Fault diagnosis method for engine gearbox based on PSO-VMD and multi-class support vector machine

Through the improved variable mode decomposition and multi-class support vector machine technology by particle swarm optimization algorithm, the characteristics of the engine gearbox vibration signal are extracted, and the problem of low accuracy of fault diagnosis under small sample data is solved, and efficient and accurate fault identification is achieved. and classification.

CN119063997APending Publication Date: 2024-12-03DALIAN UNIV OF TECH
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411124226.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

In the case of small sample data, the accuracy of engine gearbox fault diagnosis is low, and the traditional methods rely on manual experience, making it difficult to apply to large-scale development.

Method used

The particle swarm optimization algorithm (PSO) is used to improve the variable mode decomposition (VMD) algorithm, combined with multiple types of support vector machines, and the accurate diagnosis of engine gearbox failures is achieved through feature extraction and classification of vibration signals.

Benefits of technology

It improves the accuracy of engine gearbox fault diagnosis, reduces overfitting, and realizes efficient and accurate fault type identification and classification, which is suitable for small sample data processing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119063997A_ABST
    Figure CN119063997A_ABST
Patent Text Reader

Abstract

The invention discloses an engine gearbox fault diagnosis method based on PSO-VMD and a multi-class support vector machine, and belongs to the technical field of engine gearbox fault diagnosis. The method comprises the following steps: optimizing two parameters, namely the number K of components of VMD and a penalty factor alpha, by using a PSO algorithm, and performing modal decomposition on a vibration signal by using an optimized VMD method to obtain a plurality of IMF components; selecting a sensitive mode by calculating the energy of the IMF component; and reconstructing the sensitive mode to complete noise reduction processing of the vibration signal. And extracting the reconstructed vibration signal to obtain a characteristic variable, inputting the extracted characteristic variable into a multi-class support vector machine with a maximum minimum interval, and selecting a suitable adjustable parameter p and a slack variable s to complete fault diagnosis of the engine gearbox. Simulation experiments show that the method can significantly improve the accuracy of fault diagnosis of the engine gearbox.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of engine gearbox fault diagnosis, and in particular relates to an engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machines. Background Art

[0002] Engines have been widely used in all walks of life, including agricultural machinery and construction machinery. Especially in the field of vehicles and power generation, engines are widely used as power propulsion devices. The working condition of its gearbox will directly affect the working condition and performance of the equipment. With the improvement and improvement of engine performance, the cost of engine gearbox failure repair and the difficulty of fault detection and repair have been greatly increased. Traditionally, regular disassembly inspection or replacement of parts for repair is used to ensure that the gearbox can work properly, but this method requires a lot of manpower and material resources, and is highly dependent on the manual experience of maintenance personnel, which is not suitable for large-scale development. In recent years, with the development of artificial intelligence technology and machine learning methods, sensors are used to collect the status information of engine gearboxes in real time, and artificial intelligence technology is combined to analyze data to diagnose engine gearbox faults without disassembling the machine. Summary of the invention

[0003] The purpose of the present invention is to provide an engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machine, which utilizes artificial intelligence technology and machine learning methods to improve the accuracy of engine gearbox fault diagnosis, so as to solve the problems of incomplete noise reduction and low fault diagnosis accuracy in the case of small sample data in the prior art.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a fault diagnosis method of an engine gearbox based on PSO-VMD and multi-class support vector machine, comprising the following steps:

[0005] S1. Use a vibration acceleration sensor to collect the original vibration signals of the engine gearbox under three working conditions: normal, broken teeth, and tooth wear;

[0006] S2. The particle swarm optimization algorithm (PSO) is used to improve the variable mode decomposition (VMD), and the optimization model of the number of components (K) and penalty factor (α) of the VMD algorithm is established using the PSO method;

[0007] S3. Perform VMD modal decomposition on the gear vibration signals under the three working conditions of normal, broken tooth and worn in step S1 according to the optimal parameters obtained in step 2 to obtain several variational modal components (IMF) of the original vibration signal;

[0008] S4. Calculate the energy of several IMFs decomposed in step 3, and select the IMF with 90% energy as the sensitive mode;

[0009] S5. Reconstruct the sensitive mode to achieve signal noise reduction, and extract features from the reconstructed signal;

[0010] S6. Input the features of the reconstructed vibration signal into the multi-class support vector machine with the maximum minimum interval to verify the classification effect of the fault diagnosis.

[0011] First, read the time series x(t) of the vibration signal x of the engine gearbox, where t = 1, 2, 3, ... N, N is the number of acquisition points of the gearbox signal:

[0012] Among them, the VMD method can construct and solve the variational problem. The VMD algorithm is used to decompose the original vibration signal to obtain a series of variational modal components. Its variational model is as follows:

[0013]

[0014] In the formula, {u k}={u1,u2,…,u K},u k represents the kth modal component after decomposition; {ω k}={ω1,ω2,…,ω K},ω k represents the center frequency corresponding to the kth modal component; K is the number of modal components to be decomposed, which is a positive integer; is the partial derivative with respect to t; δ(t) is the Dirac distribution; * represents convolution; H is the original signal.

[0015] In order to obtain the optimal solution of the above constraints, the Lagrange multiplication operator λ and the penalty factor α are introduced to convert the constrained model into an unconstrained model, which can be expressed as:

[0016]

[0017] VMD uses the alternating multiplication operator (ADMM) to continuously λ n+1 Perform iterative updates to obtain the optimal solution of the above formula. The specific implementation process is as follows:

[0018] 1) Initialization λ 1 and n, and set n to 0

[0019] 2) The number of iterations n = n + 1;

[0020] 3) Iteratively update parameters λ 1 , the formula is as follows:

[0021]

[0022] In the formula, are the corresponding Fourier transforms respectively; τ represents the noise tolerance.

[0023] 4) Repeat steps 2) and 3) above until the iteration stop condition is met:

[0024]

[0025] In the formula, ε>0 is the discrimination accuracy.

[0026] 5) End the loop and obtain K variational mode components (IMFs).

[0027] In the step S2, since the selection of the number of components (K) and the penalty factor (α) of the variational mode decomposition (VMD) algorithm will affect the result of the variational mode component (IMF) obtained after classification, the particle swarm optimization algorithm (PSO) can be used to improve the variational mode decomposition (VMD), establish the variational mode decomposition algorithm, and iteratively train and search for the optimal parameters of VMD: the number of components (K) and the penalty factor (α):

[0028] (1) The mathematical model of particle swarm optimization algorithm is as follows:

[0029]

[0030] X i(d+1) =X id +V id (8)

[0031] In the formula, X id is the position of the ith particle, V id is the velocity of the ith particle, P id,pbest is the optimal position searched by the i-th particle, P d,gbest is the optimal position searched by the group, rand() represents a random number in the range [0,1], c1, c2 are different learning factors, ω VMD is the inertia weight, which is non-negative and the larger the value, the stronger the global optimization ability and the weaker the local optimization ability, and vice versa;

[0032] (2) Calculate the fitness value of each particle; the entropy value of the envelope signal can reflect the degree of information contained in the signal, as shown in the following formula:

[0033]

[0034] Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; a jrepresents the envelope value of the signal at the jth point; p j Indicates the envelope normalization of the signal.

[0035] The kurtosis of the envelope signal can measure the sensitivity of changes in fault information in the vibration signal, as shown in the following formula:

[0036]

[0037] Where: μ represents the mean amplitude of signal a.

[0038] Therefore, the fusion of the minimum mean envelope information entropy and envelope kurtosis is used as the fitness function of the PSO algorithm, as shown in the following formula:

[0039]

[0040] Where Z is the fitness function.

[0041] In step S4, the energy of each variational modal component is calculated, and the formula is as follows:

[0042]

[0043] Where: E k It represents the proportion of the total energy occupied by the kth modal signal.

[0044] Afterwards, the IMF with 90% energy is selected as the sensitive mode, and the sensitive mode is reconstructed to achieve signal noise reduction.

[0045] In the step S5, feature extraction is performed on the reconstructed signal, and the kurtosis, skewness, envelope entropy and sample entropy of the reconstructed signal are calculated as feature variables.

[0046] Kurtosis can describe the peak degree of the signal, which reflects the impact characteristics of the vibration signal. Kurtosis is sensitive to impact. Generally, the kurtosis value should be around 3. If it deviates too much from 3, it means that the mechanical equipment has a certain impact vibration and there may be some hidden dangers of failure. The formula is as follows:

[0047]

[0048] Where: μ represents the mean amplitude of signal a.

[0049] The skewness reflects the asymmetry of the vibration signal. Usually, the vibration signal is symmetrical about the x-axis, and the skewness should be close to 0. If the friction or collision in one direction of the equipment is large, it will cause the vibration to be asymmetric, making the skewness larger. The formula is as follows:

[0050]

[0051] Where: μ represents the mean amplitude of signal a.

[0052] Envelope entropy reflects the degree of information contained in the signal and represents the sparse characteristics of the original signal. When there is more noise and less feature information in the IMF, the envelope entropy value is larger. Conversely, the envelope entropy value is smaller. The formula is as follows:

[0053]

[0054] Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; aj represents the envelope value of the jth point of the signal; pj represents the envelope normalization of the signal.

[0055] Sample entropy is an entropy measurement method based on the amount of information in two or more patterns in time series data. It measures the diversity of data by calculating the amount of information in each pattern in the time series and weighted averaging these information. The formula is as follows:

[0056]

[0057] Where: N is the number of samples, m is the dimension of embedding selected to construct a new state vector, B i For x i and x j The number of similarity tolerance parameters whose distance is less than a given similarity tolerance parameter.

[0058] In step S6, the features are input into a multi-class support vector machine with the minimum interval maximized to verify the classification effect of the fault diagnosis.

[0059] To solve the binary classification subproblem between classes k and l and maximize the margin between their data, the goal of multi-objective optimization is as follows:

[0060]

[0061] Where: x i is the i-th eigenvector, y i is the category label, w is the normal vector of the classification plane, and p is an adjustable parameter. Selecting a suitable p can globally expand the margins between different categories and increase the lower limit of the margins.

[0062] Considering the indivisible situation, the slack variable s is introduced, and the formula is as follows:

[0063]

[0064] Using non-negative slack variables To handle data that is not linearly separable;

[0065] By substituting the slack variable s, the SVM multi-class method is transformed into an unconstrained optimization problem, which is formulated as follows:

[0066]

[0067] Where:

[0068]

[0069] The beneficial effects of the present invention are:

[0070] 1. When VMD is used to decompose the original vibration signal, the selection of the number of components (K) and the penalty factor (α) will affect the result of the variational modal component (IMF) obtained after classification. Improper selection may cause a modal component to contain multiple components, resulting in incomplete decomposition; or cause the frequency of the modal center to overlap, thereby generating false components. The present invention uses PSO to adaptively select the appropriate number of components (K) and penalty factor (α), solving the problem that VMD parameters cannot be adaptive, making the decomposition result more reasonable and accurate, thereby better achieving noise reduction of the actual signal.

[0071] 2. The commonly used multi-class support vector machine model has low accuracy in engine gearbox fault diagnosis. The present invention follows the principle of maximizing the minimum interval and proposes a new multi-class support vector machine model. By selecting appropriate adjustable parameters p and slack variables s, the generalization performance can be widely improved. It not only improves the distinction between different categories, but also effectively reduces the occurrence of overfitting, thereby greatly improving the accuracy of engine gearbox fault diagnosis.

[0072] The original vibration signal can be denoised using the time-frequency domain analysis method. Since the number of components (K) and penalty factor (α) of the variational mode decomposition (VMD) algorithm are determined before signal decomposition, the result of their selection will directly affect the result of variational mode component (IMF) obtained after classification. Therefore, the particle swarm optimization algorithm (PSO) is used to improve the variational mode decomposition (VMD), denoise the original vibration signal, and complete the extraction of characteristic parameters.

[0073] Since the support vector machine (SVM) has a strong generalization ability, it has a good classification effect when processing a small number of samples. Therefore, the present invention improves the traditional multi-class support vector machine and inputs the extracted feature parameters into the multi-class support vector machine with the maximum minimum interval, thereby realizing the recognition and classification of fault types. This method can better complete the classification of gearbox fault types. It can be seen that the study of an engine gearbox fault diagnosis system with high-efficiency and high-precision monitoring capabilities is of great significance to the development of engine gearboxes. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 It is an implementation flow chart of the present invention.

[0075] Figure 2 Schematic diagram of the original vibration signal of the gearbox broken tooth condition.

[0076] Figure 3 Schematic diagram of the optimal number of components (K) and penalty factor (α) obtained by PSO-VMD.

[0077] Figure 4 Schematic diagram of VMD decomposition when the optimal parameters are selected.

[0078] Figure 5 Schematic diagram of the classification accuracy of a multi-class support vector machine that maximizes the minimum margin. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention is introduced more clearly, concisely and completely in conjunction with the accompanying drawings. The described embodiments are part of the embodiments of the present invention, but not all of the embodiments. All embodiments according to the present invention belong to the protection scope of the present invention.

[0080] Figure 1 The complete process of the engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machine according to the present invention is shown, which includes the following steps:

[0081] Step 1: Use a vibration acceleration sensor to collect the original vibration signals of the engine gearbox under three working conditions: normal, broken teeth, and tooth wear;

[0082] In this step, the time series x(t) of the vibration signal x of the engine gearbox is read, where t=1, 2, 3, ... N, and N is the number of acquisition points of the gearbox signal:

[0083] Step 2: Particle swarm optimization (PSO) algorithm is used to improve variable mode decomposition (VMD), and the optimization model of the number of components (K) and penalty factor (α) of VMD method is established by using PSO algorithm;

[0084] In this step, since the selection of the number of components (K) and the penalty factor (α) of the variational mode decomposition (VMD) algorithm will affect the results of the variational mode component (IMF) obtained after classification, the particle swarm optimization algorithm (PSO) can be used to improve the variational mode decomposition (VMD).

[0085] VMD constructs and solves the variational problem. The VMD algorithm is used to decompose the original vibration signal to obtain a series of variational modal components. The variational model is as follows:

[0086]

[0087] In the formula, {u k}={u1,u2,…,u K},u k represents the kth modal component after decomposition; {ω k}={ω1,ω2,…,ω K},ω k represents the center frequency corresponding to the kth modal component; K is the number of modal components to be decomposed, which is a positive integer; is the partial derivative with respect to t; δ(t) is the Dirac distribution; * represents convolution; H is the original signal.

[0088] In order to obtain the optimal solution of the above constraints, the Lagrange multiplication operator λ and the penalty factor α are introduced to convert the constrained model into an unconstrained model, which can be expressed as:

[0089]

[0090] Alternating multiplication operator (ADMM) λ n+1 Perform iterative updates to obtain the optimal solution. The specific implementation process is as follows:

[0091] 1) Initialization λ 1 and n, and set to 0

[0092] 2) The number of iterations n = n + 1;

[0093] 3) Iteratively update parameters λ 1 , the formula is as follows:

[0094]

[0095] In the formula, are their corresponding Fourier transforms respectively; τ represents the noise tolerance.

[0096] 4) Repeat steps 2) and 3) above until the iteration stop condition is met:

[0097]

[0098] In the formula, ε>0 is the discrimination accuracy.

[0099] 5) End the loop and obtain K variational mode components (IMFs).

[0100] Afterwards, the particle swarm optimization algorithm (PSO) is used to improve the variable mode decomposition (VMD), and the optimization model of the number of components (K) and penalty factor (α) of the variational mode decomposition algorithm is established to iteratively find the optimal parameters. The mathematical model of the particle swarm optimization algorithm is as follows

[0101]

[0102] X i(d+1) =X id +V id (28)

[0103] In the formula, X id is the position of the ith particle, V id is the velocity of the ith particle, P id,pbest is the optimal position searched by the i-th particle, P d,gbest is the optimal position searched by the group, rand() represents a random number in the range [0,1], c1, c2 are different learning factors, ω VMD is the inertia weight, which is non-negative and the larger the value, the stronger the global optimization ability and the weaker the local optimization ability, and vice versa;

[0104] Afterwards, the fitness value of each particle is calculated; the entropy value of the envelope signal can reflect the degree of information contained in the signal, as shown in formula (49), and the kurtosis of the envelope signal can measure the sensitivity of the change of fault information in the vibration signal, as shown in formula (50). Therefore, the fusion of the minimum mean envelope information entropy and the envelope kurtosis is used as the fitness function of the PSO algorithm, as shown in formula (51):

[0105]

[0106] Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; a j represents the envelope value of the signal at the jth point; p j Indicates the envelope normalization of the signal.

[0107]

[0108] Where: μ represents the mean amplitude of signal a.

[0109]

[0110] Where Z is the fitness function.

[0111] Step 3: Perform VMD modal decomposition on the gear vibration signals under the three working conditions of normal, broken tooth and worn in step 1 according to the optimal parameters obtained in step 2 to obtain several variational modal components (IMF) of the original vibration signal;

[0112] In this step, the number of components (K) and penalty factor (α) are set according to the optimal parameters selected by PSO-VMD, and the original vibration signal is decomposed into K variational mode components (IMFs).

[0113] Step 4: Calculate the energy of several IMFs decomposed in step 3, and select the IMF with 90% energy as the sensitive mode;

[0114] In this step, the energy of each variational modal component is calculated, and the IMF with 90% energy is selected as the sensitive mode. The energy calculation formula is as follows;

[0115]

[0116] Where: E k It represents the proportion of the total energy occupied by the kth modal signal.

[0117] Step 5: Reconstruct the sensitive mode to reduce the noise of the signal and extract features from the reconstructed signal;

[0118] In this step, the sensitive mode is reconstructed to obtain a new vibration signal, and the noise reduction of the original vibration signal is completed. The kurtosis, skewness, envelope entropy and sample entropy of the reconstructed signal are calculated and used as feature variables.

[0119] Kurtosis can describe the peak degree of the signal, which reflects the impact characteristics of the vibration signal. Kurtosis is sensitive to impact. Generally, the kurtosis value should be around 3. If it deviates too much from 3, it means that the mechanical equipment has a certain impact vibration and there may be some hidden dangers of failure. The formula is as follows:

[0120]

[0121] Where: μ represents the mean amplitude of signal a.

[0122] The skewness reflects the asymmetry of the vibration signal. Usually, the vibration signal is symmetrical about the x-axis, and the skewness should be close to 0. If the friction or collision in one direction of the equipment is large, it will cause the vibration to be asymmetric, making the skewness larger. The formula is as follows:

[0123]

[0124] Where: μ represents the mean amplitude of signal a.

[0125] Envelope entropy reflects the degree of information contained in the signal and represents the sparse characteristics of the original signal. When there is more noise and less feature information in the IMF, the envelope entropy value is larger. Conversely, the envelope entropy value is smaller. The formula is as follows:

[0126]

[0127] Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; aj represents the envelope value of the jth point of the signal; pj represents the envelope normalization of the signal.

[0128] Sample entropy is an entropy measurement method based on the amount of information in two or more patterns in time series data. It measures the diversity of data by calculating the amount of information in each pattern in the time series and weighted averaging these information. The formula is as follows:

[0129]

[0130] Where: N is the number of samples, m is the dimension of embedding selected to construct a new state vector, B i For x i and x j The number of similarity tolerance parameters whose distance is less than a given similarity tolerance parameter.

[0131] Step 6: Input the feature vector of the reconstructed vibration signal into the multi-class support vector machine with the maximum minimum margin to verify the classification effect of fault diagnosis.

[0132] In this step, selecting appropriate adjustable parameters p and slack variables s can widely improve the generalization performance, which not only improves the discrimination between different categories, but also effectively reduces the occurrence of overfitting and greatly improves the accuracy of engine gearbox fault diagnosis.

[0133] To solve the binary classification subproblem between classes k and l and maximize the margin between their data, the goal of multi-objective optimization is as follows:

[0134]

[0135] Where: x i is the i-th eigenvector, y i is the category label, w is the normal vector of the classification plane, and p is an adjustable parameter. Selecting a suitable p can globally expand the margins between different categories and increase the lower limit of the margins.

[0136] Considering the indivisible situation, the slack variable s is introduced, and the formula is as follows:

[0137]

[0138] Using non-negative slack variables To handle data that is not linearly separable;

[0139] By substituting the slack variable s, the SVM multi-class method is transformed into an unconstrained optimization problem, which is formulated as follows:

[0140]

[0141] Where:

[0142]

[0143] Finally, the status of three types of engine gearboxes are classified by inputting different feature variables into a multi-class support vector machine with minimum margin maximization.

[0144] Example 1

[0145] The present invention is described in detail through simulation examples, and the effectiveness of the experimental scheme in the present invention is analyzed. This embodiment selects three sets of data sets of the engine gearbox, the sampling rate of each set of data sets is 10kHz, and the sampling time is 10s; the motor speed is 1420rpm / min, the number of teeth of the small gear is 15, and the number of teeth of the large gear is 110; the meshing frequency = (1420 / 60) * 15 = 355Hz, and the actual meshing frequency is about 365Hz in the spectrum analysis; the gearbox has three working states: broken teeth, normal, and three teeth worn.

[0146] The vibration signal of the gearbox contains a large number of amplitude-frequency modulation signals, so the VMD method can be used for variational mode decomposition. The "broken tooth" data set is selected for simulation. The original vibration signal is as follows: Figure 2 As shown in the figure, the number of components (K) and penalty factor (α) of the VMD method are adaptively selected by the particle swarm optimization algorithm, as shown in Figure 3 As shown, the optimal parameters of VMD are: the number of components (K) is 6223, and the penalty factor (α) is 20.

[0147] Because there are 100,000 data in the working condition data set of each gearbox, they are evenly divided into 90 groups, each with 1,024 data. The original vibration signal is decomposed using the obtained optimal parameters to obtain 20 variational modal components, such as Figure 4 As shown. According to the energy calculation formula, the energy and energy proportion of each variational modal component are obtained, as shown in Table 1. The IMF with an energy proportion of about 90% is selected as the sensitive mode, and the sensitive mode is reconstructed to obtain the vibration signal after noise reduction.

[0148] Table 1 Energy and energy proportion of variational modal components of the “broken tooth” dataset

[0149]

[0150] According to the method of the embodiment, the above steps are repeated to obtain data of three different gearbox working conditions (broken teeth, normal, and three teeth worn), a total of 270 groups of data reconstructed signals, and the kurtosis, skewness, envelope entropy and sample entropy of the reconstructed signals are calculated respectively, which are used as feature variables and input into the multi-class support vector machine with the maximum minimum interval. The classification accuracy of the multi-class support vector machine of the present invention is 98.2%. Figure 5 shown.

[0151] In summary, through a large number of simulation experiments, it is proved that the fault diagnosis method of the present invention can more effectively improve the accuracy of engine gearbox fault diagnosis.

[0152] With the above embodiments of the present invention as inspiration, through the above description, relevant personnel in the field can make several changes and modifications according to the technical solution of the present invention without departing from the technical idea of ​​the present invention, and these changes and modifications should also be regarded as the protection scope of the present invention. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A fault diagnosis method for engine gearbox based on PSO-VMD and multi-class support vector machine, characterized in that: The following steps are involved: S1. Use a vibration acceleration sensor to collect the original vibration signals of the engine gearbox under three working conditions: normal, broken teeth, and tooth wear; S2. The particle swarm optimization algorithm PSO is used to improve the variable mode decomposition VMD, and the optimization model of the number of components K and the penalty factor α of the VMD algorithm is established using the PSO method: S2.1 uses particle swarm optimization algorithm PSO to improve variable mode decomposition VMD, specifically: S2.1.1 reads the time series x(t) of the vibration signal x of the engine gearbox, where t = 1, 2, 3, ... N, N is the number of acquisition points of the gearbox signal: S2.1.2 The VMD algorithm is used to decompose the original vibration signal to obtain a series of variational modal components. The variational model is as follows: In the formula, {u k }={u1,u2,…,u K },u k represents the kth modal component after decomposition; {ω k }={ω1,ω2,…,ω K },ω k represents the center frequency corresponding to the kth modal component; K is the number of modal components to be decomposed, which is a positive integer; is the partial derivative of t; δ(t) is the Dirac distribution; * represents convolution; H is the original signal; S2.1.3 introduces the Lagrange multiplication operator λ and the penalty factor α to obtain the optimal solution for the above constraints, and converts the constrained model into an unconstrained model, which is expressed as: S2.1.4 Using the alternating multiplication operator ADMM Perform iterative updates to obtain the optimal solution; the specific implementation process is as follows: Initialization and n, and set n to 0; the number of iterations n = n + 1; Iteratively update parameters The formula is as follows: In the formula, are the corresponding Fourier transforms respectively; τ represents the noise tolerance; Repeat the above steps until the iteration stop condition is met: Where, ε is the discrimination accuracy and ε>0; the loop ends and K variational modal IMFs are obtained; S2.2 establishes an optimization model for the number of components K and penalty factor α of the variational mode decomposition algorithm, specifically: S2.2.1 The mathematical model of particle swarm optimization algorithm is as follows: X i(d+1) =X id +V id (8) In the formula, X id is the position of the ith particle, V id is the velocity of the ith particle, P id,pbest is the optimal position searched by the i-th particle, P d,gbest is the optimal position searched by the group, rand() represents a random number in the range [0,1], c1, c2 are different learning factors, ω VMD is the inertia weight; S2.2.2 Calculate the fitness value of each particle: The entropy value of the envelope signal reflects the degree of information contained in the signal, as shown in formula (9): Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; a j represents the envelope value of the signal at the jth point; p j Indicates the envelope normalization of the signal; The kurtosis of the envelope signal measures the sensitivity of the change of fault information in the vibration signal, as shown in formula (10): Where: μ represents the mean amplitude of signal a; Therefore, the fusion of the minimum envelope information entropy and envelope kurtosis is used as the fitness function Z of the PSO algorithm, as shown in formula (11): S3: performing VMD modal decomposition on the gear vibration signals under the three working conditions of normal, broken tooth and worn in step S1 according to the optimal parameters obtained in step S2 to obtain several variational modal components IMF of the original vibration signal; S4: Calculate several variational modal components IMF obtained by decomposition in step S3, and select the IMF with 90% energy as the sensitive mode; S5: Reconstruct the sensitive mode to achieve signal noise reduction, and extract features from the reconstructed signal; S6: Input the features of the reconstructed signal into the multi-class support vector machine with the maximum minimum interval to complete the fault diagnosis of the engine gearbox.

2. The engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machine according to claim 1 is characterized in that: In step S4, several variational modal components IMF are calculated, and the IMF with 90% energy is selected as the sensitive mode, and the formula is as follows: Where: E k It represents the proportion of the total energy occupied by the kth modal signal.

3. The engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machine according to claim 1, characterized in that: In the step S5, feature extraction is performed on the reconstructed signal: the kurtosis, skewness, envelope entropy and sample entropy of the reconstructed signal are calculated as feature variables; The kurtosis describes the peak degree of the signal, and the formula is as follows: Where: μ represents the mean amplitude of signal a; The skewness reflects the asymmetry of the vibration signal, and the formula is as follows: Where: μ represents the mean amplitude of signal a; The envelope entropy reflects the degree to which the signal contains information, and the formula is as follows: Where: MEE(k) represents the envelope information entropy of the kth IMF; M represents the length of the signal; aj represents the envelope value of the jth point of the signal; pj represents the envelope normalization of the signal; Sample entropy measures the diversity of data by calculating the amount of information for each pattern in the time series and weighted averaging these amounts of information. The formula is as follows: Where: N is the number of samples, m is the dimension of embedding selected to construct a new state vector, B i For x i and x j The number of similarity tolerance parameters whose distance is less than a given similarity tolerance parameter.

4. The engine gearbox fault diagnosis method based on PSO-VMD and multi-class support vector machine according to claim 1, characterized in that: In the step S6, the features are input into a multi-class support vector machine with the minimum interval maximized, and the adjustable parameter p and the slack variable s are selected to complete the fault diagnosis of the engine gearbox; To solve the binary classification subproblem between classes k and l and maximize the margin between their data, the goal of multi-objective optimization is as follows: Where: x i is the i-th eigenvector, y i is the category label, w is the normal vector of the classification plane, and p is an adjustable parameter. Selecting a suitable p can globally expand the margins between different categories and increase the lower limit of the margins. Considering the indivisible situation, the slack variable s is introduced, and the formula is as follows: Using non-negative slack variables To handle data that is not linearly separable; By substituting the slack variable s, the SVM multi-class method is transformed into an unconstrained optimization problem, which is formulated as follows: Where:

Citation Information

Cited By

  • Circuit breaker unknown fault diagnosis method based on IGOA-SVDD

    CN120103133A

  • Acoustic real-time monitoring method for faults of carrier rollers of belt type coal conveyor based on grouping collaboration

    CN120800797A

  • Grouping-based cooperative acoustic real-time monitoring method for coal belt conveyor roller fault

    CN120800797B

  • Gear transmission system vibration source identification method and system, electronic equipment and medium

    CN121163887A

  • Rolling bearing fault identification method under time-varying rotating speed working condition and terminal equipment

    CN122016317A