Rotating machine fault feature optimization extraction method based on variational mode extraction and comprehensive detection index
By combining variational mode extraction with comprehensive detection indicators, particle swarm optimization algorithm and sample entropy method, the feature extraction of rotating machinery faults is optimized, which solves the problem of difficult parameter selection in traditional methods and realizes high-precision classification and intelligent diagnosis of rotating machinery faults.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA YANGTZE POWER
- Filing Date
- 2026-01-22
- Publication Date
- 2026-05-19
AI Technical Summary
In the diagnosis of rotating machinery faults, traditional methods are difficult to accurately extract early and subtle fault features, and the selection of parameters for variational mode extraction methods is difficult, resulting in low diagnostic accuracy.
A variational mode extraction method is adopted, which combines the comprehensive detection index and the particle swarm optimization algorithm to optimize the parameters. A feature extraction method for rotating machinery vibration signals is designed by combining sample entropy. The variational mode extraction parameters are optimized by the particle swarm optimization algorithm. The fault feature set is calculated by the comprehensive detection index and the sample entropy. Finally, the feature set is fed into the probabilistic neural network for fault classification.
It improves the accuracy of rotating machinery fault classification, enhances the ability to distinguish fault characteristics, realizes intelligent diagnosis of rotating machinery faults, and has higher fault identification rate and robustness.
Smart Images
Figure CN122065005A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power engineering technology, and in particular to an optimized extraction method for fault features of rotating machinery based on variational mode extraction and comprehensive detection indicators. Background Technology
[0002] Rotating machinery, such as rolling bearings, electric motors, and generators, is widely used and crucial in modern industry. However, they typically operate in harsh and complex environments, resulting in a high probability of equipment failure. Therefore, accurate fault diagnosis of rotating machinery is of great significance. Traditional fault diagnosis methods often rely on obvious features in the later stages of a fault, while early, weak fault features are easily masked by noise, requiring high-precision feature extraction for identification. To address this, various time-frequency analysis methods, such as short-time Fourier transform and wavelet transform, have been proposed. However, traditional analysis methods do not consider subsequent classification effects during parameter optimization, leading to low diagnostic accuracy. Furthermore, existing methods such as empirical mode decomposition and variational mode decomposition (VMD) suffer from difficulties in parameter selection and mode aliasing, resulting in inaccurate fault feature extraction and consequently affecting the efficiency and accuracy of fault diagnosis. Therefore, researching feature extraction methods that balance decomposition effectiveness and classification performance has become an urgent need in the field of rotating machinery fault diagnosis.
[0003] Variational Mode Extraction (VME) is a signal processing technique based on the VMD principle. By setting appropriate parameters, it can extract specific modes from vibration signals with high decomposition efficiency. However, the key parameters of VME, the quadratic penalty factor and the approximate center frequency, are usually determined empirically in advance, which may lead to inaccurate decomposition results.
[0004] Particle Swarm Optimization (PSO) is a parameter optimization technique based on swarm intelligence. It iteratively searches for the optimal solution in the solution space by simulating the foraging behavior of bird flocks. In the extraction of fault features from rotating machinery, the PSO algorithm is mainly used to optimize key parameters of signal decomposition methods, thereby improving the signal decomposition effect and enhancing the ability of signal features to distinguish faults.
[0005] The Synthetic Detection Index (SDI) is a comprehensive index used to quantify the sensitivity of feature parameters in a set of candidate fault features. Increasing the SDI value can improve the sensitivity of feature parameters and enhance the fault differentiation capability.
[0006] Sample entropy (SE) is a metric for measuring the complexity of a time series; a higher entropy value indicates a more complex series. Sample entropy can be used to analyze mechanical vibration signals. For example, when a bearing is operating normally, its vibration signal exhibits certain regularity, and the sample entropy value is relatively stable. When the bearing experiences damage or other faults, the complexity of the vibration signal increases, and the sample entropy value changes. By monitoring the sample entropy value, mechanical faults can be detected in a timely manner.
[0007] Given that the VME method has high decomposition efficiency for vibration signals of rotating machinery, but faces the problem of difficult parameter selection, this application proposes an optimized extraction method for rotating machinery fault features based on variational mode extraction and comprehensive detection index. Summary of the Invention
[0008] The technical problem to be solved by this invention is to provide an optimized extraction method for rotating machinery fault features based on variational mode extraction and comprehensive detection index. The comprehensive detection index is used as the objective function of the particle swarm optimization algorithm to optimize the variational mode extraction parameters. In addition, a new method for extracting rotating machinery vibration signals is designed by combining sample entropy. This method can obtain the optimal fault features, thereby improving the accuracy of fault classification and promoting the intelligent diagnosis of rotating machinery faults.
[0009] To achieve the above objectives, this application provides an optimized extraction method for fault features of rotating machinery based on variational mode extraction and comprehensive detection indicators, comprising the following steps: S1. Decompose the input vibration signal using the variational mode extraction method and calculate the corresponding sample entropy value; S2. Construct the fitness function of the particle swarm optimization algorithm using the comprehensive detection index; S3. Optimize the parameters of the variational mode extraction method by combining the comprehensive detection index and the particle swarm optimization algorithm; S4. Extract the decomposed vibration signal using variational modes under optimal parameters; S5. Calculate the sample entropy of each modal component to obtain the fault feature set; S6. Input the fault feature set into the probabilistic neural network for fault classification.
[0010] In S1, the input vibration signal is decomposed using the variational mode extraction method, including the following steps: S1-1, Initialization , and Let n=0, where n represents the number of iterations. For approximate center frequency, Represents the expected mode. For Lagrange multipliers; S1-2. Let n = n + 1, and execute the algorithm; S1-3, For all Update according to the following two formulas respectively , and ; (1) (2) In the formula, For frequency, These are the update parameters in the variational mode extraction algorithm; S1-4, Setting the accuracy value until the convergence condition is met: (3) Stop the loop and output the demodal component; otherwise, return to S1-2 to continue the iteration.
[0011] In S1, the steps for calculating the sample entropy are as follows: S1-5, Assumptions It is a length of L Time series; S1-6. Construct vectors using S according to equation (4): (4) In the formula, ; m For the embedded dimension; S1-7, Define Vectors S ( i )and S ( j The distance between them is , i≠j, That is, the maximum difference between corresponding elements, as shown in equation (5): (5) In the formula, It is a vector S ( i )and S ( j The element index of ) has a value range of . ; S1-8, Given threshold t ( t >0), calculate Quantity: (6) S1-9. Calculate the average value of the results obtained from S1-8, as shown in equation (7): (7) S1-10, For each dimension from 1 to the maximum value m Repeat S1-6 to S1-9; S1-11. Calculate the sample entropy value of the time series, as shown in equation (8): (8) In the formula, It is the first i vectors S ( In the embedding dimension, m Time and other vectors The distance is less than the threshold t probability density estimation, The length of the time series. For the embedded dimension.
[0012] In S2, the fitness function of the particle swarm optimization algorithm is constructed using the comprehensive detection index, including the following steps: S2-1, Calculate the detection index; Let s1 and s2 be the sample entropy values calculated from two different fault state signals, and s 1. s 2 satisfies a normal distribution , , The detection index values corresponding to these two faults are shown in equation (9): (9) in, For detection index, and They represent s The mean and standard deviation of 1 and They represent s The mean and standard deviation of 2; S2-2, Calculate the recognition rate: Assumption The recognition rate is calculated using the following formula: (10) in, To calculate the recognition rate, Let z represent the difference in sample entropy between the two fault states, and denot z = z. Represents an integral infinitesimal element; S2-3. Calculate the comprehensive testing index: (11) in, For comprehensive testing index,M The total sample entropy of each fault state signal. The number of fault types to be distinguished. Indicates the first i The first fault state k The average entropy of each sample Indicates the first i The first fault state k The standard deviation of the entropy of a sample Indicates the first j The first fault state k The average entropy of each sample Indicates the first j The first fault state k The standard deviation of the entropy of a sample.
[0013] In S3, the parameters of the variational mode extraction method are optimized by combining the comprehensive detection index and the particle swarm optimization algorithm, including the following steps: The parameter combination of the variational mode extraction method is continuously updated using the particle swarm optimization algorithm, and S1-4 is repeated until the change in the fitness function value is less than 0.1 after 10 consecutive iterations or the maximum number of iterations is reached. Then the iteration is stopped, and the optimal parameter combination of the variational mode extraction method is obtained. The comprehensive detection index value and the optimal parameter combination at this time are recorded. This parameter combination will be used for subsequent vibration signal decomposition.
[0014] The particle velocity update formula in the particle swarm optimization algorithm is: (12) in, Indicates in t+ 1st moment i The particle in the first j Speed in the dimension Indicates in t Time of the first i The particle in the first j Speed in the dimension; It is inertial weight, which controls the effect of previous speed on current speed; This represents the individual learning factor, which controls the weights by which a particle learns to its own historical best position. This represents the group learning factor, which controls the weights by which particles learn to gravitate towards the optimal position in the group. and It is a random value that is uniformly distributed within the range [0, 1]. Indicates the first i The particle in the first j The optimal position of an individual in the dimension; Indicates the first j The global optimal position of the particle swarm in dimensionality; Indicates in t Time of the first i The particle in the first j The position in the dimension.
[0015] The particle position update formula is: (13) In the formula, Indicates in t+ 1st moment i The particle in the first j The position in the dimension.
[0016] In S6, probabilistic neural networks perform fault classification, including the following steps: S6-1. Input the fault feature set and corresponding fault labels; S6-2, The pattern layer calculates the Gaussian distance between the input features and the training samples: (14) In the formula, The geometric distance between the input features and the training samples in the feature space. For bandwidth parameters, It is the dimension of the feature vector; S6-3. The summation layer calculates the probability density estimates for each category by summing the outputs of samples of the same type: (15) In the formula, The sum of weighted similarities for class j. Let be the prior probability of class j. For all training samples of class j; S6-4. The decision-making level selects the category with the highest posterior probability as the prediction result.
[0017] The decision-making layer selects the category with the highest posterior probability as the prediction result, including the following steps: S6-41. Calculate the posterior probability of each category: (16) In the formula, The input feature vector to be classified. For the j-th fault category, The number of training samples. For a given feature Category The probability of; S6-42. Select the category with the highest posterior probability as the prediction result: (17) In the formula, For predicting categories, To maximize the parameter search operation.
[0018] Before S1, there are steps to input the vibration signal of the rotating machinery and to initialize the relevant parameters; Vibration data of rotating machinery is obtained from the condition monitoring system, and the vibration data covers various operating conditions of the equipment. Initialize relevant parameters, including the population size for the particle swarm optimization algorithm and the parameter optimization range for the variational mode extraction method; the optimization parameter for the variational mode extraction method is the initial center frequency. and penalty coefficient When using the particle swarm optimization algorithm, each particle represents and A set of parameter combinations ( ),in , .
[0019] Compared with the prior art, the above-conceptual technical solution conceived in this application has the following beneficial effects: 1. This invention optimizes two parameters of the variational mode extraction method using the particle swarm optimization algorithm and uses the comprehensive detection index as the fitness function of the particle swarm optimization algorithm; then, the optimized variational mode extraction method is used to process the original vibration signal of rotating machinery to obtain a series of intrinsic mode components; then, the sample entropy value of the intrinsic mode components is calculated to construct a fault feature set; finally, the fault feature set is fed into a probabilistic neural network for fault classification.
[0020] 2. This invention uses the Comprehensive Detection Index (SDI) as the objective function of the Particle Swarm Optimization (PSO) algorithm to optimize the Variational Mode Extraction (VME) parameters, and combines it with Sample Entropy (SE) to design a new method for extracting vibration signals from rotating machinery. This method can effectively process vibration signals under different fault conditions of rotating machinery, obtain the optimal fault features, thereby improving the accuracy of fault classification and promoting the intelligent diagnosis of rotating machinery faults.
[0021] 3. The method of this invention has a higher fault identification rate and better robustness, which is beneficial to the realization of intelligent fault diagnosis of rotating machinery. Its advantages include: a. Precise and efficient parameter optimization, solving the parameter selection problem of traditional methods; b. Strong fault feature discrimination, accurately capturing early and subtle faults; c. High fault classification accuracy and strong robustness; d. Balancing decomposition effect and classification performance, forming a closed-loop optimization; e. Strong engineering applicability, adaptable to complex working conditions and intelligent operation and maintenance needs. Attached Figure Description
[0022] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.
[0023] Figure 1 The method design steps of this invention are described below.
[0024] Figure 2 This is a flowchart of the variational mode extraction (VME) algorithm of the present invention.
[0025] Figure 3 This is a flowchart of the sample entropy (SE) calculation process of the present invention.
[0026] Figure 4 This is a flowchart of the particle swarm optimization (PSO) algorithm of the present invention.
[0027] Figure 5 This is a flowchart of the probabilistic neural network (PNN) classification process of the present invention. Detailed Implementation
[0028] To more clearly illustrate the purpose, technical solution, and beneficial effects of this application, a further detailed description of this application is provided below in conjunction with illustrations and specific embodiments. It should be specifically noted that the specific embodiments described below are only for illustrating the technical content of this application and do not constitute a limitation on the scope of protection of this application.
[0029] Regarding the description of the embodiments: The terms "exemplary" and "for example" appearing in this application are only used to illustrate the technical solutions through specific examples. It should be particularly emphasized that any implementation method or design scheme marked as "exemplary" or "for example" should not be construed as having an advantage over other solutions. Such expressions are only used to present the technical concepts more intuitively.
[0030] Example 1: See Figure 1 This embodiment provides an optimized extraction method for rotating machinery fault features based on variational mode extraction and comprehensive detection indicators, including the following steps: S1. The input vibration signal is decomposed using the variational mode extraction (VME) method, and the corresponding sample entropy value is calculated. S2. Construct the fitness function of the particle swarm optimization algorithm using the Comprehensive Detection Index (SDI); S3. Optimize the parameters of the variational mode extraction (VME) method by combining the comprehensive detection index (SDI) and the particle swarm optimization algorithm (PSO); S4. Use variational mode extraction (VME) under optimal parameters to decompose the input vibration signal; S5. Calculate the sample entropy (SE) of each modal component to obtain the fault feature set; S6. Input the fault feature set into a probabilistic neural network (PNN) for fault classification.
[0031] This invention uses the Comprehensive Detection Index (SDI) as the objective function of the Particle Swarm Optimization (PSO) algorithm to optimize the Variational Mode Extraction (VME) parameters. Combined with Sample Entropy (SE), a novel feature extraction method for rotating machinery vibration signals is designed. This method can effectively process vibration signals under different fault states in rotating machinery, obtaining optimal fault features, thereby improving fault classification accuracy and advancing the intelligent diagnosis of rotating machinery faults. Compared with traditional feature extraction methods, this method has a higher fault recognition rate and better robustness, which is beneficial for the realization of intelligent fault diagnosis in rotating machinery.
[0032] See Figure 2 In S1, the input vibration signal is decomposed using the variational mode extraction method, including the following steps: S1-1, Initialization , and Let n=0, where n represents the number of iterations. For approximate center frequency, Represents the expected mode. For Lagrange multipliers; S1-2. Let n = n + 1, and execute the algorithm; S1-3, For all Update according to the following two formulas respectively , and ; (1) (2) In the formula, For frequency, These are the update parameters in the variational mode extraction algorithm; S1-4, Setting the accuracy value until the convergence condition is met: (3) Stop the loop and output the demodal component; otherwise, return to S1-2 to continue the iteration.
[0033] The VME algorithm, through steps of parameter initialization, iterative updates, convergence judgment, and output of modal components, achieves high-quality modal decomposition of rotating machinery vibration signals, representing a core technological step in achieving accurate vibration signal decomposition. Its core value lies in transforming complex mixed signals into independent, pure IMFs, providing crucial support for subsequent fault feature extraction and classification. Compared to traditional decomposition processes, it exhibits stronger convergence stability, parameter adaptability, and modal independence.
[0034] See Figure 3 In S1, the steps for calculating the sample entropy are as follows: S1-5, Assumptions It is a length of L Time series; S1-6. Construct vectors using S according to equation (4): (4) In the formula, ; m For the embedded dimension; S1-7, Define Vectors S ( i )and S ( j The distance between them is , i≠j, That is, the maximum difference between corresponding elements, as shown in equation (5): (5) In the formula, It is a vector S ( i )and S ( j The element index of ) has a value range of . ; S1-8, Given threshold t ( t >0), calculate Quantity: (6) S1-9. Calculate the average value of the results obtained from S1-8, as shown in equation (7): (7) S1-10, For each dimension from 1 to the maximum value m Repeat S1-6 to S1-9; S1-11. Calculate the sample entropy value of the time series, as shown in equation (8): (8) In the formula, It is the first i vectors S ( In the embedding dimension, m Time and other vectors The distance is less than the threshold t probability density estimation, The length of the time series. For the embedded dimension.
[0035] The sample entropy calculation process, through standardized mathematical logic, achieves efficient transformation from modal components to quantified fault features. This is a crucial transformation step in the entire fault diagnosis solution, directly determining the accuracy of fault classification and the ability to identify early faults. Its core value lies in improving the sensitivity and stability of fault features, providing high-quality feature support for subsequent SDI calculation, parameter optimization, and PNN classification, thus solving the problems of poor data quality and insufficient algorithm compatibility in traditional solutions. Compared to traditional quantified feature calculation processes, it offers stronger anti-interference capabilities, universality, and fault sensitivity. Its fundamental, scenario-specific, and systematic characteristics make it an indispensable foundational step in the entire fault diagnosis solution, directly ensuring the effectiveness of subsequent VME decomposition, parameter optimization, and fault classification.
[0036] In S2, the fitness function of the particle swarm optimization (PSO) algorithm is constructed using the Comprehensive Detection Index (SDI), including the following steps: S2-1. Calculate the Detection Index (DI); Let s1 and s2 be the sample entropy values calculated from two different fault state signals, and s 1. s 2 satisfies a normal distribution , , The detection index values corresponding to these two faults are shown in equation (9): (9) in, For detection index, and They represent s The mean and standard deviation of 1 and They represent s The mean and standard deviation of 2; S2-2. Calculate the Discrimination Rate (DR): Assumption The recognition rate is calculated using the following formula: (10) in, To calculate the recognition rate, Let z represent the difference in sample entropy between the two fault states, and denot z = z. Represents an integral infinitesimal element; S2-3. Calculate the comprehensive testing index: (11) in, For comprehensive testing index, M The total sample entropy of each fault state signal. The number of fault types to be distinguished. Indicates the first i The first fault state k The average entropy of each sample Indicates the first i The first fault state k The standard deviation of the entropy of a sample Indicates the first j The first fault state k The average entropy of each sample Indicates the first j The first fault state k The standard deviation of the entropy of a sample.
[0037] The fitness function is based on the Comprehensive Detection Index (SDI). A higher SDI value indicates a stronger sensitivity of the corresponding sample entropy to faults, resulting in a higher fault identification rate. The Comprehensive Detection Index (SDI) is the sum of the SDI values of all sample entropies under all fault conditions. A higher SDI value indicates higher sensitivity of the sample entropy; therefore, increasing the SDI value can make the sample entropy more sensitive. This invention uses the SDI value of the sample entropy (SE) as the fitness function, aiming to maximize the SDI value, optimizing the initial center frequency and penalty coefficient of variational mode extraction (VME), and ultimately improving the sensitivity of the sample entropy.
[0038] See Figure 4 In S3, the parameters of the variational mode extraction (VME) method are optimized by combining the Integrated Detection Index (SDI) and the Particle Swarm Optimization (PSO) algorithm, including the following steps: The parameter combination of the variational mode extraction method is continuously updated using the particle swarm optimization algorithm, and S1-4 is repeated until the change in the fitness function value is less than 0.1 after 10 consecutive iterations or the maximum number of iterations is reached. Then the iteration is stopped, and the optimal parameter combination of the variational mode extraction method is obtained. The comprehensive detection index value and the optimal parameter combination at this time are recorded. This parameter combination will be used for subsequent vibration signal decomposition.
[0039] The particle velocity update formula in the particle swarm optimization algorithm is: (12) in, Indicates in t+ 1st moment i The particle in the first j Speed in the dimension Indicates in tTime of the first i The particle in the first j Speed in the dimension; It is inertial weight, which controls the effect of previous speed on current speed; This represents the individual learning factor, which controls the weights by which a particle learns to its own historical best position. This represents the group learning factor, which controls the weights by which particles learn to gravitate towards the optimal position in the group. and It is a random value that is uniformly distributed within the range [0, 1]. Indicates the first i The particle in the first j The optimal position of an individual in the dimension; Indicates the first j The global optimal position of the particle swarm in dimensionality; Indicates in t Time of the first i The particle in the first j The position in the dimension.
[0040] The particle position update formula is: (13) In the formula, Indicates in t+ 1st moment i The particle in the first j The position in the dimension.
[0041] The PSO (Problem Solving and Optimization) process, through its collaborative, goal-oriented, and stable convergence design, achieves efficient and accurate optimization of VME (Variable Parameter Environment) parameters. It is the core driving force in the entire fault diagnosis solution, encompassing parameter optimization, high-quality decomposition, and high-precision classification. Its core value lies in deeply integrating fault classification requirements into the parameter search process, solving the problems of difficult parameter selection and poor adaptability in traditional VME methods. Compared to traditional parameter optimization processes, it directly addresses the issue of ambiguous parameter initialization boundaries, offering superior optimization efficiency, accuracy, and adaptability.
[0042] In S4, the input vibration signal is decomposed using variational mode extraction (VME) with optimal parameters, including the following steps: S4-1, Initialization , and ,make n=0 ,in n Indicates the number of iterations. For approximate center frequency, Represents the expected mode. It is a Lagrange multiplier.
[0043] S4-2, Order n=n+1 For all Update according to the following two formulas respectively , and : (14) (15) In the formula, For frequency, These are the update parameters in the VME algorithm.
[0044] S4-3, Set the accuracy value until the convergence condition is met: (16) Stop the loop; otherwise, return to S4-2 to continue the iteration.
[0045] Step S4 decomposes the vibration signal through optimal parameter-driven variational mode extraction (VME). Its core function is to transform the theoretical results of parameter optimization into high-quality signal decomposition results, achieving signal purification and separation, fault feature enhancement, and improved algorithm adaptability. This provides crucial support for subsequent fault feature extraction (S5) and accurate classification (S6). Compared with traditional methods, this approach features strong adaptability, excellent robustness, and strong practical engineering adaptability.
[0046] In S5, the sample entropy (SE) of each modal component is calculated to obtain the fault feature set, including the following steps: S5-1. Take each intrinsic mode component (IMF) output from S4 as an independent time series, and let a single time series be... ,in N The sequence length is given.
[0047] S5-2. Construct the embedding dimension according to the following formula: m vector: (17) In the formula, .
[0048] S5-3, Define Vectors S ( i )and S ( j () i Not equal to j The distance between them is d [ S ( i ) ，S ( j [], which is the maximum difference between corresponding elements: (18) In the formula, It is a vector S ( i )and S ( j The element index of ) has a value range of . .
[0049] S5-4, Given a threshold t ( t>0 ),calculate d [ S ( ) ，S ( j )] <t Quantity: (19) In the formula, It is the first i vectors S ( In the embedding dimension, m Time and other vectors The distance is less than the threshold t probability density estimation, The length of the time series. For the embedded dimension.
[0050] S5-5. Calculate the average value of the results obtained in step S5-4: (20) S5-6 for each dimension from 1 to the maximum value m Repeat steps S5-2 to S5-5.
[0051] The estimated sample entropy of the time series is obtained: (twenty one) Step S5 constructs a fault feature set by calculating the sample entropy (SE) of each modal component after optimal parameter decomposition. The core of this step is to transform high-quality modal components into highly identifiable fault features, achieving quantification of modal components and screening and condensation of fault features, thus providing accurate input for probabilistic neural network (PNN) classification. Compared with traditional methods, this approach features strong feature sensitivity, outstanding robustness, simple and efficient computational logic, and wide applicability in engineering.
[0052] See Figure 5 In S6, probabilistic neural networks (PNNs) are used for fault classification, including the following steps: S6-1. Input the fault feature set and corresponding fault labels; S6-2, The pattern layer calculates the Gaussian distance between the input features and the training samples: ;(twenty two) In the formula, The geometric distance between the input features and the training samples in the feature space. For bandwidth parameters, It is the dimension of the feature vector; S6-3. The summation layer calculates the probability density estimates for each category by summing the outputs of samples of the same type: ;(twenty three) In the formula, For the first j The weighted sum of similarity between classes. For the first j Prior probability of a class For all training samples of class j; S6-4. The decision-making level selects the category with the highest posterior probability as the prediction result.
[0053] Furthermore, the decision-making layer selects the category with the highest posterior probability as the prediction result, including the following steps: S6-41. Calculate the posterior probability of each category: ;(twenty four) In the formula, The input feature vector to be classified. For the j-th fault category, The number of training samples. For a given feature Category The probability of; S6-42. Select the category with the highest posterior probability as the prediction result: (25) In the formula, For predicting categories, To maximize the parameter search operation.
[0054] Step S6, through optimal parameter-driven variational mode extraction, achieves accurate, efficient, and interference-resistant decomposition of vibration signals, and is the core execution link of the entire fault diagnosis process. Its core value lies in transforming the theoretical results of parameter optimization into a high-quality fault feature carrier, providing crucial support for subsequent feature extraction and classification. Compared to traditional decomposition schemes, its adaptive, robust, and highly adaptable characteristics enable it to meet the complex fault diagnosis needs of rotating machinery in industrial scenarios, making it the core transformation link from signal to feature in the entire fault diagnosis method.
[0055] Before S1, there are steps to input the vibration signal of the rotating machinery and to initialize the relevant parameters; Vibration data of rotating machinery is obtained from the condition monitoring system, and the vibration data covers various operating conditions of the equipment. Initialize relevant parameters, including the population size for the particle swarm optimization algorithm and the parameter optimization range for the variational mode extraction method; the optimization parameter for the variational mode extraction method is the initial center frequency. and penalty coefficient When using the particle swarm optimization algorithm, each particle represents and A set of parameter combinations ( ),in , .
[0056] The above two steps are the fundamental preliminary steps in the entire fault diagnosis process, converting physical signals into digital data that the algorithm can process, clarifying the algorithm's search boundaries, and improving parameter optimization efficiency. They enable seamless integration of data acquisition and algorithm execution. Compared with traditional solutions, they feature high data quality purity, fast algorithm convergence speed, strong method generalization ability, and high process stability.
[0057] While specific embodiments of the present invention have been described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and not intended to limit the scope of the invention. Modifications and variations made by those skilled in the art in accordance with the spirit of the invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A method for optimizing the extraction of fault features in rotating machinery based on variational mode extraction and comprehensive detection indicators, characterized in that, Includes the following steps: S1. Decompose the input vibration signal using the variational mode extraction method and calculate the corresponding sample entropy value; S2. Construct the fitness function of the particle swarm optimization algorithm using the comprehensive detection index; S3. Optimize the parameters of the variational mode extraction method by combining the comprehensive detection index and the particle swarm optimization algorithm; S4. Extract the decomposed vibration signal using variational modes under optimal parameters; S5. Calculate the sample entropy of each modal component to obtain the fault feature set; S6. Input the fault feature set into the probabilistic neural network for fault classification.
2. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, In S1, the input vibration signal is decomposed using the variational mode extraction method, including the following steps: S1-1, Initialization , and Let n=0, where n represents the number of iterations. For approximate center frequency, Represents the expected mode. For Lagrange multipliers; S1-2. Let n = n + 1, and execute the algorithm; S1-3, For all Update according to the following two formulas respectively , and ; ；(1) ；(2) In the formula, For frequency, These are the update parameters in the variational mode extraction algorithm. The center frequency of the (n+1)th iteration. This is the estimated value for the (n+1)th iteration in the frequency domain; Parameters for controlling frequency selectivity, bandwidth, and historical information suppression strength; S1-4, Setting the accuracy value until the convergence condition is met: ；(3) Stop the loop and output the demodal component; otherwise, return to S1-2 and continue iterating. In the formula: The convergence threshold, This is the estimated value for the nth iteration in the frequency domain.
3. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 2, characterized in that, In S1, the steps for calculating the sample entropy are as follows: S1-5, Assumptions It is a length of L Time series; S1-6, Use according to formula (4) S Constructing vectors: ； (4) In the formula, ; m For the embedded dimension; S1-7, Define Vectors S ( i )and S ( j The distance between them is , i≠j, That is, the maximum difference between corresponding elements, as shown in equation (5): ； (5) In the formula, It is a vector S ( i )and S ( j The element index of ) has a value range of . ; S1-8, Given threshold t ( t >0), calculate Quantity: ；(6) S1-9. Calculate the average value of the results obtained from S1-8, as shown in equation (7): ； (7) S1-10, For each dimension from 1 to the maximum value m Repeat S1-6 to S1-9; S1-11. Calculate the sample entropy value of the time series, as shown in equation (8): ； (8) In the formula, It is the first i vectors S ( In the embedding dimension, m Time and other vectors The distance is less than the threshold t probability density estimation, The length of the time series. For the embedded dimension.
4. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, In S2, the fitness function of the particle swarm optimization algorithm is constructed using the comprehensive detection index, including the following steps: S2-1, Calculate the detection index; let... s 1. s 2 represents the sample entropy values calculated from two different fault state signals, and s 1. s 2 satisfies a normal distribution , , The detection index values corresponding to these two types of faults are shown in equation (9): ；(9) in, For detection index, and They represent s The mean and standard deviation of 1 and They represent s The mean and standard deviation of 2; S2-2, Calculate the recognition rate: Assumption The recognition rate is calculated using the following formula: ；(10) in, To calculate the recognition rate, Let z represent the difference in sample entropy between the two fault states, and denot z = z. Represents an integral infinitesimal element; S2-3. Calculate the comprehensive testing index: ；(11) in, For comprehensive testing index, M The total sample entropy of each fault state signal. The number of fault types to be distinguished. Indicates the first i The first fault state k The average entropy of each sample Indicates the first i The first fault state k The standard deviation of the entropy of a sample Indicates the first j The first fault state k The average entropy of each sample Indicates the first j The first fault state k The standard deviation of the entropy of a sample.
5. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 2, characterized in that, In S3, the parameters of the variational mode extraction method are optimized by combining the comprehensive detection index and the particle swarm optimization algorithm, including the following steps: The parameter combination of the variational mode extraction method is continuously updated using the particle swarm optimization algorithm, and S1-4 is repeated until the change in the fitness function value is less than 0.1 after 10 consecutive iterations or the maximum number of iterations is reached. Then the iteration is stopped, and the optimal parameter combination of the variational mode extraction method is obtained. The comprehensive detection index value and the optimal parameter combination at this time are recorded. This parameter combination will be used for subsequent vibration signal decomposition.
6. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 5, characterized in that, The particle velocity update formula in the particle swarm optimization algorithm is: ；(12) in, Indicates in t+ 1st moment i The particle in the first j Speed in the dimension Indicates in t Time of the first i The particle in the first j Speed in the dimension; It is inertial weight, which controls the effect of previous speed on current speed; This represents the individual learning factor, which controls the weights by which a particle learns to its own historical best position. This represents the group learning factor, which controls the weights by which particles learn to gravitate towards the optimal position in the group. and It is a random value that is uniformly distributed within the range [0, 1]. Indicates the first i The particle in the first j The optimal position of an individual in the dimension; Indicates the first j The global optimal position of the particle swarm in dimensionality; Indicates in t Time of the first i The particle in the first j The position in the dimension.
7. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 6, characterized in that, The particle position update formula is: ； (13) In the formula, Indicates in t+ 1st moment i The particle in the first j The position in the dimension.
8. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, In S4, the vibration signal of the decomposed input is extracted using variational mode under optimal parameters, including the following steps: S4-1, Initialization , and ,make n=0 ,in n Indicates the number of iterations. For approximate center frequency, Represents the expected mode. For Lagrange multipliers; S4-2, Order n=n+1 For all Update according to the following two formulas respectively , and : ；(14) ；(15) In the formula: For frequency, These are the update parameters in the variational mode extraction algorithm; S4-3, Set the accuracy value until the convergence condition is met: ；(16) Stop the loop; otherwise, return to S4-2 to continue the iteration.
9. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, In S5, the sample entropy of each modal component is calculated to obtain the fault feature set, including the following steps: S5-1. Take each intrinsic mode component of the output of S4 as an independent time series, and let a single time series be... ,in N The sequence length; S5-2. Construct the embedding dimension according to the following formula: m vector: ；(17) In the formula, ; m For the embedded dimension; S5-3, Define Vectors S ( i )and S ( j The distance between them is , i≠j That is, the maximum difference between corresponding elements: ；(18) In the formula, It is a vector S ( i )and S ( j The element index of ) has a value range of . ; S5-4, Given a threshold t ( t>0 ),calculate d [ S ( ) ，S ( j )] <t Quantity: ；(19) In the formula, It is the first i vectors S ( In the embedding dimension, m Time and other vectors The distance is less than the threshold t probability density estimation, The length of the time series. For the embedded dimension; S5-5. Calculate the average value of the results obtained in step S5-4: ；(20) S5-6 for each dimension from 1 to the maximum value m Repeat S5-2 to S5-5; The estimated sample entropy of the time series is obtained: ；(21) In the formula: This represents the estimated sample entropy.
10. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, In S6, probabilistic neural networks perform fault classification, including the following steps: S6-1. Input the fault feature set and corresponding fault labels; S6-2, The pattern layer calculates the Gaussian distance between the input features and the training samples: ；(22) In the formula, The geometric distance between the input features and the training samples in the feature space. For bandwidth parameters, It is the dimension of the feature vector; S6-3. The summation layer calculates the probability density estimates for each category by summing the outputs of samples of the same type: ；(23) In the formula, For the first j The weighted sum of similarity between classes. For the first j Prior probability of a class For all training samples of class j; S6-4. The decision-making level selects the category with the highest posterior probability as the prediction result.
11. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 8, characterized in that, The decision-making layer selects the category with the highest posterior probability as the prediction result, including the following steps: S6-41. Calculate the posterior probability of each category: ；(24) In the formula, The input feature vector to be classified. For the j-th fault category, The number of training samples. For a given feature Category The probability of; S6-42. Select the category with the highest posterior probability as the prediction result: ；(25) In the formula, For predicting categories, To maximize the parameter search operation.
12. The method for optimizing the extraction of fault features of rotating machinery based on variational mode extraction and comprehensive detection index as described in claim 1, characterized in that, Before S1, there are steps to input the vibration signal of the rotating machinery and to initialize the relevant parameters; Vibration data of rotating machinery is obtained from the condition monitoring system, and the vibration data covers various operating conditions of the equipment. Initialize relevant parameters, including the population size for the particle swarm optimization algorithm and the parameter optimization range for the variational mode extraction method; the optimization parameter for the variational mode extraction method is the initial center frequency. and penalty coefficient When using the particle swarm optimization algorithm, each particle represents and A set of parameter combinations ( ),in , .