Axial flow pump rubber bearing wear state monitoring method and system based on vibration signal analysis

Abstract

The invention belongs to the technical field of rubber bearing wear state monitoring, and particularly relates to an axial flow pump rubber bearing wear state monitoring method based on vibration signal analysis. The method comprises the following steps: acquiring and preprocessing a signal; constructing features; dividing wear stages; and intelligent early warning is realized. The system comprises an LMS data acquisition system, a PC (Personal Computer) and an MATLAB (Matrix Laboratory) program which are connected in sequence. According to the method, multi-scale feature extraction, entropy feature construction, feature fusion and classification model construction of the vibration signals are combined, the wear states (healthy, slight wear and serious wear states) of the rubber bearing of the rubber axial flow pump are accurately monitored and recognized, and the running safety and reliability of the axial flow pump are effectively guaranteed. According to the invention, the wear state of the rubber bearing of the axial flow pump can be monitored in real time in a complex environment, and a guarantee is provided for reliable operation of the axial flow pump.

Description

Technical Field

[0001] The present invention belongs to the technical field of rubber bearing wear state monitoring, and in particular relates to a method and system for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis. Background Art

[0002] Axial-flow pump rubber bearings play a vital role in a variety of hydraulic engineering and industrial applications. These bearings replace traditional oil-lubricated metal bearings, reducing potential environmental pollution and complying with environmental protection requirements. Axial-flow pump rubber bearings not only offer excellent vibration damping performance but also eliminate the need for complex oil supply systems, simplifying maintenance and reducing costs. However, because axial-flow pump rubber bearings typically operate under harsh conditions and experience long-term mixed lubrication or even boundary lubrication, they are susceptible to wear. This wear directly impacts the efficiency and reliability of the axial-flow pump system. Excessive wear without timely maintenance can lead to decreased pump performance and even system failure. Therefore, real-time monitoring of axial-flow pump rubber bearing wear is crucial for maintaining equipment reliability and extending its service life.

[0003] Currently, condition monitoring of axial-flow pump rubber bearings primarily relies on sensors to collect operational data from the equipment and assess its health. Common sensor types include temperature sensors, unidirectional piezoelectric accelerometers, proximity sensors, and acoustic emission sensors. Vibration signal analysis is a core method for condition monitoring. The key lies in accurately reflecting the equipment's operating status by constructing a health index (HI). Common condition monitoring methods can be categorized as end-to-end and feature-based. End-to-end monitoring methods utilize deep learning to automatically extract features from time-domain signals, avoiding the loss of state information associated with manual feature construction. However, these models lack interpretability and are prone to overfitting. Feature-based monitoring methods rely on the construction and fusion of signal features, but traditional feature construction methods often struggle to fully reflect changes across wear stages. Existing vibration signal analysis often relies on a single feature, making it difficult to fully characterize the complex characteristics of different wear states. Summary of the Invention

[0004] The present invention solves the problem that the existing technology often analyzes vibration signals based on only a single feature and is unable to fully characterize the complex characteristics of different wear states. It provides a method-level system for monitoring the wear state of axial flow pump rubber bearings based on vibration signal analysis. Through feature extraction and fusion technology, efficient monitoring of the wear state of axial flow pump rubber bearings is achieved.

[0005] The technical solutions claimed in the present invention are as follows:

[0006] A method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis comprises the following steps:

[0007] S1: Signal acquisition and preprocessing: The vibration signal of the axial flow pump rubber bearing is collected and decomposed into several intrinsic mode functions through empirical mode decomposition, and the intrinsic mode function components closely related to the wear state are screened out;

[0008] S2: Feature construction: Use time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and use principal component analysis to reduce the dimensionality of multi-dimensional health features to form a low-dimensional health index;

[0009] S3: Wear stage classification: Design a classification and recognition model based on support vector machine, use the classification and recognition model to classify the wear status of the low-dimensional health index in S2 in real time, and realize the wear stage classification of the axial flow pump rubber bearing; the wear status includes: healthy state, slight wear state and severe wear state

[0010] S4: Intelligent warning: Set a preset threshold for wear status and issue a warning when the preset threshold is exceeded based on the wear status classification results in S3.

[0011] Preferably, in S1, a vibration sensor is used to collect vibration signals, the sampling rate is set to 8 kHz, and 1 second of data is collected every 5 seconds; the screening in S1 adopts a combined indicator of kurtosis and correlation coefficient, and the formula is:

[0012] Cri=a·Cor+b·Kur

[0013] Where: Cor represents the correlation coefficient; Kur represents the kurtosis; a and b represent the weight coefficients, and usually a=b=0.5.

[0014] Preferably, the time domain features in S2 include mean, root mean square, peak-to-peak value, skewness, and kurtosis; the frequency domain features use fast Fourier transform to obtain spectral features; the time-frequency domain features use wavelet packet decomposition to obtain multi-band energy and calculate corresponding entropy features.

[0015] Preferably, the wavelet packet decomposition performs time-frequency decomposition on the original time domain signal, and the principle is as follows:

[0016]

[0017] Among them: A jk and W jk denote the approximation coefficient and wavelet coefficient respectively, and k denotes the offset at scale j; represents the scaling function.

[0018] The signal after wavelet packet decomposition contains the energy distribution of the signal in each frequency band, which can reflect the health status of the bearing. The energy of a specified narrow frequency band is calculated as follows:

[0019]

[0020] Where: Coef represents the wavelet coefficient W; k represents the number of wavelet coefficients in each wavelet packet decomposition; i and j represent the decomposition node and decomposition layer number respectively; on this basis, the relative energy, that is, the normalized energy coefficient, is obtained, which is defined as:

[0021] ρ ji =E ji / E sum

[0022] Where: E sum =∑ j ∑ i E ji , represents the total energy of all wavelet packet decompositions, it is obvious that ∑ρ ji =1, so ρ can be understood as the probability distribution of energy Eji, and the entropy criterion is introduced to describe the health state of rubber;

[0023] The Shannon entropy is introduced and defined as follows:

[0024] H shan =-∑ j ∑ i [ρ ji ]log[ρ ji ]

[0025] The larger the Hshan value, the more uniform the energy distribution in the corresponding frequency band. The smaller the value, the spectrum energy is concentrated near a specific value, reflecting whether the signal has a fault characteristic frequency. At the same time, the spectral flux change rate is obtained, which is defined as follows:

[0026] S flux =∑ j ∑ i |ρ ji -ρ j+1,j+1 |

[0027] Where: ij represents the normalized energy coefficient; i, j represent the decomposition node and decomposition layer number respectively.

[0028] Preferably, the dimension reduction in S2 uses energy entropy features based on wavelet packets, combines them with spectrum features to perform feature fusion, and calculates indicators such as singular spectrum entropy.

[0029] Preferably, when training the classification and recognition model in S3, sample data under various wear states are obtained through supervised learning, and the model is trained to achieve the best classification effect; the classification and recognition model uses the low-dimensional health index after multi-dimensional feature fusion as input, and divides different wear states through the optimal hyperplane.

[0030] Preferably, the classification formula of the classification recognition model is:

[0031]

[0032] y i (ω T (x i +b))≥1-ζ i

[0033] ζ i ≥0,i=1,2,...,n

[0034] Where: linear function w T x i +b represents the boundary function of SVM; x i represents the feature vector, the labels are -1 and 1; w represents the weight vector, b represents the offset of the separating hyperplane; ζ i represents the slack variable, ζ i ≥0, so that the interval plus the slack variable is greater than or equal to 1; C represents the penalty parameter, C takes a positive value, and when the value is large, the penalty for misclassification increases, and when the C value is small, the penalty for misclassification decreases.

[0035] The present invention also provides an axial flow pump rubber bearing wear status monitoring system based on vibration signal analysis, including an LMS data acquisition system connected in sequence for collecting and transmitting vibration signals, a PC that receives the vibration signals, analyzes and processes them in real time, and transmits the analyzed and processed signals, and a MATLAB program that receives the signals transmitted by the PC, monitors the wear status, and outputs the wear status.

[0036] In the above system, the PC decomposes the collected vibration signal into several intrinsic mode functions through empirical mode decomposition, and screens out the intrinsic mode function components closely related to the wear status; uses time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and uses principal component analysis to reduce the dimension of the multi-dimensional health features to form a low-dimensional health index; based on the support vector machine classification and recognition model, the classification and recognition model is used to classify the wear status of the low-dimensional health index in real time, realizing the wear stage division of the axial flow pump rubber bearing.

[0037] In the above system, the wear status includes: healthy status, slightly worn status and severely worn status.

[0038] Beneficial effects:

[0039] The present invention provides a wear state monitoring method and system for axial flow pump rubber bearings based on vibration signal analysis, the method comprising: collecting vibration signals of axial flow pump rubber bearings, decomposing the collected vibration signals into a number of intrinsic mode functions (IMFs) through empirical mode decomposition (EMD), and screening out intrinsic mode function components closely related to the wear state, using the low-frequency IMF components after EMD decomposition to represent the stable state of the bearing, and the high-frequency IMF components to reflect the mutation information in the wear process, screening out the IMF components closely related to the wear state, filtering out noise and interference information, and improving the accuracy of subsequent wear identification; using time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and comprehensively reflecting the signal features of the rubber axial flow pump rubber bearings under different wear states. Since different features will generate large redundant information in high-dimensional space, in order to reduce the computational complexity and improve the description ability of the features for the wear state, the multi-dimensional health features are reduced in dimensionality through principal component analysis to form a low-dimensional health index, which improves computational efficiency by suppressing redundant information while retaining useful feature information. These low-dimensional health indices can not only greatly improve the wear state of the rubber bearings, but also greatly improve the wear state of the rubber bearings. The data redundancy is greatly reduced and the physical interpretability of the features is improved, making the monitoring system more stable and robust. The low-dimensional health index after dimensionality reduction can more clearly reveal the signal change pattern of the rubber bearing in the healthy, slightly worn and severely worn stages, providing a reliable feature basis for state classification. A classification and recognition model based on support vector machine (SVM) is designed, and the classification and recognition model is used to classify the wear state of the low-dimensional health index in real time to realize the wear stage division of the axial flow pump rubber bearing. The SVM classification model can construct a highly adaptable decision boundary, dividing different wear states into three states: healthy, slightly worn and severely worn, and realizing real-time identification of different wear states. A preset threshold for the wear state is set. Based on the wear state classification result in S3, an early warning is issued when the preset threshold is exceeded, which can provide clear status information for the operator. When the wear state reaches the preset threshold, the system will automatically issue an early warning message to prompt the operator to maintain and replace the bearing in time to avoid safety hazards caused by excessive wear, help maintenance personnel identify the wear state at an early stage, and reasonably arrange maintenance cycles, thereby effectively reducing maintenance costs and extending the service life of the bearing.

[0040] In terms of time domain characteristics, statistical features such as root mean square (RMS), skewness, and peak value are extracted to describe the overall state of the vibration signal. In the frequency domain, fast Fourier transform (FFT) is used to obtain spectral features, extracting information such as dominant frequency and spectral energy to characterize changes in the wear state. To capture the energy variations in frequency bands at different scales during the wear process, wavelet packet decomposition (WPD) is introduced. By calculating entropy-based features such as Shannon entropy in different frequency bands, an entropy-based multidimensional health index (MD-HI) is constructed. This index accurately reflects changes in signal complexity and provides an important basis for wear state monitoring.

[0041] When training the classification and recognition model, supervised learning is used to obtain sample data under various wear states, and the model is trained to achieve the best classification effect. The classification and recognition model uses a low-dimensional health index after multi-dimensional feature fusion as input, and uses an optimal hyperplane to segment different wear states, ensuring efficient and accurate state recognition in practical applications.

[0042] In summary, the present invention combines multi-scale feature extraction of vibration signals, entropy feature construction, feature fusion, and classification model construction to accurately monitor and identify the wear status (healthy, slightly worn, and severely worn) of rubber bearings in rubber axial flow pumps, effectively ensuring the safety and reliability of axial flow pump operation. The method of the present invention enables real-time monitoring of the wear status of axial flow pump rubber bearings in complex environments, ensuring the reliable operation of the axial flow pump. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a flow chart of a method for monitoring the wear status of rubber bearings in an axial flow pump based on vibration signal analysis according to an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the rubber bearing structure of an axial flow pump according to an embodiment of the present invention.

[0045] Figure 3 Schematic diagram of an axial flow pump rubber bearing wear status monitoring system based on vibration signal analysis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that those skilled in the art may make several changes and modifications without departing from the scope of the present invention. These all fall within the scope of protection of the present invention.

[0047] The present invention monitors the wear state of the axial flow pump rubber bearing, and the axial flow pump rubber bearing structure, such as Figure 2 As shown, it includes a transmission shaft 2, an axial flow pump rubber bearing 1, two stators 3, an impeller 4, a cylinder wall 5 and multiple unidirectional piezoelectric acceleration sensors 6; the axial flow pump rubber bearing 1 is arranged at the front end of the transmission shaft 2; one end of the transmission shaft 2 is fixedly connected to the impeller 4, and the other end is supported by the axial flow pump rubber bearing 1; the impeller 4 is fixed on the cylinder wall 5 at a distance away from the transmission shaft 2; the two stators 3 are symmetrically arranged at the front end of the outer surface of the transmission shaft 2, one end is fixedly connected to the outer surface of the transmission shaft 2, and the other end is fixedly connected to the cylinder wall 5; the multiple unidirectional piezoelectric acceleration sensors 6 are respectively arranged on the surface of the cylinder wall and the transmission shaft 2.

[0048] The present invention provides a method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis, such as Figure 1 As shown, the following steps are included:

[0049] S1: Signal acquisition and preprocessing: The vibration signal of the axial flow pump rubber bearing is collected and decomposed into several intrinsic mode functions through empirical mode decomposition, and the intrinsic mode function components closely related to the wear state are screened out;

[0050] S2: Feature construction: Use time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and use principal component analysis to reduce the dimensionality of multi-dimensional health features to form a low-dimensional health index;

[0051] S3: Wear stage classification: Design a classification and recognition model based on support vector machine, use the classification and recognition model to classify the wear status of the low-dimensional health index in S2 in real time, and realize the wear stage classification of the axial flow pump rubber bearing; the wear status includes: healthy state, slight wear state and severe wear state

[0052] S4: Intelligent warning: Set a preset threshold for wear status and issue a warning when the preset threshold is exceeded based on the wear status classification results in S3.

[0053] Example 1

[0054] Signal Acquisition and Preprocessing: A vibration sensor was used to collect signals, with a sampling rate of 8kHz and 1 second of data collected every 5 seconds. The vibration signal of the axial flow pump's rubber bearing was obtained from the vibration sensor. Empirical Mode Decomposition (IMF) was used to select appropriate IMFs and normalize the signal to remove high-frequency noise and drift.

[0055] Kurtosis is an important indicator for monitoring bearing fault pulse signals. Therefore, the screening index of the eigenfunction is the combined index of kurtosis and correlation coefficient. The formula is:

[0056] Cri=a·Cor+b·Kur

[0057] Where: Cor is the correlation coefficient, Kur is the kurtosis, a and b are weight coefficients, and usually a=b=0.5.

[0058] Feature extraction:

[0059] Time domain characteristics: including mean, root mean square, peak-to-peak value, skewness, kurtosis, etc.;

[0060] Frequency domain features: Fast Fourier transform (FFT) is used to obtain spectrum features;

[0061] Time-frequency domain features: Wavelet packet decomposition (WPD) is used to obtain multi-band energy and calculate the corresponding entropy features.

[0062] The signal of the entire life cycle of a bearing is non-stationary, and its characteristics change over time. Therefore, in addition to the commonly used time domain analysis and frequency domain analysis, time-frequency domain analysis is required to extract the characteristics of time-frequency information. Wavelet packet decomposition is used to decompose the original time domain signal into time and frequency. The principle is as follows:

[0063]

[0064] Among them: A jk and W jk denote the approximation coefficient and wavelet coefficient respectively, and k denotes the offset at scale j; represents the scaling function.

[0065] The signal after wavelet packet decomposition contains the energy distribution of the signal in each frequency band, which can reflect the health status of the bearing. The energy of a specified narrow frequency band is calculated as follows:

[0066]

[0067] Where: Coef represents the wavelet coefficient W; k represents the number of wavelet coefficients in each wavelet packet decomposition; i and j represent the decomposition node and decomposition layer number respectively; on this basis, the relative energy can be obtained, that is, the normalized energy coefficient, which is defined as:

[0068] ρ ji =E ji / E sum

[0069] Where: E sum =∑ j ∑ i E ji , represents the total energy of all wavelet packet decompositions, it is obvious that ∑ρ ji =1, so ρ can be understood as the probability distribution of energy Eji, and the entropy criterion is introduced to describe the health state of rubber;

[0070] The Shannon entropy is introduced and defined as follows:

[0071] H shan =-∑ j ∑ i [ρ ji ]log[ρ ji ]

[0072] The larger the Hshan value, the more uniform the energy distribution in the corresponding frequency band. The smaller the value, the more concentrated the spectrum energy is near a specific value. This can reflect whether the signal has a fault characteristic frequency. At the same time, the spectral flux change rate can be obtained, which is defined as follows:

[0073] S flux =∑ j ∑i |ρ ji -ρ j+1,j+1 |

[0074] Where: ij represents the normalized energy coefficient; i, j represent the decomposition node and decomposition layer number respectively.

[0075] Finally, the extracted statistical features are shown in Table 1.

[0076] Table 1. Extracted features

[0077]

[0078] Feature fusion:

[0079] PCA is used to reduce the dimensionality of multidimensional features to form a low-dimensional health index for wear status classification.

[0080] Wear status identification: Use the SVM model to classify the health index and achieve real-time identification of wear status.

[0081] Example 2:

[0082] Signal Acquisition and Preprocessing: A vibration sensor was used to collect signals, with a sampling rate of 8kHz and 1 second of data collected every 5 seconds. The vibration signal of the axial flow pump's rubber bearing was obtained from the vibration sensor. Empirical Mode Decomposition (IMF) was used to select appropriate IMFs and normalize the signal to remove high-frequency noise and drift.

[0083] Kurtosis is an important indicator for monitoring bearing fault pulse signals, so the screening index of the eigenfunction is a combined index of kurtosis and correlation coefficient. The formula is:

[0084] Cri=a·Cor+b·Kur

[0085] Where: Cor is the correlation coefficient, Kur is the kurtosis, a and b are weight coefficients, and usually a=b=0.5.

[0086] Feature extraction:

[0087] Time domain characteristics: including mean, root mean square, peak-to-peak value, skewness, kurtosis, etc.;

[0088] Frequency domain features: Fast Fourier transform (FFT) is used to obtain spectrum features;

[0089] Time-frequency domain features: Wavelet packet decomposition (WPD) is used to obtain multi-band energy and calculate the corresponding entropy features.

[0090] The signal of the entire life cycle of a bearing is non-stationary, and its characteristics change over time. Therefore, in addition to the commonly used time domain analysis and frequency domain analysis, time-frequency domain analysis is required to extract the characteristics of time-frequency information. Wavelet packet decomposition is used to decompose the original time domain signal into time and frequency. The principle is as follows:

[0091]

[0092] Among them: A jk and W jk denote the approximation coefficient and wavelet coefficient respectively, and k denotes the offset at scale j; represents the scaling function.

[0093] The signal after wavelet packet decomposition contains the energy distribution of the signal in each frequency band, which can reflect the health status of the bearing. The energy of a specified narrow frequency band is calculated as follows:

[0094]

[0095] Where: Coef represents the wavelet coefficient W; k represents the number of wavelet coefficients in each wavelet packet decomposition; i and j represent the decomposition node and decomposition layer number respectively; on this basis, the relative energy can be obtained, that is, the normalized energy coefficient, which is defined as:

[0096] ρ ji =E ji / E sum

[0097] Where: E sum =∑ j ∑ i E ji , represents the total energy of all wavelet packet decompositions, it is obvious that ∑ρ ji =1, so ρ can be understood as the probability distribution of energy Eji, and the entropy criterion is introduced to describe the health state of rubber;

[0098] The Shannon entropy is introduced and defined as follows:

[0099] H shan =-∑ j ∑ i [ρ ji ]log[ρ ji ]

[0100] The larger the Hshan value, the more uniform the energy distribution in the corresponding frequency band. The smaller the value, the more concentrated the spectrum energy is near a specific value. This can reflect whether the signal has a fault characteristic frequency. At the same time, the spectral flux change rate can be obtained, which is defined as follows:

[0101] S flux =∑ j ∑ i |ρji -ρ j+1,j+1 |

[0102] Where: ij represents the normalized energy coefficient; i, j represent the decomposition node and decomposition layer number respectively.

[0103] Finally, the extracted statistical features are shown in Table 1.

[0104] Multidimensional feature fusion:

[0105] The energy entropy feature based on wavelet packets is used in combination with spectrum features (such as characteristic frequency and energy distribution) to perform feature fusion and calculate indicators such as singular spectrum entropy.

[0106] Support Vector Machine Classification Model:

[0107] SVM is applied to classify the reduced dimensionality features to identify healthy, slightly worn, and severely worn states.

[0108] The classification formula is:

[0109]

[0110] y i (ω T (x i +b))≥1-ζ i

[0111] ζ i ≥0,i=1,2,...,n

[0112] Where: linear function w T x i +b represents the boundary function of SVM, x i represents the feature vector, the labels are -1 and 1, w represents the weight vector, and b represents the offset of the separating hyperplane; ζ i represents the slack variable, ζ i ≥0, so that the interval plus the slack variable is greater than or equal to 1; C represents the penalty parameter, C takes a positive value, and when the value is large, the penalty for misclassification increases, and when the C value is small, the penalty for misclassification decreases.

[0113] Example 3

[0114] Signal Acquisition and Preprocessing: Vibration sensors were used to collect signals, with a sampling rate set to 8kHz and 1 second of data collected every 5 seconds. The vibration signals of the axial flow pump's rubber bearings were obtained from the vibration sensors. Empirical mode decomposition (EMD) was combined with the constructed eigenvalue function's screening criteria to identify eigenvalues that best reflect bearing wear and degradation, removing high-frequency noise and drift from the signals. EMD further optimized the extracted features, reducing the impact of environmental interference on monitoring.

[0115] Kurtosis is an important indicator for monitoring bearing fault pulse signals, so the screening index of the eigenfunction is a combined index of kurtosis and correlation coefficient. The formula is:

[0116] Cri=a·Cor+b·Kur

[0117] Where: Cor is the correlation coefficient, Kur is the kurtosis, a and b are weight coefficients, and usually a=b=0.5.

[0118] Feature extraction:

[0119] Time domain characteristics: including mean, root mean square, peak-to-peak value, skewness, kurtosis, etc.;

[0120] Frequency domain features: Fast Fourier transform (FFT) is used to obtain spectrum features;

[0121] Time-frequency domain features: Wavelet packet decomposition (WPD) is used to obtain multi-band energy and calculate the corresponding entropy features.

[0122] The signal of the entire life cycle of a bearing is non-stationary, and its characteristics change over time. Therefore, in addition to the commonly used time domain analysis and frequency domain analysis, time-frequency domain analysis is required to extract the characteristics of time-frequency information. Wavelet packet decomposition is used to decompose the original time domain signal into time and frequency. The principle is as follows:

[0123]

[0124] Among them: A jk and W jk denote the approximation coefficient and wavelet coefficient respectively, and k denotes the offset at scale j; represents the scaling function.

[0125] The signal after wavelet packet decomposition contains the energy distribution of the signal in each frequency band, which can reflect the health status of the bearing. The energy of a specified narrow frequency band is calculated as follows:

[0126]

[0127] Where: Coef represents the wavelet coefficient W; k represents the number of wavelet coefficients in each wavelet packet decomposition; i and j represent the decomposition node and decomposition layer number respectively; on this basis, the relative energy can be obtained, that is, the normalized energy coefficient, which is defined as:

[0128] ρ ji =E ji / E sum

[0129] Where: E sum =∑ j ∑ i E ji, represents the total energy of all wavelet packet decompositions, it is obvious that ∑ρ ji =1, so ρ can be understood as the probability distribution of energy Eji, and the entropy criterion is introduced to describe the health state of rubber;

[0130] The Shannon entropy is introduced and defined as follows:

[0131] H shan =-∑ j ∑ i [ρ ji ]log[ρ ji ]

[0132] The larger the Hshan value, the more uniform the energy distribution in the corresponding frequency band. The smaller the value, the more concentrated the spectrum energy is near a specific value. This can reflect whether the signal has a fault characteristic frequency. At the same time, the spectral flux change rate can be obtained, which is defined as follows:

[0133] S flux =∑ j ∑ i |ρ ji -ρ j+1,j+1 |

[0134] Where: ij represents the normalized energy coefficient; i, j represent the decomposition node and decomposition layer number respectively.

[0135] Finally, the extracted statistical features are shown in Table 1.

[0136] Based on the above feature extraction, signal noise reduction processing is added to output the health status index in a more stable manner, which is suitable for application scenarios with large environmental noise or obvious signal fluctuations.

[0137] Multidimensional feature fusion:

[0138] The energy entropy feature based on wavelet packets is used in combination with spectrum features (such as characteristic frequency and energy distribution) to perform feature fusion and calculate indicators such as singular spectrum entropy.

[0139] Support Vector Machine Classification Model:

[0140] SVM is applied to classify the reduced dimensionality features to identify healthy, slightly worn, and severely worn states.

[0141] Classification formula:

[0142]

[0143] y i (ω T (x i +b))≥1-ζ i

[0144] ζ i ≥0,i=1,2,...,n

[0145] Where: linear function w T x i +b represents the boundary function of SVM, x i represents the feature vector, the labels are -1 and 1, w represents the weight vector, and b represents the offset of the separating hyperplane; ζ i represents the slack variable, ζ i ≥0, so that the interval plus the slack variable is greater than or equal to 1; C represents the penalty parameter, C takes a positive value, and when the value is large, the penalty for misclassification increases, and when the C value is small, the penalty for misclassification decreases.

[0146] Examples 1-3, the three examples gradually increase the complexity of feature extraction, the diversity of fusion methods and the sophistication of signal processing, so that the entire monitoring method from basic application to refined processing, gradually improves the recognition accuracy and robustness of wear status.

[0147] The present invention also provides an axial flow pump rubber bearing wear state monitoring system based on vibration signal analysis, such as Figure 3 As shown, it includes an LMS data acquisition system for collecting vibration signals and transmitting them, a PC that receives the vibration signals, analyzes and processes them in real time, and transmits the analyzed and processed signals, and a MATLAB program that receives the signals transmitted by the PC, monitors the wear status, and outputs the wear status.

[0148] In this system, a PC decomposes the collected vibration signal into several intrinsic mode functions (IMFs) through empirical mode decomposition (EMD), and selects IMF components closely related to the wear state. A multi-dimensional health signature is constructed using time-domain, frequency-domain, and time-frequency-domain feature extraction methods. Principal component analysis is used to reduce the dimensionality of these multi-dimensional health signatures to form a low-dimensional health index. A support vector machine-based classification and recognition model is used to classify the wear state of the low-dimensional health index in real time, enabling the classification of axial flow pump rubber bearing wear stages. These wear states include healthy, mild, and severe wear.

Claims

1. A method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis, characterized in that: The steps include: S1: Signal acquisition and preprocessing: The vibration signal of the axial flow pump rubber bearing is collected and decomposed into several intrinsic mode functions through empirical mode decomposition, and the intrinsic mode function components closely related to the wear state are screened out; S2: Feature construction: Use time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and use principal component analysis to reduce the dimensionality of multi-dimensional health features to form a low-dimensional health index; S3: Wear stage classification: Design a classification and recognition model based on support vector machine, use the classification and recognition model to classify the wear status of the low-dimensional health index in S2 in real time, and realize the wear stage classification of the axial flow pump rubber bearing; the wear status includes: healthy state, slight wear state and severe wear state S4: Intelligent warning: Set a preset threshold for wear status and issue a warning when the preset threshold is exceeded based on the wear status classification results in S3.

2. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to claim 1 is characterized in that: In S1, a vibration sensor is used to collect vibration signals, the sampling rate is set to 8kHz, and 1 second of data is collected every 5 seconds. The screening described in S1 uses a combined indicator of kurtosis and correlation coefficient, and the formula is: Cri=a·Cor+b·Kur Where: Cor represents the correlation coefficient; Kur represents the kurtosis; a and b represent the weight coefficients, and usually a=b=0.

5.

3. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to claim 2 is characterized in that: The time domain features described in S2 include mean, root mean square, peak-to-peak value, skewness, and kurtosis; the frequency domain features use fast Fourier transform to obtain spectral features; the time-frequency domain features use wavelet packet decomposition to obtain multi-band energy and calculate the corresponding entropy features.

4. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to claim 3 is characterized in that: The wavelet packet decomposition performs time-frequency decomposition on the original time domain signal. The principle is as follows: Among them: A jk and W jk denote the approximation coefficient and wavelet coefficient respectively, and k denotes the offset at scale j; represents the scaling function; The signal after wavelet packet decomposition contains the energy distribution of the signal in each frequency band, which can reflect the health status of the bearing. The energy of a specified narrow frequency band is calculated as follows: Where: Coef represents the wavelet coefficient W; k represents the number of wavelet coefficients in each wavelet packet decomposition; i and j represent the decomposition node and decomposition layer number respectively; on this basis, the relative energy, that is, the normalized energy coefficient, is obtained, which is defined as: ρ ji =And ji / AND sum Where: E sum =∑ j ∑ i E ji , represents the total energy of all wavelet packet decompositions, it is obvious that ∑ρ ji =1, so ρ can be understood as the probability distribution of energy Eji, and the entropy criterion is introduced to describe the health state of rubber; The Shannon entropy is introduced and defined as follows: H shan =-∑ j ∑ i [r ji ]log[ρ ji ] The larger the Hshan value, the more uniform the energy distribution in the corresponding frequency band. The smaller the value, the spectrum energy is concentrated near a specific value, reflecting whether the signal has a fault characteristic frequency. At the same time, the spectral flux change rate is obtained, which is defined as follows: S flux =∑ j ∑ i |r ji -r j+1,j+1 | Where: ij represents the normalized energy coefficient; i, j represent the decomposition node and decomposition layer number respectively.

5. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to any one of claims 1 to 4, characterized in that: The dimensionality reduction described in S2 uses the energy entropy feature based on wavelet packets, combined with the spectrum feature for feature fusion, and calculates the singular spectrum entropy index.

6. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to claim 5 is characterized in that: When training the classification and recognition model in S3, sample data under various wear states are obtained through supervised learning, and the model is trained to achieve the best classification effect; the classification and recognition model uses the low-dimensional health index after multi-dimensional feature fusion as input, and divides different wear states through the optimal hyperplane.

7. The method for monitoring the wear state of an axial flow pump rubber bearing based on vibration signal analysis according to claim 6, characterized in that: The classification formula of the classification recognition model is: y i (oh T (x i +b))≥1-ζ i g i ≥0,i=1,2,...,n Where: linear function w T x i +b represents the boundary function of SVM; x i represents the feature vector, the labels are -1 and 1; w represents the weight vector, b represents the offset of the separating hyperplane; ζ i represents the slack variable, ζ i ≥0, so that the interval plus the slack variable is greater than or equal to 1; C represents the penalty parameter, C takes a positive value, and when the value is large, the penalty for misclassification increases, and when the C value is small, the penalty for misclassification decreases.

8. A wear status monitoring system for axial flow pump rubber bearings based on vibration signal analysis, characterized in that: The invention comprises an LMS data acquisition system connected in sequence for collecting vibration signals and transmitting them, a PC for receiving vibration signals, performing real-time analysis and processing on them and transmitting the analyzed and processed signals, and a MATLAB program for receiving signals transmitted by the PC, monitoring the wear status and outputting the wear status.

9. The axial flow pump rubber bearing wear state monitoring system based on vibration signal analysis according to claim 8 is characterized in that: The PC decomposes the collected vibration signal into several intrinsic mode functions through empirical mode decomposition, and screens out the intrinsic mode function components closely related to the wear status; uses time domain, frequency domain and time-frequency domain feature extraction methods to construct multi-dimensional health features, and uses principal component analysis to reduce the dimensionality of the multi-dimensional health features to form a low-dimensional health index; based on the support vector machine classification and recognition model, the classification and recognition model is used to classify the wear status of the low-dimensional health index in real time, realizing the wear stage division of the axial flow pump rubber bearing.

10. The axial flow pump rubber bearing wear state monitoring system based on vibration signal analysis according to claim 9 is characterized in that: The wear status includes: healthy status, slight wear status and severe wear status.

Images