Centrifugal pump cavitation fault diagnosis method based on information fusion

Abstract

The invention discloses an information-fused centrifugal pump cavitation fault diagnosis method, which comprises the following steps of: data acquisition: installing an acoustic sensor and an acceleration sensor at key parts of a centrifugal pump, and acquiring an acoustic signal and a vibration signal; data preprocessing: sequentially carrying out noise filtering and normalization processing on the collected sound signals and vibration signals to obtain preprocessed signals; enhancing fusion, and performing weighted fusion coefficient adaptive adjustment on the preprocessed signal; clustering fusion: the weighted fusion signals are clustered according to data attributes, fusion of different sensor data is completed in combination with data difference and complementarity, and fused information is obtained; performing MFDFA analysis, inputting the online preprocessed data into a digital-analog linkage prediction model, and outputting a predicted value of the residual life of the equipment; and fault judgment: inputting the online preprocessed data into the digital-analog linkage prediction model, and outputting a predicted value of the residual life of the equipment. According to the method, the cavitation fault of the centrifugal pump can be accurately recognized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of equipment state monitoring and fault diagnosis, more particularly to an information fusion centrifugal pump cavitation fault diagnosis method. BACKGROUND

[0002] The centrifugal pump is a general mechanical device that realizes fluid delivery through the centrifugal force generated by the impeller rotation, and is widely used in industrial scenes such as oil exploitation, chemical reaction, and urban water supply. During the long-term operation of the centrifugal pump, the local pressure in the pump is lower than the saturated vapor pressure of the fluid, which is prone to cavitation phenomenon. Cavitation can cause the pump body to vibrate intensively, the running noise to increase, and the delivery efficiency to decrease. In severe cases, it can cause erosion of the impeller surface, cracking of the pump shell, and even cause the equipment to stop or safety accidents, so it is of great significance to accurately diagnose the cavitation fault of the centrifugal pump. When the centrifugal pump is running, cavitation problems are prone to occur, and cavitation often brings great problems, causing the working efficiency of the centrifugal pump to decrease, the pump body to vibrate intensively, and the noise to increase. In severe cases, it can cause the equipment to malfunction or even safety problems. The classical mechanical fault diagnosis method mainly measures one signal (such as a vibration signal, a flow signal) for processing and analysis, but it is often difficult to detect whether the centrifugal pump has cavitation fault, and misdiagnosis or missed diagnosis may occur. SUMMARY

[0003] In view of the problems in the prior art, the purpose of the present application is to provide an information fusion centrifugal pump cavitation fault diagnosis method, which can accurately identify the cavitation fault of the centrifugal pump.

[0004] The present application adopts the following technical solutions:

[0005] An information fusion centrifugal pump cavitation fault diagnosis method, comprising the following steps:

[0006] S1, data acquisition, installing an acoustic sensor and an acceleration sensor at key positions of the centrifugal pump, and collecting the acoustic signals and vibration signals generated by the pump body when the centrifugal pump is running;

[0007] S2, data preprocessing, sequentially performing noise filtering and normalization processing on the acoustic signals and vibration signals collected in step S1 to obtain preprocessed signals;

[0008] S3, enhanced fusion, introducing an adaptive weighting algorithm to adaptively adjust the weighting fusion coefficient of the preprocessed signals in step S2, specifically including weight initialization, sensor information online learning and weight adjustment, to obtain a weighted fusion signal;

[0009] S4, cluster fusion: clustering the weighted fusion signal obtained in step S3 according to data attributes, and combining data difference and complementarity to complete the fusion of different sensor data to obtain fused information;

[0010] S5, MFDFA analysis, input online pretreatment data into digital-analog linkage prediction model, output device residual life prediction value;

[0011] S6, fault judgment, input online pretreatment data into digital-analog linkage prediction model, output device residual life prediction value.

[0012] Beneficial effects

[0013] The centrifugal pump is extracted by the neural network, and the centrifugal pump cavitation fault characteristics are obtained by using the information fusion method, so that the centrifugal pump running state analysis is more accurate and comprehensive, and the early centrifugal pump cavitation fault recognition accuracy is improved.

[0014] By using the recurrent neural network to fuse multi-source information, the weight of sound signals and vibration signals in correctly identifying early faults under centrifugal pump cavitation failure can be more accurately allocated, and the problem of early fault recognition accuracy caused by weight allocation problem can be avoided.

[0015] The centrifugal pump vibration characteristics and sound characteristics are extracted, and a large amount of prior knowledge, signal processing theory and practical experience are not required, so that the stability and practicability of the intelligent feature extraction method are improved. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 It is an installation schematic diagram of the tester of an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0018] As shown in the figure, the present application discloses an information fusion centrifugal pump cavitation fault diagnosis method, which comprises the following steps:

[0019] S1, data acquisition, sound sensors and acceleration sensors are installed at key positions of the centrifugal pump, and sound signals and vibration signals generated by the pump body are collected when the centrifugal pump is running;

[0020] Sound signal collection: There are some factors such as rotation of shaft and impeller and movement of fluid during the operation of centrifugal pump, which can produce sound. The sound will be different when the centrifugal pump operates under different working conditions, especially when the fault occurs, the sound signal decomposition will produce specific frequency components and amplitude fluctuations. The instruments commonly used for sound signal collection are ultrasonic sensors and microphones. The former is suitable for high-frequency sound signal collection and can detect faults such as cavitation, wear and imbalance of centrifugal pump. The latter is suitable for low-frequency sound signal collection and can capture low-frequency vibration and airflow noise of the pump.

[0021] Vibration signal collection: Vibration signal can reflect the running state of centrifugal pump. By analyzing the vibration signal, whether mechanical failure, bearing damage and other problems occur can be analyzed. The instrument commonly used for vibration signal collection is acceleration sensor. The sensor is installed at a suitable position to collect signals.

[0022] S2, data preprocessing: After the sound signal and vibration signal collection is completed, the signal needs to be preprocessed, so that the subsequent data analysis is accurate and effective. Because the working environment of centrifugal pump is relatively harsh, there are other mechanical equipment sound, fan sound and electrical interference in the working environment. Therefore, the collected signal needs to be filtered to remove external interference and noise.

[0023] The sound signal and vibration signal collected in step S1 are sequentially subjected to noise filtering and normalization processing to obtain the preprocessed signal;

[0024] S3, strengthening fusion: introducing an adaptive weighting algorithm to adaptively adjust the weighting fusion coefficient of the preprocessed signal in step S2, which specifically includes weight initialization, sensor information online learning and weight adjustment, to obtain a weighted fusion signal;

[0025] S4, clustering fusion: clustering the weighted fusion signal obtained in step S3 according to data attributes, and combining data difference and complementarity to complete the fusion of different sensor data to obtain the fused information;

[0026] S5, MFDFA analysis: inputting the online preprocessed data into the digital-analog interactive prediction model to output the residual life prediction value of the device;

[0027] S6, fault judgment: inputting the online preprocessed data into the digital-analog interactive prediction model to output the residual life prediction value of the device.

[0028] Sound signal and vibration signal are two important signals that can reflect the running state of the pump body. Vibration signal can reflect the mechanical movement and vibration of the device, and sound signal can more sensitively capture the change of fluid flow state. Therefore, based on the fusion analysis of the two signals, the present application can provide more reliable information for the cavitation fault diagnosis of centrifugal pump.

[0029] The present application has high diagnostic accuracy: the present application fuses acoustic signals (sensitive to capture the characteristics of early bubble collapse of cavitation) and vibration signals (reflecting the characteristics of mechanical damage in the later stage of cavitation), and extracts multi-fractal features by combining the MFDFA algorithm, so that the whole cycle of cavitation fault can be recognized - the recognition accuracy of early cavitation, the recognition accuracy of medium-term cavitation, which is greatly improved compared with the single vibration signal diagnosis accuracy;

[0030] The weight distribution is accurate: the weight of the two types of signals is dynamically adjusted by the adaptive weighting algorithm, avoiding early fault misdiagnosis or late fault misdiagnosis caused by fixed weight, and the weight adjustment error is small.

[0031] It is practical: the present application does not need to rely on a large amount of prior knowledge (such as manually labeled fault samples) or complex signal processing theory, and through the standardized sensor installation, preprocessing and analysis process, ordinary technical personnel can operate after simple training; at the same time, the algorithm can be integrated into the existing equipment monitoring system (such as PLC control system), realizing real-time diagnosis and early warning, and reducing the equipment maintenance cost.

[0032] In step S2, the noise filtering adopts at least one of a low-pass filter, a high-pass filter or a band-pass filter; wherein the low-pass filter is used to suppress high-frequency noise irrelevant to the equipment running frequency, and is especially suitable for removing high-frequency components irrelevant to the equipment running frequency;

[0033] The high-pass filter is used to suppress low-frequency noise in the working environment, and is especially suitable for removing low-frequency noise in the working environment;

[0034] The band-pass filter retains the effective signal within a certain frequency range and suppresses the noise components outside the range. When processing the sound signal and vibration signal of the centrifugal pump, using the band-pass filter can realize the retention of the signal components related to the frequency of the centrifugal pump.

[0035] The type, installation position and other factors of the sensor may cause certain differences in the amplitude and energy value of the collected signal. In order to avoid the influence of these differences on the analysis result, it is necessary to normalize the data. In step S2, the normalization processing includes at least one of amplitude normalization, time domain normalization or standardization; wherein the amplitude normalization adjusts the signal amplitude to be in the same range, so that the data has comparability; the time domain normalization uniformly adjusts the time axis of the periodic signal, and eliminates the time length difference; the standardization makes the signal have the same scale and mean value through the calculation of "(signal value-signal mean value) / signal standard deviation".

[0036] In step S3, the calculation formula of weight initialization is:

[0037]

[0038] wherein, is the initialization of weight value; is the information of the th sensor; is the observation model coefficient; is the sensor observation value, on the basis of which the online learning of sensor information data and weight adjustment are carried out.

[0039] The weight adjustment formula is:

[0040]

[0041] wherein, is the weight adjustment of the sensor; is the weighting coefficient; is the weighted estimation equation.

[0042] After the online learning and weight adjustment are completed, the adaptive adjustment of the weighted fusion coefficient is carried out according to the adjusted weight for the data of each sensor, and the calculation formula is:

[0043]

[0044] wherein, is the adaptive adjustment of the weighted fusion coefficient; is the adjustment times; is the time domain.

[0045] In step S4, the calculation formula of clustering is:

[0046]

[0047] wherein, is the clustering of the same attribute information data; is the data membership function; is the clustering center;

[0048] After the data clustering is completed, the data of different sensors are fused, and when the fusion is carried out, the difference and complementarity of the data need to be considered, and a suitable method is selected for the fusion processing, and the fusion calculation formula is:

[0049]

[0050] wherein, is the information data fusion processing; is the fusion iteration; is the matrix initialization; is the initialization condition, and the fused information is .

[0051] The MFDFA algorithm is a multiple analysis algorithm, and the signal after weighted fusion of the sound signal and the vibration signal , i=1,2,···, The analysis is performed, the cumulative deviation of the signal is calculated, a new signal can be obtained, and the cumulative deviation calculation formula of the signal is as follows:

[0052]

[0053] In the formula, is the mean value of the signal time sequence, then the sub-interval division is performed, and the division steps are as follows:

[0054] The is divided into sub-intervals with equal length, the sub-interval length is , and there are sub-intervals, is the data length of , in order to make be an integer, it is necessary to divide again in reverse, and finally 2 sub-intervals are obtained.

[0055] In an embodiment of the present application, in step S5, the mean square error calculation comprises:

[0056] The data of the i-th sub-interval is fitted by using the least square method, and is obtained;

[0057] When i=1,2,··· , the mean square error is

[0058]

[0059] When i= =1+1, +2,···2 , the mean square error is

[0060]

[0061] The average value of the mean square errors of the 2 intervals is obtained, and the order fluctuation function is obtained:

[0062]

[0063] If the signal has a self-similar feature, the fluctuation function is equal to ​​​The power law relation is:

[0064] .

[0065] Herein is the generalized Hurst index, representing the slope of the least square fitting to the scale , which varies with the change of , then has the characteristics of multiple fractals, and vice versa, has the characteristics of single fractal;

[0066] The generalized Hurst index and the relationship between the quality index is:

[0067]

[0068] By using Legendre transform, the singular index of the signal and the multiple fractal spectrum :

[0069]

[0070]

[0071] The characteristic vectors , , , and are obtained by calculation, and the extreme point reflects the degree of irregularity of the signal, if it is large, the irregularity degree of the signal is high, and the fluctuation is violent; if the fractal spectrum width is large, the multiple fractal characteristics of the signal are obvious, the left end point and the right end point correspond to the singular index of the maximum and minimum fluctuation respectively, and the difference between the left and right end points of the multiple fractal spectrum reflects the proportion of the peak value of the violent fluctuation and the peak value of the stable fluctuation, and the greater the proportion is, the greater the signal fluctuation is. The , , , and are obtained by calculation as the characteristic vectors of the signal, so as to judge whether the gas cavitation fault of the centrifugal pump occurs.

[0072] The above merely provides the preferred embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art, according to the technical solution of the present application and the improved concept thereof, makes equivalent replacement or change within the technical range disclosed by the present application, and should be covered within the protection scope of the present application.

Claims

1. A centrifugal pump cavitation fault diagnosis method based on information fusion, characterized in that: Includes the following steps: S1. Data acquisition: Install acoustic sensors and acceleration sensors at key parts of the centrifugal pump to collect acoustic and vibration signals generated by the pump body during operation. S2. Data preprocessing: The acoustic signal and vibration signal acquired in step S1 are subjected to noise filtering and normalization in sequence to obtain the preprocessed signal. S3. Enhance fusion by introducing an adaptive weighting algorithm to adaptively adjust the weighted fusion coefficients of the signal preprocessed in step S2. This includes weight initialization, online learning of sensor information and weight adjustment to obtain a weighted fusion signal. S4. Clustering and Fusion: The weighted fusion signals obtained in step S3 are clustered according to data attributes, and then the data differences and complementarities are combined to complete the fusion of data from different sensors and obtain fused information. S5 and MFDFA analysis inputs online preprocessed data into the numerical simulation-linked prediction model and outputs the predicted value of the remaining life of the equipment. S6. Fault diagnosis: Input the online preprocessed data into the digital-analog linkage prediction model and output the predicted value of the remaining life of the equipment.

2. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: In step S2, the noise filtering employs at least one of a low-pass filter, a high-pass filter, or a band-pass filter; wherein, the low-pass filter is used to suppress high-frequency noise unrelated to the operating frequency of the equipment, the high-pass filter is used to suppress low-frequency noise in the working environment, and the band-pass filter is used to retain signal components related to the operating frequency of the centrifugal pump.

3. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: In step S2, the normalization process includes at least one of amplitude normalization, time-domain normalization, or standardization; wherein amplitude normalization adjusts the signal amplitude to make it within the same range, making the data comparable; time-domain normalization uniformly adjusts the time axis of periodic signals to eliminate time length differences; standardization uses the calculation of "(signal value - signal mean) / signal standard deviation" to make the signals have the same scale and mean.

4. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: In step S3, the formula for calculating the weight initialization is: In the formula, To initialize the weight values; For the first Information from each sensor; These are the coefficients of the observation model; Based on sensor observations, online learning and weight adjustment of sensor information data are performed.

5. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 4, characterized in that: The weight adjustment formula is: In the formula, Weight adjustment for sensors; These are weighting coefficients; This is the weighted estimation equation.

6. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: After completing online learning and weight adjustment, the weighted fusion coefficients of each sensor's data are adaptively adjusted based on the adjusted weights. The calculation formula is as follows: In the formula, For adaptive adjustment of the weighted fusion coefficients; To adjust the number of times; For the time domain.

7. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: In step S4, the clustering calculation formula is as follows: In the formula, Clustering data with the same attribute; For data membership functions; As cluster center; After data clustering is completed, data from different sensors are fused. During fusion, the differences and complementarities of the data need to be considered, and the fusion calculation formula is as follows: In the formula, For information and data fusion processing; For integration and iteration; Initialize the matrix; As an initialization condition, the fused information is .

8. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 1, characterized in that: In step S5, the formula for calculating the cumulative signal deviation is: In the formula, The mean of the signal time series is... Divide into subintervals of equal length, with each subinterval having a length of [length missing]. There are a total of Sub-intervals, for Data length, from Divide it again in reverse, and finally get 2. Sub-intervals.

9. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 8, characterized in that: In step S5, the mean square error calculation includes: The least squares method is used to fit the first... Data from each sub-interval, yielding... ; when =1, 2, ... When, the mean square error is when = +1, +2, ... 2 The mean square error is 2 By calculating the average mean square error over several intervals, we can obtain... First-order wave function: If the signal exhibits self-similarity, then the wave function and The power-law relationship is as follows: 。 10. The centrifugal pump cavitation fault diagnosis method based on information fusion according to claim 9, characterized in that: In steps S5-S6, The generalized Hurst exponent characterizes the least squares fit. With scale The slope, Follow It changes with the changes, then It exhibits multiple fractal characteristics, otherwise it exhibits single fractal characteristics; Generalized Hurst index With quality index The relationship is: The singular exponent of a signal can be obtained using the Legendre transform. and multifractal spectrum : The feature vector is obtained through calculation. , , , and Extreme point This reflects the degree of signal irregularity. A larger fractal width indicates a higher degree of signal irregularity and more drastic fluctuations; When the signal is large, its multifractal characteristics become more pronounced, especially at the left endpoint. and right endpoint These correspond to the singularity indices with the largest and smallest fluctuations, respectively, and the difference between the left and right endpoints of the multifractal spectrum. It reflects the proportion of peaks with drastic signal fluctuations to peaks with stable fluctuations; the larger the proportion, the greater the signal volatility.

Images