Personalized variational modal decomposition method, cavitation calibration method, system, device and medium

Abstract

The application discloses a kind of individualized variational modal decomposition method, cavitation calibration method, system, equipment and medium;Individualized variational modal decomposition method carries out variational modal decomposition to original cavitation vibration signal, and obtains final denoising signal and best mode by adaptive wavelet threshold denoising;Cavitation calibration method will the root mean square value of best mode and effective cavitation residual value as discrete state set and carry out smoothing processing;First, second tangent equation is constructed, and the intersection horizontal coordinate of two tangent equations is solved as nascent cavitation point;Cavitation point is used as critical judgment limit, and sample calibration is carried out to final denoising signal, and the final denoising signal data set with high confidence label is output;The system includes pipeline system, data acquisition system and control system, and the pipeline system provides physical characteristics;Data acquisition system converts the physical characteristics into digital signal and outputs to control system;Control system is controlled by algorithm or artificial initiative electric equipment.

Description

Technical Field

[0001] This invention relates to the technical field of rotating machinery fault diagnosis, and in particular to a personalized variational mode decomposition method, cavitation calibration method, system, equipment and medium. Background Technology

[0002] Centrifugal pumps are widely used fluid transport equipment in industrial fields. During operation, they are highly susceptible to cavitation faults, leading to impeller cavitation, increased vibration, decreased efficiency, and even equipment damage. Cavitation fault diagnosis technology based on centrifugal pump cavitation vibration signals has become an important method for diagnosing rotating machinery faults due to its advantages such as non-contact operation and strong real-time performance. However, centrifugal pump cavitation vibration signals are usually superimposed with background noise, fluid pulsation, and mechanical vibration interference, resulting in a low signal-to-noise ratio. Traditional denoising algorithms struggle to effectively extract weak cavitation impact features, limiting the accuracy of early fault identification.

[0003] While existing variational mode decomposition (VMD) methods can decompose non-stationary signals, the number of mode decompositions they produce is limited. and penalty factor All of these require manual pre-setting, making it difficult to adaptively optimize for different working conditions. This can easily lead to modal aliasing or over-decomposition, with cavitation features being masked by noise. At the same time, the identification of nascent cavitation points has long relied on empirical curves or subjective thresholds, lacking objective quantitative calibration standards based on vibration energy. This results in high noise and poor balance in the labels of subsequent machine learning sample sets, affecting the performance of the classifier.

[0004] Existing centrifugal pump anti-cavitation systems are mostly offline test benches, which can only complete benchmark performance tests. They generally lack real-time signal processing and closed-loop control capabilities, and cannot automatically trigger pressurization, alarm, or shutdown linkage protection in the early stage of cavitation, making it difficult to meet the needs of safe and stable operation in industrial sites.

[0005] In summary, there is an urgent need for a signal processing method that can adaptively extract weak features, a calibration criterion for objectively calibrating cavitation faults, and an anti-cavitation system with real-time closed-loop control capabilities. Summary of the Invention

[0006] To address the problems in centrifugal pump cavitation fault feature extraction based on vibration signals, such as the difficulty of adaptive optimization of conventional denoising algorithm parameters leading to weak features being easily interfered with by strong background noise, the lack of objective quantitative standards for identifying initial cavitation points, and the lack of real-time anti-cavitation closed-loop control in existing centrifugal pump systems, this invention provides a personalized variational mode decomposition method, cavitation calibration method, system, equipment, and medium.

[0007] To effectively solve the above problems, the present invention provides the following technical solution:

[0008] In a first aspect, the present invention provides a personalized variational mode decomposition method, the method comprising the following steps:

[0009] Step 1: Acquire raw cavitation vibration signals while the centrifugal pump is running. ;

[0010] Step 2: Process the original cavitation vibration signal Perform variational mode decomposition based on the golden section optimization and output the set of extracted modes;

[0011] Step 3: Calculate the Hurst exponent of each sub-mode in the extracted mode set using the R / S analysis method. And based on this, adaptively distinguish the information-dominant mode. With noise-dominant mode ;

[0012] Step 4: For the noise-dominant mode Adaptive wavelet thresholding based on cross-correlation coefficient is used to obtain the denoised noise-dominant mode. ;

[0013] Step 5: Transform the information into a dominant mode. With the noise-dominant mode after denoising The summation outputs the final denoised signal. .

[0014] Further, step 2 specifically involves: using the original cavitation vibration signal as a residual. And calculate its power spectrum. Extract the power spectrum amplitude with the largest value. Local maxima peak frequency And generate a set of candidate penalty factors in its neighborhood. Then, construct an effective candidate penalty factor set based on the candidate penalty factor set. and initial search interval Construct a multi-objective dynamic fitness function, iteratively optimize within the initial search interval using the golden section method, and determine the optimal penalty factor; perform VMD based on the optimal penalty factor to obtain the optimal submode. And add to the modality set The reconstructed signal is obtained. and output the extracted mode set;

[0015] Among them, the It is a local maximum and satisfies:

[0016]

[0017] The candidate penalty factor set This is obtained through the following method: at each peak frequency Candidate penalty factors are generated within ±30% of the neighborhood. :

[0018]

[0019] in, ;

[0020] For the candidate penalty factor After removing duplicate values ​​and sorting them in ascending order, the set of candidate penalty factors can be obtained. :

[0021]

[0022] in, ;

[0023] The set of effective candidate penalty factors Specifically, the cross-correlation coefficient of ;

[0024] If all If all are less than 0.9, then take the value directly. The maximum value corresponding to As the optimal penalty factor Otherwise, continue with the following steps;

[0025] The initial search interval Specifically:

[0026]

[0027] in, ;

[0028] The multi-objective dynamic fitness function is specifically as follows:

[0029]

[0030] in, Measuring the impact characteristic strength of a mode, To measure the contribution of the current mode to the residual signal energy, To measure the information correlation between the modal and residual signals, For any penalty factor The corresponding modal extraction formula is: ; and The weights are as follows:

[0031]

[0032]

[0033] in, This is the current residual signal-to-noise ratio, as detailed below:

[0034]

[0035] in, The value is a very small positive number that includes the system's floor noise level; in this embodiment, it is preferably 10. -6 .

[0036] This multi-objective dynamic fitness function simultaneously considers the prominence of cavitation impact characteristics, the integrity of energy retention, and the correlation with residual information, ensuring that the search direction is directed towards the feature that best characterizes cavitation. convergence.

[0037] Calculate separately and The fitness of the search range is determined, and the search range is updated based on the comparison results. :

[0038] like , like ,

[0039] Repeat the above steps, stopping the iteration when any of the following conditions is met: current interval length. The number of iterations reached 50; the fitness changed after 5 consecutive iterations. .

[0040] After stopping iteration, take the current search interval. The midpoint is the optimal penalty factor. :

[0041]

[0042] Step 2.5: Perform VMD based on the optimal penalty factor to obtain the optimal submode. And add to the modality set The reconstructed signal is obtained, and the extracted mode set is output.

[0043] For the present ,by implement VMD yields the optimal submode. And add to the modality set :

[0044]

[0045]

[0046]

[0047] Sum all extracted sub-modes to obtain the reconstructed signal. :

[0048]

[0049] Calculate the original cavitation vibration signal respectively With reconstructed signal Find the power spectrum and calculate the correlation coefficient. :

[0050]

[0051]

[0052] like Then update the residuals. Then return to step 2.1 and repeat the above steps until... ;when At that time, output the set of extracted modes. .

[0053] Step 3: Process the extracted mode set output in Step 2 The Hurst exponent for each submode was calculated using the R / S (rescaled range) analysis method. This allows for adaptive differentiation between information-dominant and noise-dominant modes. A classification threshold of 0.5 is used.

[0054] (1) If This indicates that the sub-mode has long-term relevance and is therefore classified as an information-dominant mode. All selected information-dominant modes are then sequentially denoted as... And retain;

[0055] (2) If This indicates that the submode exhibits anti-persistence and is therefore classified as a noise-dominant mode. All noise-dominant modes are denoted sequentially as follows: Then proceed to the next step of processing.

[0056] Step 4: Analyze the noise-dominant mode obtained in Step 3. Adaptive wavelet thresholding denoising based on cross-correlation coefficient is performed. The specific calculation process is as follows:

[0057] Calculate the noise-dominant mode Compared with the original cavitation vibration signal Cross correlation coefficient :

[0058]

[0059] The “dB5” wavelet basis was selected for the above. Perform 5-level discrete wavelet decomposition to obtain each decomposition level (denoted as the 1st level). layer, wavelet coefficients .

[0060] Extract the first Layer wavelet coefficients The median absolute deviation is used to estimate the true noise standard deviation, and a system is constructed based on... Adaptive threshold :

[0061]

[0062] in, The length of the wavelet coefficients in the current decomposition layer.

[0063] An improved smooth semi-soft thresholding function is used for wavelet coefficients at each layer. Perform point-by-point calculations to obtain the denoised wavelet coefficients. :

[0064]

[0065] Based on the denoised wavelet coefficients Inverse wavelet transform is performed to reconstruct the signal, obtaining the denoised noise-dominant mode. .

[0066] Step 5: Add the dominant mode of information retained in Step 3 to the dominant mode of noise after denoising in Step 4 to obtain the final denoised signal. :

[0067]

[0068] Furthermore, the personalized variational mode decomposition method also includes a step of selecting the optimal mode.

[0069] Specifically, calculate each information-dominant mode. Compared with the original cavitation vibration signal Cross correlation coefficient :

[0070]

[0071] The cross-correlation coefficient of the information-dominant mode Cross-correlation coefficient with the noise-dominant mode in the aforementioned steps Perform a global comparison and select the optimal mode corresponding to the global maximum value (if the global maximum value comes from...). If the set is defined, then the optimal mode is the corresponding information-dominant mode. If the global maximum value comes from If the set is selected, the optimal mode is the corresponding denoised noise-dominant mode. ), denoted as This mode is the characteristic component that best represents the cavitation impact intensity after filtering out high-frequency random noise. It is output and saved for subsequent vibration energy calculation.

[0072] Secondly, embodiments of the present invention also provide a cavitation calibration method based on the personalized variational mode decomposition, comprising the following specific steps:

[0073] Step 1: Conduct an offline variable-condition benchmark test of the centrifugal pump. Under rated operating conditions, maintain constant flow rate and speed, and gradually reduce the inlet pressure by adjusting valves or a vacuum pump until the head decreases by 3%. During this pressure reduction process, data are collected. Original cavitation vibration signals at different pressure nodes were recorded, along with the corresponding inlet pressures. The personalized variational mode decomposition method described in the first aspect is invoked to obtain the final denoised signal at each pressure node. and optimal mode Calculate the optimal mode at each node. of (Root Mean Square) value and corresponding (Effective cavitation margin) value:

[0074]

[0075]

[0076] in, This refers to the inlet pressure of the centrifugal pump. The saturated vapor pressure of the liquid is expressed in Pa. The inlet flow rate is m / s; The density of the liquid is kg / m³. 3 ; The acceleration due to gravity is 9.8 m / s². 2 .

[0077] It is used as a set of discrete states that reflect the cavitation evolution process. Since the test is a step-down process, the elements in the set are arranged according to... The values ​​are arranged in a monotonically decreasing order.

[0078] Step 2: Smooth the discrete state set using a Savitzky-Golay (SG) polynomial conformal filter. Let the half-width of the filter window be... (Preferred 2~7), for any node (represent The smoothed curve function value is extracted and calculated synchronously using the following discrete convolution formula. and its first derivative Second derivative :

[0079]

[0080]

[0081]

[0082] in, These are the standard convolution coefficients for smoothing and first- and second-order derivatives of the SG filter, respectively, determined by a preset polynomial order (preferably 2-4) and the half-width of the filter window. Determined by the least squares method.

[0083] By sliding convolution over each discrete point, a smooth and continuous cavitation characteristic curve function is constructed. And its derivative. For the boundary points at the beginning and end of the sequence, the continuation is performed using mirror symmetry or edge replication methods. For example... Figure 3 As shown, the cavitation characteristic curve function Manifested as x-axis A two-dimensional curve with the vertical axis as the ordinate.

[0084] Step 3: Define the tangent point search feature index, namely the relative rate of change of the curve slope index. ( (where the value is a very small positive number containing the system's base noise level), the first tangent equation and the second tangent equation are constructed using an adaptive double-ended tangent anchoring algorithm.

[0085] The specific implementation steps of this algorithm are as follows:

[0086] (1) In non-cavitation high Region (preferably in the set of discrete states) Searching for the first 1 / 3 of the data segment of the sequence minimum point Construct the first tangent equation (correspond Figure 3 Tangent line ① in the middle), slope ,intercept .

[0087] (2) At low cavitation characteristic curves where the cavitation characteristic curve increases sharply Region (preferably in the set) Searching for the last third of the data segment of the sequence Maximum point Construct the equation of the second tangent line. (correspond Figure 3 Tangent line ② in the middle), slope ,intercept .

[0088] Step 4: Find the intersection point of the two tangents obtained in Step 3 (i.e., Figure 3 (Point A in the diagram) x-coordinate :

[0089]

[0090] That is, the initial cavitation point .

[0091] Step 5: Based on the initial cavitation point The result obtained in step one Final denoised signals under different pressure nodes Perform sample calibration: If The final denoised signal corresponding to this node Marked as "non-empty"; if The final denoised signal corresponding to this node This is labeled "cavitation fault". This hard boundary partition is intended to provide a clear binary supervision signal for the machine learning classifier, thereby enhancing the classifier's sensitivity to nascent cavitation features.

[0092] To ensure sample balance, an equal number of samples are randomly drawn from the two types of final denoised signals. Finally, a final denoised signal dataset with high-confidence labels is output for offline training of the subsequent machine learning classifier.

[0093] Thirdly, the present invention also provides a centrifugal pump anti-cavitation system based on the personalized variational mode decomposition method and cavitation calibration method, comprising: a pipeline system, a data acquisition system, and a control system; wherein, the pipeline system provides physical characteristics including fluid circulation and adjustable operating conditions through sensors; the data acquisition system converts the physical characteristics into digital signals and outputs them to the control system; the control system receives the digital signals and controls the electrical equipment in the pipeline system in real time through algorithms or manually.

[0094] Furthermore, the control system includes an intelligent console 19, which embeds an edge computing module and a programmable logic control module. The edge computing module is used to implement the personalized variational mode decomposition method and the cavitation calibration method, and to receive vibration and pressure signals transmitted by the data acquisition system and output the system operation status identification result. The programmable logic control module controls the equipment in the pipeline system based on the identification result. The pipeline system includes a pressure stabilizing tank 5, a cavitation tank 11, and a pipe connecting the two to form a closed loop, as well as a test pump 2, an air compressor 13, a vacuum pump 16, an outlet solenoid valve 4, a pressure stabilizing tank pressure relief solenoid valve 6, a first manual valve 7, a loop solenoid valve 9, a second manual valve 10, a booster solenoid valve 12, a vacuum solenoid valve 14, a cavitation tank pressure relief solenoid valve 17, and an inlet solenoid valve 18 connected to the test pump 2. The data acquisition system includes a pressure sensor 3, an electromagnetic flowmeter 8, an inlet pressure sensor 20, a unidirectional vibration acceleration sensor 21, and a dynamic signal acquisition instrument 22 connected to the test pump 2.

[0095] Fourthly, the present invention also provides a closed-loop control method based on the centrifugal pump anti-cavitation system, comprising the following two stages: an offline benchmark testing and model training stage and an online real-time monitoring and anti-cavitation linkage stage:

[0096] Specifically, the offline benchmark testing and model training phase includes:

[0097] Step A1: Extract the dimensionless time-domain feature parameters of the final denoised signal dataset;

[0098] Step A2: Train the classifier model offline and optimize its parameters on the training set, and deploy the classifier model that meets the accuracy requirements to the edge computing module;

[0099] The online real-time monitoring and anti-cavitation linkage stage specifically includes:

[0100] Step B1: During online monitoring, the edge computing module automatically denoises and extracts features from the collected signals, and inputs them into the deployed classifier model to output the status recognition results;

[0101] Step B2: If the identification result is "cavitation fault", the programmable logic control module controls the booster equipment to start to increase the inlet pressure;

[0102] Step B3: If the identification result is "non-cavitation" and the system is currently in pressurization operation, the programmable logic control module controls the shutdown of the pressurization equipment and enters the hydraulic stability lock-in period; if it is not in pressurization operation, it directly enters the next monitoring cycle.

[0103] Step B4: During system operation, if the continuously read inlet pressure reaches the maximum safe pressure threshold, the programmable logic control module will prioritize the emergency shutdown of the booster equipment, system alarm, and shutdown actions.

[0104] Furthermore, in the offline benchmarking and model training phase, the dimensionless time-domain feature parameters include waveform factor, peak factor, impulse factor, margin factor, and kurtosis factor.

[0105] During the online real-time monitoring and anti-cavitation linkage phase, the classifier model is a support vector machine model; during the hydraulic stability lock-in period, the system only records data, and the programmable logic control module blocks new pressurization trigger commands.

[0106] Compared with the prior art, the present invention has the following beneficial effects:

[0107] First, in terms of signal denoising, the personalized variational mode decomposition (IVMD) denoising method proposed in this invention achieves adaptive optimization of the variational mode decomposition penalty factor by constructing a multi-objective dynamic fitness function and the golden section method; by combining Hurst exponent classification and adaptive wavelet threshold denoising, the dependence of traditional algorithms on manual empirical parameters is reduced, and the effective extraction of weak cavitation impact features is achieved under strong background noise.

[0108] Second, regarding the calibration of primary cavitation, the method for automatic calibration and sample set construction of primary cavitation based on vibration energy proposed in this invention utilizes the discrete state set of effective cavitation margin and root mean square (NPSHa-RMS), combined with Savitzky-Golay polynomial conformal filtering and adaptive double-ended tangent anchoring algorithm, to overcome the limitations of traditional calibration relying on subjective experience, achieve objective quantitative definition of the critical point of primary cavitation, and provide training samples with high confidence labels for classifier models;

[0109] Third, regarding closed-loop system control, the centrifugal pump anti-cavitation system and its closed-loop control method constructed in this invention run a classifier model through an edge computing module to output identification results and link with a programmable logic control module; when a cavitation fault is identified, the booster device is automatically turned on to increase the inlet pressure, and when the pressure reaches the highest safety threshold, emergency shutdown of the booster device, system alarm, and shutdown actions are prioritized, thereby realizing online monitoring and active anti-cavitation of centrifugal pump cavitation faults. Attached Figure Description

[0110] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0111] Figure 1 The overall flowchart of a personalized variational mode decomposition (IVMD) method provided by the present invention;

[0112] Figure 2 The flowchart of the variational mode decomposition algorithm based on the golden section optimization provided by this invention;

[0113] Figure 3 This invention provides a schematic diagram of the automatic calibration principle for the initial cavitation point.

[0114] Figure 4 A schematic diagram of the overall structure of a centrifugal pump anti-cavitation system provided by the present invention;

[0115] Figure 5 A flowchart for offline training of the cavitation fault classifier of the system provided by the present invention;

[0116] Figure 6 A flowchart of the online monitoring and anti-cavitation linkage control of the system provided by the present invention;

[0117] Figure 7 This is a structural diagram of the electronic device provided by the present invention;

[0118] Among them, 1-Variable frequency speed-regulating three-phase asynchronous motor; 2-Test pump; 3-Outlet pressure sensor; 4-Outlet solenoid valve; 5-Pressure stabilizing tank; 6-Pressure stabilizing tank pressure relief solenoid valve; 7-First manual valve; 8-Electromagnetic flow meter; 9-Circuit solenoid valve; 10-Second manual valve; 11-Cavitation tank; 12-Boosting solenoid valve; 13-Air compressor; 14-Vacuum solenoid valve; 15-Gas-liquid separator; 16-Vacuum pump; 17-Cavitation tank pressure relief solenoid valve; 18-Inlet solenoid valve; 19-Intelligent control console; 20-Inlet pressure sensor; 21-Unidirectional vibration acceleration sensor; 22-Dynamic signal acquisition instrument. Detailed Implementation

[0119] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0120] In a first aspect, embodiments of the present invention provide a personalized variational mode decomposition (IVMD) method for denoising cavitation vibration signals, such as... Figure 1 As shown, the specific steps include:

[0121] Step 1: Physical signal acquisition.

[0122] Pressure sensors are installed on the inlet and outlet pipelines, and a unidirectional vibration accelerometer is installed on the centrifugal pump casing closest to the impeller. While the centrifugal pump is running, vibration signals are acquired at a high frequency (preferably 20480Hz, sampling time preferably 1-3s), which is used as the raw cavitation vibration signal containing cavitation characteristics and environmental noise. The original cavitation vibration signal Length is .

[0123] Step 2: Process the original cavitation vibration signal Perform variational mode decomposition (VMD) based on the golden ratio optimization, and output the set of extracted modes. For example... Figure 2 As shown, the specific steps include:

[0124] Step 2.1: Use the original cavitation vibration signal as a residual and calculate its power spectrum;

[0125] Specifically, the original cavitation vibration signal As residual Initialize the extracted mode set to empty ( Modal counting .

[0126]

[0127] For the present Perform a Fast Fourier Transform (FFT) to calculate its power spectrum. :

[0128]

[0129] Step 2.2: Extract the power spectrum amplitude of the largest value. The peak frequency of local maxima is determined, and a set of candidate penalty factors is generated in its neighborhood.

[0130] Specifically, the power spectrum is taken The largest amplitude Peak frequency ,in To ensure that the frequency corresponds to the actual energy concentration point rather than the noise peak, each It must be a local maximum and satisfy:

[0131]

[0132] To ensure a reasonable number and to cover the engineering experience range of optimal VMD parameters, at each peak frequency Candidate penalty factors are generated within ±30% of the neighborhood. :

[0133]

[0134] For the candidate penalty factor After removing duplicate values ​​and sorting them in ascending order, we obtain the set of candidate penalty factors. :

[0135]

[0136] Step 2.3: Perform VMD on each candidate penalty factor set to obtain temporary modes, calculate the cross-correlation coefficient between each temporary mode and the reference mode, and retain the penalty factors that meet the threshold conditions to construct an effective penalty factor set and an initial search interval;

[0137] Specifically, regarding the current Based on the set of candidate penalty factors Perform mode decomposition of each number of times VMD yields temporary modes :

[0138]

[0139] and each peak frequency exist =1.0 (i.e.) The temporary modes corresponding to the specified frequency () are used as the reference modes of that frequency group, denoted as . .

[0140] Calculate each temporary mode and the reference mode of its frequency group Cross-correlation coefficient between ,reserve of To form an effective set of candidate penalty factors .

[0141]

[0142] in, The formula for the cross-correlation coefficient is as follows:

[0143]

[0144] This yields a set of effective candidate penalty factors. :

[0145]

[0146] If all If all are less than 0.9, then take the value directly. The maximum value corresponding to As the optimal penalty factor Otherwise, continue with the following steps.

[0147] Pick The minimum and maximum values ​​form the initial search interval. :

[0148]

[0149] Step 2.4: Construct a multi-objective dynamic fitness function that takes into account the prominence of cavitation impact characteristics, the integrity of energy retention, and the correlation of residual information. Use the golden section method to iteratively optimize within the initial search interval to determine the optimal penalty factor.

[0150] Calculate the initial search interval interval length According to the golden ratio In the interval Insert two measuring points and :

[0151]

[0152] For the present ,by , Execute separately VMD yields the corresponding mode. and For any penalty factor The corresponding modal extraction formula is:

[0153]

[0154] Constructing a multi-objective dynamic fitness function :

[0155]

[0156] in, (Envelope spectrum kurtosis entropy) measures the intensity of the impact characteristics of a mode; the smaller the value, the more obvious the impact characteristics.

[0157] The envelope signal is obtained by performing a Hilbert transform on the modes. Then its power spectrum was calculated. Then normalize to obtain the probability distribution. The details are as follows:

[0158]

[0159]

[0160]

[0161] in, For modality The Hilbert transform.

[0162] (Energy retention rate) measures the contribution of the current mode to the residual signal energy; a higher value indicates a more important mode. The formula is:

[0163]

[0164] (Mutual information) measures the correlation between the modal and residual signals; a higher value indicates a stronger correlation. It is calculated using histogram estimation (preferably with 32 bins).

[0165]

[0166] in, For modality With current residual The normalized probability of the two-dimensional joint histogram. and These are the corresponding marginal probabilities.

[0167] and The weights are as follows:

[0168]

[0169]

[0170] in, This is the current residual signal-to-noise ratio, as detailed below:

[0171]

[0172] in, The value is a very small positive number that includes the system's floor noise level; in this embodiment, it is preferably 10. -6 .

[0173] This multi-objective dynamic fitness function simultaneously considers the prominence of cavitation impact characteristics, the integrity of energy retention, and the correlation with residual information, ensuring that the search direction is directed towards the feature that best characterizes cavitation. convergence.

[0174] Calculate separately and The fitness of the search range is determined, and the search range is updated based on the comparison results. :

[0175] like , like ,

[0176] Repeat the above steps, stopping the iteration when any of the following conditions is met: current interval length. The number of iterations reached 50; the fitness changed after 5 consecutive iterations. .

[0177] After stopping iteration, take the current search interval. The midpoint is the optimal penalty factor. :

[0178]

[0179] Step 2.5: Perform VMD based on the optimal penalty factor to obtain the optimal submode. And add to the modality set The reconstructed signal is obtained, and the extracted mode set is output.

[0180] For the present ,by implement VMD yields the optimal submode. And add to the modality set :

[0181]

[0182]

[0183]

[0184] Sum all extracted sub-modes to obtain the reconstructed signal. :

[0185]

[0186] Calculate the original cavitation vibration signal respectively With reconstructed signal Find the power spectrum and calculate the correlation coefficient. :

[0187]

[0188]

[0189] like Then update the residuals. Then return to step 2.1 and repeat the above steps until... ;when At that time, output the set of extracted modes. .

[0190] Step 3: Process the extracted mode set output in Step 2 The Hurst exponent for each submode was calculated using the R / S (rescaled range) analysis method. This allows for adaptive differentiation between information-dominant and noise-dominant modes. A classification threshold of 0.5 is used.

[0191] (1) If This indicates that the sub-mode has long-term relevance and is therefore classified as an information-dominant mode. All selected information-dominant modes are then sequentially denoted as... And retain;

[0192] (2) If This indicates that the submode exhibits anti-persistence and is therefore classified as a noise-dominant mode. All noise-dominant modes are denoted sequentially as follows: Then proceed to the next step of processing.

[0193] Step 4: Analyze the noise-dominant mode obtained in Step 3. Adaptive wavelet thresholding denoising based on cross-correlation coefficient is performed. The specific calculation process is as follows:

[0194] Calculate the noise-dominant mode Compared with the original cavitation vibration signal Cross correlation coefficient :

[0195]

[0196] The “dB5” wavelet basis was selected for the above. Perform 5-level discrete wavelet decomposition to obtain each decomposition level (denoted as the 1st level). layer, wavelet coefficients .

[0197] Extract the first Layer wavelet coefficients The median absolute deviation is used to estimate the true noise standard deviation, and a system is constructed based on... Adaptive threshold :

[0198]

[0199] in, The length of the wavelet coefficients in the current decomposition layer.

[0200] An improved smooth semi-soft thresholding function is used for wavelet coefficients at each layer. Perform point-by-point calculations to obtain the denoised wavelet coefficients. :

[0201]

[0202] Based on the denoised wavelet coefficients Inverse wavelet transform is performed to reconstruct the signal, obtaining the denoised noise-dominant mode. .

[0203] Step 5: Add the dominant mode of information retained in Step 3 to the dominant mode of noise after denoising in Step 4 to obtain the final denoised signal. :

[0204]

[0205] Furthermore, the personalized variational mode decomposition method also includes a step of selecting the optimal mode.

[0206] Specifically, calculate each information-dominant mode. Compared with the original cavitation vibration signal Cross correlation coefficient :

[0207]

[0208] The cross-correlation coefficient of the information-dominant mode Cross-correlation coefficient with the noise-dominant mode in the aforementioned steps Perform a global comparison and select the optimal mode corresponding to the global maximum value (if the global maximum value comes from...). If the set is defined, then the optimal mode is the corresponding information-dominant mode. If the global maximum value comes from If the set is selected, the optimal mode is the corresponding denoised noise-dominant mode. ), denoted as This mode is the characteristic component that best represents the cavitation impact intensity after filtering out high-frequency random noise. It is output and saved for subsequent vibration energy calculation.

[0209] Secondly, embodiments of the present invention also provide a cavitation calibration method based on the personalized variational mode decomposition, comprising the following specific steps:

[0210] Step 1: Conduct an offline variable-condition benchmark test of the centrifugal pump. Under rated operating conditions, maintain constant flow rate and speed, and gradually reduce the inlet pressure by adjusting valves or a vacuum pump until the head decreases by 3%. During this pressure reduction process, data are collected. Original cavitation vibration signals at different pressure nodes were recorded, along with the corresponding inlet pressures. The personalized variational mode decomposition method described in the first aspect is invoked to obtain the final denoised signal at each pressure node. and optimal mode Calculate the optimal mode at each node. of (Root Mean Square) value and corresponding (Effective cavitation margin) value:

[0211]

[0212]

[0213] in, This refers to the inlet pressure of the centrifugal pump. The saturated vapor pressure of the liquid is expressed in Pa. The inlet flow rate is m / s; The density of the liquid is kg / m³. 3 ; The acceleration due to gravity is 9.8 m / s². 2 .

[0214] It is used as a set of discrete states that reflect the cavitation evolution process. Since the test is a step-down process, the elements in the set are arranged according to... The values ​​are arranged in a monotonically decreasing order.

[0215] Step 2: Smooth the discrete state set using a Savitzky-Golay (SG) polynomial conformal filter. Let the half-width of the filter window be... (Preferred 2~7), for any node (represent The smoothed curve function value is extracted and calculated synchronously using the following discrete convolution formula. and its first derivative Second derivative :

[0216]

[0217]

[0218]

[0219] in, These are the standard convolution coefficients for smoothing and first- and second-order derivatives of the SG filter, respectively, determined by a preset polynomial order (preferably 2-4) and the half-width of the filter window. Determined by the least squares method.

[0220] By sliding convolution over each discrete point, a smooth and continuous cavitation characteristic curve function is constructed. And its derivative. For the boundary points at the beginning and end of the sequence, the continuation is performed using mirror symmetry or edge replication methods. For example... Figure 3As shown, the cavitation characteristic curve function Manifested as x-axis A two-dimensional curve with the vertical axis as the ordinate.

[0221] Step 3: Define the tangent point search feature index, namely the relative rate of change of the curve slope index. ( (where the value is a very small positive number containing the system's base noise level), the first tangent equation and the second tangent equation are constructed using an adaptive double-ended tangent anchoring algorithm.

[0222] The specific implementation steps of this algorithm are as follows:

[0223] (1) In non-cavitation high Region (preferably in the set of discrete states) Searching for the first 1 / 3 of the data segment of the sequence minimum point Construct the first tangent equation (correspond Figure 3 Tangent line ① in the middle), slope ,intercept .

[0224] (2) At low cavitation characteristic curves where the cavitation characteristic curve increases sharply Region (preferably in the set) Searching for the last third of the data segment of the sequence Maximum point Construct the equation of the second tangent line. (correspond Figure 3 Tangent line ② in the middle), slope ,intercept .

[0225] Step 4: Find the intersection point of the two tangents obtained in Step 3 (i.e., Figure 3 (Point A in the diagram) x-coordinate :

[0226]

[0227] That is, the initial cavitation point .

[0228] Step 5: Based on the initial cavitation point The result obtained in step one Final denoised signals under different pressure nodes Perform sample calibration: If The final denoised signal corresponding to this node Marked as "non-empty"; if The final denoised signal corresponding to this node This is labeled "cavitation fault". This hard boundary partition is intended to provide a clear binary supervision signal for the machine learning classifier, thereby enhancing the classifier's sensitivity to nascent cavitation features.

[0229] To ensure sample balance, an equal number of samples are randomly drawn from the two types of final denoised signals. Finally, a final denoised signal dataset with high-confidence labels is output for offline training of the subsequent machine learning classifier.

[0230] Thirdly, embodiments of the present invention also provide a centrifugal pump anti-cavitation system based on the personalized variational mode decomposition method and cavitation calibration method, such as... Figure 4 As shown in the accompanying drawings of this embodiment of the invention, the relative positions of the various devices are only schematic diagrams illustrating the pipeline connection logic and do not represent the actual height, proportions, or spatial constraints in the physical installation. The system specifically includes a pipeline system, a data acquisition system, and a control system; the pipeline system provides physical characteristics, including fluid circulation and adjustable operating conditions, through sensors; the data acquisition system converts these physical characteristics into digital signals and outputs them to the control system; the control system receives the digital signals and, through real-time linkage via algorithms or manual active control, controls the electrical equipment in the pipeline system.

[0231] Specifically, the piping system includes a pressure stabilizing tank 5, a cavitation tank 11, and pipes connecting the two to form a closed loop. The top of the pressure stabilizing tank 5 is connected to an exhaust pressure relief pipe, and the top of the cavitation tank 11 is connected to a gas phase pressure regulating main pipe. The two are controlled by the pressure stabilizing tank pressure relief solenoid valve 6 and the cavitation tank pressure relief solenoid valve 17, respectively. The air compressor 13 is connected to the gas phase pressure regulating main pipe of the cavitation tank 11 via a booster solenoid valve 12; the gas-liquid separator 15 is externally connected to a vacuum pump 16 and is connected to the gas phase pressure regulating main pipe of the cavitation tank 11 via a vacuum solenoid valve 14; manual valves (first manual valve 7, second manual valve 10) and various control solenoid valves (outlet solenoid valve 4, loop solenoid valve 9, inlet solenoid valve 18) are fixedly installed on the loop pipes, forming a closed loop after connection to the test pump 2.

[0232] The data acquisition system includes an outlet pressure sensor 3, an electromagnetic flowmeter 8, an inlet pressure sensor 20, a unidirectional vibration acceleration sensor 21, and a dynamic signal acquisition device 22. The unidirectional vibration acceleration sensor 21 is installed on the outer casing of the test pump 2, closest to the impeller. The outlet pressure sensor 3, inlet pressure sensor 20, and unidirectional vibration acceleration sensor 21 are all connected to the dynamic signal acquisition device 22 for unified data acquisition. The electromagnetic flowmeter 8 is directly connected to the intelligent control console 19.

[0233] The control system includes an intelligent control console 19. The intelligent control console 19 embeds an edge computing module (e.g., an industrial control computer equipped with a processor and memory) and a programmable logic control module (e.g., a PLC controller). The edge computing module implements the aforementioned personalized variational mode decomposition method and cavitation calibration method, and receives and processes various acquired data (including vibration signals and inlet / outlet pressure signals) transmitted by the dynamic signal acquisition instrument 22 in real time to output the current system operating status identification result. The programmable logic control module communicates with the edge computing module and, based on the identification result and the real-time feedback data from the electromagnetic flowmeter 8 and the inlet pressure sensor 20, integrates and controls the variable frequency speed-regulating three-phase asynchronous motor 1, air compressor 13, vacuum pump 16, and various solenoid valves in the pipeline system connected to the test pump 2.

[0234] During system operation, fluid flows from cavitation tank 11 to pressure stabilizing tank 5 via test pump 2, and then flows back from pressure stabilizing tank 5 to cavitation tank 11, forming a closed loop. The two tanks can eliminate pressure pulses during fluid flow, making the collected data more accurate. It should be noted that, to highlight the core control architecture of this system, the description of conventional auxiliary pipelines such as water injection into the loop is omitted in this embodiment; those skilled in the art should understand that before the system is started, the closed loop has been injected with experimental fluid at a predetermined level through conventional operations, and a gas-liquid free liquid surface that meets the pressure regulation requirements has been established in each tank, and it is in a sealed initial reference state.

[0235] Based on the centrifugal pump anti-cavitation system constructed above, the complete workflow of this embodiment is mainly divided into two stages: "offline benchmark testing and model training stage" and "online real-time monitoring and anti-cavitation linkage stage," which are described in detail below. Figure 5 and Figure 6 The flowchart shown below illustrates the specific implementation steps as follows:

[0236] Step 1: Perform offline variable operating condition benchmark testing on the test pump:

[0237] (1) Except for the pressure relief solenoid valve 6 of the pressure stabilizing tank, the pressure boosting solenoid valve 12, the vacuum solenoid valve 14 and the pressure relief solenoid valve 17 of the cavitation tank, open all other main circuit valves (including the outlet solenoid valve 4, the first manual valve 7, the circuit solenoid valve 9, the second manual valve 10 and the inlet solenoid valve 18), start the variable frequency speed control three-phase asynchronous motor 1 to make the test pump 2 run at low speed, discharge the air in the pipeline to the top of the tank and check the system sealing.

[0238] (2) Gradually adjust the variable frequency speed-regulating three-phase asynchronous motor 1 to the rated speed of the test pump 2, and adjust the opening of the circuit solenoid valve 9 through the intelligent control console 19 until the rated flow of the test pump 2 is reached. After the test pump 2 runs stably, collect the inlet and outlet pressure signals and vibration signals at the same time, and calculate the current reference head.

[0239] (3) After starting the vacuum pump 16, open the vacuum solenoid valve 14 to reduce the inlet pressure. When the inlet pressure reaches the required level, close the vacuum pump 16 and the vacuum solenoid valve 14. During this process, the opening of the fine-tuning circuit solenoid valve 9 keeps the flow rate constant at the rated value. After the test pump 2 runs stably, the inlet and outlet pressure signals and vibration signals under the current working conditions are collected at the same time, and the current head is calculated in real time.

[0240] (4) After the current pressure node data is collected, repeat step (3) to gradually reduce the inlet pressure until the calculated head of test pump 2 decreases by 3% compared with the reference head.

[0241] (5) Call the aforementioned cavitation calibration method to process the signals collected at each pressure node during the pressure reduction process to obtain the final denoised signal dataset with high confidence labels.

[0242] Step 2: Data Preprocessing and Offline Classifier Training

[0243] (1) In order to obtain enough data for offline training of the classifier, the final denoised signal dataset with high confidence labels needs to be preprocessed (in this embodiment, 10x downsampling is preferred) to ensure that the number of data samples under each label after preprocessing is not less than 100. For the preprocessed dataset, 80% is randomly selected in a stratified manner as the training set and 20% is selected as the test set.

[0244] (2) Regarding the selection of the machine learning classifier, this embodiment can use a deep learning network or a traditional machine learning model. Considering the computing power limitations and real-time requirements of the edge computing module, this embodiment prefers SVM (Support Vector Machine) as the cavitation fault classifier, and it is necessary to further extract feature parameters from the preprocessed dataset obtained in step (1). This embodiment prefers five dimensionless time-domain feature parameters, including: waveform factor, peak factor, impulse factor, margin factor, and kurtosis factor (the above parameters are all conventional dimensionless statistical parameters in the field of signal processing).

[0245] (3) Constructing the classifier model. This embodiment preferably uses an SVM model with a radial basis function (RBF) kernel. Set the penalty coefficient. nuclear parameters Grid search space (preferred) The optimal hyperparameter combination is obtained by globally optimizing the training set using grid search combined with 10-fold hierarchical cross-validation. This optimal combination is then used to train the final SVM model on the training set. The model is then validated using a test set. If the accuracy does not reach 90%, the types of feature parameters need to be readjusted or the penalty coefficient increased. With kernel parameters The model is retrained in the grid search space until the accuracy reaches 90% or higher. Finally, the successfully trained SVM model is output and imported into the edge computing module of the intelligent console 19 to complete the update and deployment of the cavitation fault identification program.

[0246] Step 3: Online real-time monitoring and cavitation control linkage:

[0247] When conducting centrifugal pump experiments using the system, vibration signals are collected (preferably at a sampling frequency of 20480Hz, a sampling time of 0.1s, and a typical monitoring cycle of 5s to 10s). The edge computing module of the intelligent control console 19 automatically denoises and extracts the five dimensionless time-domain feature parameters, which are then input into the cavitation fault identification program. Based on the output results, the programmable logic control module executes the following closed-loop control logic:

[0248] (1) If the output result is “non-empty”, it indicates that the system is running normally, the programmable logic control module does not trigger the linkage action, and the system directly enters the next regular monitoring cycle.

[0249] (2) If the output result is "cavitation fault", the programmable logic control module controls the pressurization solenoid valve 12 to open and controls the air compressor 13 to start, filling the cavitation tank 11 with air to increase the inlet pressure of the test pump 2. During pressurization, the edge computing module maintains the signal acquisition and identification of the above-mentioned normal monitoring cycle, and the programmable logic control module reads the data of the electromagnetic flowmeter 8 at the same time, and keeps the flow rate required for the experiment constant by adjusting the opening of the loop solenoid valve 9.

[0250] (3) Once the output of the cavitation fault identification program changes to "non-cavitation", the programmable logic control module immediately controls the air compressor 13 and the booster solenoid valve 12 to close. To avoid frequent start-stop oscillations of the system under critical conditions, the system automatically enters a hydraulic stability lock-in period (preferably 30s~60s) after the boosting stops: during this lock-in period, the system only records data, and the programmable logic control module blocks new boosting trigger commands; after the lock-in period ends, the command blocking is released, and the normal closed-loop response logic to the cavitation fault identification result is restored.

[0251] (4) For the highest safety considerations, during the operation of the air compressor 13, the system continuously reads the data of the inlet pressure sensor 20. If the inlet pressure has reached the highest safety pressure threshold (preferably 1 standard atmosphere), regardless of the output of the cavitation fault identification program, the programmable logic control module will prioritize the execution of emergency protection actions, immediately control the air compressor 13 and the booster solenoid valve 12 to close, trigger the system alarm and control the system to stop.

[0252] like Figure 7As shown, this application provides an electronic device including a memory 101 for storing one or more programs and a processor 102. When the one or more programs are executed by the processor 102, they implement the method as described in any of the first aspects above.

[0253] The system also includes a communication interface 103. The memory 101, processor 102, and communication interface 103 are electrically connected directly or indirectly to each other to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The memory 101 can be used to store software programs and modules, and the processor 102 executes various functional applications and data processing by executing the software programs and modules stored in the memory 101. The communication interface 103 can be used for signaling or data communication with other node devices.

[0254] The memory 101 may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0255] The processor 102 can be an integrated circuit chip with signal processing capabilities. The processor 102 can be a general-purpose processor 102, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0256] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can also be implemented in other ways. The method and system embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0257] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0258] On the other hand, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by processor 102, the computer program implements the methods described in any of the first aspects above. If the functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0259] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized variational mode decomposition method, characterized in that, Includes the following steps: Step 1: Acquire raw cavitation vibration signals while the centrifugal pump is running. ; Step 2: Process the original cavitation vibration signal Perform variational mode decomposition based on the golden section optimization and output the set of extracted modes; Step 3: Calculate the Hurst exponent of each sub-mode in the extracted mode set using the R / S analysis method. And based on this, adaptively distinguish the information-dominant mode. With noise-dominant mode ; Step 4: For the noise-dominant mode Adaptive wavelet thresholding based on cross-correlation coefficient is used to obtain the denoised noise-dominant mode. ; Step 5: Transform the information into a dominant mode. With the noise-dominant mode after denoising The summation outputs the final denoised signal. .

2. The personalized variational mode decomposition method according to claim 1, characterized in that, Step 2 specifically involves: using the original cavitation vibration signal as a residual. And calculate its power spectrum. ; Extract the power spectrum amplitude with the largest value Local maxima peak frequency And generate a set of candidate penalty factors in its neighborhood. ; Then, construct an effective candidate penalty factor set based on the candidate penalty factor set. and initial search interval Construct a multi-objective dynamic fitness function, iteratively optimize within the initial search interval using the golden section method, and determine the optimal penalty factor; perform VMD based on the optimal penalty factor to obtain the optimal submode. And add to the modality set The reconstructed signal is obtained. and output the extracted mode set; Among them, the It is a local maximum and satisfies: The candidate penalty factor set This is obtained through the following method: at each peak frequency Candidate penalty factors are generated within ±30% of the neighborhood. : ;in, ; For the candidate penalty factor After removing duplicate values ​​and sorting them in ascending order, the set of candidate penalty factors can be obtained. : ;in, The set of effective candidate penalty factors Specifically, the cross-correlation coefficient of If all If all are less than 0.9, then take the value directly. The maximum value corresponding to As the optimal penalty factor Otherwise, continue with the subsequent steps; the initial search interval Specifically: ;in, The multi-objective dynamic fitness function is specifically as follows: ;in, Measuring the impact characteristic strength of a mode, To measure the contribution of the current mode to the residual signal energy, To measure the information correlation between the modal and residual signals, For any penalty factor The corresponding modal extraction formula is: ; and The weights are as follows: ;in, This is the current residual signal-to-noise ratio, as detailed below: ;in, The optimal penalty factor is a minimal positive number that includes the system's basis noise level. Specifically, the current search range Midpoint: .

3. The personalized variational mode decomposition method according to claim 1, characterized in that, In step 4, the adaptive wavelet thresholding denoising based on the cross-correlation coefficient specifically involves: extracting the first... Layer wavelet coefficients The median absolute deviation is used to estimate the true noise standard deviation, and a cross-correlation coefficient based on the noise-dominant mode and the original cavitation vibration signal is constructed. Adaptive threshold : ;in, The length of the wavelet coefficients in the current decomposition layer; An improved smooth semi-soft thresholding function is used for wavelet coefficients at each layer. Perform point-by-point calculations to obtain the denoised wavelet coefficients. : Based on the denoised wavelet coefficients The denoised noise-dominant mode can be obtained by performing an inverse wavelet transform. .

4. The personalized variational mode decomposition method according to claim 1, characterized in that, The method also includes a step of selecting the optimal mode: specifically, calculating the cross-correlation coefficient between each dominant mode and the original vibration signal and performing a global comparison, and extracting and saving the optimal mode corresponding to the global maximum value.

5. A cavitation calibration method, characterized in that, Includes the following steps: Step 1: Conduct offline variable operating condition benchmark tests on the centrifugal pump, and collect the original cavitation vibration signals at different pressure nodes during the pressure reduction process; obtain the final denoised signal and optimal mode of each node based on the personalized variational mode decomposition method described in claims 1 to 4, calculate the root mean square value of the optimal mode and the corresponding effective cavitation margin value, and use the root mean square value and the effective cavitation margin value as a discrete state set. Step 2: Smooth the discrete state set using a Savitzky-Golay polynomial conformal filter to construct a smooth and continuous cavitation characteristic curve function. and its first derivative Second derivative ; Step 3: Define the tangent point search feature index and use the adaptive double-ended tangent anchoring algorithm to construct the first tangent equation and the second tangent equation respectively; Step 4: Solve for the x-coordinate of the intersection point of the two tangent line equations, and use it as the initial cavitation point; Step 5: Using the initial cavitation point as the critical decision boundary, perform sample calibration on the final denoised signal under each pressure node, and output the final denoised signal dataset with high confidence labels.

6. The method according to claim 5, characterized in that, In step 2, the cavitation characteristic curve function is specifically defined as follows: x-axis The two-dimensional curve is represented by the ordinate; in step 3, the tangent point search feature index specifically refers to: The adaptive double-ended tangent anchoring algorithm specifically refers to: in non-empty high... Area search local minimum point Construct the first tangent equation slope ,intercept ; At low cavitation characteristic curves, the cavitation characteristic curve increases sharply. Area search Maximum point Construct the second tangent equation slope ,intercept ; Among them, the non-cavitation high The region is specifically a set of discrete states. The first 1 / 3 of the data segment of the sequence; the low The specific area is The last third of the data segment of the sequence.

7. A centrifugal pump anti-cavitation system, characterized in that, include: The system comprises a pipeline system, a data acquisition system, and a control system; wherein the pipeline system provides physical characteristics, including fluid circulation and adjustable operating conditions, through sensors; the data acquisition system converts the physical characteristics into digital signals and outputs them to the control system; the control system receives the digital signals and uses algorithms to link in real time or manually control the electrical equipment in the pipeline system.

8. A centrifugal pump anti-cavitation system according to claim 7, characterized in that, The control system includes an intelligent console (19), which is embedded with an edge computing module and a programmable logic control module. The edge computing module is used to implement the personalized variational mode decomposition method according to any one of claims 1 to 4 and the cavitation calibration method according to claim 5 or 6, and is used to receive vibration and pressure signals transmitted by the data acquisition system and output the system operation status identification result. The programmable logic control module controls the equipment in the pipeline system based on the identification result. The pipeline system includes a pressure stabilizing tank (5), a cavitation tank (11), and a pipe connecting the two to form a closed loop. The circuit includes a test pump (2), an air compressor (13), a vacuum pump (16), an outlet solenoid valve (4), a pressure stabilizing tank pressure relief solenoid valve (6), a first manual valve (7), a circuit solenoid valve (9), a second manual valve (10), a booster solenoid valve (12), a vacuum solenoid valve (14), a cavitation tank pressure relief solenoid valve (17), and an inlet solenoid valve (18). The data acquisition system includes a pressure sensor (3), an electromagnetic flowmeter (8), an inlet pressure sensor (20), a unidirectional vibration acceleration sensor (21), and a dynamic signal acquisition instrument (22) connected to the test pump (2).

9. A closed-loop control method for a centrifugal pump anti-cavitation system based on claim 7 or 8, characterized in that, It includes the following two phases: the offline benchmarking and model training phase, and the online real-time monitoring and anti-cavitation linkage phase: Specifically, the offline benchmark testing and model training phase includes: Step A1: Extract the dimensionless time-domain feature parameters of the final denoised signal dataset; Step A2: Train the classifier model offline and optimize its parameters on the training set, and deploy the classifier model that meets the accuracy requirements to the edge computing module; The online real-time monitoring and anti-cavitation linkage stage specifically includes: Step B1: During online monitoring, the edge computing module automatically denoises and extracts features from the collected signals, and inputs them into the deployed classifier model to output the status recognition results; Step B2: If the identification result is "cavitation fault", the programmable logic control module controls the booster equipment to start to increase the inlet pressure; Step B3: If the identification result is "non-cavitation" and the system is currently in pressurization operation, the programmable logic control module controls the shutdown of the pressurization equipment and enters the hydraulic stability lock-in period; if it is not in pressurization operation, it directly enters the next monitoring cycle. Step B4: During system operation, if the continuously read inlet pressure reaches the maximum safe pressure threshold, the programmable logic control module will prioritize the emergency shutdown of the booster equipment, system alarm, and shutdown actions.

10. A closed-loop control method according to claim 9, characterized in that, In the offline benchmarking and model training phase, the dimensionless time-domain feature parameters include waveform factor, peak factor, impulse factor, margin factor, and kurtosis factor. During the online real-time monitoring and anti-cavitation linkage phase, the classifier model is a support vector machine model; during the hydraulic stability lock-in period, the system only records data, and the programmable logic control module blocks new pressurization trigger commands.

Images