment 1 of the present invention.

[0045] Figure 2 This is a schematic diagram of the cross-sectional shape of the rail;

[0046] Figure 3 This is a schematic diagram of the surface profile of the rail;

[0047] Figure 4 This is a schematic diagram of the rail surface contour drawn by sampling the rail cross-section using a two-dimensional sensor to obtain the rail profile coordinates;

[0048] Figure 5 This is a schematic diagram of the rail surface profile drawn using the transformed coordinates;

[0049] Figure 6 This is a schematic diagram of the data for the top section of the rail.

[0050] Figure 7 This is a diagram illustrating abnormal data.

[0051] Figure 8 This is a comparison of the convergence rate and standard deviation of the entropy value for a 1 / f noise sequence using NCDF and the discrete entropy using ordinary linear mapping.

[0052] Figure 9 This diagram illustrates sequences / vectors of different complexities and their complexity values; where, Figure 9 (a) is a schematic diagram of sequences / vectors with different complexities; Figure 9 (b) is a diagram showing different complexity values;

[0053] Figure 10 for A schematic diagram showing how entropy changes with parameter q;

[0054] Figure 11 This is a schematic diagram showing the distribution of RDE values ​​for various damaged and undamaged profiles as q→1. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The purpose of this invention is to provide a method, system, and device for identifying rail surface damage based on improved discrete entropy, so as to automatically and accurately extract various damage features, mine valuable information from damage data, and provide decision-making assistance and information support for railway staff.

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Example 1

[0059] This embodiment provides a rail surface damage identification method based on improved discrete entropy, such as... Figure 1 As shown, the method includes the following steps:

[0060] Step 101: Standardize the original rail profile data sequence using the normal cumulative distribution function (NCDF). Specifically, the NCDF is used to map the original profile data sequence to a normalized sequence with values ​​in the range [0, 1].

[0061] Over time, rails will gradually wear down on their load-bearing surfaces. To assess rail wear, it is necessary to measure the rail cross-section and evaluate its difference from the profile of a standard rail to obtain the rail wear value. Rail damage in my country mainly includes several forms such as side wear, stripe damage, peeling, and fish-scale damage, accounting for more than 80% of all rail damage.

[0062] For the collection of rail profile data, the rail inspection vehicle used is equipped with three data acquisition hosts and has four seats. The driver is responsible for driving safety and controlling the vehicle speed at different times each time. The speed control is performed twice at low speed, twice at medium speed, and twice at high speed.

[0063] According to GB2585-2007 "Hot-rolled steel rails for railways", the rail cross-sectional shape is as follows: Figure 2 As shown in the figure. According to the national standard definition, its surface profile is described using coordinates as shown in Table 1, and its shape is drawn as shown in the figure. Figure 3 As shown.

[0064] Table 1. National Standard Definition of Surface Profile Coordinates

[0065]

[0066] The rail profile coordinates were obtained by sampling the rail cross-section using a two-dimensional sensor on site, as shown in Table 2. The shape is plotted as follows. Figure 4 As shown.

[0067] Table 2 shows the rail profile coordinates obtained by sampling the rail cross-section using a two-dimensional sensor.

[0068]

[0069]

[0070] Because the coordinate system of the measurement data is determined by the sensor's position and orientation at the time of each frame's acquisition, each frame of data needs to be rotated and translated before it can be compared with the standard profile to confirm the amount of rail wear. Furthermore, since the sensor position is affected by vehicle movement and vibration, the rotation and translation parameters for each frame are different. The next step is to find the rotation and translation parameters that best match the standard profile and use these parameters for coordinate transformation. The transformed track profile coordinates are shown in Table 3, and the shape is plotted as follows... Figure 5 As shown.

[0071] Table 3. Converted orbital profile coordinates

[0072]

[0073] Observe the data of its top section as follows Figure 6 As shown. The original rail profile data may contain some unusable anomalies, generally due to surface contamination encountered during the initial laser scanning of the rail surface. To address this, this invention can first use a simple Euclidean distance metric to measure the difference between the profile data to be analyzed and the standard profile data, thus eliminating data samples with significant discrepancies. For example... Figure 7 The data in the four intervals between 2900 and 3600 on the horizontal axis shown are outliers. These were removed, and the original data underwent preprocessing. Then, data standardization based on the normal cumulative distribution function (NCDF) is performed to obtain the standardized sequence. If a standard linear mapping method is used, when the maximum or minimum value of the original data deviates too far from its mean or median, most of the original data may be assigned to only a few categories, which will greatly reduce the accuracy of the representation. However, if NCDF is used to process the data, this situation can be avoided.

[0074] On the other hand, using NCDF (Non-Constant Entropy Calculation) also provides better robustness for discrete entropy algorithms when dealing with data samples of varying lengths. For example... Figure 8 As shown, for pink noise (1 / f noise), the discrete entropy value using NCDF reaches its maximum value and has a small standard deviation when the data length is approximately 1000. However, the discrete entropy value using ordinary linear mapping only gradually stabilizes when the data length exceeds 30000, but its standard deviation is significantly larger than that using NCDF.

[0075] Step 102: Generate a vector set based on the standardized sequence. This specifically includes: quantizing the standardized sequence; and reconstructing the phase space of the quantized sequence to generate a vector set.

[0076] For a sequence normalized using NCDF, its amplitude is quantized at c levels. The quantization process is as follows: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] The value is mapped to a positive integer z from 1 to c. i c ,in, This is the i-th element after quantization. The `round` function here rounds the value to the nearest class. For the choice of parameter `c`, a larger `c` is preferable. However, if `c` is too large, the algorithm's computation time will be very long, and the pattern construction process will be sensitive to noise. Conversely, if `c` is too small, it may map two different amplitude values ​​to the same class, leading to inaccurate quantization results. Therefore, a balance should be found between pursuing accuracy and reducing the impact of noise when choosing parameter `c`.

[0077] Phase space reconstruction is used to convert one-dimensional quantized sequences Mapping to a high-dimensional space yields a vector representation, allowing us to capture various patterns within the sequence and reveal its intrinsic characteristics. From Extract the embedding vector with dimension m and time delay d, denoted as . Each embedding vector and discrete pattern Correspondingly, j = 0, 1, ..., m-1, where the number of all possible patterns corresponding to each vector is cm.

[0078] In the phase space reconstruction process, determining the appropriate vector dimension *m* and time delay *d* is crucial. In physics, fractal-based methods are global approaches for estimating the intrinsic dimension (ID) of nonlinear attractors. Here, this invention employs... Algorithm and Grassberger-Procaccia (GP) algorithm.

[0079] for The algorithm assumes that v(r) represents the number of boxes (i.e., hypercubes) with side length r that need to be covered in set Ω, and the ID of set Ω can be represented as:

[0080]

[0081] For the Grassberger-Procaccia algorithm:

[0082]

[0083] in,

[0084]

[0085] Then, the embedding dimension m is estimated using Takens' theorem, and m is also estimated using the pseudo nearest neighbor (FNN) method.

[0086] m≥2ID+1

[0087] This invention estimates the IDs of chaotic mapping sequences (here, a logical mapping (a = 3.6)) and random system sequences (fractional-order Gaussian noise).

[0088] For estimating the time delay d, Fraser, AM, and Swinney, HL proposed the Mutual Information (MI) method. This invention studies the MI method on the aforementioned data, and the results, along with the embedding dimension, are presented in Table 4.

[0089] Table 4 provides estimates of the dimension m and time delay d for the two types of data.

[0090]

[0091] Therefore, the embedding dimension of the detection data analyzed in this invention should satisfy m≥3, and the time delay should satisfy τ=1.

[0092] Step 103: Assign each vector in the vector group to each discrete pattern, and calculate the weighted probability distribution of each discrete pattern based on the complexity of each vector.

[0093] Each vector in the vector group in phase space corresponds to a specific discrete mode. For each mode, its corresponding weighted probability distribution is calculated using the complexity of each vector itself. The calculation method is as follows: Where w i This represents the weight of the i-th vector (pattern). It measures the magnitude of the vector's amplitude fluctuation. Sequences / vectors of different complexities have different complexity values, such as... Figure 9 As shown.

[0094] Thus, for each possible pattern, its weighted probability distribution is:

[0095]

[0096] in It is the characteristic function of u. hour hour The corresponding weights are:

[0097]

[0098] Therefore, the weighted discrete pattern can effectively extract different complex features from the original sequence, characterize the inherent fluctuation information of the research object, and reduce information loss. Experiments of this invention show that the data mining and feature extraction algorithm has stronger robustness in the face of interference and noise.

[0099] Step 104: Based on the weighted probability distribution, use... Entropy calculation improves discrete entropy values.

[0100] For non-stationary data with some correlation and long-range correlation, ordinary Shannon entropy cannot describe the system well. To improve it, in the 1960s... Shannon entropy was generalized to obtain Entropy. In this step, the present invention employs... Instead of ordinary Shannon entropy, the final improved discrete entropy value is calculated using the form of entropy. The introduction of the parameter q can accommodate different probability distributions of the data. The formula for calculating entropy is:

[0101]

[0102] Where q>0, q≠1, adjusting the value of parameter q can adapt to the influence of different probability intervals of the research object on the results.

[0103] When q approaches 1, we have

[0104]

[0105] At this time Entropy is Shannon entropy.

[0106] When q = 2 Entropy becomes collision entropy.

[0107] As q approaches infinity, H ∞ = -log(max p) i );

[0108] When q approaches 0, H0 = log|X|, where |X| is the number of values ​​that the random variable X can take.

[0109] Figure 10 Here is a schematic diagram showing how the Rényi entropy changes with parameter q: when q > 0, Entropy is a convex function of P, and it is obtained when the probability distribution is equal. The maximum value of entropy does not change with the change of q.

[0110] Finally, the improved discrete entropy (RDE) can be expressed as:

[0111]

[0112] The number of quantization levels is c, the vector dimension is m, and q takes values ​​of q>0 and q≠1.

[0113] Generally speaking, the RDE value is larger when the system characteristics are relatively disordered and the frequencies of various patterns are evenly distributed, while the RDE is smaller when the system characteristics are ordered and the frequencies of various patterns are unevenly distributed. RDE measures the degree of disorder of the analyzed object / system.

[0114] Step 105: Obtain the entropy distribution range of various types of damage.

[0115] Step 106: Identify rail surface damage based on the improved discrete entropy value and the entropy value distribution range.

[0116] Based on the improved discrete entropy value, the type of damage to the rail surface is determined according to the entropy value distribution range of various types of damage obtained in step 105.

[0117] The various rail surface damage data collected in the experiment included three main categories: side wear damage, stripe damage, and peeling and fish scale damage. For these three categories, this invention randomly selected 20 data samples from each type of damage profile (no-damage standard profile, side wear damage, stripe damage, peeling, and fish scale damage) from the collected dataset. For each experiment, 20 samples were randomly selected, analyzed, replaced, and repeated 10 times, and the average RDE value (q→1) was calculated. The experimental results are as follows: Figure 11As shown in the figure, the distribution range of RDE for various damaged profiles and undamaged profiles is different and all fall within a relatively small range.

[0118] Table 5 lists the distribution range of RDE values ​​for various damage profiles. Therefore, the RDE value calculated from the raw data can be used to determine which damage type it belongs to by looking up the table.

[0119] Table 5. Distribution range of RDE values ​​for various profile data.

[0120] mean min max Intactrailprofile 1.0663 1.0660 1.0665 Sidewear 1.3752 1.3495 1.4211 Stripewear 1.7332 1.7289 1.7991 Peeled and fishscalewear 1.5705 1.5528 1.5983

[0121] Example 2

[0122] In order to implement the method corresponding to Embodiment 1 above and achieve the corresponding functions and technical effects, a rail surface damage identification system based on improved discrete entropy is provided below.

[0123] The system includes:

[0124] The standardization module is used to standardize the original rail profile data sequence using a normal cumulative distribution function.

[0125] The vector group generation module is used to generate vector groups based on standardized sequences; specifically, it includes: a quantization unit for quantizing the standardized sequence; and a phase space reconstruction unit for reconstructing the phase space of the quantized sequence to generate vector groups.

[0126] The weighted probability distribution calculation module is used to map each vector in the vector group to each discrete pattern, and calculate the weighted probability distribution of each discrete pattern according to the complexity of each vector.

[0127] An improved discrete entropy value calculation module is used to calculate the improved discrete entropy value based on the weighted probability distribution using Rényi entropy.

[0128] The entropy distribution range acquisition module is used to obtain the entropy distribution range of various types of damage.

[0129] The damage identification module is used to identify rail surface damage based on the improved discrete entropy value and the entropy value distribution range.

[0130] Example 3

[0131] Embodiment 3 of the present invention provides an electronic device, including a memory and a processor. The memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the rail surface damage identification method based on improved discrete entropy of Embodiment 1.

[0132] The aforementioned electronic device may be a server.

[0133] Example 4

[0134] Embodiment 4 of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the rail surface damage identification method based on improved discrete entropy of Embodiment 1.

[0135] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0136] This article uses specific examples to illustrate the principles and implementation methods of the invention. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of the present invention. The described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

Claims

1. A rail surface damage identification method based on improved discrete entropy, characterized in that, The method comprises: normal cumulative distribution function is used to normalize the original rail profile data sequence; a vector group is generated based on the normalized sequence; each vector in the vector group is corresponded to each discrete pattern, and a weighted probability distribution of each discrete pattern is calculated according to the complexity of each vector; an improved discrete entropy value is calculated by using Rényi entropy according to the weighted probability distribution; an entropy value distribution range of each type of damage is obtained; the rail surface damage is identified based on the improved discrete entropy value and the entropy value distribution range; the calculation formula of the weighted probability distribution is as follows: wherein, denotes the weighted probability distribution of each discrete pattern denotes the indicator function of the discrete pattern u to be matched, denotes the complexity weight of the i-th vector, denotes the i-th discrete pattern, denotes the i-th vector of quantization levels c and length m, and N denotes the number of vectors.​ the calculation formula of the improved discrete entropy value is as follows: wherein, denotes an improved discrete entropy value, denotes original profile data, denotes a dimension of a vector, denotes a quantization level, denotes an adjustment parameter.

2. The improved discrete entropy based rail surface damage detection method of claim 1, wherein, the normal cumulative distribution function is used to normalize the original rail profile data sequence, specifically comprising: the normal cumulative distribution function is used to map the original rail profile data sequence to a unitized sequence with a value range of [0, 1].

3. The improved discrete entropy based rail surface damage detection method of claim 1, wherein, a vector group is generated based on the normalized sequence, specifically comprising: the normalized sequence is quantized; the quantized sequence is reconstructed in phase space to generate a vector group.

4. A rail surface damage identification system based on improved discrete entropy, characterized in that, The system comprises: a normalization processing module for normalizing the original rail profile data sequence by using the normal cumulative distribution function; a vector group generation module for generating a vector group based on the normalized sequence; a weighted probability distribution calculation module for corresponding each vector in the vector group to each discrete pattern, and calculating a weighted probability distribution of each discrete pattern according to the complexity of each vector; an improved discrete entropy value calculation module for calculating an improved discrete entropy value by using Rényi entropy according to the weighted probability distribution; an entropy value distribution range acquisition module for obtaining an entropy value distribution range of each type of damage; a damage identification module for identifying the rail surface damage based on the improved discrete entropy value and the entropy value distribution range; the calculation formula of the weighted probability distribution is as follows: wherein, denotes the weighted probability distribution of each discrete pattern denotes the indicator function of the discrete pattern u to be matched, denotes the complexity weight of the i-th vector, denotes the i-th discrete pattern, denotes the i-th vector of quantization levels c and length m, and N denotes the number of vectors.​ the calculation formula of the improved discrete entropy value is as follows: wherein, denotes an improved discrete entropy value, denotes original profile data, denotes a dimension of the vector, denotes a quantization level, denotes an adjustment parameter.

5. The improved discrete entropy based rail surface damage identification system as claimed in claim 4, wherein, the vector group generation module specifically comprises: a quantization unit for quantizing the normalized sequence; a phase space reconstruction unit for reconstructing the quantized sequence in phase space to generate a vector group.

6. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the rail surface damage identification method based on improved discrete entropy according to any one of claims 1-3.

7. A computer readable storage medium characterized in that, The computer program is stored in the memory and is executed by the processor to realize the rail surface damage identification method based on improved discrete entropy according to any one of claims 1-3.

Images