A method and system for monitoring a desilting electric pump based on a digital twin mechanism model

Abstract

The application discloses a kind of based on digital twin mechanism model's sand pumping electric pump monitoring method and system.The method comprises: the three-axis vibration signal of the stator shell of sand pumping electric pump and the three-phase stator voltage signal and three-phase stator current signal of frequency converter output end are collected by sensing hardware;Based on the reduced order solid-liquid two-phase flow mechanism model in edge computing gateway, according to the current speed and instruction state of sand pumping electric pump, the theoretical stator current and theoretical torque current of sand pumping electric pump are calculated as prediction reference value;Actual torque current component is obtained by using field programmable logic gate array to carry out park transformation to three-phase stator current signal, and envelope spectrum analysis is carried out to three-axis vibration signal to extract high-frequency feature;An extended Kalman filter observer is constructed, the residual of actual torque current component and theoretical torque current is calculated, and the residual, the difference value of residual and high-frequency feature are input into pre-trained multilayer perception machine network;By multilayer perception machine network, the current overflow silt volume fraction and impeller radial wear depth are output by solving.The application can realize non-contact quantitative monitoring of the wear state and overflow concentration of electric pump in harsh fluid environment.

Description

Technical Field

[0001] This application relates to the field of industrial internet and water conservancy sediment discharge equipment monitoring technology, specifically to a sediment discharge electric pump monitoring method and system based on a digital twin mechanism model. Background Technology

[0002] Submersible sand-discharging pumps are widely used in water conservancy project dredging and mine tailings transportation. During operation, the impeller, volute, and other flow-through components of the sand-discharging pump are directly subjected to high-speed scouring by high-concentration hard particles such as quartz sand and pebbles, resulting in extremely rapid wear. Impeller wear can lead to a precipitous drop in pump efficiency, causing rotor eccentric vibration, and in severe cases, motor burnout.

[0003] In existing technologies, the internal condition monitoring of sand-discharging electric pumps relies on low-frequency vibration sensors installed at the motor bearings or temperature sensing elements in the stator windings. When external vibration or temperature reaches the alarm threshold, the internal impeller has already suffered irreversible damage or the flow channel is blocked. The objective physical environment of high-concentration sediment and high water pressure precludes the possibility of placing displacement sensors or visual sensors inside the pump cavity flow channel. Deep learning prediction models trained on historical operating data will deviate from physical constraints when faced with sudden changes in mud concentration or unrecorded non-standard operating conditions, generating false alarms and exhibiting poor generalization ability.

[0004] There is an urgent need for a method that can overcome the shortcomings of pure data models and break through the limitations of physical monitoring, so as to achieve non-contact quantitative monitoring of the internal wear status and overcurrent concentration of sand discharge pumps in high-concentration sediment environments. Summary of the Invention

[0005] This application provides a monitoring method and system for sand-discharging electric pumps based on a digital twin mechanism model, which solves the technical problems in the prior art where the internal operating status of sand-discharging electric pumps is difficult to measure directly and the model has poor generalization ability.

[0006] This application provides a monitoring method for a sand-discharging electric pump based on a digital twin mechanism model, comprising the following steps: acquiring triaxial vibration signals of the stator housing of the sand-discharging electric pump and three-phase stator voltage and current signals at the output of the frequency converter through sensing hardware; calculating the theoretical stator current and theoretical torque current of the sand-discharging electric pump as prediction reference values ​​based on the reduced-order solid-liquid two-phase flow mechanism model in the edge computing gateway, according to the current speed and command status of the sand-discharging electric pump; and performing a Parker transform on the three-phase stator current signal at a frequency greater than or equal to 20kHz using a field-programmable gate array to obtain the actual torque current component i. q The envelope spectrum of the triaxial vibration signal was analyzed to extract high-frequency features; an extended Kalman filter observer was constructed to calculate the actual torque current component i. qThe residual of the theoretical torque current is calculated, and the residual, the difference value of the residual, and the high-frequency feature are input into a pre-trained multilayer perceptron network. The multilayer perceptron network is used to solve the problem and output the current flow sediment volume fraction and impeller radial wear depth.

[0007] This application provides a monitoring system for a sand-discharging electric pump, comprising: a sensing hardware module disposed on the stator housing and the output terminal of the frequency converter of the sand-discharging electric pump, used to collect the triaxial vibration signal of the stator housing of the sand-discharging electric pump and the three-phase stator voltage signal and three-phase stator current signal of the frequency converter output terminal; a benchmark extrapolation module deployed in an edge computing gateway and connected to the sensing hardware module, used to run a reduced-order solid-liquid two-phase flow mechanism model, and calculate the theoretical stator current and theoretical torque current of the sand-discharging electric pump as prediction benchmark values ​​based on the current speed and command status of the sand-discharging electric pump; and a feature decoupling module, embedded with a field-programmable gate array and connected to the sensing hardware module, used to perform Parker transformation on the three-phase stator current signal at a frequency greater than or equal to 20kHz to obtain the actual torque current component i. q The envelope spectrum analysis of the triaxial vibration signal is performed to extract high-frequency features; a fusion solution module is communicatively connected to the benchmark derivation module and the feature decoupling module to construct an extended Kalman filter observer and calculate the actual torque current component i. q The residual of the theoretical torque current is calculated, and the residual, the difference of the residual, and the high-frequency feature are input into a pre-trained multilayer perceptron network to solve the problem, and output the current flow sediment volume fraction and impeller radial wear depth.

[0008] This application provides a computer device, including a memory and a processor; the memory stores a computer program, and when the computer program is executed, it implements the sand-discharging electric pump monitoring method based on the digital twin mechanism model provided in the foregoing embodiments.

[0009] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed, it implements the sand-discharging electric pump monitoring method based on a digital twin mechanism model provided in the foregoing embodiments.

[0010] This application embodiment establishes a theoretical operating benchmark by deploying a reduced-order solid-liquid two-phase flow mechanism model, extracts characteristic residuals representing transient resistance changes using an extended Kalman filter observer, and inputs physical residual constraints into a multilayer perceptron network. This successfully decouples two different physical causes, namely, sudden changes in mud concentration and cumulative impeller wear, overcoming the overfitting defect of pure data models and realizing non-contact quantitative monitoring in harsh fluid environments. Attached Figure Description

[0011] Figure 1 This is the main flowchart of the monitoring method for sand-discharging electric pumps based on a digital twin mechanism model provided in the embodiments of this application.

[0012] Figure 2 This is a flowchart of obtaining the comprehensive mechanical resistance torque and calculating the theoretical stator current provided in the embodiments of this application.

[0013] Figure 3 This is a flowchart of the extraction of high-frequency features and actual torque current components provided in the embodiments of this application.

[0014] Figure 4 This is a flowchart of constructing an extended Kalman filter observer and performing residual calculations, provided in an embodiment of this application.

[0015] Figure 5 This is a schematic diagram of the overall structure of the sand-draining electric pump monitoring system provided in the embodiments of this application.

[0016] Figure 6 This is a structural block diagram of the computer device provided in the embodiments of this application.

[0017] Explanation of reference numerals in the attached figures: In the diagram: 501 - Sensing hardware module, 502 - Benchmark extrapolation module, 503 - Feature decoupling module, 504 - Fusion calculation module, 600 - Computer equipment, 601 - Processor, 602 - Memory, 603 - Internal data bus, S101 - Acquire sensing signals, S102 - Calculate prediction benchmark, S103 - Extract high-frequency features, S104 - Calculate feature residuals, S105 - Network calculation output, S201 - Read prior parameters, S202 - Obtain theoretical values ​​for clear water operation, S203 - Perform resistance and torque correction, S301 - Read sampling signals, S302 - Separate torque and current, S303 - Extract high-frequency features, S401 - Initialize observation parameters, S402 - Generate state prediction, S403 - Calculate residuals, S404 - Correct state, S405 - Output residual difference. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] Example 1

[0020] like Figure 1As shown in the figure, this application provides a method for monitoring a sand-discharging electric pump based on a digital twin mechanism model. The method runs in an edge computing gateway deployed in an industrial site, where a real-time operating system with a Linux kernel and dedicated hardware acceleration logic run collaboratively. The method includes the following processing steps.

[0021] In step S101, the triaxial vibration signal of the stator housing of the sand-discharging electric pump and the three-phase stator voltage and current signals at the output of the frequency converter are acquired through sensing hardware. In this embodiment, the sensing hardware layer includes a triaxial piezoelectric accelerometer rigidly connected to the top and bottom of the stator housing of the sand-discharging electric pump motor via threads. The frequency response range of this triaxial piezoelectric accelerometer covers 0.5Hz to 10kHz. A three-phase Hall current transformer and voltage transformer with a sampling frequency set to 20kHz are connected to the output of the frequency converter. By using a piezoelectric accelerometer rigidly connected via threads and a high-frequency Hall transformer, raw data covering wideband mechanical high-frequency harmonics and high-speed electrical transient fluctuations can be acquired without loss, providing a reliable data base for subsequent capture of millisecond-level transient resistance changes.

[0022] In step S102, based on the reduced-order solid-liquid two-phase flow mechanism model in the edge computing gateway, the theoretical stator current and theoretical torque current of the sand-discharging electric pump are calculated as prediction reference values ​​according to the current speed and command status of the pump. This model is constructed based on the energy conservation and fluid momentum theorem, employing a reduced-order processing method. Figure 2 The process is as follows: The edge computing gateway executes step S201 to obtain the rated flow rate, impeller geometry parameters, and basic density parameters of the working medium of the sand-draining electric pump. Based on the current speed, command status, and impeller geometry parameters, it calculates the reference hydraulic torque under clean water conditions. Step S202 is executed to calculate the reference hydraulic torque under clean water conditions based on the current speed, command status, and impeller geometry parameters, thus obtaining the theoretical value for clean water conditions. Based on this, step S203 combines the basic density parameters with a preset solid particle suspension coefficient to perform nonlinear correction on the reference hydraulic torque, obtaining the comprehensive mechanical resistance torque.

[0023] To obtain the accurate corrected mechanical shaft power and thus the comprehensive mechanical resistance torque, the power derivation logic is implemented in the reduced-order mechanism model through the following corrected equation: Among them, P actual This indicates the current actual corrected shaft power of the electric pump; P th This represents the theoretical shaft power under standard clear water conditions, which is a known function built into the system. The coefficient representing the influence of sediment density is used to characterize the increase in work done by the increase in fluid specific gravity; C v K represents the volume fraction of sand content;w Indicates the flow channel volume loss coefficient; Indicates the amount of thinning of the impeller's characteristic dimensions; This represents the angular velocity of the rotor. In this embodiment, and K w The constant coefficient, The value can be calibrated to 1.65 depending on the specific silt material. Based on the above mechanism equations, the system can deduce and calculate the theoretical three-phase stator current and theoretical torque current i that the electric pump should possess under the current speed and command conditions. q(th) By setting up a reduced-order mechanism model, not only was the theoretical performance envelope calculated, but more importantly, normal power fluctuations caused by conventional fluid dynamics were filtered out in advance, laying a solid physical reference benchmark for subsequent stripping of nonlinear anomalies.

[0024] To improve the dynamic adaptability of the comprehensive mechanical resistance torque correction, more in-depth calculation logic is executed in step S203. The transient fluctuation frequency of the three-phase stator current signal is extracted, and the solid particle suspension coefficient is updated in real time based on the difference between the transient fluctuation frequency and the passing frequency of the sand-discharging pump blades. Using the updated solid particle suspension coefficient and the reduced-order Bernoulli equation of hydrodynamics, the additional frictional torque of the solid-liquid two-phase flow in the impeller channel is calculated. The reference hydraulic torque is added to the additional frictional torque to generate the comprehensive mechanical resistance torque. The update cycle of this comprehensive mechanical resistance torque is strictly synchronized with the sampling cycle of the three-phase stator current signal. It should be noted that the above-mentioned synchronization cycle limitation is only a preferred example; those skilled in the art can also implement this application using frequency doubling or frequency division synchronization logic.

[0025] In step S103, the three-phase stator current signal is subjected to Parker transformation using a field-programmable gate array at a frequency greater than or equal to 20kHz to obtain the actual torque current component i. q The system performs envelope spectrum analysis on the triaxial vibration signals to extract high-frequency features. The edge computing fusion layer employs an industrial gateway integrating a field-programmable gate array (FPGA) and a multi-core processor. The FPGA serves as a dedicated hardware decoding unit to process high-frequency voltage and current signals. Combined with... Figure 3 The extraction process shown includes step S301, where high-frequency sampled three-phase stator voltage and current signals are continuously read through the direct memory access channel in the field-programmable gate array (FPGA). Step S302 involves extracting the three-phase stator current signal i. a i b i c The fundamental phase is determined, and a coordinate system rotation transformation is performed to separate the actual torque current component i corresponding to the quadrature axis. q and excitation current component i dIn step S303, envelope spectrum analysis is performed on the triaxial vibration signal at the bottom. Specifically, this involves performing a Hilbert transform on the triaxial vibration signal to generate an analytic signal, calculating the modulus of this analytic signal to obtain the envelope signal, and then performing a Fast Fourier Transform on the envelope signal. This step extracts the rotational frequency amplitude A. 1X With the frequency amplitude A of the blade BPF As a high-frequency characteristic, by introducing a field-programmable gate array to perform hard decoding at the 20kHz level, it is possible to separate the transient torque current component i, which directly characterizes the short-term surge in fluid resistance, without delay. q It can identify load distortion caused by sudden changes in high-concentration sediment at the millisecond level.

[0026] In step S104, an extended Kalman filter observer is constructed to calculate the actual torque current component i. q With theoretical torque current i q(th) The residuals, their differences, and high-frequency features are input into a pre-trained multilayer perceptron network. Figure 4 The process involves extending the Kalman filter observer's operation based on a state prediction and update mechanism.

[0027] In step S401, the state vector and covariance matrix of the extended Kalman filter observer are initialized. Specifically, the state vector consists of internal variables characterizing the transient physical operation of the sand-discharging pump. Combined with the aforementioned mechanistic equations, it specifically includes core physical parameters such as the actual torque current component, excitation current component, rotor angular velocity, and the volume fraction of flowing sediment. The covariance matrix includes an error covariance matrix, a process noise covariance matrix, and a measurement noise covariance matrix. The process noise covariance matrix is ​​used to physically characterize the system dynamic uncertainties caused by fluid turbulence and the random impact of high-concentration sediment particles on the impeller; the measurement noise covariance matrix is ​​used to physically characterize the high-frequency switching harmonics of the frequency converter and the electromagnetic interference introduced by sensor measurements.

[0028] In step S402, within each discrete time step t, the prior state estimate for the current time is predicted using the optimal state estimate from the previous time step and the reduced-order solid-liquid two-phase flow mechanism model. The preferred mathematical expression for this Kalman prediction derivation process is: in, F represents the vector of prior state estimates at the current k-th step; k Represents the state transition matrix; B represents the posterior estimate of the optimal state at the previous time step; k Represents the control input model matrix; u k This represents the control vector. In step S403, the residual generation logic is executed. The formula for establishing and calculating the feature residuals is as follows: Among them, E r Represents the torque current residual vector; i q This represents the actual torque-current component obtained through on-site measurement and coordinate transformation decoupling; i q(th) This represents the theoretical torque current state prior prediction value output by the aforementioned mechanism model. In step S404, based on this residual E... r Calculate the observation update matrix and the Kalman gain. Use the Kalman gain and residuals to correct the prior state estimate and obtain the posterior state estimate for the current time step. In step S405, calculate the rate of change of the residuals between two adjacent discrete time steps to form the difference value within the historical sliding time window. .

[0029] In step S105, the current flow sediment volume fraction and impeller radial wear depth are calculated using a multilayer perceptron network. Since the preceding mechanistic model offsets the fundamental hydraulic nonlinear changes, the calculated residual E... r Residual difference and the high-frequency eigenvector of vibration [A] 1X A BPF The input node is fed into a pre-trained multilayer perceptron network. The residuals and high-frequency features are mapped to high-frequency perturbation components; the difference values ​​of the residuals are mapped to low-frequency drift components. A weighted sum of the high-frequency perturbation components and low-frequency drift components is calculated in the fully connected layer of the multilayer perceptron network. After forward propagation through a nonlinear activation function, the output layer of the multilayer perceptron network retains two nodes and directly outputs the corresponding overflow sediment volume fraction C. v The percentage value and the radial wear depth of the impeller The millimeter-level values ​​are achieved. Based on the fusion of extended Kalman filter observers in the time dimension, the deep learning network operates under constraints with clear physical boundaries, accurately attributing high-frequency disturbances to changes in sand content and low-frequency residual drift to impeller wear, thus avoiding the overfitting risk faced by pure neural networks in bivariate coupling.

[0030] Furthermore, to ensure the stable operation of the industrial gateway and sensing system, this application includes a bridging module hidden within the backend support architecture. The system includes a power supply module that provides isolated and stable operating voltage to the sensing hardware and industrial gateway. A signal preprocessing circuit is provided between the three-phase Hall current and voltage transformers and the piezoelectric accelerometer and the acquisition channel. This preprocessing circuit is configured to perform low-pass filtering for noise reduction, opto-electric isolation, and differential voltage amplification on the analog signal output from the front end. It should be noted that while opto-isolators and differential amplifiers are used as examples of signal conditioning, those skilled in the art can achieve high-quality signal transmission using alternatives such as magnetic isolation or instrumentation amplifiers. By deploying the preprocessing circuit and independent power supply module, the high-power electromagnetic interference generated during inverter operation can be shielded, improving the signal-to-noise ratio of the acquired data.

[0031] Example 2

[0032] Based on the operating method described in Embodiment 1, Embodiment 2 of this application provides a sand-draining electric pump monitoring system as the physical and virtual architecture carrier of the aforementioned method.

[0033] like Figure 5 As shown, the overall structure of the sand-draining electric pump monitoring system includes a sensing hardware module 501, a benchmark inference module 502, a feature decoupling module 503, and a fusion calculation module 504.

[0034] The sensing hardware module 501 is installed on the stator housing of the sand-dredging pump and the output terminal of the frequency converter. It is used to collect the triaxial vibration signals of the stator housing of the sand-dredging pump and the three-phase stator voltage and three-phase stator current signals of the frequency converter output terminal. The sensing hardware module 501 includes a piezoelectric accelerometer and a Hall effect current transformer. The piezoelectric accelerometer is fixed to the drive end bearing housing and the outside of the stator housing of the sand-dredging pump to collect high-frequency vibration data; the Hall effect current transformer is clipped onto the power supply cable between the frequency converter and the sand-dredging pump to collect high-frequency electrical parameters without contact.

[0035] The benchmark simulation module 502 is deployed in the edge computing gateway and connected to the sensing hardware module 501 via an industrial fieldbus. This module is configured to run a reduced-order solid-liquid two-phase flow mechanism model and calculate theoretically predicted benchmark values ​​based on the current rotational speed and command status. The feature decoupling module 503 has an embedded field-programmable gate array (FPGA) and is connected to the sensing hardware module 501 at high speed. To ensure safe data acquisition and high-frequency computation, the feature decoupling module 503 includes a hardware decoding circuit and an isolation buffer. The hardware decoding circuit acquires the voltage and current data on the high-voltage side through the isolation buffer and performs coordinate system rotation transformation separation calculations within the FPGA using a parallel multiply-accumulate unit array to extract the actual torque current component i. q and vibrational envelope spectrum characteristics.

[0036] The fusion calculation module 504 is communicatively connected to the benchmark extrapolation module 502 and the feature decoupling module 503. It is used to construct an extended Kalman filter observer, calculate the residual between the actual and theoretical values, and input the data matrix into a multilayer perceptron network for inverse calculation, outputting the flow sediment volume fraction and impeller radial wear depth. In the preferred architecture of this embodiment, the benchmark extrapolation module 502 and the fusion calculation module 504 share a microcontroller motherboard with a system-on-a-chip. This microcontroller motherboard has an industrial Ethernet interface and is configured to send the monitoring results output from the calculation to an external control terminal.

[0037] In a preferred variant, the sand-discharging electric pump monitoring system extends outward to connect to a digital twin presentation layer architecture. This digital twin presentation layer receives the impeller radial wear depth values ​​and state matrix output from the microcontroller motherboard via an industrial communication protocol. Based on the received quantized parameters, the digital twin presentation layer uses a rendering engine to update the vertex coordinates of the sand-discharging electric pump's 3D model in real time. Through the dynamic updating of the 3D vertex coordinates, the 3D impeller model can be visualized by synchronously scaling down to reflect the internal wear condition of the actual physical entity. By adding a digital twin presentation architecture based on an industrial communication protocol, secure transmission of lower-level data to higher-level applications is achieved, providing on-site maintenance personnel with an intuitive and three-dimensional display of the internal damage morphology, and improving the interactive efficiency of predictive maintenance of the equipment.

[0038] Example 3

[0039] Based on Embodiments 1 and 2, Embodiment 3 of this application provides a computer device.

[0040] like Figure 6 As shown, the computer device 600 includes a processor 601 and a memory 602. The processor 601 and memory 602 are connected via an internal data bus 603. The memory 602 stores executable computer program instructions. When these computer program instructions are called and executed by the processor 601, the steps of the sand-draining electric pump monitoring method based on a digital twin mechanism model, as detailed in Embodiment 1, are implemented. In this embodiment, the processor 601 is a general-purpose microprocessor, a digital signal processor, or an application-specific integrated circuit; the memory 602 is a random access memory or non-volatile flash memory. Through the computational execution of this hardware architecture, concurrent processing of massive amounts of high-frequency data and reverse computation of deep neural networks can be completed quickly and efficiently.

[0041] Example 4

[0042] Embodiment 4 of this application provides a computer-readable storage medium that internally stores a computer program or a sequence of executable instructions. When the computer program is executed by a processor in a computer device or an edge computing gateway, it implements the sand-discharging electric pump monitoring method based on a digital twin mechanism model as described in detail in Embodiment 1. The computer-readable storage medium can take any tangible and non-transient data storage form, including read-only memory, programmable logic device firmware, magnetic hard disk, solid-state drive, and optical disk. Persistently storing the core monitoring algorithm through a non-transient medium ensures that the algorithm logic is not lost in the event of unexpected power outages in complex industrial environments, and supports remote firmware upgrades and periodic iterative distribution of the mechanism model.

[0043] In summary, the monitoring technology solution for sediment discharge electric pumps based on a digital twin mechanism model provided in this application establishes a physical operating benchmark by employing a reduced-order solid-liquid two-phase flow mechanism model, and captures millisecond-level transient resistance fluctuations by combining high-frequency hard decoding technology of field-programmable gate arrays. An extended Kalman filter observer is used to generate a residual sequence that characterizes nonlinear anomalies, and this residual is used as a constraint condition with strict physical boundaries input into a multilayer perceptron for bivariate decoupling. This system-level deep collaboration effectively filters out power fluctuation interference from conventional fluid dynamics, avoids the overfitting phenomenon that is prone to occur in pure neural networks under bivariate coupling, overcomes the physical obstacle of easily damaged built-in sensors in high-concentration sediment environments, and realizes non-contact quantitative monitoring of the wear state and sediment concentration inside the closed flow channel of the electric pump.

Claims

1. A monitoring method for a sand-discharging electric pump based on a digital twin mechanism model, characterized in that, include: The three-axis vibration signal of the stator housing of the sand-discharging electric pump, as well as the three-phase stator voltage signal and three-phase stator current signal at the output of the frequency converter are collected by the sensing hardware. Obtain the rated flow rate, impeller geometric parameters, and basic density parameters of the working medium of the sand-discharging electric pump; Based on the reduced-order solid-liquid two-phase flow mechanism model in the edge computing gateway, the reference hydraulic torque under clear water conditions is calculated according to the current speed and command status of the sand discharge electric pump and the impeller geometric parameters. By combining the basic density parameter with the preset solid particle suspension coefficient, the benchmark hydraulic torque is nonlinearly corrected to obtain the comprehensive mechanical resistance torque; The comprehensive mechanical resistance torque is mapped to the electromagnetic model of the motor, and the corresponding theoretical stator current and theoretical torque current are calculated as prediction reference values. The three-phase stator current signal is subjected to Parker transformation using a field-programmable gate array at a frequency greater than or equal to 20 kHz to obtain the actual torque current component. i q The envelope spectrum analysis of the triaxial vibration signal was performed to extract high-frequency features. Construct an extended Kalman filter observer to calculate the actual torque current component. i q The residual with the theoretical torque current; Calculate the rate of change of the residual between two adjacent discrete time steps, and use it as the difference value of the residual; The residual, the difference value of the residual, and the high-frequency feature are then input into a pre-trained multilayer perceptron network. The current volume fraction of sediment in the flow and the radial wear depth of the impeller are calculated and output through the multilayer sensor network.

2. The method as described in claim 1, characterized in that, The operation of combining the basic density parameter with the preset solid particle suspension coefficient to perform nonlinear correction on the reference hydraulic torque to obtain the comprehensive mechanical resistance torque includes: Extract the transient fluctuation frequency of the three-phase stator current signal; The solid particle suspension coefficient is updated based on the difference between the transient fluctuation frequency and the passing frequency of the blades of the sand-discharging electric pump. Based on the updated solid particle suspension coefficient and the Bernoulli reduced-order equation of hydrodynamics, the additional frictional torque of solid-liquid two-phase flow in the impeller channel is calculated. The reference hydraulic torque is added to the additional frictional torque to generate the comprehensive mechanical resistance torque; The update cycle of the integrated mechanical resistance torque is synchronized with the sampling cycle of the three-phase stator current signal.

3. The method as described in claim 1, characterized in that, The three-phase stator current signal is subjected to Parker transformation using a field-programmable gate array at a frequency greater than or equal to 20kHz to obtain the actual torque current component. i q The operation of extracting high-frequency features from the triaxial vibration signal by envelope spectrum analysis includes: The three-phase stator voltage signal and the three-phase stator current signal are read through the direct memory access channel in the field-programmable gate array; The fundamental phase of the three-phase stator current signal is extracted, and a coordinate system rotation transformation is performed to separate the actual torque current component corresponding to the quadrature axis. i q ; The triaxial vibration signal is subjected to Hilbert transform to generate an analytical signal, the modulus of the analytical signal is calculated to obtain the envelope signal, and the envelope signal is subjected to fast Fourier transform to extract the high-frequency features.

4. The method as described in claim 3, characterized in that, The extended Kalman filter observer is constructed to calculate the actual torque current component. i q The operation of the residual with respect to the theoretical torque current includes: Initialize the state vector and covariance matrix of the extended Kalman filter observer; Within each discrete time step t, the optimal state estimate of the previous time step and the reduced-order solid-liquid two-phase flow mechanism model are used to predict the prior state estimate of the current time step. Calculate the actual torque current component i q The residual is generated by the difference between the predicted torque current value mapped in the prior state estimate and the predicted torque current value. Calculate the observation update matrix based on the residuals, and then calculate the Kalman gain; The state prior estimate is corrected using the Kalman gain and the residual to obtain the state posterior estimate at the current time.

5. The method as described in claim 1, characterized in that, The operation of calculating and outputting the current flow sediment volume fraction and impeller radial wear depth through the multilayer sensor network includes: The residual and the high-frequency feature are mapped to high-frequency disturbance components; The difference value of the residual is mapped to a low-frequency drift component; In the fully connected layer of the multilayer perceptron network, calculate the weighted sum of the high-frequency perturbation component and the low-frequency drift component; The corresponding percentage values ​​of the volume fraction of the flowing sediment and the millimeter-level values ​​of the radial wear depth of the impeller are output through a nonlinear activation function.

6. A monitoring system for a sand-discharging electric pump, characterized in that, include: The sensing hardware module is installed on the stator housing of the sand-discharging electric pump and the output terminal of the frequency converter. It is used to collect the triaxial vibration signal of the stator housing of the sand-discharging electric pump and the three-phase stator voltage signal and three-phase stator current signal of the output terminal of the frequency converter. The benchmark simulation module, deployed in the edge computing gateway and connected to the sensing hardware module, is used to run a reduced-order solid-liquid two-phase flow mechanism model to obtain the rated flow rate, impeller geometric parameters, and basic density parameters of the working medium of the sand discharge electric pump. Based on the current speed and command status of the sand-discharging electric pump and the impeller geometric parameters, calculate the reference hydraulic torque under clear water conditions; By combining the basic density parameter with the preset solid particle suspension coefficient, the benchmark hydraulic torque is nonlinearly corrected to obtain the comprehensive mechanical resistance torque; The comprehensive mechanical resistance torque is mapped to the electromagnetic model of the motor, and the corresponding theoretical stator current and theoretical torque current are calculated as prediction reference values. The feature decoupling module, which embeds a field-programmable gate array (FPGA) and is connected to the sensing hardware module, is used to perform Parker transformation on the three-phase stator current signal at a frequency greater than or equal to 20kHz to obtain the actual torque current component. i q The envelope spectrum analysis of the triaxial vibration signal was performed to extract high-frequency features. The fusion calculation module, which is communicatively connected to the benchmark extrapolation module and the feature decoupling module, is used to construct an extended Kalman filter observer and calculate the actual torque current component. i q The residual with the theoretical torque current; Calculate the rate of change of the residual between two adjacent discrete time steps, and use it as the difference value of the residual; The residual, the difference value of the residual, and the high-frequency feature are input into a pre-trained multilayer perceptron network for calculation, and the current flow sediment volume fraction and impeller radial wear depth are output.

7. The system as described in claim 6, characterized in that, The feature decoupling module includes a hardware decoding circuit and an isolation buffer; The hard decoding circuit acquires the three-phase stator voltage signal and the three-phase stator current signal through the isolation buffer, and performs coordinate system rotation transformation separation calculations through the parallel multiply-accumulate unit array inside the field programmable gate array.

8. A computer device, characterized in that, Including memory and processor; The memory stores a computer program, which, when executed by the processor, implements the method as described in claim 1.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in claim 1.

Images