A centrifugal pump multi-fault identification method based on vibration intensity and statistical indicators

Abstract

The application provides a centrifugal pump multi-fault identification method based on vibration intensity and statistical indexes, comprising: building a test bench to realize normal, rotor imbalance, rotor misalignment and mechanical looseness operating states, collecting vibration acceleration and vibration displacement signals under each operating state; screening the most sensitive signal measuring points; calculating the statistical indexes of the screened signal measuring points to obtain a feature set; performing standardization and dimension reduction processing on the feature set to obtain a target feature set; dividing a plurality of measured sample groups into a training set and a test set to obtain a training feature matrix and a test feature matrix; performing parameter optimization to obtain globally optimal radial basis kernel parameters g and penalty parameters C, inputting the training feature matrix into an identification model to complete data training, and then inputting the test feature matrix into the identification model to obtain a centrifugal pump multi-fault identification result. The application solves the problem that normal, rotor imbalance, rotor misalignment and mechanical looseness of a vertical centrifugal pump cannot be classified and identified.

Description

Technical Field

[0001] This invention relates to the field of centrifugal pump fault identification technology, and in particular to a method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators. Background Technology

[0002] Single-stage, single-suction vertical centrifugal pumps are a key component of ship piping systems, responsible for transporting fluids to maintain the normal operation of various auxiliary equipment. Ship rolling, equipment vibration, and prolonged operation under certain loads can easily cause component failures in centrifugal pump units, especially rotor system failures and mechanical loosening, thus reducing the service life and efficiency of the centrifugal pump unit. In industrial production scenarios, the sensor data acquired from rotating machinery often exhibits characteristics such as channel imbalance and noise interference. Rotating machinery fault diagnosis methods primarily rely on single-fault identification, and signal characteristic representation indicators are limited. Currently, no mature and reliable multi-fault identification system for centrifugal pumps is in operation. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a multi-fault identification method for centrifugal pumps based on vibration intensity and statistical indicators, solving the problem of accurately and reliably identifying the four operating states of a vertical centrifugal pump: normal operation, rotor imbalance, rotor misalignment, and mechanical loosening.

[0004] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0005] A method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators includes:

[0006] Step S1: Establish a multi-fault signal synchronous test bench for vertical centrifugal pump units to simulate centrifugal pump fault states. Implement four operating states: normal, rotor imbalance, rotor misalignment, and mechanical loosening. Collect five vibration acceleration signals and one vibration displacement signal under each operating state. Calculate the vibration intensity of each measuring point under each operating state and select the most sensitive signal measuring points that reflect the fault state.

[0007] Step S2: Calculate the statistical indicators of the selected signal measurement points to obtain a feature set composed of several statistical indicators; use the weighted kernel principal component analysis method to standardize and reduce the dimensionality of the feature set to obtain the target feature set; divide the several sets of samples measured under the four operating states into a training set and a test set, wherein the training set constitutes a training feature matrix and the test set constitutes a test feature matrix.

[0008] Step S3: Use an optimization algorithm to find the parameters and obtain the globally optimal radial basis kernel parameter g and penalty parameter C. Input the training feature matrix into the recognition model to complete the data training, and then input the test feature matrix into the recognition model to obtain the centrifugal pump multi-fault recognition result.

[0009] Furthermore, the test bench in step S1 includes a centrifugal pump, a motor, and a pipeline system. The centrifugal pump is a single-stage, single-suction vertical centrifugal pump. The motor is mounted on the upper end of the centrifugal pump via a support, and the motor shaft is connected to the pump shaft of the centrifugal pump. The centrifugal pump is started, stopped, and speed-controlled via a three-phase AC frequency converter. The pipeline system includes a check valve, an inlet valve, a vertical centrifugal pump, a frequency converter motor, an electromagnetic flowmeter, an outlet valve, and connecting pipes.

[0010] Furthermore, in step S1, the centrifugal pump synchronously collects the vibration acceleration of the motor bracket, the vibration displacement of the pump shaft in the X and Y directions, the vibration acceleration of the pump inlet flange, the vibration acceleration of the pump outlet flange, the vibration acceleration of the pump body, and the vibration acceleration of the bracket feet under four operating states, and collects more than or equal to 100 sets of data under each operating state.

[0011] Furthermore, the vibration acceleration is measured by an ICP-type uniaxial vibration acceleration sensor, and the vibration displacement is measured by an LD980-Y integrated eddy current sensor.

[0012] Further:

[0013] The formula for calculating vibration intensity is:

[0014] The formula for calculating frequency domain signals is:

[0015] The formula for calculating the amplitude spectrum is:

[0016] The formula for calculating harmonic frequency is:

[0017] The vibration intensity of the vibration displacement is given by the formula Calculated;

[0018] The vibration intensity of the vibration acceleration is:

[0019] Where N is the total number of discrete signals, v(n) is the nth discrete velocity signal, x(n) is the measured vibration signal at N points, and f s The value in k is the signal sampling frequency. a For greater than Nf a / f s The smallest integer, k b For greater than Nfb / f s The largest integer.

[0020] Furthermore, the feature set in step S2 includes several time-domain indicators, several frequency-domain indicators, and several time-frequency-domain energy feature indicators.

[0021] Furthermore, in step S2, the formula is as follows: The calculation is standardized, and the sample correlation coefficient matrix is ​​obtained from the formula. The calculations yielded s = 1, 2, ..., 20, k = 1, 2, ..., 100; the Jacobi method was used to solve for the eigenvalues ​​(λ1, λ2, ..., λ) of the correlation coefficient matrix R. p ) and the corresponding eigenvector a i =(a i1 ,a i2 ,···,a ip ), i = 1, 2, ..., p;

[0022] Feature contribution rate is given by formula The calculations show that, based on the principle of sorting the feature parameters from highest to lowest contribution rate and having a separability index greater than 0.85, the top nine feature vectors by contribution rate form the target feature set.

[0023] Furthermore, in step S3, a genetic optimization algorithm is used to optimize the parameters.

[0024] The beneficial effects of this invention are:

[0025] This invention employs signal processing technology and intelligent diagnostic models to establish a multi-fault identification method for centrifugal pumps, solving the problem of difficulty in classifying and identifying four operating states of centrifugal pumps: normal operation, rotor imbalance, rotor misalignment, and mechanical loosening, thus achieving the technical effect of multi-fault classification and identification for centrifugal pumps. Attached Figure Description

[0026] Figure 1 This is a flowchart of the multi-fault identification method for centrifugal pumps based on vibration intensity and statistical indicators according to the present invention.

[0027] Figure 2 This is a schematic diagram of a centrifugal pump multi-fault simulation and signal acquisition test bench in an embodiment of the present invention;

[0028] Figure 3 This is a comparison of vibration intensity at various measuring points under centrifugal pump failure conditions in this embodiment of the invention.

[0029] Figure 4 This is the result of the original feature parameter weight distribution in the embodiments of the present invention;

[0030] Figure 5This is the feature extraction result of the training samples for extracting multiple fault conditions using the WKPCA method in this embodiment of the invention;

[0031] Figure 6 The fitness curve of the GA-SVM model used in this embodiment of the invention;

[0032] Figure 7 This is the identification result of a centrifugal pump multi-fault test sample using the GA-SVM model in an embodiment of the present invention;

[0033] Explanation of reference numerals in the attached figures:

[0034] 1-Check valve, 2-Inlet valve, 3-Centrifugal pump, 4-Variable frequency motor, 5-Electromagnetic flow meter, 6-Outlet valve, 7-Connecting pipe, 8-Three-phase AC frequency converter, 9-Data acquisition instrument, 10-Display;

[0035] M1 - Motor bracket vibration acceleration sensor; M2 - Pump shaft X and Y direction vibration displacement sensor; M3 - Pump inlet flange vibration acceleration sensor; M4 - Pump outlet flange vibration acceleration sensor; M5 - Pump body vibration acceleration sensor; M6 - Bracket foot vibration acceleration sensor. Detailed Implementation

[0036] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the following description is provided in conjunction with the accompanying drawings. Figure 1-7 Detailed descriptions of specific embodiments of the present invention are provided below. Numerous specific details are set forth in the following description to provide a thorough understanding of the invention. However, the invention can be practiced in many ways other than those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0037] A method for identifying multiple faults in a centrifugal pump based on vibration intensity and statistical indicators according to an embodiment of the present invention includes:

[0038] Step S1: Establish a multi-fault signal synchronous test bench for vertical centrifugal pump units to simulate centrifugal pump fault states. Implement four operating states: normal, rotor imbalance, rotor misalignment, and mechanical loosening. Collect five vibration acceleration signals and one vibration displacement signal under each operating state. Calculate the vibration intensity of each measuring point under each operating state and select the most sensitive signal measuring points that reflect the fault state.

[0039] like Figure 2As shown, the centrifugal pump unit multi-fault signal synchronous test bench includes a centrifugal pump 3, a motor, and a running pipeline system. The centrifugal pump 3 is a single-stage, single-suction vertical centrifugal pump with an integrated installation structure with the support. The variable frequency motor 4 is fixedly installed on the upper end of the pump through the support, and the motor shaft and pump shaft are connected by a coupling. The centrifugal pump 3 achieves start-up, shutdown, and speed control through parameter adjustment of a three-phase AC frequency converter 8. The centrifugal pump running pipeline system consists of a check valve 1, an inlet valve 2, an electromagnetic flowmeter 5, an outlet valve 6, and connecting pipes 7. The centrifugal pump signal measurement points are arranged in the following positions: vertically to the motor support, horizontally to the pump shaft in the X and Y directions (non-contact), vertically to the pump inlet flange, vertically to the pump outlet flange, horizontally to the pump body, and vertically to the support feet.

[0040] Among them, the specific speed of the vertical centrifugal pump is n s =67.09, rated flow rate is Q d =100m 3 / h, head H d =80m, speed n=2950r / min, pumping pipeline is DN80 pipeline; the vertical centrifugal pump motor is a variable frequency motor 4, model Y200L2-2, power 37KW; the frequency converter is an ABB ACS800 series frequency converter, output power 90kW; the flow meter is an electromagnetic flow meter 5, flow range 0-200m 3 / h, accuracy 0.5 grade; Data acquisition instrument 9 uses INV3020C model, 24-bit acquisition card, 16 channels; Display 10 uses 21-inch display screen; Vibration acceleration sensor uses INV9822 type ICP single-axis vibration acceleration sensor, sensitivity 10mV / ms -2 The vibration displacement sensor adopts the LD980-Y integrated eddy current sensor, with a displacement range of 0-2mm and an accuracy of 0.1%.

[0041] Vibration intensity is the vibration velocity of a centrifugal pump. The signal measured in the experiment is a discrete signal, and the calculation formula is as follows: The formula for frequency domain signals is The amplitude spectrum is calculated using the formula Harmonic frequencies are given by the formula Calculated;

[0042] The vibration intensity of the vibration displacement is given by the formula Calculated;

[0043] The intensity of vibration acceleration is given by the formula The calculation yields N, where N is the total number of discrete signals, v(n) is the nth discrete velocity signal, x(n) is the measured vibration signal at N points, and f... s The value in k is the signal sampling frequency. a For greater than Nfa / f s The smallest integer, k b For greater than Nf b / f s The largest integer;

[0044] The vibration intensity of the vertical centrifugal pump at each measuring point was calculated by averaging five sets of data under four conditions: normal operation, rotor imbalance, rotor misalignment, and mechanical loosening. Figure 3 As shown. The signal measuring sensors include a motor support vibration acceleration sensor M1, a pump shaft X and Y direction vibration displacement sensor M2, a pump outlet flange vibration acceleration sensor M3, a pump inlet flange vibration acceleration sensor M4, a pump body vibration acceleration sensor M5, and a machine foot vibration acceleration sensor M6. Among them, the signal from the pump shaft vibration displacement sensor M2 is most sensitive to rotor imbalance and misalignment, with its vibration intensity increasing by 53.9% and 65.3% respectively compared to the normal state. The signal from the machine foot vibration acceleration sensor M6 is most sensitive to mechanical loosening, with its vibration intensity increasing by 73.6%. Therefore, the signals from the pump shaft X and Y direction vibration displacement sensor M2 and the machine foot vibration acceleration sensor M6 are selected to reflect the fault characteristic information of the vertical centrifugal pump.

[0045] Step S2: Calculate the statistical indicators of the selected signal measurement points to obtain a feature set composed of several statistical indicators; use weighted kernel principal component analysis to standardize and reduce the dimensionality of the feature set to obtain the target feature set; divide the several sets of samples measured under the four operating states into a training set and a test set, wherein the training set constitutes a training feature matrix and the test set constitutes a test feature matrix, as follows:

[0046] Rotor imbalance, misalignment, and mechanical loosening faults all exhibit different characteristics. A multi-domain, multi-category fault feature extraction method is used to form a fault feature set. The feature indicators include 10 time-domain indicators, 4 frequency-domain indicators, and 6 time-frequency-domain energy feature indicators, namely: mean, kurtosis, peak value, variance, harmonic mean, waveform indicator, impulse indicator, peak value indicator, margin indicator, skewness indicator, mean frequency, center of gravity frequency, root mean square frequency, frequency standard deviation, and the feature energy E1 to E6 of the first 6 IMF components under different characteristic time scales.

[0047] To address the issue of aliasing among the principal components of different faults and reduce or eliminate information redundancy, dimensionality reduction is performed on the original high-dimensional fault features. This involves calculating the weight index of each feature parameter in the original fault feature set, such as... Figure 4 As shown.

[0048] The data standardization process, with 20 original fault feature parameters and 100 samples, is performed using the formula: The sample correlation coefficient matrix is ​​calculated using the formula. The calculations yielded s = 1, 2, ..., 20, k = 1, 2, ..., 100;

[0049] The Jacobi method is used to solve for the eigenvalues ​​(λ1, λ2, ..., λ) of the correlation coefficient matrix R. p ) and the corresponding eigenvector a i =(a i1 ,a i2 ,···,a ip ), i = 1, 2, ..., p; the characteristic contribution rate is given by the formula The calculations show that, based on the principle of sorting the feature parameters from high to low contribution rates and having a separability index greater than 0.85, the top 9 feature vectors by contribution rate form a feature matrix that reduces the dimensionality of 20 feature values ​​to 9 principal elements.

[0050] Depend on Figure 4 Analysis shows that the nine feature parameters that meet the conditions are mean, kurtosis, peak value, harmonic mean, mean frequency, E1, E2, E5, and E6, with weight values ​​of (0.21515, 0.2938, 0.317025, 0.35055, 0.100925, 0.319, 0.1441, 0.114, 0.1457), forming a 400*9 feature matrix as the target feature set for fault classification.

[0051] The various data features are weighted to enhance the role of certain fault features in fault classification. The two-dimensional distribution of the kernel principal component feature points corresponding to the four operating state categories is as follows: Figure 5 As shown. Finally, the fault feature set is divided into a training set of 200 groups, with 50 groups for each operating state, forming a 200*9 training feature matrix; and a test set of 200 groups, with 50 groups for each operating state, forming a 200*9 test feature matrix.

[0052] Step S3: Use an optimization algorithm to find the parameters and obtain the globally optimal radial basis kernel parameter g and penalty parameter C. Input the training feature matrix into the recognition model to complete the data training, and then input the test feature matrix into the recognition model to obtain the centrifugal pump multi-fault recognition result.

[0053] Depend on Figure 6 It can be seen that the fitness of the GA algorithm reaches its maximum value in the 24th generation, and the optimal solution obtained is g = 1.18, C = 45.63. The fault classification accuracy based on the GA-SVM model reaches 100%, and the model identification results are as follows. Figure 7 As shown.

[0054] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators, characterized in that, include: Step S1: Establish a multi-fault signal synchronous test bench for vertical centrifugal pump units to simulate centrifugal pump fault states. Implement four operating states: normal, rotor imbalance, rotor misalignment, and mechanical loosening. Collect five vibration acceleration signals and one vibration displacement signal under each operating state. Calculate the vibration intensity of each measuring point under each operating state and select the most sensitive signal measuring points that reflect the fault state. Step S2: Calculate the statistical indicators of the selected signal measurement points to obtain a feature set composed of several statistical indicators; use the weighted kernel principal component analysis method to standardize and reduce the dimensionality of the feature set to obtain the target feature set; divide the several groups of samples measured under the four operating states into a training set and a test set, wherein the training set constitutes a training feature matrix and the test set constitutes a test feature matrix. Step S3: Use an optimization algorithm to find the parameters and obtain the globally optimal radial basis kernel parameter g and penalty parameter C. Input the training feature matrix into the recognition model to complete the data training, and then input the test feature matrix into the recognition model to obtain the centrifugal pump multi-fault recognition result. The test bench mentioned in step S1 includes a centrifugal pump, a motor, and a pipeline system. The centrifugal pump is a single-stage, single-suction vertical centrifugal pump. The motor is mounted on the upper end of the centrifugal pump via a support. The motor shaft of the motor is connected to the pump shaft of the centrifugal pump. The centrifugal pump is started, stopped, and speed-controlled via a three-phase AC frequency converter. The pipeline system includes a check valve, an inlet valve, a vertical centrifugal pump, a frequency converter motor, an electromagnetic flowmeter, an outlet valve, and connecting pipes.

2. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 1, characterized in that, In step S1, the centrifugal pump synchronously collects the vibration acceleration of the motor bracket, the vibration displacement of the pump shaft in the X and Y directions, the vibration acceleration of the pump inlet flange, the vibration acceleration of the pump outlet flange, the vibration acceleration of the pump body, and the vibration acceleration of the bracket feet under four operating states, and collects more than or equal to 100 sets of data under each operating state.

3. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 1, characterized in that, The vibration acceleration was measured by an ICP-type uniaxial vibration acceleration sensor, and the vibration displacement was measured by an LD980-Y integrated eddy current sensor.

4. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 3, characterized in that: The formula for calculating vibration intensity is: ; The formula for calculating frequency domain signals is: ; The formula for calculating the amplitude spectrum is: ; The formula for calculating harmonic frequency is: ; The vibration intensity of the vibration displacement is given by the formula Calculated; The intensity of vibration acceleration is given by the formula Calculated; Where N is the total number of discrete signals. For the nth discrete velocity signal, To measure the vibration signal at N points, The value in the middle is the signal sampling frequency. greater than The smallest integer, greater than The largest integer.

5. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 1, characterized in that, The feature set in step S2 includes several time-domain indicators, several frequency-domain indicators, and several time-frequency-domain energy feature indicators.

6. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 1, characterized in that, In step S2, weighted kernel principal component analysis is used for standardization and dimensionality reduction. The feature vectors are sorted from high to low according to the principle that the separability index is greater than 0.85, and the target feature set is obtained.

7. The method for identifying multiple faults in centrifugal pumps based on vibration intensity and statistical indicators according to claim 1, characterized in that, In step S3, a genetic optimization algorithm is used to optimize the parameters.

Images