A turnout fault detection method, a terminal device and a storage medium

Abstract

The switch machine action fault classification method comprises the following steps: acquiring action current curve data of a switch machine device; using a wavelet packet decomposition algorithm to extract a frequency domain feature vector of the current curve, and fusing a time domain feature vector of the curve; and establishing a multi-task joint sparse representation switch machine action current curve fault classification algorithm model. Through the above steps, the fault classification of the current monitoring fault signal is realized, thereby realizing the monitoring of the working state of the switch and ensuring the safe working of the switch. The multi-task joint sparse representation classification algorithm is innovatively applied to intelligent analysis of the action current curve of the switch machine, thereby providing support for realizing automatic alarm of switch faults, and improving the intelligent level of subway operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of turnout detection, specifically to a turnout fault detection method, terminal device, and storage medium. Background Technology

[0002] As one of the main pieces of track signaling equipment, turnouts primarily guide the direction of train movement to enable track switching and cross-track operation, playing a crucial role in ensuring train safety and operational efficiency. The Guangzhou Metro network is vast, with complex and diverse geological conditions. Different operating modes and train equipment conditions pose significant challenges to the application of high-speed turnouts. Furthermore, the heavy traffic and frequent turnout switching on metro lines lead to wear and tear over extended periods. Since some turnouts are exposed to the elements, they are prone to blockages and changes in their mechanical condition, resulting in malfunctions, failures, or incomplete operation, potentially causing train derailment and other traffic accidents.

[0003] For a long time, my country has still used traditional fault-based and planned maintenance methods for the maintenance of subway signaling equipment, and the same applies to the maintenance of turnouts. Currently, the on-site maintenance and inspection of turnouts mainly relies on maintenance personnel periodically reviewing the turnout switch machine operation data collected by microcomputer monitoring. By comparing the current and voltage curves obtained by microcomputer monitoring with the monitoring data of normally operating turnouts, fault identification is achieved, thereby determining the operating status of the turnout. Summary of the Invention

[0004] In order to overcome the above-mentioned shortcomings of the prior art, the present invention aims to provide a turnout fault detection method, which can realize the detection of turnout faults.

[0005] A second objective of the present invention is to provide a storage device.

[0006] A third objective of this invention is to provide a terminal device.

[0007] To achieve its objectives, the present invention employs the following technical solution:

[0008] A method for classifying faults in a turnout switch machine includes the following steps:

[0009] Obtain the operating current curve data of the turnout switch machine;

[0010] The wavelet packet decomposition algorithm is used to extract the frequency domain feature vector of the current curve, and the time domain feature vector of the curve is fused together.

[0011] A fault classification algorithm model based on the operating current curve of a turnout switch machine using a multi-task joint sparse representation is established.

[0012] As a preferred approach, a fault classification algorithm model for turnout switch machine operating current curves is established by supplementing the model with kernel function method. This model incorporates multi-task joint sparse representation.

[0013] Preferably, the process of obtaining the operating current curve data of the turnout switching machine includes the following steps:

[0014] The hexadecimal data is converted to decimal data and stored using single-precision floating-point numbers. Based on the sampling frequency, the time of each turnout operation is standardized, with the longest curve data time as the standard, and zeros are added to the end of curve data that is not long enough.

[0015] As a preferred option

[0016] A fault classification algorithm model based on the operating current curve of a turnout switch machine using a multi-task joint sparse representation is established, including the following steps:

[0017] Each curve feature classification subtask is represented as a sparse representation model, thus constructing multiple classification tasks with sparse representations; by establishing l in the model 1，2 Hybrid regularization constraints are integrated across multiple tasks to construct a multi-task joint sparse representation model.

[0018] As a preferred option

[0019] The frequency domain feature vector of the current curve is extracted using the wavelet packet decomposition algorithm, and then the time domain feature vector of the curve is fused. The steps include:

[0020] Suppose that the current sample data contains J types of faults, and each fault curve is described by K types of features. For each action current curve data sample, stack all the features to construct a time-domain feature vector. After performing wavelet packet decomposition on the current curve data, stack all the wavelet frequency-domain features to construct a wavelet frequency-domain feature vector.

[0021] The feature matrix used to represent the k-th feature The feature matrix representing the k-th feature of the j-th fault category, m k p represents the dimension of the k-th feature. j This represents the number of training samples contained in the j-th class. It is the total number of samples in the entire training set, y k This is the k-th feature of the test sample.

[0022] As a preferred approach, the establishment of a fault classification algorithm model for turnout switch machine operating current curves, supplemented by the kernel function method, includes the following steps:

[0023] The following formula is assumed to be the mathematical model for multi-task joint sparse representation:

[0024]

[0025] Among them, the kernel matrix h k =φ k (y k The kernel function for the k-th feature subtask is κ. k (m,n)=φ k (m) T φ k (n), where φ is the mapping function from low dimension to high dimension, and the sparse coefficient matrix is...

[0026] Using kernel function Let m be the set of all possible alignment combinations of curves m and n. Where σ is 2, μ is the sample mean, and the temporal similarity of the curves is calculated; in the wavelet frequency domain, Euclidean distance is used as the kernel function. μ is the sample mean;

[0027] Treating the columns of the kernel matrix as data samples, the above model is solved using APG (Accelerated Proximal Gradient) algorithm to obtain the sparse coefficient matrix W. The class is determined by the reconstruction error, and the class j of the current test sample is the one with the smallest reconstruction error. * As shown in the following formula:

[0028]

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] Through the above steps, fault classification of currently monitored fault signals is achieved, thereby enabling monitoring of the turnout's operational status and ensuring its safe operation. An innovative approach is proposed: applying a multi-task joint sparse representation classification algorithm to the intelligent analysis of turnout switch machine operating current curves. This provides support for automatic turnout fault alarms and improves the level of intelligent subway operation and maintenance. Intelligent judgment of abnormal turnout current curves and fault classification can, to a certain extent, meet the on-site needs for accuracy, timeliness, and situational awareness in turnout maintenance, providing a basis for condition-based maintenance of switch machine equipment. Attached Figure Description

[0031] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1This is a diagram of data storage format;

[0033] Figure 2 This is a schematic diagram of multi-task joint sparse representation.

[0034] Figure 3 This is a flowchart of the fault classification method for turnout switching machine described in this invention. Detailed Implementation

[0035] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features of the present invention can be combined with each other. Many specific details are set forth in the following description to provide a thorough understanding of the present invention; the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0036] The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention.

[0037] See Figures 1 to 3 As shown, the present invention provides a turnout fault detection method, which includes the following steps:

[0038] S1. Obtain the operating current curve data of the turnout switching machine; including the following steps:

[0039] The hexadecimal data is converted to decimal data and stored using single-precision floating-point numbers. Based on the sampling frequency, the time of each turnout operation is standardized, with the longest curve data time as the standard, and zeros are added to the end of curve data that is not long enough.

[0040] S2. Extract the frequency domain feature vector of the current curve using the wavelet packet decomposition algorithm, and fuse the time domain feature vector of the curve, including the following steps:

[0041] Suppose that the current sample data contains J types of faults, and each fault curve is described by K types of features. For each action current curve data sample, stack all the features to construct a time-domain feature vector. After performing wavelet packet decomposition on the current curve data, stack all the wavelet frequency-domain features to construct a wavelet frequency-domain feature vector.

[0042] The feature matrix used to represent the k-th feature The feature matrix representing the k-th feature of the j-th fault category, m k p represents the dimension of the k-th feature.j This represents the number of training samples contained in the j-th class. It is the total number of samples in the entire training set, y k This is the k-th feature of the test sample.

[0043] S3. The establishment of a fault classification algorithm model for turnout switch machine operating current curves using a kernel function method for multi-task joint sparse representation includes the following steps:

[0044] Each curve feature classification subtask is represented as a sparse representation model, thus constructing multiple classification tasks with sparse representations; by establishing l in the model 1，2 Hybrid regularization constraints are integrated across multiple tasks to construct a multi-task joint sparse representation model;

[0045] The following formula is assumed to be the mathematical model for multi-task joint sparse representation:

[0046]

[0047] Among them, the kernel matrix h k =φ k (y k The kernel function for the k-th feature subtask is κ. k (m,n)=φ k (m) T φ k (n), where φ is the mapping function from low dimension to high dimension, and the sparse coefficient matrix is...

[0048] Using kernel function Let m be the set of all possible alignment combinations of curves m and n. Where σ is 2, μ is the sample mean, and the temporal similarity of the curves is calculated; in the wavelet frequency domain, Euclidean distance is used as the kernel function. μ is the sample mean;

[0049] Treating the columns of the kernel matrix as data samples, the above model is solved using APG (Accelerated Proximal Gradient) algorithm to obtain the sparse coefficient matrix W. The class is determined by the reconstruction error, and the class j of the current test sample is the one with the smallest reconstruction error. * As shown in the following formula:

[0050]

[0051] Sparse representation, in this context, refers to using a relatively small number of linear combinations of basic signals to represent most or all of the original signal. These basic signals, called atoms, are selected from an overcomplete dictionary; an overcomplete dictionary is formed by aggregating atoms whose number exceeds the dimension of the signal. Therefore, any signal has different sparse representations depending on the set of atoms. Suppose we represent a dataset X using an M*N matrix, where each row represents a sample and each column represents an attribute of the sample. Generally, this matrix is ​​dense, meaning most elements are not zero. The meaning of sparse representation is to find a coefficient matrix A (K*N) and a dictionary matrix B (M*K) such that B*A reconstructs X as closely as possible, and A is as sparse as possible. A is the sparse representation of X.

[0052] Wavelet analysis decomposes a time series into low-frequency and high-frequency components. In this decomposition, information lost in the low-frequency components is captured by the high-frequency components. In the next level of decomposition, the low-frequency information from the previous level is again decomposed into low-frequency and high-frequency components, with the lost low-frequency information being captured by the high-frequency components. This process continues, allowing for deeper decompositions. Wavelet packet decomposition decomposes not only the low-frequency components but also the high-frequency components. Therefore, wavelet packet decomposition is a more widely used wavelet decomposition method, applied to signal decomposition, coding, denoising, and compression.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for detecting turnout faults, characterized in that, Includes the following steps: Obtain the operating current curve data of the turnout switch machine; The wavelet packet decomposition algorithm is used to extract the frequency domain feature vector of the current curve, and the time domain feature vector of the curve is fused together. A multi-task joint sparse representation fault classification algorithm model for turnout switch machine operating current curves is established: Each curve feature classification subtask is represented as a sparse representation model, thus constructing multiple classification tasks with sparse representations; by establishing in the model... Hybrid regularization constraints are integrated across multiple tasks to construct a multi-task joint sparse representation model; The establishment of a fault classification algorithm model for turnout switch machine operating current curves using a kernel function method for multi-task joint sparse representation includes the following steps: The following formula is assumed to be the mathematical model for multi-task joint sparse representation: Among them, the kernel matrix , , No. The kernel function for each feature subtask is: , Let be the mapping function from low dimension to high dimension, and be the sparse coefficient matrix. ; Using kernel function , For curves , The set of all possible alignment combinations ,in The value of is 2. To calculate the temporal similarity of the curves, we use the sample mean; in the wavelet frequency domain, we use Euclidean distance as the kernel function. , The mean of the sample; Treating the columns of the kernel matrix as data samples, the above model is solved using the APG proximal acceleration gradient algorithm to obtain the sparse coefficient matrix. The category is determined by the reconstruction error; the category with the smallest reconstruction error is the current test sample category. As shown in the following formula: 。 2. The turnout fault detection method according to claim 1, characterized in that, Obtaining the operating current curve data of the turnout switch machine includes the following steps: The hexadecimal data is converted to decimal data and stored using single-precision floating-point numbers. Based on the sampling frequency, the time of each turnout operation is standardized, with the longest curve data time as the standard, and zeros are added to the end of curve data that is not long enough.

3. The turnout fault detection method according to claim 1, characterized in that, The frequency domain feature vector of the current curve is extracted using the wavelet packet decomposition algorithm, and then the time domain feature vector of the curve is fused. The steps include: Suppose the current sample data contains Fault types, each fault curve extracted The features are described in various forms; for each action current curve data sample, all features are stacked to construct a time-domain feature vector; after wavelet packet decomposition of the current curve data, all wavelet frequency domain features are stacked to construct a wavelet frequency domain feature vector. Used to indicate the first The feature matrix of a type of feature Representing the The first type of fault The feature matrix of a type of feature Representing the The dimension of the feature Indicates the first The number of training samples contained in a class sample. It is the total number of samples in the entire training set. For the test sample One characteristic.

4. A terminal device, characterized in that, It includes a processor, a memory, and a program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the program, implements the turnout fault detection method as described in any one of claims 1 to 3.

5. A readable storage medium, characterized in that, The readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the readable storage medium to perform the turnout fault detection method as described in any one of claims 1 to 3.

Images