A variable frequency vane pump operation parameter prediction and cavitation state monitoring method and system based on three-phase current multi-domain feature fusion

Abstract

This invention provides a method and system for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of multi-domain features of three-phase current. The system includes an offline model training phase and an online intelligent monitoring phase. In the offline phase, three-phase stator currents under multiple operating conditions are collected, and time-domain statistics, STFT frequency domain features, wavelet envelopes, and three-phase composite features are extracted and fused. Frequency regression models, flow regression models, and frequency-flow condition grouping cavitation classification models are trained, and the parameters are fixed. In the online phase, three-phase currents are collected in real time, multi-domain fused features are constructed, and the fixed parameters are called to obtain the predicted drive frequency and continuous flow rate. Based on these two values, the current frequency-flow condition group is determined, and the corresponding cavitation classification model is called to output the cavitation level. Non-intrusive multi-state monitoring is achieved using only the motor power supply circuit current.

Description

Technical Field

[0001] This invention relates to the fields of fluid machinery condition monitoring, motor current characteristic analysis, soft measurement of operating parameters, and intelligent fault diagnosis, and particularly to a method and system for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain characteristics. Background Technology

[0002] Vane pumps rely on rotating impellers to transfer energy to fluids and are core fluid machinery in agricultural irrigation, water conservancy projects, industrial circulation, energy power, and process transportation systems. During long-term operation, factors such as flow rate deviation from the design point, inlet pressure changes, fluid pulsation, and seal clearance degradation can lead to cavitation, wear, and performance degradation in vane pumps. Variable frequency drives (VFDs) adjust the vane pump speed by changing the power supply frequency to meet energy-saving and flexible fluid supply requirements. However, VFD operation creates a strong coupling relationship between drive frequency, speed, flow rate, head, load torque, and cavitation state. The current response characteristics of the same equipment differ significantly at different frequencies and flow rates, making condition monitoring difficult.

[0003] Currently, there are two main technical approaches for monitoring the operational status of impeller pumps. The first approach uses external sensors, which acquire hydraulic or mechanical status signals by installing flow meters in the pipeline, pressure and vibration sensors on the pump body or bearing housing, or acoustic sensors around the pump. While these sensors can reflect the hydraulic state within the pump, the placement of measurement points is constrained by installation space, signal transmission is susceptible to attenuation due to structural transmission paths and interference from ambient noise, and long-term exposure of the sensor body to corrosive fluids or high-temperature, high-pressure environments leads to decreased reliability. Therefore, their adoption is costly in scenarios involving submersible pumps, high temperatures, high pressures, or where pipeline modifications are inconvenient. The second approach uses motor current characteristic analysis, utilizing existing current sensors in the power supply circuit to obtain the root mean square value or power spectrum characteristics of the single-phase current, indirectly sensing changes in the impeller pump load. However, under variable frequency drive conditions, the current's dominant frequency shifts with the drive frequency, and changes in flow rate cause a redistribution of overall load levels and frequency band energy. Cavitation bubble collapse manifests as local transient impacts and envelope modulation. These three states act on different time-frequency scales, and a single time-domain statistic or a single spectral feature cannot simultaneously distinguish them. Furthermore, the current reference amplitude and spectral structure under different driving frequencies and flow conditions are significantly different. If a unified classifier is used for cavitation identification, changes in the operating condition reference are easily misjudged as cavitation features, leading to an increase in the false alarm rate of monitoring.

[0004] Therefore, existing technologies have not yet solved the following problems: how to construct multi-scale complementary feature representations using only three-phase current; how to distinguish between operating condition benchmark differences and cavitation state differences under variable frequency and variable flow conditions; and how to establish an integrated data processing link for frequency identification, flow prediction, and cavitation monitoring. A non-intrusive multi-domain feature fusion and operating condition grouping identification method is needed to improve the monitoring accuracy and adaptability under complex variable operating conditions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of multi-domain features of three-phase current. By fusing time-domain statistical features, short-time Fourier transform frequency domain features, wavelet envelope features, and three-phase composite features, the method achieves drive frequency identification, continuous flow prediction, and cavitation degree identification.

[0006] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0007] A method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain features includes an offline model training stage and an online intelligent monitoring stage, with the two stages connected by solidifying model parameters;

[0008] The offline model training phase includes the following steps:

[0009] Collect the three-phase stator current and corresponding status labels of the variable frequency vane pump under multiple operating conditions. The status labels include drive frequency label, flow rate label and cavitation level label.

[0010] The three-phase stator currents are windowed and preprocessed to obtain standardized three-phase current samples.

[0011] Time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features are extracted from the standardized three-phase current samples and spliced ​​together to form a multi-domain fusion feature vector.

[0012] Using multi-domain fusion feature vectors as input, frequency regression model and traffic regression model are trained respectively;

[0013] The dataset is divided into sub-datasets according to driving frequency and flow conditions. An independent cavitation classification model is trained in each frequency-flow combination to form a frequency-flow condition grouped cavitation classification structure.

[0014] Solidify the trained model parameters;

[0015] The online intelligent monitoring phase includes the following steps:

[0016] The three-phase stator current in the power supply circuit of the variable frequency vane pump drive motor is collected in real time, and window segmentation and preprocessing are performed to obtain online standardized three-phase current samples.

[0017] Extract the same multi-domain fusion feature vector as the one used in the offline model training phase from the online standardized three-phase current samples;

[0018] Input the multi-domain fusion feature vector into the trained frequency regression model, call the fixed parameters, and output the driving frequency prediction value; input the multi-domain fusion feature vector into the trained traffic regression model, call the fixed parameters, and output the continuous traffic prediction value.

[0019] Based on the predicted driving frequency and continuous flow rate, the frequency-flow condition group to which the current sample belongs is determined. The multi-domain fusion feature vector is input into the cavitation classification model corresponding to the determined condition group, the fixed parameters are called, and the cavitation level is output.

[0020] Furthermore, the STFT frequency domain features extract the main frequency and frequency band energy ratio to characterize the energy redistribution caused by variable frequency drive conditions and flow cavitation; the wavelet envelope features extract the envelope spectrum peak value through Hilbert transform to enhance the weak modulation information induced by cavitation; the wavelet envelope features include at least one of the envelope root mean square value, envelope peak value, envelope kurtosis, and envelope spectrum features; the time domain statistical features include at least one of the mean, standard deviation, root mean square value, peak-to-peak value, skewness, and kurtosis of each phase current to characterize the amplitude level, dispersion, and impact characteristics of the three-phase current caused by load changes in the vane pump.

[0021] Furthermore, the three-phase composite characteristics include the three-phase composite root mean square value, the average value of Clarke current vector amplitude, the root mean square value of Clarke current vector amplitude, the proportion of positive sequence components, the proportion of negative sequence components, and the proportion of zero sequence components.

[0022] Furthermore, the cavitation level label is calibrated based on the head reduction percentage, which is the rate of decrease of the current head relative to the maximum head under the same frequency and flow conditions.

[0023] Furthermore, in the online intelligent monitoring phase, the frequency-flow condition group to which the current sample belongs is determined based on the predicted driving frequency value and the predicted continuous flow value; if the predicted driving frequency value or the predicted continuous flow value exceeds the boundary of the trained condition group, it is assigned to the nearest trained condition group, and a confidence warning is output to prompt condition extrapolation.

[0024] Furthermore, both the frequency regression model and the flow regression model are extreme random tree regression models; each extreme random tree regression model includes multiple regression trees, each regression tree randomly selects features from candidate features and randomly generates multiple node splitting thresholds within their value range, and then selects the splitting method according to the decrease of the variance of the target value within the node; the driving frequency prediction value or the continuous flow prediction value is the average value output of each regression tree.

[0025] Furthermore, the cavitation classification model is an extreme random tree classification model trained separately for each frequency-flow condition group; during the training phase, an independent multi-class classifier is established for each frequency-flow combination; during the online phase, the corresponding classifier is matched according to the predicted value of the driving frequency and the predicted value of the continuous flow, and the multi-domain fusion feature vector is input into the matched classifier and the cavitation level is output.

[0026] A system for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of multi-domain features of three-phase current includes:

[0027] The three-phase current acquisition module is used to acquire the three-phase stator current in the power supply circuit of the variable frequency vane pump drive motor.

[0028] The signal processing module is used to perform window segmentation and preprocessing on the three-phase stator current and output standardized three-phase current samples.

[0029] A multi-domain feature construction module is used to extract time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features from standardized three-phase current samples.

[0030] The feature fusion module is used to concatenate the time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features to form a multi-domain fused feature vector;

[0031] The frequency prediction module is used to receive the multi-domain fused feature vector and output the driving frequency prediction value;

[0032] The traffic prediction module is used to receive the multi-domain fusion feature vector and output continuous traffic prediction values;

[0033] The frequency-flow grouping module is used to determine the current frequency-flow condition group based on the predicted driving frequency value and the predicted continuous flow value.

[0034] The cavitation status identification module is used to receive the frequency-flow condition group identifier and the multi-domain fusion feature vector, call the cavitation classification model parameters corresponding to the frequency-flow condition group, and output the cavitation level.

[0035] The status output module is used to display, store, or send drive frequency prediction values, continuous flow prediction values, and cavitation levels.

[0036] The beneficial effects of this invention are as follows:

[0037] 1. The method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain characteristics described in this invention can realize online monitoring of the operating status of variable frequency vane pumps by utilizing only the three-phase stator current in the power supply circuit of the drive motor, without the need to add flow meters, pressure sensors, vibration sensors or acoustic sensors to the pump body or pipeline, thereby reducing installation complexity and maintenance costs.

[0038] 2. The method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain features described in this invention integrates time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features into a unified multi-domain fusion feature vector. This allows information such as drive frequency shift, load amplitude change, frequency band energy redistribution, cavitation transient impact, and interphase coupling to be expressed complementaryly within the same feature space, avoiding information loss from single time-domain or frequency-domain analysis.

[0039] 1. The method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain features described in this invention divides the data set into subsets according to the driving frequency and flow conditions. Within each frequency-flow combination, a cavitation classifier is trained independently, enabling each classifier to learn cavitation features under similar speed and flow benchmarks. This decouples the changes in operating benchmarks from changes in cavitation status, thereby reducing the false alarm rate of cavitation identification under variable frequency and variable flow operating conditions.

[0040] 4. The variable frequency vane pump operation parameter prediction and cavitation state monitoring method based on three-phase current multi-domain feature fusion described in this invention uses an extreme random tree regression model to estimate the driving frequency and continuous flow rate respectively, and uses the random partitioning and averaging integration mechanism of multiple decision trees to process high-dimensional nonlinear multi-domain features, so that the frequency and flow rate soft measurement results have good robustness and interpretability.

[0041] 5. The method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of multi-domain features of three-phase current described in this invention calculates and solidifies the global mean and global standard deviation of the current of each phase based on the full training data during the offline training stage. During the online monitoring stage, the solidified statistics are directly called to perform standardization on the real-time window data, so that the statistical distribution of the online input data is consistent with the offline training distribution, and the distribution shift caused by the fluctuation of window-level statistics is suppressed.

[0042] 6. The variable frequency vane pump operation parameter prediction and cavitation state monitoring method based on three-phase current multi-domain feature fusion described in this invention can be deployed on industrial computers, edge computing devices or embedded controllers after training, and can be connected to programmable logic controllers, distributed control systems or remote operation and maintenance platforms, thereby improving the engineering applicability and deployment flexibility of the monitoring system. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings described below are some embodiments of the present invention. For those skilled in the art, it is obvious that other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is a flowchart of the method for predicting operating parameters and monitoring cavitation status of variable frequency vane pumps based on the fusion of three-phase current multi-domain features, as described in this invention.

[0045] Figure 2 This is a schematic diagram illustrating the construction and fusion of three-phase current multi-domain features in this invention.

[0046] Figure 3 This is a schematic diagram of the extreme random tree frequency and flow regression model structure of the present invention.

[0047] Figure 4 This is a schematic diagram illustrating the cavitation degree identification based on frequency-flow condition grouping according to the present invention.

[0048] Figure 5 This is a schematic diagram illustrating the offline training, deployment, and online monitoring process of the model of this invention. Detailed Implementation

[0049] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0050] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "axial," "radial," "vertical," "horizontal," "inner," and "outer," etc., indicating orientation or positional relationships based on the orientation or positional relationships shown in the accompanying drawings, are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0051] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0052] like Figure 1 and Figure 5 As shown, the method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of multi-domain features of three-phase current, as described in this invention, includes an offline model training stage and an online intelligent monitoring stage. The offline model training stage completes the training and parameter fixing of multi-domain feature configuration, frequency regression model, flow regression model, and frequency-flow condition grouping cavitation classification model based on historically acquired data. The online intelligent monitoring stage, based on real-time acquired three-phase stator current, loads the model parameters fixed in the offline model training stage, performs multi-domain feature calculation, regression prediction, condition grouping, and cavitation identification, and outputs the monitoring results. The two stages are connected through the fixing of model parameters and a unified feature processing flow.

[0053] I. Offline Model Training Phase

[0054] Step S01: Multi-condition training data acquisition and label calibration

[0055] On the variable frequency vane pump test bench, the vane pump is driven by a drive motor. The variable frequency drive unit is connected to the drive motor via a power supply line used to adjust the power supply frequency. A three-phase current acquisition module is installed in the power supply circuit between the variable frequency drive unit and the drive motor. It includes three current sensors, a signal conditioning unit, an analog-to-digital converter unit, and a synchronization clock unit, used to synchronously acquire the three-phase stator current. , , The instantaneous value of , where k is the index of the discrete sampling point.

[0056] The hydraulic circulation loop of the test bench includes a vane pump, a storage tank, a regulating valve, and connecting pipelines. A flow meter is installed on the outlet pipeline of the vane pump, and inlet and outlet pressure sensors are installed at the inlet and outlet flanges of the vane pump, respectively. These flow meters and pressure sensors are only used to provide realistic labels for samples during the offline training phase; they are no longer required for the online intelligent monitoring phase.

[0057] Three-phase stator currents were collected under various driving frequencies and flow rates. Each sample simultaneously had three types of tags: driving frequency tag, flow rate tag, and cavitation level tag.

[0058] The drive frequency label is directly calibrated based on the output frequency of the variable frequency drive. The flow rate label is calibrated based on the actual flow rate measured by the flow meter. The cavitation level label is calibrated based on the head drop percentage ΔH, where ΔH is expressed as:

[0059] ,

[0060] Wherein: H max The maximum head is given under the same frequency and flow conditions, in meters (m); H is the current head, in meters (m); ΔH is the percentage decrease in head, dimensionless and expressed as a percentage.

[0061] Cavitation levels are categorized into multiple classes based on the percentage decrease in head, ranging from weak cavitation to stronger cavitation. Labels can be generated from the time-frequency characteristics of the current and physically verified using the head decrease distribution, or calibrated based on head decrease, net positive suction head (NPSH), gas volume fraction, or manual assessment.

[0062] In this embodiment, as shown in Table 1, training data were established at three driving frequencies: 30Hz, 40Hz, and 50Hz. The design flow rate at 50Hz was 55m³ / s. 3 / h is 1.0Q d Based on the similarity law of vane pumps, the 1.0Q at 30Hz and 40Hz is... d Set to 33.0m respectively 3 / h and 44.0m 3 / h; 0.7Q per frequency. d0.8Q d 0.9Q d 1.0Q d 1.1Q d and 1.2Q d Six flow rate conditions were implemented, forming 18 frequency-flow rate condition groups. Multiple cavitation sampling points were set up for each condition, and three-phase current, flow rate, and inlet / outlet pressure were collected simultaneously. Under the same frequency and flow rate conditions, the maximum head H was used as the reference. max As a baseline, the percentage decrease in head is calculated; the cavitation labels are set from Class 0 to Class 4, which respectively indicate the trend of development from weak cavitation or near-normal state to stronger cavitation.

[0063] Table 1 Test conditions in the examples

[0064] Drive frequency <![CDATA[1.0Q d ]]> Flow range Flow conditions Cavitation point for each operating condition 30 Hz <![CDATA[33.0 m 3 / h]]> <![CDATA[23.1-39.6 m 3 / h]]> <![CDATA[0.7Q d -1.2Q d ]]> 7 40 Hz <![CDATA[44.0 m 3 / h]]> <![CDATA[30.8-52.8 m 3 / h]]> <![CDATA[0.7Q d -1.2Q d ]]> 7 50 Hz <![CDATA[55.0 m 3 / h]]> <![CDATA[38.5-66.0 m 3 / h]]> <![CDATA[0.7Q d -1.2Q d ]]> 7

[0065] Step S02: Window Segmentation and Preprocessing

[0066] First, the continuously acquired records are grouped according to driving frequency, flow conditions, and cavitation level. Each complete continuous acquisition record is used as an indivisible data partitioning unit, dividing each group into training and test sets. Model hyperparameters are determined only using the training set for cross-validation; the test set is not used for model training, feature selection, parameter adjustment, or threshold determination. After dataset partitioning, windowing is performed on the continuous acquisition records in both the training and test sets. Each window contains N consecutive sampling points, with a step size of S sampling points between adjacent windows. All windows obtained from the same complete continuous acquisition record must belong to the same subset to prevent adjacent windows from entering the training and test sets separately, causing data leakage. If this constraint is violated, the window partitioning position is readjusted or the sample subsets are reassigned until adjacent windows of the same continuous acquisition sequence exist in only a single subset.

[0067] For the phase current within each window (x=a,b,c,k=0,1,…,N-1) Perform Z-score standardization:

[0068] ,

[0069] in: This represents the value of the kth sampling point of the x-phase current during the offline training phase, in A. μ is the standardized dimensionless value. x is the global mean of the full training data for phase x current, in A; ε is the global standard deviation of the full training data of phase x current, in A; ε is the zero constant.

[0070] Simultaneously calculate and solidify the global mean value μ of the three-phase current. a μ b μ c and global standard deviation σ a σ b σ c These are stored as preset statistics in the multi-domain feature configuration parameters, which can be directly called during the online intelligent monitoring phase instead of being calculated in real time based on the online window.

[0071] Step S03: Multi-domain feature construction and fusion, such as Figure 2 As shown, the details are as follows:

[0072] For each standardized three-phase current window sample, time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features are constructed, as shown in Table 2, where:

[0073] Time-domain statistical characteristics: Calculate the mean, standard deviation, root mean square value, peak-to-peak value, skewness, and kurtosis of the current in each phase to describe the load amplitude level, fluctuation degree, and impact.

[0074] Short-Time Fourier Transform (STFT) Frequency Domain Characteristics: Performing a short-time Fourier transform on the x-th phase current yields the time spectrum. Where m is the time window position index; ω is the angular frequency; taking the square of the modulus of the time spectrum yields the time-frequency energy spectrum: Based on the time-frequency energy spectrum, the dominant frequency, spectral energy, spectral entropy, and energy proportion of a preset frequency band are extracted. The dominant frequency is used to characterize the frequency conversion drive conditions, while the spectral energy and frequency band energy ratio are used to characterize the energy redistribution caused by flow rate and cavitation.

[0075] Wavelet envelope features: Wavelet packet decomposition is performed on the current of each phase to calculate the energy and energy ratio of each frequency band; Hilbert transform is performed on the reconstructed wavelet packet signal to obtain the envelope signal and extract the root mean square value of the envelope, the peak value of the envelope, the kurtosis of the envelope and the peak value of the envelope spectrum, which are used to enhance the weak modulation and transient impact information induced by cavitation.

[0076] Characteristics of three-phase synthesis:

[0077] The three-phase composite characteristics include the three-phase root mean square value, which is the arithmetic mean of the root mean square values ​​of the currents in each phase; and the three-phase composite root mean square value. The mean value of the Clarke current vector magnitude I(k); the root mean square value of the Clarke current vector magnitude I(k); the proportions of the positive-sequence component, the negative-sequence component, and the zero-sequence component. These characteristics are used to describe the overall electromagnetic state of the three phases and the phase imbalance.

[0078] Calculate the root mean square value of the current in each phase:

[0079] ,

[0080] in: This represents the root mean square value of the x-phase current, in amperes (A).

[0081] Calculate the three-phase combined root mean square value:

[0082] ,

[0083] in: This is the three-phase composite root mean square value, in A.

[0084] The Clarke transform is used to construct the three-phase composite current vector, and the mean and root mean square values ​​of the vector amplitudes, as well as the relevant proportions of the positive-sequence, negative-sequence, and zero-sequence components, are extracted, as follows:

[0085] Performing the Clarke transform on the three-phase stator current yields the α-axis current component i in a two-phase stationary coordinate system. α (k) and β-axis current component i β (k):

[0086] ,

[0087] ,

[0088] The magnitude of the three-phase combined current vector I(k) is:

[0089] ,

[0090] In the formula: This represents the value of the kth sampling point of the α-axis current, in A. I(k) is the value of the kth sampling point of the β-axis current, in A; I(k) is the value of the kth sampling point of the magnitude of the three-phase composite current vector, in A.

[0091] The mean and root mean square values ​​of the vector magnitudes, as well as the correlation ratios of the positive, negative, and zero-order components, are extracted based on I(k).

[0092] The above four types of features are concatenated and scaled to form a multi-domain fusion feature vector F:

[0093] ,

[0094] Wherein: F time F is a time-domain statistical eigenvector; STFT F is the frequency domain eigenvector of the STFT; WPE The wavelet envelope eigenvector; is the three-phase composite feature vector; F is the multi-domain fusion feature vector.

[0095] Table 2

[0096] Feature domain Example features Key characterization information Time-domain statistical domain Mean, Standard Deviation, RMS, Peak-to-Peak Value, Skewness, Kurtosis Load amplitude, fluctuation and impact STFT frequency domain Main frequency, spectral energy, spectral entropy, and bandwidth energy ratio Drive frequency, harmonics and energy distribution wavelet envelope domain Band energy, energy percentage, envelope RMS, envelope kurtosis Non-stationary modulation and cavitation impact Three-phase synthesis domain Three-phase RMS, Clarke amplitude, sequence component ratio Overall load and coupling

[0097] Step S04: Training the frequency regression model

[0098] like Figure 3 As shown, an extreme random tree frequency regression model is constructed using the multi-domain fusion feature vector F as input. Let the model consist of T regression trees, and let the multi-domain fusion feature vector F of the t-th tree for sample i be... i The output is h t (F i If ), then the integrated regression output is:

[0099] ,

[0100] in: The output of the ensemble regression is for sample i; T is the number of regression trees, dimensionless; h t (F i ) represents the output of the t-th regression tree for sample i.

[0101] When the regression target is the driving frequency That is, the predicted value of the driving frequency. The unit is Hz. Each regression tree randomly selects features from candidate features and randomly generates multiple splitting thresholds within their value range. Then, the splitting method is selected based on the decrease in the variance of the target value within the node. Averaging across multiple trees reduces the prediction volatility of a single tree.

[0102] The frequency regression model was trained using the training set, and its hyperparameters were tuned using cross-validation within the training set. The frequency prediction performance was then evaluated on the test set. Evaluation metrics included mean absolute error, root mean square error, and coefficient of determination.

[0103] Step S05: Training the Flow Regression Model

[0104] Using the multi-domain fusion feature vector F as input, an extreme random tree flow regression model is constructed. The model structure and ensemble regression calculation formula are consistent with step S04. When the regression target is flow, That is, the continuous flow forecast value The unit is m 3 / h.

[0105] The traffic regression model was trained using the training set, and its hyperparameters were tuned using cross-validation within the training set. The traffic prediction performance was then evaluated on the test set. Evaluation metrics included mean absolute error, root mean square error, coefficient of determination, and relative error.

[0106] Step S06: Training of Frequency-Flow Condition Grouping and Cavitation Classification Model

[0107] The dataset is divided into sub-datasets based on driving frequency and traffic conditions. An independent cavitation classifier is built within each frequency-traffic combination, forming a Groupwise ExtraTrees classification structure, such as... Figure 4 As shown.

[0108] Let the probability that sample i in the output of the t-th classification tree belongs to class c be... Then the ensemble probability and predicted category are:

[0109] ,

[0110] in: Let be the ensemble probability that sample i belongs to class c, which is dimensionless; The probability that sample i belongs to class c is the output of the t-th classification tree, which is dimensionless; Let be the predicted category of sample i; C is the set of categories.

[0111] Within each frequency-flow operating condition group, cavitation levels are divided into multiple categories, arranged from weak cavitation to stronger cavitation. Each operating condition group has its own independent training and test sets, and independent cavitation classifiers are trained. The triggering conditions and branching logic are as follows: if the number of samples in a certain operating condition group is lower than a preset threshold, then that operating condition group is merged with adjacent operating condition groups before training the classifier; if this constraint is violated, then the operating condition group boundaries are redefined or the training samples for that operating condition group are expanded.

[0112] The cavitation classification performance of each operating condition group was evaluated based on the test set. Evaluation metrics included perfect accuracy, neighbor tolerance accuracy, macro-average F1, ROC-AUC, and mean precision. Neighbor tolerance accuracy was considered engineering-acceptable when the predicted class differed from the true class by no more than one level.

[0113] Table 3 shows the model parameters in one embodiment.

[0114] Table 3

[0115] parameter Frequency Regression Traffic return Cavitation classification Input feature dimension 206 206 206 Model type ExtraTrees Regressor ExtraTrees Regressor Groupwise ExtraTrees Classifier Number of trees 500 500 600 per group Number of working condition groups - - 18 Classification criteria Squared error Squared error Gini Maximum number of candidate features 1.0 1.0 sqrt Category weight - - balanced Random Seed 42 42 42

[0116] Step S07: Model Validation and Parameter Consolidation

[0117] The trained frequency regression model, flow regression model, and cavitation classification model for each frequency-flow operating condition group were evaluated using a test set. After successful validation, the optimal model parameters were solidified. These optimal parameters included multi-domain feature configuration parameters, frequency regression model parameters, flow regression model parameters, cavitation classification model parameters for each frequency-flow operating condition group, operating condition group boundaries, and alarm thresholds. The multi-domain feature configuration parameters included the global mean value μ of the three-phase current. x and global standard deviation σ x .

[0118] In a test verification example, the root mean square error of frequency prediction for 30 Hz, 40 Hz, and 50 Hz was less than 0.004 Hz, and the frequency identification accuracy was 100%; the mean absolute error of flow prediction at the three frequencies was 0.7655 m. 3 / h, 0.9714 m 3 / h and 0.6080 m 3 / h, with determination coefficients of 0.9602, 0.9416, and 0.9921, respectively; the complete accuracy rate for the five cavitation states was 92.782%, the adjacent tolerance accuracy rate was 96.054%, and the macro-average F1 was approximately 0.929, as shown in Table 4.

[0119] Table 4

[0120] Monitoring tasks Example performance Frequency prediction At 30, 40, and 50 Hz, the RMSE is less than 0.004 Hz, achieving a recognition accuracy of 100%. Traffic prediction <![CDATA[MAE is 0.6080-0.9714 m 3 / h, R 2 is 0.9416-0.9921]]> Cavitation identification The perfect accuracy was 92.782%, the adjacent tolerance accuracy was 96.054%, and the Macro-F1 was approximately 0.929.

[0121] II. Online Intelligent Monitoring Phase

[0122] Step S08: Real-time acquisition of three-phase stator current

[0123] In the power supply circuit of the drive motor, the three-phase current acquisition module synchronously acquires the three-phase stator current. , , The sampling frequency, anti-aliasing filter parameters, and signal conditioning method are consistent with step S01 of the offline model training phase.

[0124] Step S09: Real-time window segmentation and preprocessing

[0125] The real-time data stream of the three-phase stator current is divided into continuous windows. Each window contains N sampling points.

[0126] Perform Z-score normalization on the phase current within each window:

[0127] ,

[0128] in: The value of the x-phase current at the k-th sampling point, collected in real time during the online intelligent monitoring phase, is expressed in A; μ x and σ x Instead of calculating it in real time based on an online window, the global mean and global standard deviation are directly called from the global mean and global standard deviation that are fixed during the offline model training phase. ε is the standardized dimensionless value; ε is the constant after removing zero.

[0129] Step S10: Online Multi-Domain Feature Construction and Fusion

[0130] For the standardized three-phase current window samples, the same multi-domain feature construction process as step S03 in the offline model training stage is adopted to extract time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features in sequence, and then concatenate them to form an online multi-domain fusion feature vector F. online .

[0131] Step S11: Frequency and Flow Forecast

[0132] The online multi-domain fusion feature vector F online Input the frequency regression model trained during the offline model training phase, call the solidified frequency regression model parameters, and output the driving frequency prediction value. .

[0133] The online multi-domain fusion feature vector F online Input the traffic regression model trained during the offline model training phase, call the solidified traffic regression model parameters, and output continuous traffic prediction values. .

[0134] Step S12: Operating condition grouping and cavitation identification

[0135] Based on the obtained predicted drive frequency value and continuous flow forecast values Determine the frequency-flow condition group to which the current sample belongs. Specifically, and Compare with the boundaries of each frequency-flow condition group determined in step S06 of the offline model training phase: If and If a sample falls within both the frequency and flow ranges of a pre-trained operating condition group, then that operating condition group is the operating condition group to which the current sample belongs; if or If the condition exceeds the boundary of all trained condition groups, it is assigned to the nearest trained condition group, and a confidence warning is output to indicate the condition extrapolation.

[0136] After determining the operating condition group to which the current sample belongs, the online multi-domain fusion feature vector F will be used. online Input the cavitation classification model corresponding to the frequency-flow operating condition group to which the current sample belongs, call the classifier parameters of this operating condition group that were fixed during the offline model training phase, output the probability distribution of the current sample belonging to each cavitation category, and take the one with the highest probability as the cavitation prediction category. .

[0137] Step S13: Result Output and Alarm

[0138] Predicted drive frequency Continuous flow forecast Cavitation prediction categories The system displays and stores the predicted confidence level. Exponential smoothing or majority voting is performed on the output results of multiple consecutive windows, with the smoothed result used as the final monitoring basis.

[0139] When the cavitation category reaches the preset level, the flow rate deviates from the allowable range, the prediction confidence level decreases abnormally, or multiple consecutive windows are abnormal, the result output module performs audible and visual alarms, control interlocks, historical record archiving, or remote communication operations.

[0140] In other embodiments, the sampling frequency, window length, and feature dimension can be adjusted according to the rated power of the vane pump, the switching frequency of the inverter, and the available computing power. The number of frequency band divisions, the number of wavelet packet decomposition layers, and the feature selection method can be increased or decreased according to signal characteristics and diagnostic requirements. In other embodiments, the frequency regression model and flow regression model are not limited to extreme random trees and can be replaced by random forests, gradient boosting trees, or other ensemble learning models. The cavitation classification model is not limited to the Groupwise ExtraTrees structure and can use other grouping strategies or ensemble learning frameworks. The frequency-flow condition grouping boundary is not limited to fixed grid division and can be dynamically determined based on cluster analysis, adaptive binning, or similarity laws. In other embodiments, the monitoring output is not limited to drive frequency, continuous flow rate, and cavitation level and can include head, efficiency, or other mechanical fault outputs. Transfer learning, domain adaptation, or incremental learning can be used to update the model parameters of different vane pump models, while the input during the online monitoring phase remains the three-phase stator current.

[0141] The variable frequency vane pump operating parameter prediction and cavitation state monitoring system based on three-phase current multi-domain feature fusion of the present invention includes a three-phase current acquisition module, a signal processing module, a multi-domain feature construction module, a feature fusion module, a frequency prediction module, a flow prediction module, a frequency-flow grouping module, a cavitation state identification module, and a state output module. The three-phase current acquisition module is used to acquire the three-phase stator current in the power supply circuit of the variable frequency vane pump drive motor. The signal processing module, connected to the three-phase current acquisition module, is used to perform window segmentation and preprocessing on the three-phase stator current and output standardized three-phase current samples. The multi-domain feature construction module, connected to the signal processing module, is used to extract time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features from the standardized three-phase current samples. The feature fusion module, connected to the multi-domain feature construction module, is used to concatenate the time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features to form a multi-phase current fusion module. The system comprises: a multi-domain fusion feature vector; a frequency prediction module connected to the feature fusion module, used to receive the multi-domain fusion feature vector and output a driving frequency prediction value; a flow prediction module connected to the feature fusion module, used to receive the multi-domain fusion feature vector and output a continuous flow prediction value; a frequency-flow grouping module connected to the frequency prediction module and the flow prediction module, used to determine the current frequency-flow condition group based on the driving frequency prediction value and the continuous flow prediction value; a cavitation status identification module connected to the frequency-flow grouping module and the feature fusion module, used to receive the frequency-flow condition group identifier and the multi-domain fusion feature vector, call the cavitation classification model parameters corresponding to the frequency-flow condition group, and output the cavitation level; and a status output module connected to the frequency prediction module, the flow prediction module, and the cavitation status identification module, used to display, store, or send the driving frequency prediction value, the continuous flow prediction value, and the cavitation level.

[0142] The three-phase current acquisition module includes a current sensor, a signal conditioning unit, an analog-to-digital conversion unit, and a synchronous clock unit; the signal processing module, the multi-domain feature construction module, the feature fusion module, the frequency prediction module, the flow prediction module, the frequency-flow grouping module, and the cavitation state identification module can all be implemented by a processor in an industrial computer, edge computing device, frequency converter, or embedded controller executing a computer program stored in its memory.

[0143] The status output module includes at least one of a data display unit, a trend storage unit, an anomaly alarm unit, and a communication unit; the communication unit is used to transmit the drive frequency, continuous flow, cavitation level, and its prediction confidence to a programmable logic controller, a distributed control system, an equipment management platform, or a remote operation and maintenance platform.

[0144] It should be understood that although this specification is described according to various embodiments, not every embodiment contains only one independent technical solution. This way of describing the specification is only for clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

[0145] The detailed descriptions listed above are merely specific illustrations of feasible embodiments of the present invention and are not intended to limit the scope of protection of the present invention. All equivalent embodiments or modifications made without departing from the spirit of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of multi-domain features of three-phase current, characterized in that, It includes an offline model training phase and an online intelligent monitoring phase, which are connected by solidifying model parameters; The offline model training phase includes the following steps: Collect the three-phase stator current and corresponding status labels of the variable frequency vane pump under multiple operating conditions. The status labels include drive frequency label, flow rate label and cavitation level label. The three-phase stator currents are windowed and preprocessed to obtain standardized three-phase current samples. Time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features are extracted from the standardized three-phase current samples and spliced ​​together to form a multi-domain fusion feature vector. Using multi-domain fusion feature vectors as input, frequency regression model and traffic regression model are trained respectively; The dataset is divided into sub-datasets according to driving frequency and flow conditions. An independent cavitation classification model is trained in each frequency-flow combination to form a frequency-flow condition grouped cavitation classification structure. Solidify the trained model parameters; The online intelligent monitoring phase includes the following steps: The three-phase stator current in the power supply circuit of the variable frequency vane pump drive motor is collected in real time, and window segmentation and preprocessing are performed to obtain online standardized three-phase current samples. Extract the same multi-domain fusion feature vector as the one used in the offline model training phase from the online standardized three-phase current samples; Input the multi-domain fusion feature vector into the trained frequency regression model, call the fixed parameters, and output the driving frequency prediction value; input the multi-domain fusion feature vector into the trained traffic regression model, call the fixed parameters, and output the continuous traffic prediction value. Based on the predicted driving frequency and continuous flow rate, the frequency-flow condition group to which the current sample belongs is determined. The multi-domain fusion feature vector is input into the cavitation classification model corresponding to the determined condition group, the fixed parameters are called, and the cavitation level is output.

2. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics as described in claim 1, is characterized in that, The STFT frequency domain features extract the main frequency and frequency band energy ratio to characterize the frequency conversion drive conditions, flow rate changes, and energy redistribution caused by cavitation; the wavelet envelope features are subjected to Hilbert transform to extract the envelope spectrum peak value to enhance the weak modulation information induced by cavitation. The wavelet envelope features include at least one of the following: root mean square value of envelope, peak value of envelope, kurtosis of envelope, and envelope spectrum features; the time-domain statistical features include at least one of the following: mean, standard deviation, root mean square value, peak-to-peak value, skewness, and kurtosis of each phase current, used to characterize the amplitude level, dispersion, and impact characteristics of the three-phase current caused by load changes in the vane pump.

3. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics as described in claim 1, is characterized in that... The three-phase composite characteristics include the three-phase composite root mean square value, the average value of Clarke current vector amplitude, the root mean square value of Clarke current vector amplitude, the proportion of positive sequence components, the proportion of negative sequence components, and the proportion of zero sequence components.

4. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics as described in claim 1, is characterized in that, The cavitation level label is calibrated based on the percentage decrease in head, which is the rate of decrease in current head relative to the maximum head under the same frequency and flow conditions.

5. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics according to claim 1, characterized in that, During the online intelligent monitoring phase, the frequency-flow condition group to which the current sample belongs is determined based on the predicted driving frequency and the predicted continuous flow. If the predicted driving frequency or the predicted continuous flow exceeds the boundary of the trained condition group, it is assigned to the nearest trained condition group, and a confidence warning is output to prompt condition extrapolation.

6. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics according to claim 1, characterized in that, Both the frequency regression model and the flow regression model are extreme random tree regression models; each extreme random tree regression model includes multiple regression trees, each regression tree randomly selects features from candidate features and randomly generates multiple node splitting thresholds within their value range, and then selects the splitting method according to the decrease of the variance of the target value within the node; the driving frequency prediction value and the continuous flow prediction value are the average values ​​output by each regression tree.

7. The method for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics according to claim 1, characterized in that, The cavitation classification model is an extreme random tree classification model trained separately for frequency-flow operating condition groups; During the training phase, an independent multi-class classifier is established for each frequency-flow combination. During the online phase, the corresponding classifier is matched based on the predicted value of the driving frequency and the predicted value of the continuous flow. The multi-domain fusion feature vector is input into the matched classifier and the cavitation level is output.

8. A system for predicting operating parameters and monitoring cavitation status of a variable frequency vane pump based on the fusion of three-phase current multi-domain characteristics according to claim 1, characterized in that, include: The three-phase current acquisition module is used to acquire the three-phase stator current in the power supply circuit of the variable frequency vane pump drive motor. The signal processing module is used to perform window segmentation and preprocessing on the three-phase stator current and output standardized three-phase current samples. A multi-domain feature construction module is used to extract time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features from standardized three-phase current samples. The feature fusion module is used to splice the time-domain statistical features, STFT frequency-domain features, wavelet envelope features, and three-phase composite features to form a multi-domain fused feature vector; The frequency prediction module is used to receive the multi-domain fused feature vector and output the driving frequency prediction value; The traffic prediction module is used to receive the multi-domain fusion feature vector and output continuous traffic prediction values; The frequency-flow grouping module is used to determine the current frequency-flow condition group based on the predicted driving frequency value and the predicted continuous flow value. The cavitation status identification module is used to receive the frequency-flow condition group identifier and the multi-domain fusion feature vector, call the cavitation classification model parameters corresponding to the frequency-flow condition group, and output the cavitation level. The status output module is used to display, store, or send drive frequency prediction values, continuous flow prediction values, and cavitation levels.

Images