A centrifugal pump state prediction and maintenance system based on digital twinning

Abstract

This invention relates to the field of centrifugal pumps and discloses a centrifugal pump condition prediction and maintenance system based on digital twins. The system includes: a control center, serving as the core control hub for centrifugal pump condition prediction and maintenance, used to receive, transmit, and record data from various units and coordinate their operation; and a parameter monitoring unit, which is signal-connected to the control center. The detection terminals of the parameter monitoring unit are deployed in the key flow channels of the centrifugal pump to collect four core parameters directly related to performance degradation. The key flow channels include the main flow channel from the impeller inlet to the volute outlet and the sealing ring gap flow channel between the impeller and the pump casing. In this invention, through the coordinated monitoring of the main flow channel and the sealing ring gap flow channel, combined with a multi-dimensional prediction model and the virtual-real mapping of digital twins, accurate prediction of centrifugal pump performance degradation is achieved. It can provide early warning of impeller wear, sealing ring leakage, and other faults 24-48 hours in advance, effectively avoiding unplanned downtime.

Description

Technical Field

[0001] This invention relates to the field of centrifugal pumps, and more particularly to a centrifugal pump condition prediction and maintenance system based on digital twins. Background Technology

[0002] Centrifugal pumps are core equipment for fluid transport in industrial production, widely used in chemical, water conservancy, power, and municipal industries. They primarily use centrifugal force generated by impeller rotation to pressurize and transport liquids, and are a crucial link in ensuring the continuous operation of production processes. With the advancement of industrial intelligence, the operational stability and energy efficiency of centrifugal pumps have an increasingly significant impact on the economy and safety of the overall production system.

[0003] In existing technologies, the operation and management of centrifugal pumps mainly rely on traditional periodic maintenance and basic parameter monitoring. Specifically, the industry generally records macroscopic parameters such as pump body vibration and outlet pressure through manual inspections, or collects data using simple sensors (such as single-point vibration sensors and temperature sensors), and judges the equipment status based on experience. Maintenance strategies are mostly fixed-cycle maintenance (such as shutdown inspection every 3 or 6 months) or passive maintenance after a failure occurs, restoring equipment performance by replacing vulnerable parts (such as sealing rings and bearings).

[0004] However, due to the lack of accurate prediction of performance degradation trends, it is difficult to provide early warning of hidden faults such as impeller wear and sealing ring leakage. Repairs are often carried out only after the fault occurs, resulting in frequent unplanned shutdowns. This not only increases maintenance costs but also seriously affects production continuity. To address these issues, a centrifugal pump status prediction and maintenance system based on digital twins is proposed. Summary of the Invention

[0005] To overcome the above shortcomings, this invention provides a centrifugal pump condition prediction and maintenance system based on digital twins, which aims to solve the problem that the existing technology cannot fully reflect the deep mechanism of performance degradation, resulting in low prediction accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a centrifugal pump condition prediction and maintenance system based on digital twins, comprising:

[0007] The control center, as the core control hub for centrifugal pump status prediction and maintenance, is used to receive, transmit, and record data from each unit and coordinate their operation.

[0008] The parameter monitoring unit is connected to the control center via a signal. The detection end of the parameter monitoring unit is installed in the key flow channel of the centrifugal pump to collect four core parameters that are directly related to performance degradation.

[0009] The key flow channels include the main flow channel from the impeller inlet to the volute outlet and the sealing ring gap flow channel between the impeller and the pump casing.

[0010] The four core parameters specifically include: flow rate parameter, head parameter, efficiency parameter, and vibration parameter;

[0011] The feature extraction unit, integrated in the control center, receives four parameter data collected by the parameter monitoring unit at the key flow channel of the centrifugal pump and extracts them to form feature parameters.

[0012] A decay prediction unit is integrated into the control center. The decay prediction unit is connected to the feature extraction unit and the control center signal, and a multi-dimensional prediction model is constructed based on feature parameters.

[0013] The maintenance matching unit is integrated into the control center and has a built-in digital twin model. The maintenance matching unit is connected to the decline prediction unit and the control center. Based on the decline level classification data obtained from the multi-dimensional prediction model, the digital twin model enables dynamic matching and optimization of maintenance strategies.

[0014] As a further description of the above technical solution:

[0015] Electromagnetic flowmeter: used to measure flow parameters, installed in the main flow channel from the impeller inlet to the volute outlet;

[0016] Paired differential pressure transmitters: used to calculate head parameters, installed at the impeller inlet section and the volute outlet section respectively;

[0017] Three-dimensional vibration sensor: used to collect vibration parameters caused by turbulent flow field, installed on the outer wall of the volute;

[0018] High-frequency vibration sensor: used to collect high-frequency vibration parameters caused by abnormal gaps, installed at the pump casing position corresponding to the sealing ring;

[0019] Temperature sensor: Used to help determine the frictional temperature rise caused by gap leakage, and installed at the pump casing position corresponding to the sealing ring.

[0020] As a further description of the above technical solution: the data acquisition process of the parameter monitoring unit specifically includes:

[0021] S1: Signal Conversion and Sampling

[0022] The analog signal output from the parameter monitoring unit is converted into a digital signal using an A / D converter. The conversion formula is as follows:

[0023]

[0024] Where: D is the digital quantity, A is the analog input, Amin / Amax is the analog range, and n is the number of bits of the A / D converter;

[0025] Flow and pressure parameters are sampled at a 1-second period, and vibration parameters are sampled at a 10kHz frequency.

[0026] S2: Data Preprocessing

[0027] The moving average filter is applied to the sampled data, and the calculation formula is as follows:

[0028]

[0029] Where: N is the window size, and X(t) is the original data at time t;

[0030] Wavelet denoising was performed on the vibration data to retain the characteristic frequencies of 20-1000Hz;

[0031] S3: Data Transmission and Storage

[0032] The preprocessed data is uploaded to the control center via industrial Ethernet;

[0033] It uses dual hard drive storage, and the data retention period is ≥1 year.

[0034] As a further description of the above technical solution: the feature extraction process of the feature extraction unit includes:

[0035] S1: Sliding Window Construction

[0036] A 5-minute sliding window (1-minute window step) was constructed for the flow and pressure parameters, with each window containing 300 sampling points;

[0037] A 1024-point sliding window (window step size 512 points) is constructed for the vibration parameters.

[0038] S2: Temporal Feature Extraction

[0039] Calculate the mean within the window Standard deviation and rate of change

[0040]

[0041]

[0042]

[0043] in Number of data points within the window (flow / stress window) Vibration window );

[0044] S3: Frequency Domain Feature Extraction

[0045] Perform an FFT transform on the vibration window data to extract the characteristic frequency amplitudes. and energy percentage

[0046]

[0047] in The number of FFT points (taken as 1024). Characteristic frequency;

[0048] S4: Feature Vector Construction

[0049] The extracted time-domain and frequency-domain features are combined into a 12-dimensional feature vector;

[0050] .

[0051] As a further description of the above technical solution: the multi-dimensional prediction model of the recession prediction unit includes:

[0052] S1: Data-driven prediction

[0053] Using an LSTM neural network, the input feature vector The time series data is used to output the parameter prediction values ​​for the next 72 hours.

[0054] LSTM

[0055] in: The time step is set to 1440, corresponding to 24 hours. The model is trained using the Adam optimizer, with the loss function being the mean squared error.

[0056] S2: Physical Model Prediction

[0057] Based on the similarity law and wear formula of centrifugal pumps, predict performance degradation:

[0058]

[0059]

[0060] in: For traffic, For Yang Cheng, For rotational speed, For the gap of the sealing ring, The wear coefficient is... Runtime;

[0061] S3: Fusion Prediction

[0062] The final prediction result is output using weighted fusion:

[0063]

[0064] Where: weight The model learns adaptively and satisfies .

[0065] As a further description of the above technical solution: the digital twin model of the maintenance matching unit includes:

[0066] Module 1: Geometric Model

[0067] A 1:1 3D model, including the impeller and volute, was constructed based on the centrifugal pump CAD drawings.

[0068] The model accuracy is ±0.1mm, and it supports dynamic updates of component dimensions;

[0069] Module 2: Flow Field Model

[0070] Based on the Navier-Stokes equations and the k-ε turbulence model, fluid flow is simulated using real-time flow rate and pressure collected by the parameter monitoring unit as boundary conditions.

[0071]

[0072] in For fluid density, For the velocity vector, For pressure, Dynamic viscosity

[0073] Module 3: Loss Model

[0074] Sealing ring wear model:

[0075] Impeller cavitation model:

[0076] in: This refers to the amount of wear. The flow velocity of the medium in the flow channel is calculated based on the flow characteristics. For pressure difference, This is the required net positive suction head (NPSH).

[0077] As a further description of the above technical solution: the dynamic matching process of the maintenance matching unit includes:

[0078] Recession Level Classification:

[0079] Performance degradation rate based on the output of the degradation prediction unit It is divided into three levels of decline:

[0080] Mild decline ( ), moderate recession ( ), severe recession ( );

[0081] Strategy generation:

[0082] In the case of mild degradation, the optimal operating parameters are simulated using a digital twin model, and "online adjustment suggestions" are output.

[0083] During moderate degradation, a "replacement list of vulnerable parts" is generated by combining component wear data from a digital twin model.

[0084] In cases of severe degradation, a digital twin model is used to simulate the performance of key components such as the impeller after replacement, and an "emergency repair plan" is generated.

[0085] As a further description of the above technical solution: the verification module of the maintenance matching unit includes:

[0086] Data Acquisition: After maintenance, parameters such as flow rate and vibration are collected at a sampling period of weight 3 (1 second / sample) to obtain actual performance data.

[0087] Effect verification: comparison Performance of digital twin model simulation Calculate the deviation rate:

[0088]

[0089] like If deemed satisfactory, a correction is triggered.

[0090] like This triggers an alert;

[0091] like Maintenance failed.

[0092] Model correction: Adjust the loss parameters of the digital twin model and update them synchronously to the decay prediction unit;

[0093]

[0094] in: The correction coefficients (0.1 for mild recession, 0.2 for moderate recession, and 0.3 for severe recession) are used to synchronize the corrected data to the multi-dimensional prediction model of the recession prediction unit.

[0095] The present invention has the following beneficial effects:

[0096] 1. In this invention, by coordinating the monitoring of the main flow channel and the sealing ring gap flow channel, combined with the multi-dimensional prediction model and the virtual-real mapping of digital twin, the performance degradation of centrifugal pumps can be accurately predicted. It can provide early warning of impeller wear, sealing ring leakage and other faults 24-48 hours in advance, effectively avoiding unplanned shutdowns.

[0097] 2. In this invention, relying on the maintenance strategy pre-simulation and closed-loop optimization mechanism of the digital twin model, the invention can dynamically match the maintenance plan according to the degradation level, which not only avoids the excessive cost of traditional fixed-cycle maintenance, but also solves the lag of passive maintenance, thereby improving the overall operating efficiency of the equipment. Attached Figure Description

[0098] Figure 1 This is a flowchart of a centrifugal pump condition prediction and maintenance system based on digital twins proposed in this invention. Detailed Implementation

[0099] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0100] Reference Figure 1 The digital twin-based centrifugal pump condition prediction and maintenance system provided by this invention is applicable to single-stage centrifugal pumps (model IS65-50-160) with a power of 55kW in fields such as chemical engineering and water conservancy. These pumps have a rated flow rate of 50 m³ / h, a head of 32 m, and a speed of 2900 r / min. The system achieves precise control over the performance degradation of centrifugal pumps through a closed-loop process of "parameter monitoring - feature extraction - degradation prediction - maintenance matching - effect verification." The core unit consists of the following components:

[0101] Control Center: Adopts an industrial-grade control cabinet (size 800×600×300mm), with a built-in Siemens S7-1500 PLC (CPU 1511C-1 PN) as the main controller, equipped with a 1TB SSD redundant storage module (RAID1 backup) and a 10.1-inch touch screen (resolution 1280×800). The control center communicates with each unit via Ethernet (Profinet protocol), with a communication latency of ≤50ms. It can display centrifugal pump operating parameters, predictive curves and maintenance suggestions in real time, and supports historical data query (retaining 365 days of records).

[0102] Parameter monitoring unit: The following sensors are installed in the main flow channel (impeller inlet to volute outlet) and the sealing ring gap flow channel of the centrifugal pump:

[0103] Electromagnetic flowmeter: Model DN65-MAG (measuring range 0-100m³ / h, accuracy ±0.5%), installed in a straight pipe section 3 times the pipe diameter (φ65mm) before the volute outlet, used to collect real-time flow parameters and output a 4-20mA current signal.

[0104] Paired differential pressure transmitter: Model DP3051 (measuring range 0-1MPa, accuracy ±0.075%), the high-pressure end is connected to the volute outlet flange, and the low-pressure end is connected to the impeller inlet flange. The pressure difference is collected through the pressure tapping pipe (φ10mm stainless steel pipe) to calculate the head parameters.

[0105] Three-dimensional vibration sensor: Model VS-3000 (frequency response 1-2000Hz, sensitivity 100mV / g), fixed to the outer wall of the volute (100mm from the outlet) by a magnetic base, to collect X, Y, and Z three-dimensional vibration acceleration, reflecting the degree of turbulence in the main flow field.

[0106] High-frequency vibration sensor: Model VS-HF5K (sampling frequency 5kHz, measurement range 0-50g), installed on the pump housing corresponding to the sealing ring (50mm away from the bearing seat), to collect high-frequency vibrations (1000-5000Hz) caused by abnormal gaps.

[0107] Temperature sensor: PT100 platinum resistance thermometer (measuring range -20℃ to 150℃, accuracy ±0.1℃), embedded in the outer pump housing of the sealing ring, to monitor frictional temperature rise caused by gap leakage.

[0108] Specifically, the distance between the electromagnetic flowmeter and the volute outlet is specified as "3 times the pipe diameter (195mm for a φ65mm pump)" to avoid measurement errors caused by turbulent flow field at the volute outlet.

[0109] The vibration sensor is fixed by M8 bolts to a pre-set threaded hole on the outer wall of the volute (hole depth 15mm) to ensure that the resonance frequency avoids the 1-2000Hz characteristic frequency band.

[0110] Further parameter acquisition and preprocessing include:

[0111] I. Signal Conversion and Sampling:

[0112] The analog signals output by the sensors (such as the 4-20mA current of the electromagnetic flowmeter and the mV-level voltage of the vibration sensor) are converted into digital signals by an A / D converter (model ADS1115, 16-bit accuracy). Flow and pressure parameters are sampled at a fixed rate of 1 second per sample, while vibration parameters are sampled at 1kHz by default. When the vibration amplitude exceeds 1g, the sampling automatically switches to a 10kHz high-frequency sampling rate (for 5 seconds).

[0113] Data preprocessing: A 5-point moving average filter is used to eliminate random noise (e.g., the flow rate data is filtered from 125.3, 126.1, 125.8, 126.5, and 125.9 to 125.9); vibration data is subjected to 3-layer noise reduction using the db4 wavelet basis, retaining the 20-1000Hz frequency band (corresponding to the characteristic frequencies of the impeller and bearing).

[0114] Data transmission and storage: The preprocessed data is encapsulated in JSON format (e.g., {"timestamp":"2025-08-01T10:00:00","flow":50.2,"head":31.8,"vibrationX":0.12}), uploaded to the control center via industrial Ethernet, stored on an SSD, and an index file is generated (named by "device number + date", e.g., "Pump01_20250801.dat").

[0115] II. Feature Extraction

[0116] Sliding window construction: For low-frequency parameters such as flow rate and head, a sliding window (containing 300 sampling points, with a step size of 1 minute) is generated every 5 minutes; for vibration data, a 1024-point window (with a step size of 0.0512 seconds and 50% overlap) is generated every 0.1024 seconds to ensure feature continuity.

[0117] Temporal feature extraction:

[0118] Taking flow rate as an example, the average value of 300 sampling points within a certain window Standard deviation rate of change (Reflecting a slight downward trend in flow rate); fluctuation coefficient of head parameters (Reflects pressure stability);

[0119] Feature vector construction:

[0120] The above features are combined into a 12-dimensional vector (e.g., ℃), which is then normalized to the [0,1] interval using Min-Max and used as the input to the prediction model.

[0121] III. Performance Degradation Prediction

[0122] Data-driven prediction:

[0123] An LSTM neural network (3 hidden layers, 64 neurons per layer) is input with a 24-hour feature vector time series (1440 time steps) and outputs parameter predictions for the next 72 hours. For example, it predicts that in the 24th hour, the flow rate will drop to 48.5 m³ / h, the head will drop to 30.2 m, and the vibration amplitude will rise to 0.2 g.

[0124] Physics model assistance:

[0125] Based on the similarity law of centrifugal pumps, when the speed decreases from 2900 r / min to 2750 r / min, the predicted flow rate decreases from 50 m³ / h to 47.4 m³ / h (which conforms to Q2=Q1×n2 / n1); combined with the seal ring wear formula, it is predicted that after 300 hours of operation, the gap will increase from 0.1 mm to 0.15 mm (wear coefficient k=0.0001 mm / (h⋅(m³ / h)²)).

[0126] Fusion prediction results:

[0127] The weighted fusion of data-driven and physical model results (weights w1=0.7, w2=0.3) outputs the comprehensive conclusion: "Flow rate will decrease by 2.5% in the next 72 hours, the sealing ring will experience slight wear, and the performance degradation rate R=12% (slight degradation)."

[0128] IV. Maintenance Strategy Matching

[0129] Steps for building a digital twin model:

[0130] Geometric model: "Based on the centrifugal pump CAD drawings (providing key dimensions: impeller diameter 160mm, volute base circle diameter 200mm), parametric modeling was performed using SolidWorks, where the impeller blade thickness is dynamically adjustable according to the wear amount of 0-1mm (each 0.1mm is a step)";

[0131] Flow field model: "The mesh generation adopts structured mesh (5mm mesh size for impeller flow channel and 8mm mesh size for volute), 2000 iteration steps, convergence criterion residual ≤1e-6, and the error between the actual flow field and the measured flow field is ≤2% (attached is a comparison figure of flow velocity distribution under a certain working condition)."

[0132] Digital twin model support:

[0133] Geometric model: A 1:1 3D model was constructed using SolidWorks (impeller diameter 160mm, volute curvature radius 200mm), with high accuracy. mm, which can dynamically update the impeller wear (e.g., from 0 mm to 0.5 mm).

[0134] Flow field model: Fluent software was used for simulation, with real-time flow rate and pressure as boundary conditions, to simulate the velocity distribution (impeller outlet velocity 12 m / s) and pressure loss (volute loss 1.2 mH2O) in the main flow channel, and the deviation from the measured data was analyzed. .

[0135] Loss model: The wear of the sealing ring is based on calculate( For the medium flow rate, (for clearance pressure difference), impeller cavitation risk is calculated as follows: Assessment (current) hour, ).

[0136] Tiered maintenance strategy:

[0137] Mild degradation (R=12%): Digital twin model simulation shows that reducing the speed from 2900 r / min to 2800 r / min can reduce impeller wear, predict a stable flow rate of 48.2 m³ / h, and improve efficiency by 1.5%. The system outputs an "online adjustment suggestion": "Adjust the inverter frequency to 46.7 Hz and maintain the inlet pressure at 0.3 MPa."

[0138] Moderate degradation (e.g., R=30%): Model simulation shows that the sealing ring gap increases to 0.2mm, resulting in a 5% increase in leakage. A "Vulnerable Parts Replacement List" is generated: "Replace HG / T 20592 standard sealing ring (material: nitrile rubber, inner diameter 50mm), estimated downtime of 2 hours, spare parts inventory is sufficient."

[0139] Severe degradation (e.g., R=60%): The model simulates impeller blade wear of 3mm, and turbulent flow field leading to a 18% decrease in head. The output "Emergency Repair Plan" is: "Replace the impeller with 2Cr13 material (model IS65-50-160), and transfer it from the spare parts warehouse (expected arrival in 12 hours). Disassembly and assembly steps are attached: ① Remove pump casing bolts → ② Extract rotor → ③ Replace impeller → ④ Realign."

[0140] V. Verification of Maintenance Effectiveness

[0141] Data Acquisition: Within one hour after maintenance, collect parameters such as flow rate and vibration at 1 second / time. For example, after replacing the sealing ring, the measured flow rate was 49.5 m³ / s. h, vibration amplitude 0.08g.

[0142] Performance Verification: Comparison of post-maintenance performance with digital twin simulation (predicted traffic 49.8m). h, vibration 0.07g), calculate the deviation rate. ( ), .

[0143] Model correction: If the deviation rate after a certain maintenance If an early warning is issued, the wear coefficient of the sealing ring in the loss model will be adjusted. (Revised from 0.00001 to 0.000108), and synchronously updated to the recession prediction unit to ensure the accuracy of subsequent predictions.

[0144] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A centrifugal pump condition prediction and maintenance system based on digital twins, characterized in that, include: The control center, as the core control hub for centrifugal pump status prediction and maintenance, is used to receive, transmit, and record data from each unit and coordinate their operation. The parameter monitoring unit is connected to the control center via signal. The detection end of the parameter monitoring unit is installed in the key flow channel of the centrifugal pump to collect four core parameters that are directly related to performance degradation. The key flow channels include the main flow channel from the impeller inlet to the volute outlet and the sealing ring gap flow channel between the impeller and the pump casing. The four core parameters specifically include: flow rate parameter, head parameter, efficiency parameter, and vibration parameter; The feature extraction unit, integrated in the control center, receives four parameter data collected by the parameter monitoring unit at the key flow channel of the centrifugal pump and extracts them to form feature parameters. A decay prediction unit is integrated into the control center. The decay prediction unit is connected to the feature extraction unit and the control center signal, and a multi-dimensional prediction model is constructed based on feature parameters. The maintenance matching unit, integrated in the control center, has a built-in digital twin model. The maintenance matching unit is connected to the degradation prediction unit and the control center. Based on the degradation level classification data obtained from the multi-dimensional prediction model, the degradation level is divided into three levels. Different maintenance methods are adopted according to different levels to achieve dynamic matching and optimization of maintenance strategies.

2. The centrifugal pump condition prediction and maintenance system based on digital twin according to claim 1, characterized in that: The detection end of the parameter monitoring unit includes: Electromagnetic flowmeter: used to measure flow parameters, installed in the main flow channel from the impeller inlet to the volute outlet; Paired differential pressure transmitters: used to calculate head parameters, installed at the impeller inlet section and the volute outlet section respectively; Three-dimensional vibration sensor: used to collect vibration parameters caused by turbulent flow field, installed on the outer wall of the volute; High-frequency vibration sensor: used to collect high-frequency vibration parameters caused by abnormal gaps, installed at the pump casing position corresponding to the sealing ring; Temperature sensor: Used to help determine the frictional temperature rise caused by gap leakage, and installed at the pump casing position corresponding to the sealing ring.

3. The centrifugal pump condition prediction and maintenance system based on digital twin according to claim 1, characterized in that: The parameter monitoring unit's data acquisition process specifically includes: S1: Signal Conversion and Sampling The analog signal output from the parameter monitoring unit is converted into a digital signal using an A / D converter. The conversion formula is as follows: Where: D is the digital quantity, A is the analog input, Amin / Amax is the analog range, and n is the number of bits of the A / D converter; Flow and pressure parameters are sampled at a 1-second period, and vibration parameters are sampled at a 10kHz frequency. S2: Data Preprocessing The moving average filter is applied to the sampled data, and the calculation formula is as follows: Where: N is the window size, and X(t) is the original data at time t; Wavelet denoising was performed on the vibration data to retain the characteristic frequencies of 20-1000Hz; S3: Data Transmission and Storage The preprocessed data is uploaded to the control center via industrial Ethernet; It uses dual hard drive storage, and the data retention period is ≥1 year.

4. The centrifugal pump condition prediction and maintenance system based on digital twin according to claim 1, characterized in that: The feature extraction process of the feature extraction unit includes: S1: Sliding Window Construction A 5-minute sliding window (1-minute window step) was constructed for the flow and pressure parameters, with each window containing 300 sampling points; A 1024-point sliding window (window step size 512 points) is constructed for the vibration parameters. S2: Temporal Feature Extraction Calculate the mean within the window Standard deviation and rate of change in Number of data points within the window (flow / stress window) Vibration window ); S3: Frequency Domain Feature Extraction Perform an FFT transform on the vibration window data to extract the characteristic frequency amplitudes. and energy percentage in The number of FFT points (taken as 1024). Characteristic frequency; S4: Feature Vector Construction The extracted time-domain and frequency-domain features are combined into a 12-dimensional feature vector; 。 5. A centrifugal pump condition prediction and maintenance system based on digital twins according to claim 4, characterized in that: The multi-dimensional prediction model of the recession prediction unit includes: S1: Data-driven prediction Using an LSTM neural network, the input feature vector The time series data is used to output the parameter prediction values ​​for the next 72 hours. LSTM in: The time step is set to 1440, corresponding to 24 hours. The model is trained using the Adam optimizer, with the loss function being the mean squared error. S2: Physical Model Prediction Based on the similarity law and wear formula of centrifugal pumps, predict performance degradation: in: For traffic, For Yang Cheng, For rotational speed, For the gap of the sealing ring, The wear coefficient is... Runtime; S3: Fusion Prediction The final prediction result is output using weighted fusion: Where: weight The model learns adaptively and satisfies .

6. The centrifugal pump condition prediction and maintenance system based on digital twin according to claim 5, characterized in that: The digital twin model of the maintenance matching unit includes: Module 1: Geometric Model A 1:1 3D model, including the impeller and volute, was constructed based on the centrifugal pump CAD drawings. The model accuracy is ±0.1mm, and it supports dynamic updates of component dimensions; Module 2: Flow Field Model Based on the Navier-Stokes equations and the k-ε turbulence model, fluid flow is simulated using real-time flow rate and pressure collected by the parameter monitoring unit as boundary conditions. in For fluid density, For the velocity vector, For pressure, Dynamic viscosity Module 3: Loss Model Sealing ring wear model: Impeller cavitation model: in: This refers to the amount of wear. The flow velocity of the medium in the flow channel is calculated based on the flow characteristics. For pressure difference, This is the required net positive suction head (NPSH).

7. A centrifugal pump condition prediction and maintenance system based on digital twins according to claim 6, characterized in that: The dynamic matching and optimization of the maintenance strategy of the maintenance matching unit includes: Recession Level Classification: Performance degradation rate based on the output of the degradation prediction unit It is divided into three levels of decline: Mild decline ( ), moderate recession ( ), severe recession ( ); Strategy generation: During mild degradation, the optimal operating parameters are simulated using a digital twin model, and "online adjustment suggestions" are output. During moderate degradation, a "replacement list of vulnerable parts" is generated by combining component wear data from a digital twin model. In the event of severe degradation, a digital twin model is used to simulate the performance after impeller replacement and output an "emergency repair plan".

8. A centrifugal pump condition prediction and maintenance system based on digital twins according to claim 7, characterized in that: The verification module of the maintenance matching unit includes: Data Acquisition: After maintenance, flow rate and vibration parameters are collected at a sampling period of weight 3 (1 second / sample) to obtain actual performance. Effect verification: comparison Performance of digital twin model simulation Calculate the deviation rate: like If deemed satisfactory, a correction is triggered. like This triggers an alert; like Maintenance failed. Model correction: Adjust the loss parameters of the digital twin model and update them synchronously to the decay prediction unit; in: The correction coefficients (0.1 for mild recession, 0.2 for moderate recession, and 0.3 for severe recession) are used to synchronize the corrected data to the multi-dimensional prediction model of the recession prediction unit.

Images