A method and device for diagnosing faults of a plunger pump of a switch machine

Abstract

The present invention relates to a method for diagnosing faults of a switch machine plunger pump, comprising the following steps: collecting vibration signals from a plunger pump housing; decomposing and processing the collected vibration signals to obtain a plurality of ISC components; screening the ISC components for signal reconstruction; extracting features of the reconstructed signal to construct a plunger pump fault diagnosis feature set; constructing a diagnostic model using an SVM algorithm; inputting the plunger pump fault diagnosis feature set into the diagnostic model for training; and applying the diagnostic model. The method of the present invention is targeted at diagnosing faults of a switch machine plunger pump under small sample conditions and can effectively decompose the plunger pump's vibration signal data and extract signal features. The plunger pump fault diagnosis accuracy is high based on decorrelation masking local feature scale decomposition. Fault pattern recognition is performed using an SVM model optimized with intelligent optimization algorithm parameters. This model accurately classifies faults of switch machine plunger pump equipment and simultaneously offers the advantages of high efficiency and low cost.

Description

Technical Field

[0001] The invention belongs to the technical field of mechanical equipment fault diagnosis, and in particular relates to a fault diagnosis method for a plunger pump of a switch machine. Background Art

[0002] With the rapid development of the railway industry, my country's high-speed rail has achieved world-leading operational scale and overall technology. To ensure safe and stable train operations, higher safety and reliability requirements are being placed on railway signaling equipment. As part of railway signaling equipment, the operating status of switch machines directly impacts the normal operation of trains. During the switch machine's switching process, plunger pumps convert mechanical energy into hydraulic energy, providing power support to enable the machine's various functions. Therefore, to ensure the safe and reliable operation of these pumps, fault monitoring and diagnosis are crucial.

[0003] Currently, fault diagnosis for switch plunger pumps is evolving from traditional empirical diagnosis to artificial intelligence diagnosis. Traditional empirical diagnosis not only consumes a large amount of manpower and financial resources, but is also inefficient and cannot guarantee the real-time operating performance of related equipment, making it unsuitable for the railway's development towards automation and intelligence. Intelligent diagnostic technology, on the other hand, collects and analyzes the vibration signals of the plunger pump's casing or end cover under different operating conditions, extracts key characteristic information about its health status, and finally inputs it into a classifier for fault pattern recognition, thus compensating for the shortcomings of traditional empirical diagnosis. Currently, fault diagnosis methods for switch plunger pumps still have the following shortcomings:

[0004] (1) The signal decomposition efficiency is low, and there is a certain modal aliasing effect in the decomposition process. The main solution to this problem is to add Gaussian white noise to the signal to be decomposed to assist in decomposition, and then offset the added white noise by integrating and averaging the decomposition results. However, this type of method has certain defects. First, the number of integrations and the amplitude of the added noise need to be set based on human experience, which is not adaptive to a certain extent. Second, the noise is relatively random. If the number of integrations is small, the white noise residue will increase, and if the number of integrations is large, the amount of calculation will increase, which will greatly reduce the signal decomposition efficiency.

[0005] (2) When extracting fault diagnosis feature sets, high-dimensional fault feature matrices have information redundancy, and too high a dimension will affect the calculation time. For entropy feature dimensionality reduction, the current scale selection mainly relies on manual observation and comparison of the feature mean and standard deviation of each state signal. This has the disadvantage of low scale selection efficiency, which in turn affects the accuracy of state recognition.

[0006] (3) Due to the particularity of field equipment, large fault data sets are difficult to obtain. Deep learning requires a large amount of training data to achieve high performance, and the hardware requirements are high during the learning process. For the current plunger pump fault diagnosis method, there is no complete core algorithm, and the existing algorithms have problems such as low fault diagnosis accuracy. Summary of the Invention

[0007] The purpose of the present invention is to provide a method and device for diagnosing faults of a plunger pump of a switch machine, which can extract signal features by decomposing the vibration signal data of the plunger pump, and then use an SVM model optimized by an intelligent optimization algorithm to perform fault pattern recognition;

[0008] In order to achieve the above object, the technical solution adopted by the present invention is:

[0009] A method for diagnosing a fault of a plunger pump of a switch machine comprises the following steps:

[0010] Collect vibration signals of the plunger pump housing;

[0011] The collected vibration signal is decomposed by using decorrelation masking local characteristic scale decomposition method to obtain several ISC components.

[0012] Filter ISC components for signal reconstruction;

[0013] Extract and reconstruct signal features and construct a feature set for plunger pump fault diagnosis;

[0014] The diagnostic model was constructed using the SVM algorithm;

[0015] The plunger pump fault diagnosis feature set is input into the diagnosis model for training;

[0016] Perform application of diagnostic models.

[0017] Preferably, the decorrelation masking local feature scale decomposition method further includes the following steps of decomposing and processing the collected vibration signal:

[0018] A1. Decompose the vibration signal x(t) using local eigenscale decomposition to obtain component PISC1.

[0019] A2 constructs the masking signal s(t) of the component PISC1;

[0020] A3. Add the masking signal s(t) to the vibration signal x(t) to construct the signal y(t); perform local eigenscale decomposition on y(t) to obtain the component ISC1; subtract the masking signal s(t) from the ISC1 component to obtain the component signal MISC1 h1(t);

[0021] A4. Remove the component signal MISC1 h1(t) from the vibration signal x(t) to obtain the next signal to be decomposed. Repeat steps A1 to A3 for the next signal to be decomposed to obtain the second component signal MISC2 h2(t).

[0022] A5. Calculate the correlation coefficient r between the component signal MISC1 h1(t) and the component signal MISC2 h2(t), perform decorrelation processing on h1(t) to obtain the decomposed signal

[0023] A6. Decompose the signal Remove from the vibration signal x(t) to obtain the signal to be decomposed Treat decomposed signal Repeat steps A1 to A4 above to decompose and obtain two components

[0024] A7. Calculate the decomposition signal and quantity If the correlation coefficient is less than the threshold δ, then is the ISC component, and vice versa and The decorrelation process is performed until the correlation coefficient values ​​between all components are less than the threshold δ.

[0025] Preferably, the masking signal s(t):

[0026]

[0027]

[0028] Where f1(i) is the instantaneous frequency, a1(i) is the instantaneous amplitude, M is the bandwidth factor, and f is the frequency of the masked signal.

[0029] Preferably, the threshold δ is 0.1.

[0030] Preferably, the step of screening the ISC components further includes the following steps: calculating the correlation coefficient between the ISC components and the vibration signal, and screening out the ISC components corresponding to the correlation coefficients being greater than 0.3.

[0031] Preferably, the step of extracting the reconstructed signal features also includes the following steps: calculating the GRCMDE value of the reconstructed signal, and calculating the GRCMDE value feature overlap under multiple scale factors, and then selecting the GRCMDE values ​​with high feature overlap under k scale factors to construct a plunger pump fault diagnosis feature set.

[0032] Preferably, the diagnostic model training step also includes the following steps: dividing the plunger pump fault diagnosis feature set into a training set and a test set according to an 8:2 ratio, using the training set data to train the diagnostic model, and using the vector weighted average algorithm INFO to optimize the penalty factor and kernel function parameters in the SVM model.

[0033] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing faults of a plunger pump of a switch machine are executed.

[0034] Another aspect of the present invention further provides a fault diagnosis device for a plunger pump of a switch machine, comprising:

[0035] Memory, for storing software applications,

[0036] The processor is used to execute the software application program, and each program of the software application program correspondingly executes the steps of the above-mentioned method for diagnosing faults of a switch machine plunger pump.

[0037] The method of the present invention is aimed at fault diagnosis of the switch machine plunger pump under small sample conditions. It can effectively decompose the vibration signal data of the plunger pump and extract signal features. The plunger pump fault diagnosis accuracy based on decorrelation masking local feature scale decomposition is high. The SVM model with optimized intelligent optimization algorithm parameters is used for fault pattern recognition. This model has high accuracy in classifying switch machine plunger pump equipment faults and also has the advantages of high efficiency and low cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The vibration waveforms of the switch machine plunger pump under four working conditions;

[0039] Figure 2 It is a logic flow chart of the DMLCD decomposition algorithm in the fault diagnosis method of the present invention;

[0040] Figure 3 The first six waveforms of the DMLCD decomposition of the four working condition signals of the plunger pump are shown;

[0041] Figure 4 is the characteristic coincidence value of the four working condition signals of the plunger pump;

[0042] Figure 5 This is a logic flow chart of the INFO-SVM algorithm in the fault diagnosis method of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described below with reference to the accompanying drawings and specific embodiments.

[0044] A method for diagnosing a fault of a plunger pump of a switch machine according to the present invention comprises the following steps:

[0045] S1, collecting the vibration signal of the plunger pump housing;

[0046] Specifically, a measuring point is arranged above the middle casing of the plunger pump. The plunger pump is driven by a motor to collect vibration signal data of the switch machine plunger pump. During the test, data are collected under four operating conditions: cylinder peeling, valve plate wear, plunger ball head wear, and normal working conditions. The labels of these four working conditions are set as "1", "2", "3", and "0" respectively. Figure 1 Shown are the time domain waveforms of vibration signals under four working conditions.

[0047] S2, using the decorrelation masking local characteristic scale decomposition method to decompose the collected vibration signal to obtain several ISC components;

[0048] Specifically, the decorrelation masked local characteristic scale decomposition (DMLCD) method is used to decompose the vibration signal data of the four working conditions of the piston pump to obtain a series of intrinsic scale functions (ISCs) of different time scales. Among them, the first six ISC components obtained by DMLCD decomposition of the vibration signal data of the four working conditions are as follows: Figure 3 As shown;

[0049] More specifically, decompose the logical process as follows Figure 2 As shown, the decorrelation masking local feature scale decomposition method performs decomposition processing step S2 on the collected vibration signal, including the following steps:

[0050] A1. Decompose the vibration signal x(t) using local eigenscale decomposition (LCD) to obtain component PISC1.

[0051] A2 constructs the masking signal s(t) of the component PISC1;

[0052] Among them, the instantaneous frequency f1(i), instantaneous amplitude a1(i) and bandwidth coefficient M corresponding to the component PISC1 are calculated by Hilbert transform to determine the frequency f of the masking signal, and then construct the masking signal s(t);

[0053]

[0054] s(t)=A u sin(2πft)

[0055] A3. Add the masking signal s(t) to the vibration signal x(t) to construct the signal y(t), where y(t) = x(t) + s(t). Perform local characteristic scale decomposition (LCD) on the signal y(t) to obtain the component ISC1. Subtract the masking signal s(t) from the ISC1 component to obtain the component signal MISC1 h1(t), i.e., MISC1 = ISC1 - s(t).

[0056] A4. Remove the component signal MISC1 h1(t) from the vibration signal x(t) to obtain the next signal to be decomposed. Repeat steps A1 to A3 for the next signal to be decomposed to obtain the second component MISC2 h2(t).

[0057] Specifically, repeat steps A1 to A3 for the next signal to be decomposed, that is, perform LCD decomposition on the next signal to be decomposed to obtain component PISC2, construct a masking signal corresponding to component PISC2, add the masking signal corresponding to component PISC2 to the signal to be decomposed and perform LCD decomposition again to obtain component ISC2, remove the masking signal contained in ISC2, and obtain component MISC2 h2(t);

[0058] A5. Calculate the correlation coefficient r between the component signal MISC1 h1(t) and the component signal MISC2 h2(t), and perform decorrelation processing on h1(t), that is: Get the decomposition signal

[0059] A6. The processed decomposition signal Remove from the vibration signal x(t) to obtain the signal to be decomposed Treat decomposed signal Repeat steps A1 to A4 above to decompose and obtain two components

[0060] Specifically, treat the decomposition signal Repeat the above steps A1 to A4 to decompose the signal to be decomposed. Perform LCD decomposition to obtain component PISC3, construct the masking signal corresponding to component PISC3, and add the masking signal corresponding to component PISC3 to the decomposed signal The LCD decomposition is performed again to obtain the component ISC3, and the masking signal contained in ISC3 is removed to obtain the component MISC3h3(t).

[0061] Remove the component MISC3h3(t) from the vibration signal x(t) to obtain the next signal to be decomposed. Perform LCD decomposition on the next signal to be decomposed to obtain component PISC4. Construct the masking signal corresponding to component PISC4. Add the masking signal corresponding to component PISC4 to the signal to be decomposed and perform LCD decomposition again to obtain component ISC4. Remove the masking signal contained in ISC4 to obtain component MISC4 h4(t). Component MISC3 h3(t) and component MISC4 h4(t) are components

[0062] A7. Calculate the decomposition signal and quantity If the correlation coefficient is less than the threshold δ, it is considered is the optimal component OISC1, and vice versa and The decorrelation process is performed until the correlation coefficient values ​​between all components are less than a threshold value δ. Preferably, the threshold value δ is 0.1. The multiple optimal components outputted are the ISC components in step S2.

[0063] S3, screening ISC components for signal reconstruction;

[0064] Specifically, the step of screening the ISC components further includes the following steps: calculating the correlation coefficient between the ISC components and the vibration signal, screening out the ISC components corresponding to the correlation coefficient being greater than 0.3, and then performing signal reconstruction on the screened ISC components.

[0065] Table 1 Correlation coefficients between ISC components and original signals

[0066]

[0067] Among them, Table 1 shows the correlation coefficients of the first six-order ISC components and the original vibration signal obtained by DMLCD decomposition of the vibration signal data of the above four working conditions. It can be seen from Table 1 that the correlation coefficients of the first two-order components and the original signal are both greater than 0.3, which can reflect the original signal information to a large extent. The first two-order ISC components are screened to reconstruct the signal.

[0068] S4, extracting reconstructed signal features and constructing a plunger pump fault diagnosis feature set;

[0069] Specifically, the step of extracting the reconstructed signal features further includes the following steps: calculating the generalized refined composite multi-scale spread entropy GRCMDE value of the reconstructed signal, and calculating the GRCMDE value feature overlap under multiple scale factors, then selecting the GRCMDE values ​​with high feature overlap under k scale factors, selecting the scale factors with high feature information discrimination according to the feature overlap, reducing the feature dimension, and constructing a plunger pump fault diagnosis feature set;

[0070] More specifically, the entropy characteristic overlap between the four working conditions under 20 scale factors is calculated, and the calculation results are as follows: Figure 4 As shown in the figure, the greater the feature overlap, the higher the discrimination of data under different working conditions. According to the order of merit, the GRCMDE values ​​under the first five scale factors are selected to construct the plunger pump fault diagnosis feature set.

[0071] The calculation formula of feature overlap is:

[0072]

[0073] Where: N is the total number of sample types; i is the i-th fault type after the feature means of different types of samples are sorted from large to small; M i is the characteristic mean of the i-th type sample; S i is the characteristic standard deviation of the i-th type sample.

[0074] S5, using SVM algorithm to build a diagnostic model;

[0075] S6. Inputting the plunger pump fault diagnosis feature set into the diagnosis model for training;

[0076] Specifically, the diagnostic model training step also includes the following steps: dividing the plunger pump fault diagnosis feature set into a training set and a test set according to an 8:2 ratio, using the training set data to train the diagnostic model, that is, the original SVM model, using the vector weighted average algorithm INFO to optimize the penalty factor and kernel function parameters in the SVM model to obtain the INFO-SVM model, that is, the diagnostic model, so that the optimized model achieves a high fault diagnosis accuracy on the test set, wherein the algorithm flow chart of INFO-SVM is shown in Figure 5 , some inventors published a plunger pump fault diagnosis based on KPEMD and INFO-SVM [J]. Journal of North University of China (Natural Science Edition). 2023, 44(3): 216-221, 228, which disclosed the steps of classifying and identifying the feature vector set by the SVM model optimized by the INFO algorithm to obtain the INFO-SVM model.

[0077] S7. Execute application of the diagnostic model.

[0078] Specifically, the test set is input into the diagnosis model to identify and classify the different operating states of the plunger pump, and the fault diagnosis results of the switch plunger pump are obtained;

[0079] More specifically, the INFO-SVM model was obtained by the above method. 300 groups of samples were selected from the characteristic values ​​of the four working conditions of the plunger pump, and a total of 1,200 groups of samples were sequentially input into the diagnosis model for fault diagnosis. In order to make the diagnosis results reliable, 10 tests were conducted on each model, and the original local scale decomposition method was compared with the traditional support vector machine model to obtain the fault diagnosis rate statistics, as shown in Table 2 below.

[0080] Table 2 Comparison of model diagnosis rate results

[0081]

[0082] As can be seen from Table 2, the INFO-SVM model obtained by the method of the present invention and the traditional SVM model have a plunger pump fault diagnosis rate of greater than 90% based on decorrelation masking local feature scale decomposition, which is higher than the diagnosis result of the model based on local feature scale decomposition, and is significantly improved compared with the traditional SVM model.

[0083] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the above-mentioned method for diagnosing faults of a plunger pump of a switch machine are executed.

[0084] Another aspect of the present invention further provides a fault diagnosis device for a plunger pump of a switch machine, comprising:

[0085] Memory, for storing software applications,

[0086] The processor is used to execute the software application program, and each program of the software application program correspondingly executes the steps of the above-mentioned method for diagnosing faults of a switch machine plunger pump.

Claims

1. A fault diagnosis method for a switch machine plunger pump, characterized in that: The following steps are involved: Collect vibration signals of the plunger pump housing; The collected vibration signal is decomposed and processed using the decorrelation masking local characteristic scale decomposition method to obtain several ISC components. Filter ISC components for signal reconstruction; Extract and reconstruct signal features and construct a feature set for plunger pump fault diagnosis; The diagnostic model was constructed using the SVM algorithm; The plunger pump fault diagnosis feature set is input into the diagnosis model for training; Perform application of diagnostic models; The decomposition and processing step of the collected vibration signal by the decorrelation masking local feature scale decomposition method further includes the following steps: A1. Using local feature scale decomposition to analyze vibration signals Decompose it to obtain component PISC1; A2. Constructing the masking signal of component PISC1 ; A3. Masking signal Add to vibration signal To build a signal ;right Perform local feature scale decomposition to obtain component ISC1; the masking signal Subtracted from the ISC1 component, the component signal MISC1 is obtained ; A4. The component signal MISC1 From the vibration signal Remove it from the equation to get the next signal to be decomposed. Repeat steps A1 to A3 for the next signal to be decomposed to get the second component signal MISC2 ; A5. Calculate component signal MISC1 MISC2 The correlation coefficient r of Perform decorrelation processing to obtain the decomposed signal ; A6. Decompose the signal From the vibration signal Remove it and get the signal to be decomposed , treat the decomposed signal Repeat steps A1 to A4 above to decompose and obtain two components ; A7. Calculate the decomposition signal and quantity If the correlation coefficient is less than the threshold δ, then is the ISC component, and vice versa and Perform decorrelation processing until the correlation coefficient values ​​between all components are less than the threshold δ; Wherein, the masking signal : ， ， in, is the instantaneous frequency, is the instantaneous amplitude, M is the bandwidth factor, and f is the frequency of the masked signal.

2. A method for diagnosing faults of a switch machine plunger pump according to claim 1, characterized in that: The threshold δ is 0.

1.

3. The fault diagnosis method for a plunger pump of a switch machine according to claim 1, characterized in that: The step of screening the ISC components further comprises the following steps: The correlation coefficient between the ISC component and the vibration signal was calculated, and the ISC component corresponding to the correlation coefficient greater than 0.3 was selected.

4. The fault diagnosis method of the plunger pump of a switch machine according to claim 1, characterized in that: The step of extracting and reconstructing signal features also The method comprises the following steps: calculating the GRCMDE value of the reconstructed signal, calculating the feature overlap of the GRCMDE values ​​under multiple scale factors, then selecting the GRCMDE values ​​with high feature overlap under k scale factors, and constructing a plunger pump fault diagnosis feature set.

5. The fault diagnosis method of the plunger pump of a switch machine according to claim 1, characterized in that: The diagnostic model training step also includes the following steps: dividing the plunger pump fault diagnosis feature set into a training set and a test set according to an 8:2 ratio, using the training set data to train the diagnostic model, and using the vector weighted average algorithm INFO to optimize the penalty factor and kernel function parameters in the SVM model.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for diagnosing faults of a switch machine plunger pump according to any one of claims 1 to 5 are executed.

7. A fault diagnosis device for a switch machine plunger pump, characterized in that: include: Memory, for storing software applications, The processor is used to execute the software application program, wherein each program of the software application program correspondingly executes the steps of the method for diagnosing faults of a switch machine plunger pump as claimed in any one of claims 1 to 5.

Images