Rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine

Abstract

The invention relates to a rolling bearing fault diagnosis method based on an improved composite multi-scale fuzzy entropy and an optimized support vector machine, and the method comprises the steps: (1) decomposing a vibration signal of a rolling bearing through employing MODWPT, and obtaining a plurality of signal components with a fixed bandwidth; (2) within a set scale factor range, constructing a coarse-grained sequence based on a local mean value for each signal component, and calculating an improved composite fuzzy entropy sequence of all signal components; (3) splicing a plurality of sequences into a feature vector according to a sequence of signal component center frequencies from low to high; (4) carrying out dimensionality reduction on the feature vectors by adopting a principal component analysis method, and constructing a rolling bearing fault diagnosis model training set and a test set; and (5) establishing a fault identification model by using a support vector machine, optimizing hyper-parameters through a whale algorithm, and inputting a test set to realize fault diagnosis of the rolling bearing. A new solution is provided for rolling bearing fault diagnosis, and the diagnosis accuracy is improved.

Description

Technical Field

[0001] This invention belongs to the field of fault diagnosis technology, specifically relating to a rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine. Background Technology

[0002] Rolling bearings are core components in rotating machinery, providing support for transmission shafts, and their health directly affects the smoothness of equipment operation. Statistics show that approximately 30% of rotating machinery failures are caused by bearings. Fault analysis based on the statistical characteristics of vibration signals is a common method for diagnosing bearing faults. However, due to the influence of noise from other equipment in the industrial environment, effective fault information in the time and frequency domain characteristics of vibration signals is masked. Rolling bearing vibration signals exhibit non-stationary and nonlinear characteristics, making linear analysis in the time and frequency domains unable to capture fault features, thus complicating the diagnosis of rolling bearing faults.

[0003] In recent years, nonlinear analysis methods have been introduced into the extraction of rolling bearing fault feature information, and nonlinear parameter identification algorithms based on entropy theory have been widely used in fault diagnosis. Entropy features such as sample entropy, permutation entropy, and fuzzy entropy are used to extract rolling bearing fault features. However, these methods only measure the complexity of signal components at a single time series scale, resulting in reduced diagnostic accuracy. Multi-scale fuzzy entropy is a generalization of fuzzy entropy across multiple time scales. By calculating the coarse-grained fuzzy entropy of each element in the original signal, it measures the complexity and effectively reduces the interference of structural noise.

[0004] In fault state identification, Support Vector Machines (SVMs) are a commonly used machine learning classification algorithm. They can extract key state information from high-dimensional data samples to complete the identification process. The goal is to find an optimal decision boundary that can maximally separate data from different categories. However, SVMs have many hyperparameters that significantly impact model performance, and the prediction effect depends heavily on the selection of these hyperparameters, requiring further optimization.

[0005] Therefore, this invention proposes a rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine. The method uses MODWPT to decompose the rolling bearing vibration signal, extracts the fault features of each frequency band component through improved composite multi-scale fuzzy entropy, uses principal component analysis for dimensionality reduction, and uses the whale algorithm to optimize the penalty factor and kernel function parameters of the support vector machine model to obtain more reliable diagnostic results.

[0007] The technical problem solved by this invention is achieved through the following technical solution: A method for diagnosing rolling bearing faults based on improved composite multi-scale fuzzy entropy and optimized support vector machine, the method comprising the following steps: S1. The maximum overlap discrete wavelet transform (MODWPT) is used to decompose the vibration signal of the rolling bearing under different states to obtain several signal components with fixed bandwidth. S2. Within a defined scaling factor range, construct a coarse-grained sequence based on local mean for each signal component, and calculate the performance of all coarse-grained sequences. m dimensional reconstruction vector and m The sum of the similarities of the +1 dimensional reconstructed vectors; keeping the lower bound of the scale factor interval at 1, and increasing the upper bound by a fixed step size, the similarity within each interval is calculated separately. m dimension, m The mean of the sum of the similarities of the +1 dimensional reconstructed vectors is used to obtain the improved composite fuzzy entropy sequence matrix of all signal components. S3. Solve for the improved composite multi-scale fuzzy entropy sequence matrix of all signals, and concatenate multiple sequences into a feature vector according to the order of the center frequency of the signal components from low to high. S4. Principal component analysis is used to reduce the dimensionality of the eigenvectors. Principal components with a cumulative contribution rate exceeding a set threshold are selected to construct the training and testing sets for the rolling bearing fault diagnosis model. S5. Use the Support Vector Machine algorithm to build a fault identification model, and use the whale optimization algorithm to search for the optimal penalty factor. C With kernel function γ The accuracy of the training set is used as the fitness function; the support vector machine state recognition model with optimized input parameters of the test set samples is used to classify rolling bearing faults.

[0008] Moreover, S1 specifically refers to: Given the number of decomposition levels, scale the filter bank of the Discrete Wavelet Transform (DWPT). Wavelet filter bank Adjusted to , ; Set the 0th level wavelet coefficients Using a low-pass filter bank The first-level nodal wavelet coefficients are obtained by cyclic filtering of the signal. Similarly, high-pass filter banks right X Cyclic filtering yields the coefficients of nodes at the same level in a binary tree. In the subsequent i-level Insert 2 between each member j-1 -1 zero; The wavelet coefficients of the previous stage Cyclic filtering yields the first i wavelet coefficients of level : ; Among them, if ,like .

[0009] Furthermore, the S2 improved composite fuzzy entropy sequence matrix is ​​specifically as follows: Set scale factor Embedded Dimension m Fuzzy Index n Similarity tolerance r The signal is processed in a coarse-grained manner to obtain group sequence ,in, Each element in the sequence is defined as follows: ; Calculate the scale factor All coarse-grained sequences of m Wei and m The sum of similarities of +1 dimension reconstructed vectors , ,Pick and mean ; The improved composite multiscale fuzzy entropy for calculating signal components is expressed as follows: ; Where X is a time series.

[0010] Moreover, in the improved composite fuzzy entropy m Sum of similarity of reconstructed vectors The calculation process is as follows: Set the embedding dimension to m , with a length of N signal Reconstructed m dimensional vector group The member expressions in the vector group are: ; In the formula, ; Calculate vectors and Maximum difference of elements : ; Using the exponential function as the ambiguity function Calculate vector and similarity : ; Refactor all m The similarity of the vectors in dimension is added together to obtain m Sum of similarity of reconstructed vectors : .

[0011] Furthermore, the specific steps of the S5 whale optimization algorithm are as follows: Set the fitness function and the number of individuals to search for prey. N Randomly generate the initial positions of all individuals. X ; The searching individual simultaneously surrounds and spirals towards its prey. Assuming the probability of the searching individual executing a hunting strategy in a single search is 50%, its position can be described by the following mathematical expression: ; in, t This represents the current iteration number. b For the spiral shape parameters, l are random numbers and , Search for the individual location vector within the current population. This represents the individual position corresponding to the current optimal solution. For the coefficient vector, This represents the distance between the individual's current location and the target. When surrounding prey, the searched individual's position is randomly updated, which can be described as: ; in, This is a vector representing the random location of an individual within the current population. When approaching prey, the best solutions obtained by all individuals are considered as candidates for the best solution in the search process, and the search individuals move closer to the optimal position. This behavior can be described by the following mathematical expression: ; The searching individual gradually approaches its prey using a spiral motion, mathematically expressed as: .

[0012] The advantages and beneficial effects of this invention are as follows: 1. This invention uses MODWPT to decompose the signal into signal components of equal bandwidth, thereby achieving decoupling of complex signals and making the effective components of the constructed feature vector more significant. 2. This invention introduces an improved composite multi-scale fuzzy entropy, which extends the fuzzy entropy to multiple time scales, extracts nonlinear fault features of vibration signals at multiple time scales, and enhances the robustness of the entropy features by calculating the mean of the similarity within the scale factor range, thus realizing the extraction of fault information from low signal-to-noise ratio signals.

[0013] 3. This invention uses the whale algorithm to optimize the support vector machine state recognition model, which solves the problem that its hyperparameters are too dependent on manual selection and improves the predictive ability of the fault diagnosis algorithm. Attached Figure Description

[0014] Figure 1 This is a flowchart of the present invention; Figure 2 This is a schematic diagram of the time-domain waveforms of all signal components after MODWPT decomposition of a sample signal in the outer ring of the present invention. Figure 3 This is an entropy sequence diagram of bearing MODWPT signal component C2 under five health states according to the present invention; Figure 4 This is a graph showing the number of optimization iterations versus prediction accuracy for the Whale Algorithm used to optimize the Support Vector Machine in this invention. Figure 5 The confusion matrix diagram is the result of inputting the test set of this invention into the optimized support vector machine model. Detailed Implementation

[0015] The present invention will be further described in detail below through specific embodiments. The following embodiments are merely descriptive and not limiting, and should not be used to limit the scope of protection of the present invention.

[0016] A method for diagnosing rolling bearing faults based on improved composite multi-scale fuzzy entropy and optimized support vector machine, the method comprising the following steps: S1. The maximum overlap discrete wavelet transform (MODWPT) is used to decompose the vibration signal of the rolling bearing under different states to obtain several signal components with fixed bandwidth. Given the number of decomposition levels, scale the filter bank of the Discrete Wavelet Transform (DWPT). Wavelet filter bank Adjusted to , ; Set the 0th level wavelet coefficients Using a low-pass filter bank The first-level nodal wavelet coefficients are obtained by cyclic filtering of the signal. Similarly, high-pass filter banks right X Cyclic filtering yields the coefficients of nodes at the same level in a binary tree. In the subsequent i-level Insert 2 between each member j-1 -1 zero; The wavelet coefficients of the previous stage Cyclic filtering yields the first i wavelet coefficients of level : ; Among them, if ,like .

[0017] S2. Within a defined scaling factor range, construct a coarse-grained sequence based on local mean for each signal component, and calculate the performance of all coarse-grained sequences. m dimensional reconstruction vector and m The sum of the similarities of the +1 dimensional reconstructed vectors; keeping the lower bound of the scale factor interval at 1, and increasing the upper bound by a fixed step size, the similarity within each interval is calculated separately. m dimension, m The mean of the sum of the similarities of the +1 dimensional reconstructed vectors is used to obtain the improved composite fuzzy entropy sequence matrix of all signal components. Set scale factor Embedded Dimension m Fuzzy Index n Similarity tolerance r The signal is processed in a coarse-grained manner to obtain group sequence ,in, Each element in the sequence is defined as follows: ; Calculate the scale factor All coarse-grained sequences of m Wei and m The sum of similarities of +1 dimension reconstructed vectors , ,Pick and mean ; The improved composite multiscale fuzzy entropy for calculating signal components is expressed as follows: ; Where X is a time series.

[0018] Moreover, in the improved composite fuzzy entropy m Sum of similarity of reconstructed vectors The calculation process is as follows: Set the embedding dimension to m , with a length of N signal Reconstructed mdimensional vector group The member expressions in the vector group are: ; In the formula, ; Calculate vectors and Maximum difference of elements : ; Using the exponential function as the ambiguity function Calculate vector and similarity : ; Refactor all m The similarity of the vectors in dimension is added together to obtain m Sum of similarity of reconstructed vectors : .

[0019] S3. Solve for the improved composite multi-scale fuzzy entropy sequence matrix of all signals, and concatenate multiple sequences into a feature vector according to the order of the center frequency of the signal components from low to high. S4. Principal component analysis is used to reduce the dimensionality of the eigenvectors. Principal components with a cumulative contribution rate exceeding a set threshold are selected to construct the training and testing sets for the rolling bearing fault diagnosis model. S5. Use the Support Vector Machine algorithm to build a fault identification model, and use the whale optimization algorithm to search for the optimal penalty factor. C With kernel function γ The training set accuracy is used as the fitness function; the support vector machine state recognition model with optimized input parameters of the test set samples is used to classify rolling bearing faults. The specific steps of the whale optimization algorithm are as follows: Set the fitness function and the number of individuals to search for prey. N Randomly generate the initial positions of all individuals. X ; The searching individual simultaneously surrounds and spirals towards its prey. Assuming the probability of the searching individual executing a hunting strategy in a single search is 50%, its position can be described by the following mathematical expression: ; in, t This represents the current iteration number. b For the spiral shape parameters, l are random numbers and , Search for the individual location vector within the current population. This represents the individual position corresponding to the current optimal solution. For the coefficient vector, This represents the distance between the individual's current location and the target. When surrounding prey, the searched individual's position is randomly updated, which can be described as: ; in, This is a vector representing the random location of an individual within the current population. When approaching prey, the best solutions obtained by all individuals are considered as candidates for the best solution in the search process, and the search individuals move closer to the optimal position. This behavior can be described by the following mathematical expression: ; The searching individual gradually approaches its prey using a spiral motion, mathematically expressed as: .

[0020] This example uses the publicly available Xi'an Jiaotong University Accelerated Life Test Bearing Dataset (XJTU-SY) to verify the rationality of the method of this invention. The rolling bearing used in the test is LDK UER204, the signal sampling frequency is 25.6kHz, the sampling interval is 1min, and the sampling time is 1.28s. Three operating conditions were set up in the test, each containing full-life vibration acceleration data for five bearings. All bearing failures occurred during the actual accelerated test, and the degree of failure gradually developed. The data used in this example comes from bearings 1-1 and 1-3 with outer ring failures, and bearings 2-1 and 3-3 with inner ring failures. Based on the root mean square of the bearing's full-life signal, the degree of failure is divided into degradation and failure. Normal samples, outer ring degradation samples, outer ring failure samples, inner ring degradation samples, and inner ring failure samples from the bearing's full-life signal set are selected for fault diagnosis. Each signal sample is 2560 bytes long. The test dataset is shown in Table 1.

[0021] Table 1 Experimental Dataset

[0022] In this embodiment, the MODWPT decomposition level is set to 3, the decomposition wavelet is 'db12', and a total of 8 signal components are obtained. The time-domain waveform of the signal component of a certain sample in the outer ring degenerates as follows. Figure 2 As shown.

[0023] The improved composite multiscale entropy of the signal components was calculated. With a scale factor of 10, an embedding dimension of 2, a fuzzy exponent of 2, and a similarity tolerance of 0.15, entropy sequences of 8 signal components were obtained. The entropy sequences of the MODWPT component C2 of the bearing signal samples for 5 health states are shown below. Figure 3As shown, the signal complexity varies under different fault states. Therefore, the proposed improved composite multi-scale entropy can effectively distinguish fault states. An entropy sequence is constructed by concatenating the signal component center frequencies from smallest to largest, resulting in a sample feature vector with a dimension of 80. Principal component analysis is used to reduce the dimensionality of the feature vector, and the feature vectors with a cumulative contribution value exceeding 98% are used as the projection matrix. The dimensionality of the reduced feature vector is 10.

[0024] A fault diagnosis model is built using the training set. The whale optimization algorithm is set to search for 50 prey individuals, with 20 iterations. The search parameters are the support vector machine penalty coefficient C and kernel function parameter γ. The search dimension is 2, with a lower bound of [0.001 0.001] and an upper bound of [1000 1000]. The fitness function is the prediction accuracy of 40-fold cross-validation using support vector machines. The optimization algorithm iteration steps minus the accuracy is shown in the figure. Figure 4 As shown, the accuracy rate reached 93.125%.

[0025] Diagnostic results as follows Figure 5 As shown, the horizontal axis represents the original label of the fault type, and the vertical axis represents the label classified after model diagnosis. It can be seen that the misdiagnosis rate of fault type is extremely low. There is a small error in the judgment of fault degree, which is related to the data label. The fault data used is not obtained by manually injecting quantifiable local defects, and its damage degree at a specific moment is unobservable. In summary, the method proposed in this invention can realize fault diagnosis in the actual operation of bearings.

[0026] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A rolling bearing fault diagnosis method based on improved composite multiscale fuzzy entropy and optimized support vector machine, characterized in that: The steps of the method are as follows: S1. The maximum overlap discrete wavelet transform (MODWPT) is used to decompose the vibration signal of the rolling bearing under different states to obtain several signal components with fixed bandwidth. S2, constructing a local mean based coarse-grained sequence for each signal component over a set scale factor range, computing a sum of all coarse-grained sequence m reconstruction vectors and m +1 dimensional reconstruction vectors similarity sum Keeping the lower bound of the scaling factor interval at 1, and increasing the upper bound by a fixed step size, calculate the scaling factor interval for each interval separately. m dimension, m The mean of the sum of the similarities of the +1 dimensional reconstructed vectors is used to obtain the improved composite fuzzy entropy sequence matrix of all signal components. S3. Solve for the improved composite multi-scale fuzzy entropy sequence matrix of all signals, and concatenate multiple sequences into a feature vector according to the order of the center frequency of the signal components from low to high. S4. Principal component analysis is used to reduce the dimensionality of the eigenvectors. Principal components with a cumulative contribution rate exceeding a set threshold are selected to construct the training and testing sets for the rolling bearing fault diagnosis model. S5. Use the Support Vector Machine algorithm to build a fault identification model, and use the whale optimization algorithm to search for the optimal penalty factor. C With kernel function γ The training set accuracy is used as the fitness function. The support vector machine state recognition model with optimized input parameters from the test set samples is used to classify rolling bearing faults.

2. The rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine according to claim 1, characterized in that: Specifically, S1 is: Given the number of decomposition levels, scale the filter bank of the Discrete Wavelet Transform (DWPT). Wavelet filter bank Adjusted to , ; Set the 0th level wavelet coefficients Using a low-pass filter bank The first-level nodal wavelet coefficients are obtained by cyclic filtering of the signal. Similarly, high-pass filter banks right X Cyclic filtering yields the coefficients of nodes at the same level in a binary tree. In the subsequent i-level Insert 2 between each member j-1 -1 zero; The wavelet coefficients of the previous stage Cyclic filtering yields the first i wavelet coefficients of level : ； Among them, if ,like .

3. The rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine according to claim 1, characterized in that: The S2 improved composite fuzzy entropy sequence matrix is ​​specifically as follows: Set scale factor Embedded Dimension m Fuzzy Index n Similarity tolerance r The signal is processed in a coarse-grained manner to obtain group sequence ,in, Each element in the sequence is defined as follows: ； Calculate the scale factor All coarse-grained sequences of m Wei and m The sum of similarities of +1 dimension reconstructed vectors , ,Pick and mean ; The improved composite multiscale fuzzy entropy for calculating signal components is expressed as follows: ； Where X is a time series.

4. The rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine according to claim 3, characterized in that: The improved composite fuzzy entropy m Sum of similarity of reconstructed vectors The calculation process is as follows: Set the embedding dimension to m , with a length of N signal Reconstructed m dimensional vector group The member expressions in the vector group are: ； In the formula, ; Calculate vectors and Maximum difference of elements : ； Using the exponential function as the ambiguity function Calculate vector and similarity : ； Refactor all m The similarity of the vectors in dimension is added together to obtain m Sum of similarity of reconstructed vectors : 。 5. The rolling bearing fault diagnosis method based on improved composite multi-scale fuzzy entropy and optimized support vector machine according to claim 1, characterized in that: The specific steps of the S5 whale optimization algorithm are as follows: Set the fitness function and the number of individuals to search for prey. N Randomly generate the initial positions of all individuals. X ; The searching individual simultaneously surrounds and spirals towards its prey. Assuming the probability of the searching individual executing a hunting strategy in a single search is 50%, its position can be described by the following mathematical expression: ； in, t This represents the current iteration number. b For the spiral shape parameters, l are random numbers and , Search for the individual location vector within the current population. This represents the individual position corresponding to the current optimal solution. For the coefficient vector, This represents the distance between the individual's current location and the target. When surrounding prey, the searched individual's position is randomly updated, which can be described as: ； in, This is a vector representing the random location of an individual within the current population. When approaching prey, the best solutions obtained by all individuals are considered as candidates for the best solution in the search process, and the search individuals move closer to the optimal position. This behavior can be described by the following mathematical expression: ； The searching individual gradually approaches its prey using a spiral motion, mathematically expressed as: 。

Images