Fault diagnosis method and system based on adaptive weighted sin geometry modal decomposition

Abstract

The application provides a kind of fault diagnosis method and system based on adaptive weighted symplectic geometry modal decomposition, adopts AWSGMD-CP algorithm to the vibration signal of rolling bearing and is decomposed, obtains WSCCs component, extracts power spectrum entropy weighted singular value as fault feature vector according to WSCCs component;Based on the obtained fault feature vector and the optimized ACO-ELM model, the fault diagnosis result is obtained;Wherein, the optimization of ACO-ELM model includes: using ACO algorithm, the parameters in ELM are iteratively optimized, and the optimal input weight and neuron threshold are obtained;The application is based on symplectic geometry similarity transformation, and an improved adaptive weighted and reconstructed symplectic geometry modal decomposition algorithm is proposed, which has good decomposition performance for vibration signals, can effectively extract rich fault features in signals, and improves the accuracy of fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method and system based on adaptive weighted symplectic geometric mode decomposition. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Rolling bearings, as a crucial component of rotating machinery, often bear time-varying loads, making them prone to unavoidable damage. Furthermore, their vibration signals have relatively low energy and typically contain significant noise and irrelevant components, leading to unclear or even obscured fault characteristics. To address this challenge, numerous researchers have conducted extensive studies, primarily focusing on improving signal processing methods and fusing and optimizing machine learning algorithms. However, most existing methods fail to effectively extract fault characteristics from bearing vibrations or lack strong adaptive capabilities and noise robustness, resulting in unsatisfactory classification performance.

[0004] To improve the feature extraction capability of vibration signals, researchers have proposed numerous signal decomposition methods. Wavelet transform decomposes time series into components of different frequency bands or scales, enabling the identification of abrupt changes in the time series. However, the selection of wavelet functions is still largely empirical and lacks adaptability; different basis functions can sometimes lead to significant differences in decomposition results.

[0005] EMD (Enhanced Mode Decomposition) adaptively decomposes a multi-component modulated signal into IMF (Integrated Mode Factor) components without requiring the setting of basis functions, but it suffers from endpoint effects and mode aliasing. LMD (Low Mode Decomposition) is an improvement on EMD, adaptively decomposing the signal into the sum of product functions PF (Power Factors), where any PF component is an amplitude-frequency modulated (AM-FM) signal. This method is widely used in bearing fault diagnosis as an adaptive decomposition method. VMD (Vibration Mode Decomposition) is an adaptive time-frequency analysis method. Due to the characteristics of Wiener filters, VMD performs well in decomposing frequency bands with similar frequencies and reducing mode confusion. To better improve the feature extraction capability of fault information, Pan et al. proposed the SGMD method based on symplectic geometry theory, which can adaptively decompose a time series into several independent mode SGCs (Synthetic Generic Components). It not only preserves the integrity of the time series but also eliminates noise and irrelevant components, requires no user-defined parameters, and has strong adaptability. However, there is a problem of reconstruction constraints during the decomposition process. The decomposed components are not completely independent, and components with similar characteristics need to be recombined. The decomposed components contain residuals and noise, and appropriate iteration termination conditions need to be set to remove these useless components. Otherwise, too many decomposed components will be generated, leading to high computational costs and over-decomposition.

[0006] After feature extraction, a suitable classifier is needed for pattern recognition, with BP and SVM being commonly used. BP is a classic and practical feedforward network that uses backpropagation to solve the problem of weights connecting multiple hidden units. However, its slow learning speed, numerous parameter settings, and extrema limit its application in fault diagnosis. SVM is a machine learning method with strong generalization ability, one of its advantages being its ability to solve classification problems with small samples and nonlinearity, making it widely used in pattern recognition. However, SVM requires more and more complex parameters to optimize and suffers from slow training speed. ELM is a single-hidden-layer feedforward neural network with fast learning speed and strong generalization ability, making it suitable for fault diagnosis. However, because the weights and thresholds in ELM are randomly set during computation, it can affect the stability and accuracy of fault diagnosis. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a fault diagnosis method and system based on adaptive weighted symplectic geometric mode decomposition. Based on symplectic geometric similarity transformation, an improved adaptive weighted and reconstructed symplectic geometric mode decomposition algorithm is proposed, which has good decomposition performance for vibration signals, can effectively extract rich fault features from the signals, and improves the accuracy of fault diagnosis.

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

[0009] The first aspect of this invention provides a fault diagnosis method based on adaptive weighted symplectic geometric mode decomposition.

[0010] A fault diagnosis method based on adaptive weighted symplectic geometric mode decomposition includes the following process:

[0011] Vibration signals of rolling bearings;

[0012] The vibration signal of the rolling bearing is decomposed using the AWSGMD-CP algorithm to obtain the WSCCs components. The power spectral entropy weighted singular values ​​are then extracted from the WSCCs components as the fault feature vector.

[0013] Based on the obtained fault feature vector and the optimized ACO-ELM model, the fault diagnosis result is obtained; the optimization of the ACO-ELM model includes: using the ACO algorithm to iteratively optimize the parameters in the ELM to obtain the optimal input weights and neuron thresholds.

[0014] As a further limitation of the first aspect of the present invention, the vibration signal of the rolling bearing is decomposed using the AWSGMD-CP algorithm to obtain WSCCs components, including:

[0015] After obtaining the initial single-component signal through SGMD decomposition, the number of iterations is limited by the cosine difference factor to distinguish the noise component; the Pearson correlation coefficient is used for component recombination, while ensuring periodic similarity and directional consistency; the variable entropy weighting matrix is ​​used to assign values ​​to the components, and the signal is denoised while retaining fault information to obtain WSCCs components.

[0016] As a further limitation of the first aspect of the invention, the initial single-component signal is obtained through SGMD decomposition, and a cosine difference factor is used to limit the number of iterations, including:

[0017] Construct a trajectory matrix based on the rolling bearing vibration signal;

[0018] Autocorrelation analysis is performed on the trajectory matrix to obtain the covariance matrix, and then the symplectic orthogonal matrix is ​​obtained.

[0019] The initial single-component reconstruction matrix is ​​obtained from the symplectic orthogonal matrix;

[0020] Diagonally average the elements in the initial single-component reconstruction matrix to obtain d sets of initial single components;

[0021] For the initial single component of group d, the sum of the components of the first k groups is calculated to obtain the superimposed component T. k ;

[0022] Calculate the cosine value between adjacent superimposed components, and construct the cosine difference factor based on the cosine value between adjacent superimposed components;

[0023] When the value of the cosine difference factor tends to be stable and less than the set threshold, the first c components are selected as the effective components, and the rolling bearing vibration signal x is decomposed into... in, Y is the sum of the noise components after the inflection point c. i Let i be the initial single component of the i-th group.

[0024] As a further limitation of the first aspect of the invention, the cosine difference factor S i ,include:

[0025] S i =|(C i+1 -C i )|<ε e

[0026] Among them, C i+1 Let C be the cosine value of the (i+1)th superposition component. i Let ε be the cosine value of the i-th superimposed component. e To set a threshold.

[0027] As a further limitation of the first aspect of the present invention, for the retained first c groups of effective components, the Pearson correlation coefficient is calculated for recombination, while ensuring the similarity of component periods and the consistency of directions, including:

[0028] Select the first initial single component Y1, and calculate the Pearson correlation coefficient with each of the remaining c-1 initial single components. Within the allowable error range, obtain the first symplectic combination component SCC1. Then, remove SCC1 from the original signal, select the first component of the removed signal, and compare it with the remaining components. This process is repeated to obtain h combination components, and x is decomposed into...

[0029] As a further limitation of the first aspect of the invention, assigning values ​​to the components using a changing entropy weighting matrix includes:

[0030] Calculate the absolute value of the slope between two adjacent points for each symplectic component;

[0031] Calculate the entropy of change for each symplectic component based on the absolute value of the slope;

[0032] Construct a weighted matrix using the value of the changing entropy;

[0033] Each symplectic combination component is weighted using a weighting matrix to obtain the weighted symplectic combination components WSCCs.

[0034] As a further limitation of the first aspect of the invention, the weighting matrix D i ,include:

[0035]

[0036] Among them, V max It is the maximum value of the entropy of change, V min It is the minimum value of the entropy of change;

[0037] The weighted symplectic combination components WSCCs are:

[0038]

[0039] As a further limitation of the first aspect of the present invention, extracting power spectral entropy weighted singular values ​​as fault feature vectors based on WSCCs components includes:

[0040] The WSCCs component is denoted as d. i (t), i = 1, 2, K, h;

[0041] For d i The row vector matrix composed of (t) is decomposed by SVD to obtain the singular value vector E. i ;

[0042] For di (t)=(d i1, d i2 ,K , d in The frequency domain signal X is obtained by Fourier transform. i (ω r ), thus obtaining the power spectrum S i (ω r ), and thus the power spectral entropy M is obtained. i ;

[0043] Using power spectral entropy M i The values ​​are used to construct the weighted matrix G. i ;

[0044] Using the weighting matrix G i For singular value eigenvectors E i After weighting, the final result is the power spectral entropy weighted singular value eigenvector T. i T i =G i ×E i .

[0045] As a further limitation of the first aspect of the invention, a weighting matrix G is constructed. i ,include:

[0046]

[0047] Among them, M max Power spectral entropy M i The maximum value in M min It is the power spectral entropy M i The minimum value in.

[0048] A second aspect of the present invention provides a fault diagnosis system based on adaptive weighted symplectic geometric mode decomposition.

[0049] A fault diagnosis system based on adaptive weighted symplectic geometric mode decomposition includes:

[0050] The data acquisition module is configured to capture the vibration signal of the rolling bearing.

[0051] The signal decomposition and feature extraction module is configured to: decompose the vibration signal of the rolling bearing using the AWSGMD-CP algorithm to obtain WSCCs components, and extract the power spectral entropy weighted singular values ​​as fault feature vectors based on the WSCCs components.

[0052] The fault diagnosis module is configured to obtain fault diagnosis results based on the obtained fault feature vector and the optimized ACO-ELM model. The optimization of the ACO-ELM model includes: using the ACO algorithm to iteratively optimize the parameters in the ELM to obtain the optimal input weights and neuron thresholds.

[0053] Compared with the prior art, the beneficial effects of the present invention are:

[0054] 1. Based on symplectic geometric similarity transformation, this invention proposes an improved adaptive weighted and reconstructed symplectic geometric mode decomposition algorithm, which has good decomposition performance for vibration signals and can effectively extract rich fault features from the signals.

[0055] 2. This invention utilizes the cosine difference factor to limit the number of iterations of the symplectic geometric mode decomposition algorithm, distinguishes noise components, and reduces feature dimensions.

[0056] 3. This invention uses the Pearson correlation coefficient as the component reconstruction evaluation standard to ensure periodic similarity and directional consistency, reconstructing mutually independent SCCs components. Based on the change entropy weighting matrix, noise is suppressed again while retaining fault information.

[0057] 4. This invention adopts the feature mapping method of power spectrum entropy weighted singular values, and fully combines the improved adaptive weighting and reconstructed symplectic geometric mode decomposition algorithm, which can effectively extract the fault features of the signal.

[0058] 5. This invention utilizes the ant colony optimization algorithm to iteratively optimize the input weights and thresholds in the extreme learning machine model, achieving excellent performance and outperforming traditional ELM, BP, and SVM models in terms of diagnostic effectiveness. Attached Figure Description

[0059] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0060] Figure 1 This is a flowchart illustrating the fault diagnosis method based on adaptive weighted symplectic geometric mode decomposition provided in Embodiment 1 of the present invention.

[0061] Figure 2 This is a schematic diagram of the signal decomposition and feature extraction process provided in Embodiment 1 of the present invention;

[0062] Figure 3 This is a flowchart illustrating the ACO-optimized ELM algorithm provided in Embodiment 1 of the present invention;

[0063] Figure 4 This is a schematic diagram of the first 6 SGCs components after SGMD decomposition provided in Embodiment 1 of the present invention;

[0064] Figure 5 This is a schematic diagram of the last 6 SGCs components after SGMD decomposition provided in Embodiment 1 of the present invention;

[0065] Figure 6 This is a schematic diagram of the four groups of WSCCs components after weighting and reconstructing the first six SGCs components provided in Embodiment 1 of the present invention. Detailed Implementation

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0067] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0068] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0069] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0070] Example 1:

[0071] like Figure 1 As shown, Embodiment 1 of the present invention provides a fault diagnosis method based on adaptive weighted symplectic geometric mode decomposition, comprising:

[0072] Step 1: Data acquisition and signal design, to obtain the original vibration signal of the rolling bearing;

[0073] Step 2: Signal decomposition and feature extraction. The AWSGMD-CP algorithm is used to decompose the original signal to obtain WSCCs components, and the power spectral entropy weighted singular values ​​are extracted as fault feature vectors.

[0074] Step 3: Model building and testing. Using the ACO algorithm, the parameters in ELM are iteratively optimized to obtain the optimal input weights and neuron thresholds. The optimized ACO-ELM model is then used to test the fault diagnosis task.

[0075] The construction process for signal decomposition and feature extraction in step 2 is as follows:

[0076] (1) Signal decomposition

[0077] The AWSGMD-CP theory is mainly based on the SGMD algorithm. After obtaining the initial single-component signals through SGMD decomposition, Pearson correlation coefficients are used for component recombination while ensuring cycle similarity and direction consistency. A large amount of noise components are contained in the initial single components and need to be distinguished. The cosine difference factor is used to limit the number of iterations to effectively distinguish the noise components. Finally, a variable entropy weighted matrix is used to assign values to the components, denoising the signals while retaining the fault information to construct the WSCCs components.

[0078] ①, Phase space reconstruction

[0079] Assume an original time series signal (x = x1, x2, L, x n ), where n represents the length of the signal, and construct the trajectory matrix X as follows:

[0080]

[0081] where d is the embedding dimension, τ is the delay time, and m = n - (d - 1)τ. The value of d is adaptively determined by calculating the power spectral density PSD of the original time series.

[0082] ②, QR decomposition

[0083] To obtain the Hamiltonian matrix, perform autocorrelation analysis on the trajectory matrix X to obtain the covariance matrix A = X T X, and the symplectic orthogonal matrix Q can be expressed as follows:

[0084]

[0085]

[0086] where W is the Hamiltonian matrix, and the matrix Q can be obtained by symplectic similarity transformation SST.

[0087] ③, Obtain the reconstruction matrix Z

[0088]

[0089] ④, Diagonal averaging

[0090] For the elements z i in the matrix Z ij (1 ≤ i ≤ m, 1 ≤ j ≤ d), let d * = min(m, d), m * = max(m, d), m = n - (d - 1)τ. If m < d, let z ij * = z ij , otherwise z ij* =z ji .

[0091] The corresponding time series Y i element y in k (k = 1, 2, ..., n), the calculation is as follows:

[0092]

[0093] Initial single-component reconstruction matrix Z i Transformed into a one-dimensional time series Y i The trajectory matrix X can be transformed into d sequences of length n using diagonal averaging: Y = Y1 + Y2 + ... + Y... d The sum of these sequences is x.

[0094] ⑤ Noise separation

[0095] Summing the components, for the initial single components (Y1, Y2, L, Y) of group d. d ), calculate the sum of the first k components to obtain the superimposed component T. k :

[0096]

[0097] Calculate the cosine value between adjacent superimposed components, and use the following formula to calculate the cosine value C of the superimposed components. i :

[0098]

[0099] By constructing a difference factor and considering the characteristics of cosine values, the difference operation can better highlight the turning point where the superimposed components tend to stabilize. The cosine difference factor S is constructed using the following formula. i :

[0100] S i =|(C i+1 -C i )|<ε e (8)

[0101] When S i The value tends to be stable and less than the threshold ε e When x approaches 0, the first c components can be selected as effective components. The part before point c contains a large amount of useful information from the original signal, while the part after point c can be considered as noise components. x is decomposed into... in, It is the sum of noise components after the inflection point c.

[0102] ⑥ Component Reorganization

[0103] The Pearson Correlation Coefficient (PCC), proposed by Carl Pearson, is a way to measure the similarity between two variables, where x = [x1, x2, ..., x...]. n ]、y=[y1,y2,L,y n The degree of linear correlation r between them xy The calculation is as follows:

[0104]

[0105] in, and x i and y i The average value, r xy For ∈[-1,1], the larger the absolute value of the correlation coefficient, the stronger the correlation, and vice versa.

[0106] For the retained first c groups of effective components, PCC is calculated for recombination, while ensuring component period similarity and directional consistency. Specifically: First, the first initial single component Y1 is selected, and PCC is calculated separately for it and the remaining c-1 initial single components. Within the allowable error range, the first symplectic combination component SCC1 is obtained. Then, SCC1 is removed from the original signal, and the first component of the removed signal is compared with the remaining components to obtain h combination components sequentially. x is decomposed into...

[0107] ⑦ Component reconstruction

[0108] To perform signal denoising while preserving fault information, a variable entropy weighting matrix is ​​used to assign values ​​to each SCCs, constructing weighted symplectic combination components (WSCCs). The amount of fault information contained in each component is then evaluated. A higher entropy value indicates more fault information in the component, and vice versa. The specific process is as follows:

[0109] To determine the rate of change, calculate each SCC using the following formula. i (i = 1, 2, K, h) The absolute value of the slope f between two adjacent points ij (j = 1, 2, ..., n-1):

[0110]

[0111] Where i represents the number of SCCs, n represents the number of sampling points contained in each SCC, and SCC i,j f represents the value of the i-th SCC component at the j-th sampling point. s The sampling frequency;

[0112] To calculate the entropy of change, combining the theories of rate of change and information entropy, the entropy V of each SCC is calculated using the following formula. i :

[0113]

[0114] Find the weighted matrix using the change entropy V i The value is used to construct the weighting matrix D using the following formula. i :

[0115]

[0116] Among them, V max It is the entropy of change V i The maximum value in, V min It is the entropy of change V i The minimum value in.

[0117] Reconstructing components using the weighting matrix D i Each SCC is weighted to obtain the weighted symplectic combination components WSCCs:

[0118] WSCC i =D i ×SCC i (13)

[0119]

[0120] (2) Feature extraction

[0121] After the signal is decomposed and reconstructed by the AWSGMD-CP algorithm, many WSCCs components are obtained, which contain a large amount of fault-related information. A suitable strategy is needed to extract fault features from the decomposed components. Singular value decomposition can extract these fault features well, accurately reveal the characteristic information of the signal, and has low computational complexity, making it suitable for feature extraction of decomposed components. At the same time, the power spectral entropy theory, which can reflect the complexity of the signal, is used to weight the singular value eigenvectors.

[0122] The specific steps are as follows:

[0123] ① Perform AWSGMD-CP decomposition on the vibration signal to obtain the WSCC component, denoted as d. i (t), i = 1, 2, K, h;

[0124] ② Regarding d i The row vector matrix composed of (t) is decomposed by SVD to obtain the singular value matrix E. i ;

[0125] E i =SVD([d1(t),d2(t),K,dh (t)] T (15)

[0126] ③ Regarding d i (t)=(d i1 ,d i2 ,K , d in The frequency domain signal X is obtained by Fourier transform. i (ω r ), ω r If it is a spectrum, then the power spectrum S i (ω r )for:

[0127]

[0128] Energy is conserved during the transformation from the time domain to the frequency domain, which can be written as:

[0129]

[0130] S i (ω r ) can be seen as d i (t) Energy distribution in the frequency domain, power spectral entropy M i Defined as:

[0131]

[0132] ④ Using the power spectral entropy value, construct the weighting matrix G using the following formula. i :

[0133]

[0134] Among them, M max It is the power spectral entropy M i The maximum value in M min It is the power spectral entropy M i The minimum value in.

[0135] ⑤ Using the weighting matrix G i The singular value eigenvector E is expressed by the following formula. i After weighting, the final result is the power spectral entropy weighted singular value eigenvector T. i :

[0136] T i =G i ×E i (20)

[0137] The complete signal decomposition and feature extraction process is as follows: Figure 2 As shown.

[0138] The specific process of model building and testing in step 3 is as follows:

[0139] Traditional ELM (Elastic Computational Model) randomly selects input weights and thresholds, which reduces the model's prediction accuracy. Therefore, appropriate algorithms need to be incorporated for optimization to avoid the impact of random selection on diagnostic results. The ACO (Aggressive Search) algorithm is a probabilistic search algorithm for intelligent populations, characterized by strong robustness, simple implementation, and wide applicability.

[0140] Based on the powerful global search capability of the ACO algorithm, the input weights and thresholds of ELM are used as each ant in the ACO algorithm to plan the optimal path and obtain the global optimal solution, thereby improving the accuracy and generalization ability of the classification model.

[0141] This invention uses the ACO algorithm to optimize the ELM model and constructs an ACO-ELM model to achieve better classification results. The process is as follows: Figure 3 As shown, the specific process is as follows:

[0142] ① Initialize parameters and calculate the probability of an ant moving between two points as follows:

[0143]

[0144] In the formula, k represents an ant; Let τ be the probability that the ant moves between the two points at time t; ij (t) represents the amount of information remaining at time t; η ij (t) is the inspiration function, described by the reciprocal of the Euclidean distance; r k Let α be the position that ant k can approach; α be the pheromone adjustment parameter; and β be the heuristic function adjustment parameter.

[0145] ② By updating generation by generation, the optimal foraging route is found, and the global optimal route of the ant colony is obtained, which is the input weight and threshold of the extreme learning machine.

[0146] ③ Based on the ant colony algorithm for finding the optimal path, the activation function of the hidden layer is appropriately selected to obtain the output matrix H of the hidden layer.

[0147] ④ Solve for the weight matrix of the output layer to perform classification and recognition.

[0148] In this embodiment, a CWRU rolling bearing fault simulation platform is built. The device includes a 2hp motor, a torque sensor, a power meter, and electronic control equipment. Electrical discharge machining technology is used to simulate motor bearing faults. The bearings are located at the motor drive end and the fan end. The sampling frequency is 12KHz. This embodiment sets 11 fault types, including 1 normal state, 5 drive end faults, and 5 fan end faults. Each fault includes 1024 sampling points. 80 sets of data are selected and assigned to the training set and the test set. The detailed dataset description is shown in Table 1.

[0149] Table 1: Detailed Description of Rolling Bearing Dataset

[0150]

[0151]

[0152] To verify the superiority of the proposed rolling bearing fault identification system based on adaptive weighted and reconstructed symplectic geometric mode decomposition, experiments were conducted on feature extraction and fault classification to set up comparison tasks.

[0153] (1) Feature extraction

[0154] To evaluate the advantages of the overall feature extraction method, the improved AWSGMD-CP was combined with power spectral entropy weighted singular value feature extraction and compared with traditional mechanical vibration signal decomposition methods EEMD, LMD, and VMD. The traditional SGMD method was also compared. Figure 4 This shows the first 6 SGCs of a set of vibration signals under normal conditions. Figure 5 The last six SGCs are shown. These last six SGCs have smaller amplitudes and represent a smaller proportion of the energy. From... Figure 4 As can be seen, components 1 and 2, and components 5 and 6 are very similar and can be recombinated. Their weighted component WSCCs are as follows: Figure 6 As shown in Table 2, to quantitatively describe the performance of different decomposition algorithms, a unified ACO-ELM classification model was used for fault diagnosis. Ten experiments were conducted, and the average value was taken as the experimental results.

[0155] Table 2: Experimental results using different feature extraction algorithms

[0156]

[0157] The improved AWSGMD-CP has a higher diagnostic accuracy than other methods, mainly because all SGCs components not only retain useful components but also eliminate useless components during the recombination, constraint iteration and weighting process, thus constructing mutually independent components while reducing the feature dimension.

[0158] (2) Fault Classification

[0159] To verify the advantages of the ACO-optimized ELM classification model, it was compared with the traditional ELM model, and several common classification models such as BP and SVM were also compared. The purpose was to evaluate the performance of the classifier and optimization algorithm. The experimental results are shown in Table 3.

[0160] Table 3: Experimental results using different classification algorithms

[0161]

[0162] As can be seen from the classification results in Table 3, the classification accuracy of the ELM model is higher than that of the other methods, and the diagnostic performance of the model after ELM parameter optimization by ACO is more outstanding. ACO achieves better classification accuracy and stronger generalization ability by iteratively searching for the optimal input weights and thresholds in ELM.

[0163] Example 2:

[0164] Embodiment 2 of the present invention provides a fault diagnosis system based on adaptive weighted symplectic geometric mode decomposition, comprising:

[0165] The data acquisition module is configured to capture the vibration signal of the rolling bearing.

[0166] The signal decomposition and feature extraction module is configured to: decompose the vibration signal of the rolling bearing using the AWSGMD-CP algorithm to obtain WSCCs components, and extract the power spectral entropy weighted singular values ​​as fault feature vectors based on the WSCCs components.

[0167] The fault diagnosis module is configured to obtain fault diagnosis results based on the obtained fault feature vector and the optimized ACO-ELM model. The optimization of the ACO-ELM model includes: using the ACO algorithm to iteratively optimize the parameters in the ELM to obtain the optimal input weights and neuron thresholds.

[0168] The working method of the system is the same as the fault diagnosis method based on adaptive weighted symplectic geometric mode decomposition provided in Example 1, and will not be repeated here.

[0169] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0170] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0171] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0172] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0173] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0174] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition, characterized in that, The method comprises the following processes: a vibration signal of a rolling bearing; the vibration signal of the rolling bearing is decomposed by using an AWSGMD-CP algorithm to obtain WSCCs components, which comprise: after obtaining the initial single-component signal through SGMD decomposition, a cosine difference factor is used to limit the number of iterations to distinguish noise components; a Pearson correlation coefficient is used for component recombination while ensuring periodic similarity and directional consistency; a change entropy weighted matrix is used to assign values to components to retain fault information and perform signal denoising to obtain the WSCCs components; the change entropy weighted matrix is used to assign values to components, which comprises: calculating the absolute value of the slope between adjacent points of each symplectic combined component; according to the absolute value of the slope, the change entropy of each symplectic combined component is calculated; a weighted matrix is constructed using the value of the change entropy; each symplectic combined component is weighted using the weighted matrix to finally obtain a weighted symplectic combined component WSCCs; Weighting matrix comprising: wherein is the maximum value in the change entropy, is the minimum value in the change entropy; the weighted symplectic combined component WSCCs is: ， ； a power spectrum entropy weighted singular value is extracted from the WSCCs component as a fault feature vector; based on the obtained fault feature vector and an optimized ACO-ELM model, a fault diagnosis result is obtained; wherein the optimization of the ACO-ELM model comprises: using an ACO algorithm to iteratively optimize the parameters in the ELM to obtain optimal input weights and neuron thresholds.

2. The fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition according to claim 1, wherein the initial single-component signal is obtained through SGMD decomposition, and a cosine difference factor is used to limit the number of iterations, which comprises: a trajectory matrix is constructed according to the vibration signal of the rolling bearing; autocorrelation analysis is performed on the trajectory matrix to obtain a covariance matrix, and then a symplectic orthogonal matrix is obtained; an initial single-component reconstruction matrix is obtained according to the symplectic orthogonal matrix; Diagonal averaging of the elements in the initial single-component reconstruction matrix yields group initial single component; For group initial single component, pre-computation sum of components to obtain superposition component ; the cosine value between adjacent superimposed components is calculated, and a cosine difference factor is constructed according to the cosine value between adjacent superimposed components; When the cosine difference factor tends to be stable and less than a set threshold, the front component as an effective component, the rolling bearing vibration signal is decomposed into , wherein is the turning point , the sum of noise components after is the first i group of initial single components.

3. The fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition according to claim 2, wherein Cosine difference factor , comprising: wherein, is the cosine value of the i+1th superimposed component, is the cosine value of the i+1th superimposed component, is a set threshold value.

4. The fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition according to claim 2, wherein For the reserved front group effective components, the Pearson correlation coefficient is calculated for reorganization while ensuring component period similarity and direction consistency, including: Select the first initial single component , and respectively calculate the Pearson correlation coefficient with the remaining initial single components, and obtain the first combined component SCC1 within the error tolerance, then remove SCC1 from the original signal, compare the first component of the removed signal with the remaining components, and sequentially obtain combined components, decomposed into .

5. The fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition according to claim 1, wherein a power spectrum entropy weighted singular value is extracted from the WSCCs component as a fault feature vector, which comprises: The WSCCs components are denoted by ; To SVD decomposition is performed on the row vector matrix composed of the singular value vector ; To Fourier transform to get frequency domain signal , get power spectrum , and then get power spectrum entropy ; Utilizing power spectral entropy of values to construct a weighting matrix ; Utilizing a weighting matrix , the singular value eigenvector is weighted, and a power spectrum entropy weighted singular value eigenvector is finally obtained , .

6. The fault diagnosis method based on adaptive weighted symplectic geometric modal decomposition according to claim 5, wherein Constructing a weighting matrix comprising: wherein is the maximum value of the power spectrum entropy in the frequency range is the minimum value of the power spectrum entropy in the frequency range 7. A fault diagnosis system based on adaptive weighted syngeometric modal decomposition, characterized in that, comprises: a data acquisition module configured to obtain a vibration signal of a rolling bearing; a signal decomposition and feature extraction module configured to decompose the vibration signal of the rolling bearing by using an AWSGMD-CP algorithm to obtain WSCCs components, which comprise: after obtaining the initial single-component signal through SGMD decomposition, a cosine difference factor is used to limit the number of iterations to distinguish noise components; a Pearson correlation coefficient is used for component recombination while ensuring periodic similarity and directional consistency; a change entropy weighted matrix is used to assign values to components to retain fault information and perform signal denoising to obtain the WSCCs components; The component is valued by using a change entropy weighted matrix, including: Calculating the absolute value of the slope between the adjacent two points of each synch combination component; According to the absolute value of the slope, the change entropy of each synch combination component is calculated; Using the value of the change entropy, a weighted matrix is constructed; Each synch combination component is weighted by using the weighted matrix, and finally a weighted synch combination component WSCCs is obtained; weighting matrix comprising: wherein is the maximum value in the change entropy, is the minimum value in the change entropy; The weighted synch combination component WSCCs is: ， ； According to the WSCCs component, the power spectrum entropy weighted singular value is extracted as a fault feature vector; The fault diagnosis module is configured to obtain a fault diagnosis result based on the obtained fault feature vector and the optimized ACO-ELM model, wherein the optimization of the ACO-ELM model includes: using the ACO algorithm to iteratively optimize the parameters in the ELM to obtain the optimal input weight and neuron threshold.

Images