A mixed flow channel differential pressure flow measuring system of a large inclined axial flow pump and a mounting method thereof

Abstract

The application provides a mixed flow channel differential pressure flow measurement system of a large-scale inclined axial flow pump and a mounting method thereof, comprising: a pump station water inlet flow channel, which is characterized in that a front section is smoothly connected to a rear section by a concrete material flow channel and a metal material flow channel; a pressure sensing assembly, which comprises pressure taking structures arranged at cross sections of the concrete material flow channel and the metal material flow channel respectively, pressure introduction pipelines in communication with the pressure taking structures, and a differential pressure sensor connected between the two pressure introduction pipelines, and is used for obtaining a fluid pressure difference signal between the two different material cross sections; a flow measurement processing unit in communication connection with the differential pressure sensor; the flow measurement processing unit is configured to receive the pressure difference signal output by the differential pressure sensor, and calculate a real-time flow of a pump set according to a pressure difference-flow mathematical model rated in advance through a device model test. The application has the characteristics of strong anti-interference ability, standardized installation process and low cost by utilizing the structural characteristics of the pump station flow channel.

Description

Technical Field

[0001] This invention belongs to the field of water conservancy engineering and pump station monitoring technology, specifically relating to a differential pressure flow measurement system for a large inclined axial flow pump and its installation method. Background Technology

[0002] Large-scale inclined axial flow pumps, due to their significant advantages such as low head, large flow rate, and shallow discharge depth, are widely used in water conservancy pumping stations in plain river networks and soft soil foundation areas, undertaking key tasks such as tide control, flood drainage, and clean water diversion. In smart water conservancy construction and energy-saving operation management of pumping stations, flow rate is a core parameter for evaluating pumping station efficiency, calculating energy consumption per unit area, and optimizing scheduling strategies.

[0003] However, the following technical bottlenecks and application challenges currently exist for online flow monitoring of such large inclined axial flow pump units: 1. Limitations of existing flow monitoring technologies The limitations of ultrasonic flow meters: The inlet channel of large inclined axial flow pumps is usually a bent elbow-shaped reinforced concrete structure, and a central diaphragm is often installed to meet structural requirements. This diaphragm greatly restricts and compresses the transducer installation area of ​​the ultrasonic flow meter. The channel cross-section is flat and wide, irregular in shape, and has a large streamline curvature. This complex channel structure cannot provide a sufficiently long straight pipe section for the ultrasonic flow meter, resulting in extremely uneven flow velocity distribution. The outlet channel exhibits obvious flow deviation characteristics and complex flow patterns, severely limiting the measurement accuracy of the acoustic method. In addition, ultrasonic flow meters are expensive (usually with a very high cost per unit), require purging of the channel during installation, and under flood drainage conditions, suspended sediment and impurities in the water can easily interfere with the acoustic signal, causing data jumps or even loss.

[0004] Point-type flow meters lack representativeness: Traditional impeller-mounted flow measurement blades or insertion-type flow meters are essentially point measurements. Due to the huge cross-section of the flow channel in large pumping stations, the velocity difference between the pipe wall and the center is significant. Relying on the velocity at a single point or a few points cannot accurately represent the average velocity of the entire cross-section, resulting in a large error in the integrated flow rate calculation. At the same time, insertion-type devices (such as flow-around tubes) require cantilever installation, have poor structural stability, and can disturb the original flow field, affecting the pump's inlet efficiency.

[0005] Traditional differential pressure methods are susceptible to clogging and signal fluctuations: While differential pressure flow measurement offers cost advantages, in practical engineering, pressure taps are directly in contact with the raw water, and often involve simultaneously measuring the total pressure directly facing the incoming flow and the lateral pressure at a specific angle, using the dynamic pressure difference between the two to calculate the velocity and flow rate. During flood drainage or water diversion, water quality is typically poor, containing large amounts of silt, aquatic plants, and other impurities, easily clogging conventional static pressure gauges and causing measurement failure. Furthermore, due to the strong turbulence and pressure pulsations caused by dynamic-static interference within the flow channel, the directly acquired raw pressure signal often contains a large amount of high-frequency noise, resulting in drastic fluctuations in the calculated flow rate value, making it difficult to use for precise control.

[0006] 2. Installation challenges of concrete flow channel pressure testing devices The inlet section of a large inclined axial flow pump is typically a concrete structure, and installing a static pressure gauge at this location presents significant construction challenges, as there is currently a lack of standardized and reliable procedures. Pre-embedded positioning is difficult: During the civil construction phase, if the pressure measuring pipe is directly pre-embedded, the pressure measuring pipe is prone to displacement or tilting due to the impact and vibration during the concrete pouring process. This results in the final pressure measuring hole not being perpendicular to the flow channel wall, generating additional dynamic pressure components, which seriously affects the accuracy of static pressure measurement.

[0007] Post-drilling carries significant risks: Post-drilling installation not only presents greater construction difficulties but also increases the risk of concrete chipping, cracking, and even damage to the internal reinforcing mesh. Furthermore, it is difficult to machine a smooth, flat pressure tapping orifice on the rough concrete surface; burrs and unevenness at the orifice can cause localized turbulence, further introducing measurement errors.

[0008] Poor sealing and seepage prevention: Existing installation methods cannot guarantee a rigid connection and sealing between the pressure testing components and the concrete structure. After long-term operation, leakage along the pipe wall is likely to occur, and once damaged, it is extremely difficult to repair.

[0009] 3. Lack of system reliability and backup methods Current flow monitoring systems largely rely on electronic sensors and power supply systems. In emergency situations where pumping stations encounter extreme weather or power system failures that cause automated monitoring to malfunction, there is often a lack of intuitive physical measurement methods that do not require external power as a backup, making it impossible to meet the flow observation needs under emergency conditions. Summary of the Invention

[0010] The purpose of this invention is to overcome the shortcomings of the aforementioned background technology and provide a differential pressure flow measurement system for the mixing channel of a large inclined axial flow pump and its installation method. Utilizing the structural characteristics of the pump station's own flow channel, it features strong anti-interference ability, standardized installation process, and low cost.

[0011] The technical solution adopted in this invention is: a differential pressure flow measurement system for the mixing channel of a large inclined axial flow pump, comprising: The pump station's inlet channel is characterized by a smooth transition from a concrete channel at the front to a metal channel at the rear. The pressure sensing component includes a pressure tapping structure disposed on the concrete flow channel section and the metal flow channel section respectively, a pressure tapping pipe connected to the pressure tapping structure, and a differential pressure sensor connected between the two pressure tapping pipes, for acquiring the fluid pressure difference signal between the two different material sections. The flow measurement and processing unit is communicatively connected to the differential pressure sensor. The flow processing unit is configured to receive the differential pressure signal output by the differential pressure sensor and calculate the real-time flow rate of the pump group based on the differential pressure-flow rate mathematical model calibrated in advance through device model testing.

[0012] In the above technical solution, the geometric structure of the water inlet channel satisfies the following constraints: The ratio of the flow area of ​​the concrete flow channel cross-section to the flow area of ​​the metal flow channel cross-section is set to a predetermined threshold or higher, so as to generate a fluid contraction pressure drop between the two cross-sections that can be precisely measured under low flow rate conditions; the shape of the concrete flow channel cross-section is rectangular or rounded rectangle, and the shape of the metal flow channel cross-section is circular.

[0013] In the above technical solution, the flow measurement processing unit has a built-in adaptive noise reduction module based on signal decomposition: This module is configured to process the raw differential pressure signal using an adaptive noise-complete ensemble empirical mode decomposition technique; This module decomposes the acquired time-series signal into several intrinsic mode function components, and automatically identifies and removes components containing channel turbulence noise and high-frequency pulsations of dynamic-static interference by calculating the complexity statistics of each component and the original signal. This module reconstructs the remaining low-frequency effective components into a high signal-to-noise ratio net differential pressure signal for subsequent flow calculation, thereby improving the stability of the measurement readings.

[0014] In the above technical solution, the pressure tapping structure is provided with several independent pressure tapping holes on the concrete material flow channel cross-section and the metal material flow channel cross-section respectively; Several pressure taps on the same cross-section are physically connected by a ring-shaped connecting pipe to form a pressure equalization structure, which is used to output the average static pressure of the cross-section to the differential pressure sensor.

[0015] The above technical solution also includes a liquid column type differential pressure display device: The device is connected in parallel between the pressure-inducing pipe of the concrete material flow channel section and the pressure-inducing pipe of the metal material flow channel section. The device is an inverted U-shaped tube structure, filled with an oily indicator liquid with a density less than water and insoluble in water. It is used to provide a backup flow indication when the flow processing unit or differential pressure sensor fails, or to visually verify the accuracy of the differential pressure sensor output on site.

[0016] In the above technical solution, the pressure difference-flow rate mathematical model is in the form of a power function: Q = K × (ΔP) a Where Q is the real-time flow rate of the pump group, ΔP is the differential pressure signal value output by the differential pressure sensor, K is the flow coefficient, a is the exponent, and K and a are constants obtained by fitting the differential pressure and flow rate data through device model tests.

[0017] In the above technical solution, the flow measurement processing unit has a built-in deep learning-based series compensation soft measurement model: The system also includes an electrical parameter acquisition interface and a blade angle acquisition interface; the electrical parameter acquisition interface is used to acquire motor current and power factor data, and the blade angle acquisition interface is used to acquire blade adjustment angle data; the electrical parameters and blade adjustment angle data are acquired through dedicated sensors or read from the existing control system of the pump station through a communication interface; The series compensation soft measurement model adopts a long short-term memory recurrent neural network architecture, receives the real-time flow rate of the pump group calculated by the differential pressure-flow mathematical model as a reference input, and uses motor current, power factor and blade adjustment angle as auxiliary correction variables. The model is configured to output a flow correction amount through the nonlinear mapping capability of a neural network, and the sum of the flow correction amount and the real-time flow of the pump group is used as the final flow output; the flow correction amount is used to automatically compensate for flow field distortion caused by changes in blade angle and flow coefficient drift under low flow conditions.

[0018] In the above technical solution, the constants K and a are obtained by calibration using the following method: On the calibrated water pump model test bench, the flow rate Q of the model pump was measured at multiple operating points. m and the corresponding pressure difference data of the model flow channel cross section; For the measured model flow Q m Converted to the prototype flow rate Q' according to the similarity criterion, the conversion formula is: Q'=(n×D) 3 ) / (n m ×D m 3 )×Q mWhere n is the prototype pump speed, D is the prototype pump impeller diameter, and n m D represents the pump speed in the model. m The impeller diameter of the model pump; Taking the common logarithm of the converted prototype flow rate Q' and pressure difference ΔP respectively, and performing linear fitting, we obtain log10(Q)=a×log10(ΔP)+b; The linear fitting result is converted into a power function form to obtain the flow coefficient K=10b and the exponent a, and K and a are written into the flow measurement processing unit.

[0019] In the above technical solution, a signal splitting device is installed on the pressure tapping pipeline to divide the pressure signal into two independent channels: The first channel is a static measurement channel, connected in series with a physical damper or voltage regulator to filter out high-frequency pressure pulsations in the flow channel before being connected to the differential pressure sensor. The signal output by the differential pressure sensor is used for flow calculation. The second channel is a dynamic diagnostic channel, which uses a straight-through undamped pipeline and connects to a high-frequency dynamic pressure sensor independent of the differential pressure sensor. This sensor is used to capture the turbulent pulsation characteristics in the flow channel to achieve online diagnosis of the blockage status of the pressure-sensing pipeline.

[0020] In the above technical solution, the flow measurement and processing unit is configured to calculate a blockage index based on the signal from the dynamic diagnostic channel to quantitatively determine the blockage status of the pressure tapping pipeline: The flow processing unit performs detrending processing on the raw signal acquired by the high-frequency dynamic pressure sensor to separate the pressure pulsation component. The flow measurement processing unit calculates the measured variance of the pressure pulsation component within a preset time window; The flow measurement and processing unit has a built-in benchmark variance model, which describes the relationship between the intensity of turbulent fluctuations and the flow rate under clean pipeline conditions. The benchmark variance model is established through self-learning in the early stage of system debugging. The formula for calculating the congestion index BI is as follows: ; This represents the measured variance.

[0021] This represents the baseline variance for the current flow rate Q. The flow measurement and processing unit executes different response strategies based on the numerical range of the blockage index BI: when BI is less than the first preset blockage threshold, the pipeline is determined to be unobstructed; when BI is between the first preset blockage threshold and the second preset blockage threshold, it is determined to be partially blocked and a maintenance warning is issued; when BI is greater than or equal to the second preset blockage threshold, it is determined to be severely blocked and an alarm is triggered; when BI is less than 0, it is determined that there may be a leak in the pipeline and a maintenance alarm is triggered.

[0022] In the above technical solution, the flow measurement and processing unit also has a built-in virtual flow takeover model under sensor failure: The system also includes a liquid level data acquisition interface for acquiring liquid level data of the inlet and outlet pools; the liquid level data is acquired through a dedicated liquid level sensor or read from the existing control system of the pump station through a communication interface. The flow processing unit defines the flow rate calculated by the differential pressure signal output by the differential pressure sensor and the differential pressure-flow mathematical model as the physical measurement flow rate. The flow measurement and processing unit runs an independent virtual flow soft measurement model in parallel. This model does not rely on the differential pressure signal, but takes the motor current and power factor collected by the electrical parameter acquisition interface, the blade adjustment angle collected by the blade angle acquisition interface, and the liquid level difference between the inlet and outlet pools measured by the liquid level sensor as inputs, and outputs the virtual flow through a pre-trained neural network model. The flow processing unit monitors the signal status of the differential pressure sensor in real time. When any of the following fault characteristics are detected, a fault flag is triggered: the sensor output signal exceeds the effective range; the rolling variance of the sensor output signal is continuously lower than the preset variance threshold for a preset duration while the motor is running; or the deviation between the physical measured flow rate and the virtual flow rate continuously exceeds the preset deviation percentage. The flow measurement processing unit also defines a start-stop transition window, which is the time interval from the detection of the pump start or stop command to the flow reaching the stability criterion; when the start-stop transition window is activated, the system automatically switches to the virtual flow output mode, and switches back to the physical measurement mode after the flow stabilizes. When the fault flag is triggered or the start-stop transition window is activated, the flow processing unit automatically switches the flow data source output by the system from physical flow measurement to virtual flow. The switching process uses a weighted smoothing algorithm to achieve a smooth transition. At the same time, a sensor fault alarm is issued but the unit operation is not stopped.

[0023] In the above technical solution, the flow measurement processing unit adopts a finite state machine management system for operating modes and data source switching: The finite state machine includes the following states and transition conditions: State 0 is the normal mode. The system outputs the sum of the physical measured flow rate and the correction amount output by the series compensation soft measurement model. At the same time, the virtual flow rate is calculated in parallel and the consistency index between the two is calculated according to a preset period. The consistency index is defined as the rolling average of the relative deviation between the physical measured flow rate and the virtual flow rate. This consistency index also serves as an online calibration reference for the differential pressure sensor. State 1 is the warning mode. When the consistency index exceeds the first preset deviation threshold, the system enters this state. The system still outputs the physical measurement flow rate but issues a calibration prompt. State 2 is the takeover mode. When the aforementioned fault flag is triggered, the system enters this state, automatically and smoothly switches to outputting virtual flow and issuing a sensor fault alarm. The transitions between states are automatically executed based on real-time criteria of the fault flag and the consistency index. When the fault condition is eliminated and the physical measurement signal returns to normal, the system automatically switches back from the takeover mode to the normal mode step by step.

[0024] In the above technical solution, a dynamic diagnostic channel is also provided on the pressure tapping pipeline. This channel is connected to a high-frequency dynamic pressure sensor to collect pressure pulsation signals. The state transition conditions of the finite state machine also include a blockage index criterion. The blockage index is calculated based on the ratio of the measured variance of the pressure pulsation signal to the reference variance, and is used to characterize the degree of blockage in the pressure tapping pipeline. When the congestion index is in the partial congestion range, the system enters state 1 warning mode and issues a maintenance prompt; when the congestion index BI reaches the severe congestion threshold, the system enters state 2 takeover mode and automatically switches to output virtual traffic.

[0025] In the above technical solution, the flow measurement and processing unit further includes a real-time device efficiency calculation module: The real-time device efficiency calculation module is configured to calculate the real-time device efficiency η of the water pump according to the following formula: Where ρ is the fluid density, g is the gravitational acceleration, and Q is the real-time flow rate output by the flow measurement and processing unit; Input active power to the motor; For net headway; The flow measurement processing unit is configured to compare the real-time device efficiency with the pre-stored design efficiency curve, and trigger an efficiency deviation alarm when the real-time efficiency is more than a preset percentage lower than the design efficiency value corresponding to the current flow.

[0026] This invention also provides an installation method for a differential pressure measurement system for a mixing channel of the large inclined axial flow pump. The pressure tapping structure of the metal channel section is achieved by setting pressure taps on the metal pipe section; the pressure tapping structure of the concrete channel section is achieved by pre-embedding a static pressure measuring head during the concrete pouring stage. The method specifically includes the following steps: Step 1: Prepare a pressure head assembly, which includes a pressure head body with internal threads, a positioning member disposed on the outside of the pressure head body, and a positioning fastener adapted to the internal threads; Step 2: Drill installation holes at the predetermined measuring point positions on the construction formwork of the concrete flow channel; Step 3: Place the pressure testing head body on one side of the inner surface of the construction template, and use the positioning fastener to pass through the mounting hole from one side of the outer surface of the construction template and screw it into the internal thread of the pressure testing head body, so that the pressure tapping end face of the pressure testing head body is tightly fitted and fixed to the inner surface of the construction template. Step 4: Rigidly connect the positioning component to the steel reinforcement skeleton inside the concrete structure, and connect and fix the pressure tapping pipe to the pressure measuring head body, so that the pressure tapping pipe is led out to the outside of the concrete structure. Step 5: Pour concrete and cure it; Step 6: After the concrete has solidified, remove the construction template and the positioning fasteners to form a pre-embedded pressure testing hole flush with the wall surface on the concrete flow channel surface; screw the pressure tap into the pre-embedded pressure testing hole and seal it.

[0027] The above technical solution also includes a measurement point optimization step prior to step two: Establish a three-dimensional numerical model of the pump station's inlet flow channel; Simulation of the flow field under rated operating conditions is performed to extract the pressure distribution cloud map of the pre-selected section. Calculate the coefficient of variation of the cross-sectional pressure distribution, and select the coordinate point with the smallest pressure fluctuation amplitude that avoids the local vortex zone as the exact location of the drilling of the construction template in step two.

[0028] In the above technical solution, the positioning fastener is a long bolt with a length greater than the thickness of the template; In step three, the depth to which the long bolt is screwed into the pressure testing head body is configured to completely fill the threaded section at the front end of the body, so as to prevent concrete slurry from seeping into the inside of the pressure testing head during the pouring process and to maintain the cleanliness of the internal threads after demolding.

[0029] The beneficial effects of this invention are: The overall technical solution of this invention fully utilizes the existing structural feature of large-scale inclined axial flow pump stations, where the inlet channel transitions from concrete to metal. A natural flow area contraction section is formed at the interface between the two different materials, thus achieving flow measurement based on the differential pressure principle without the need for additional dedicated flow measurement devices or modifications to the channel structure. This solution organically integrates flow measurement functionality with the existing civil engineering structure and unit installation process of the pump station, effectively reducing the engineering investment and construction complexity of the flow measurement system. It is particularly suitable for the simultaneous construction of new pump stations and the retrofitting of existing pump stations. Compared with traditional ultrasonic flow meters and electromagnetic flow meters, the differential pressure flow measurement system of this invention has no moving parts, requires low maintenance, and has a simple and reliable measurement principle, maintaining good measurement accuracy even under the operating conditions of large-diameter, low-flow-velocity pump stations.

[0030] By constraining the geometry of the inlet channel, the ratio of the flow area of ​​the concrete channel cross-section to that of the metal channel cross-section is ensured to be no less than a preset area ratio threshold. This guarantees a sufficient magnitude of fluid contraction pressure drop between the two cross-sections even under low flow velocity conditions. This design ensures the measurability of the differential pressure signal from the outset, avoiding insufficient measurement sensitivity due to excessively small pressure differences. Furthermore, designing the concrete channel cross-section as rectangular or rounded rectangles and the metal channel cross-section as circular conforms to the engineering realities of the pump station's inlet channel transitioning from the civil engineering section to the unit section, meeting flow measurement requirements without requiring deliberate alteration of the channel morphology.

[0031] The flow measurement processing unit incorporates an adaptive noise reduction module based on signal decomposition. This module employs adaptive noise complete ensemble empirical mode decomposition (EMD) technology to process the original differential pressure signal, effectively separating and eliminating turbulent noise in the flow channel and high-frequency pulsation components generated by impeller dynamic-static interference. The module automatically identifies noise components by calculating the complexity statistics of each intrinsic mode function component and the original signal, eliminating the need for manual setting of filtering parameters and demonstrating strong adaptability. The reconstructed net differential pressure signal after noise reduction significantly improves the signal-to-noise ratio, effectively enhancing the stability of flow measurement readings and reducing measurement errors caused by signal fluctuations.

[0032] The pressure tapping structure employs a design with several independent pressure taps on the same cross-section, connected by a ring-shaped pipeline to form a pressure equalization structure. This effectively eliminates single-point pressure tapping deviations caused by uneven flow velocity distribution or local eddies within the flow channel. The pressure equalization structure outputs the average static pressure of the cross-section to the differential pressure sensor, rather than the instantaneous pressure at a specific point. This improves the representativeness and stability of the differential pressure measurement, allowing the measurement results to more accurately reflect the overall flow state of the cross-section.

[0033] The liquid column differential pressure display device serves as a mechanical backup for the electronic measurement system. It provides intuitive flow indication in case of failure in the flow processing unit or differential pressure sensor, ensuring that operators can still determine the pump's flow status by observing the liquid column height difference at the oil-water interface even in the event of electronic equipment failure. This device can also be used for visual verification of the electronic measurement system's output values ​​during on-site inspections, providing an independent comparative reference for the accuracy of the differential pressure sensor. The inverted U-shaped tube structure, combined with an oily indicating liquid with a density less than water and insoluble in water, creates a direct linear correlation between the liquid column height difference and the pressure difference. The readings are intuitive, the principle is simple, and it operates without a power supply, improving the overall reliability and redundancy of the flow measurement system.

[0034] The pressure-flow rate mathematical model is set as a power function, which conforms to the theoretical relationship between pressure drop and flow rate in the contraction section in fluid mechanics. The model structure is simple and the physical meaning is clear. The flow coefficient K and the exponent a are obtained through device model test calibration, which can accurately reflect the hydraulic characteristics of the actual flow channel and avoid the errors that may be caused by pure theoretical calculations. The power function model has low computational load, which is convenient for real-time calculation in the flow measurement and processing unit, and meets the real-time requirements of pump station operation monitoring for flow data.

[0035] The deep learning-based series compensation soft-sensor model employs a long short-term memory recurrent neural network architecture, enabling it to learn and memorize the temporal characteristics during pump operation. The model uses the flow rate obtained from differential pressure measurement as a reference input, and motor current, power factor, and blade adjustment angle as auxiliary correction variables. Through the nonlinear mapping capability of the neural network, it outputs a flow rate correction, effectively compensating for flow field distortion caused by blade angle changes and flow coefficient drift under low flow conditions. The acquisition of electrical parameters and blade angle data is designed to allow for both acquisition via dedicated sensors and reading from the existing control system of the pump station via a communication interface. This avoids hardware redundancy and investment waste caused by repeatedly setting up sensors, and improves the system's compatibility and integration with the existing automation system of the pump station.

[0036] By conducting multi-condition tests on a certified pump model test bench, and converting the model flow rate to the prototype flow rate according to similarity criteria, followed by logarithmic linear fitting, the flow coefficient K and exponent a in the power function model can be accurately calibrated. This calibration method is based on the principle of hydraulic similarity, and the test results are traceable and credible. The obtained parameters can be reliably used for flow rate calculation of the prototype pump station.

[0037] A signal splitter is installed on the pressure tapping pipeline to divide the pressure signal into a static measurement channel and a dynamic diagnostic channel, thus decoupling the functions of flow measurement and pipeline condition diagnosis. The static measurement channel filters out high-frequency pulsations through a physical damper or voltage regulator before connecting to a differential pressure sensor, providing a stable differential pressure signal for flow calculation. The dynamic diagnostic channel uses a straight-through undamped pipeline to connect to a high-frequency dynamic pressure sensor, preserving the complete characteristic information of turbulent pulsations within the flow channel, providing a data basis for online diagnosis of pressure tapping pipeline blockage. The two channels are independent and do not interfere with each other, enabling the system to simultaneously possess both accurate flow measurement and condition self-diagnosis functions.

[0038] The method for calculating the blockage index based on dynamic diagnostic channel signals can quantitatively characterize the degree of blockage in pressure-feeding pipelines by comparing the ratio of the measured pressure pulsation variance to the baseline variance. When a pipeline is blocked, the transmission of pressure pulsations is hindered, the measured variance decreases, and the blockage index increases accordingly. This method utilizes turbulent pulsation characteristics as an indicator signal of pipeline patency, requires no additional detection equipment, and the diagnostic process is entirely online. It can issue early warnings in the early stages of blockage development, providing maintenance personnel with sufficient time to handle the situation and avoiding measurement inaccuracies caused by pipeline blockage. When the blockage index is less than 0, it can also identify potential pipeline leaks, further expanding the coverage of the diagnostic function.

[0039] The virtual flow takeover model under sensor failure provides software-level redundancy for the flow measurement system. This model uses motor current, power factor, blade adjustment angle, and inlet / outlet water level difference as inputs, and outputs virtual flow through a pre-trained neural network, completely independent of differential pressure sensor signals. When the differential pressure sensor fails or the flow is unstable during the pump start-up / shutdown transition, the system automatically switches to virtual flow output mode, ensuring the continuity of flow data. The switching process uses a weighted smoothing algorithm to achieve a smooth transition, avoiding the impact of data jumps on the downstream control system. During a failure, the system issues an alarm but does not stop the unit's operation, ensuring the availability of the flow measurement function without affecting the normal scheduling of the pump station.

[0040] The operation mode and data source switching of the flow measurement system are managed using a finite state machine, enabling the system to have clear state divisions and standardized transition logic. In normal mode, the system outputs the physical flow measurement after series compensation correction and calculates the consistency index in real time as a reference for online sensor calibration. In early warning mode, the system prompts for calibration requirements without affecting data output. In takeover mode, the system automatically switches to virtual flow and issues a fault alarm. The transitions between states are entirely based on preset criteria and executed automatically. After the fault is cleared, the system can gradually switch back to normal mode, achieving autonomous management of the flow measurement system's operating status and reducing reliance on manual intervention.

[0041] By incorporating the blockage index into the state transition criteria of the finite state machine, the system can comprehensively consider two failure modes: sensor malfunction and pressure tapping pipeline blockage. When the blockage index enters the partial blockage range, the system enters an early warning mode to prompt maintenance. When the blockage index reaches the severe blockage threshold, the system automatically switches to a takeover mode to output virtual flow. This design organically combines pipeline health status with data source switching logic, further enhancing the flow measurement system's ability to cope with complex fault scenarios.

[0042] The real-time device efficiency calculation module utilizes existing flow, level, and electrical parameter data from the flow measurement system to calculate the real-time device efficiency of the pump set online according to the standard definition formula for pump efficiency, and compares it with the pre-stored design efficiency curve. When the real-time efficiency is significantly lower than the design value, an efficiency deviation alarm is triggered, prompting operators to monitor the pump set's operating status or perform necessary maintenance. This function expands the value of the flow measurement system from simple flow measurement to the economic assessment of pump station operation, providing data support for energy conservation, consumption reduction, and optimized scheduling of pump stations.

[0043] The present invention provides a method for installing a static pressure gauge in a concrete flow channel. Before concrete pouring, the gauge assembly is pre-embedded in the construction template. Positioning fasteners are used to fix the pipe body from the outside of the template, and the positioning components are rigidly connected to the reinforcing steel frame, ensuring the gauge remains stable during concrete pouring and curing. After the concrete solidifies, the template and positioning fasteners are removed, creating a pressure gauge hole flush with the channel wall. This method synchronizes gauge installation with concrete flow channel construction, avoiding structural damage and waterproofing issues that may occur from drilling into hardened concrete later. Furthermore, the flush end face of the pressure gauge hole with the channel wall does not interfere with the flow field, ensuring the accuracy of the pressure signal.

[0044] In the measurement point location determination stage, a measurement point optimization step based on a three-dimensional fluid dynamics numerical model is introduced. Pressure distribution cloud maps of pre-selected cross-sections are extracted through simulation, the coefficient of variation of the pressure distribution is calculated, and coordinate points with minimal pressure fluctuations and avoiding local vortex zones are selected as measurement point locations. This method utilizes numerical simulation technology to pre-assess flow field characteristics, ensuring that measurement points are located in the region with the most uniform and stable static pressure distribution. This improves the representativeness of the pressure sampling signal from the source and reduces measurement errors caused by improper measurement point location selection.

[0045] By selecting long bolts with a length greater than the template thickness as positioning fasteners, and ensuring they are screwed into the pressure testing head tube (i.e., the main body) to completely fill the threaded section at the front end of the tube, cement slurry can be effectively prevented from seeping into the pressure testing head during concrete pouring. After demolding, the internal threads of the pressure testing head remain clean, allowing for direct installation and connection of the pressure tapping nozzle to the pressure tapping pipeline. This simplifies subsequent installation and commissioning work, improves construction efficiency, and enhances the forming quality of the pressure testing holes. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the architecture of the present invention; Figure 2 This is a schematic diagram of the structure of the present invention; Figure 3 This is a schematic diagram of the computational model of the present invention; Figure 4 This is a partial schematic diagram of the present invention; Figure 5 This is a partial schematic diagram b of the present invention; Figure 6 The logarithmic fitting curve of the flow rate-pressure difference relationship in Example 1; Figure 7 The logarithmic fitting curve of the flow rate-pressure difference relationship in Example 2; Figure 8 The logarithmic fitting curve of the flow-pressure difference relationship in Example 3 is shown.

[0047] Among them, 1-concrete flow channel section A, 2-metal flow channel section B, 3-pressure tap, 4-pressure tapping pipe, 5-reducer, 6-stop valve, 7-pressure stabilizer, 8-pressure gauge valve, 9-differential pressure gauge, 10-light oily measuring fluid, 11-differential pressure sensor, 12-local flow measurement unit, 13-intermediate diaphragm, 14-inlet flow channel, 15-outlet flow channel, 16-(large inclined axial flow) pump unit; 17-static pressure tap, 18-static pressure connector, 19-static pressure panel, 20-anchor bar, 21-installation bolt, 22-positioning bolt, 23-construction template, 24-nut. Detailed Implementation

[0048] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments to facilitate a clear understanding of the present invention, but these descriptions do not constitute a limitation on the present invention.

[0049] This invention provides a differential pressure flow measurement system for the mixing channel of a large inclined axial flow pump, such as... Figure 1 and Figure 2 As shown, the system makes full use of the inherent concrete-metal transition structure of the inlet channel 14 of the inclined axial flow pump to achieve high-precision flow monitoring without adding additional flow measurement devices.

[0050] The system mainly includes: the water inlet channel 14 of the pump station, the pressure sensing component, and the flow measurement and processing unit.

[0051] The pump station’s inlet channel 14 is characterized by a smooth transition from a concrete channel in the front section to a metal channel in the rear section. This structure is an inherent feature of large inclined axial flow pumps. The elbow-shaped inlet channel 14 in the front section is cast in place with reinforced concrete, and the metal pipe in the rear section is directly connected to the pump impeller chamber.

[0052] The pressure sensing component includes pressure tapping structures respectively disposed on the concrete material flow channel cross-section and the metal material flow channel cross-section, pressure tapping pipe 4 connected to the pressure tapping structure, and differential pressure sensor connected between the two pressure tapping pipes 4, for acquiring fluid pressure difference signals between the two cross-sections of different materials.

[0053] The flow processing unit is communicatively connected to the pressure sensing component and is configured to receive the fluid static pressure signal, calculate the static pressure difference between the concrete flow channel cross section and the metal flow channel cross section, and calculate the real-time flow rate of the pump group based on the pressure difference-flow mathematical model calibrated in advance through device model tests.

[0054] The measurement principle of this invention is based on Bernoulli's equation and the continuity equation in fluid mechanics. According to Bernoulli's equation, along the same streamline, an increase in flow velocity leads to a decrease in static pressure. Let the area of ​​the concrete flow channel cross-section A1 be S. A The area of ​​the metal flow channel cross section B2 is S. B Because of S A Significantly greater than S B (This embodiment requires an area ratio of not less than 3 times.) When the fluid flows from section A to section B, according to the continuity equation Q=v×S, the flow velocity will inevitably increase, and correspondingly, the static pressure P at section B will increase. B The static pressure P below section A A .

[0055] The static pressure difference between the two sections is ΔP = P A -P B There is a definite physical relationship between the pressure difference and the flow rate Q. Theoretical analysis shows that for a fully developed pipe flow, the pressure difference is proportional to the square of the flow rate. However, in actual engineering, due to the complex geometry of the elbow-shaped inlet channel 14, the existence of vortex structures in the water flow, and the wall boundary layer effect, this ideal relationship needs to be calibrated and corrected through model tests.

[0056] Compared with existing technologies, the measurement of this invention is directly based on the definite physical relationship between static pressure difference and flow rate, avoiding scale effect errors and interpolation errors caused by indirect calculations based on water level and performance curves; static pressure measurement is not sensitive to flow velocity distribution, and the use of unequal cross-section contraction design makes the flow field stable and the pressure difference representative; no additional complex structure is required, and the measurement points are arranged using existing flow channels, which significantly reduces equipment and construction costs; static pressure measurement is not affected by water quality, dirt, or air bubbles, and is suitable for drainage water quality environments with strong anti-interference ability.

[0057] like Figure 1 As shown, the components and functions of the flow measurement system in this embodiment are as follows: The pipeline connection and signal transmission configuration of the flow measurement system of this invention are as follows: Pressure tapping line 4 is made of seamless stainless steel pipe with a specification of DN25, and is responsible for transmitting the static pressure signals from section A and section B to the instrument system respectively. Pressure tapping line 4 is connected to the pressure measuring head through reducer 5 to achieve a smooth transition of pipe diameter.

[0058] The shut-off valve 6 is installed at key nodes of each pressure tapping line 4 to isolate the line during maintenance or sensor replacement, ensuring the safety and convenience of system maintenance.

[0059] The regulator 7 is located in the pressure-sensing pipeline 4 upstream of the differential pressure sensor 11. It is a sealed pressure buffer container used to attenuate the high-frequency pulsation of fluid pressure and improve the stability of the differential pressure measurement signal.

[0060] Pressure gauge valve 8 is installed on the pressure tapping lines 4 on both sides of differential pressure sensor 11 and inverted U-tube differential pressure gauge 9, and is used for venting, zero point calibration and range switching operations before commissioning.

[0061] The measuring instrument system of the present invention adopts a dual configuration of automatic and manual measurement, including two sets of differential pressure measuring devices connected in parallel: a differential pressure sensor 11 and an inverted U-tube differential pressure gauge 9.

[0062] Differential pressure sensor 11 is the main measuring element of the system. An industrial-grade intelligent differential pressure transmitter is selected, with a recommended accuracy class of 0.1. The measuring range is determined according to the maximum design differential pressure value, with a margin of 20% to 30%. The high-pressure side of the sensor is connected to the pressure tapping pipe 4 at the pressure measuring section A of the concrete flow channel, and the low-pressure side is connected to the pressure tapping pipe 4 at the pressure measuring section B of the metal flow channel of the water pump.

[0063] The sensor should be installed above the two pressure measurement sections to facilitate pipeline venting. A pressure regulator 7 is installed upstream of the sensor to attenuate pressure pulsations, and three-valve manifolds (pressure gauge valves 8) are installed on both sides for zero-point calibration and online maintenance. The sensor outputs a 4-20mA standard analog signal, transmitted via differential pressure sensor 11 cable (RVVP2×1.5mm). 2 The shielded cable transmits the signal to the local current measurement unit 12.

[0064] The inverted U-shaped differential pressure gauge 9 serves as the system's backup measuring element and calibration device, connected in parallel to the pressure tapping line 4 via pressure gauge valve 8. This device is made of transparent acrylic tube with an inner diameter of 15 to 20 mm and a total height of approximately 800 to 1000 mm. The tube is engraved with differential pressure graduations (range 0 to 2.0 m water column, graduation value 0.01 m). The internal packing density is approximately 0.8 g / cm³. 3 10g of light oily measuring fluid.

[0065] During operation, the static pressure signals from sections A and B enter the left and right legs of the U-tube, respectively. The height difference Δh between the oil and water interfaces directly reflects the pressure difference between the two sections. Operators can read the Δh value and manually calculate the flow rate by substituting it into the pressure difference-flow rate formula. This serves as an emergency measurement method in case of electronic instrument failure or as a comparison basis for periodic calibration.

[0066] This liquid column differential pressure display device provides a backup flow indication when the flow processing unit or the differential pressure sensor fails, or it is used for on-site visual verification of the accuracy of the differential pressure sensor output. During normal operation, this device serves as a redundancy verification method, allowing operators to periodically compare electronic instrument readings with the liquid column indication to promptly detect sensor zero drift or calibration offset.

[0067] For initial commissioning, the following procedure must be followed: Open pressure gauge valves 8 on both sides to allow the water pressure to be measured to enter the U-tube; open the top vent valve to release air from the pipe until the oil-water interface stabilizes; after closing the vent valve, the Δh value can be read. The local flow measurement unit 12 is powered by AC220V and has an RS485 MODBUS protocol communication port and a 4-20mA analog output function. It can upload flow data to the computer monitoring system via the communication bus.

[0068] Figure 1 The lower left is a top view of the pressure measuring section A1 of the concrete flow channel, showing the planar outline of the elbow-shaped water inlet channel 14 and the pre-embedded position of the pressure measuring head. Figure 1 The lower right corner shows a frontal view of the pressure measurement section B2 of the water pump's metal flow channel, illustrating the arrangement of pressure measurement points on the impeller and metal pipe section.

[0069] This invention sets forth specific constraints on the geometry of the inlet channel 14. The ratio of the flow area of ​​the concrete channel cross-section to the flow area of ​​the metal channel cross-section is set to a predetermined threshold, preferably three times or more.

[0070] The determination of the area ratio threshold is based on the following technical considerations: According to Bernoulli's equation, the pressure difference ΔP is proportional to the square of the velocity change, which in turn depends on the area ratio of the two cross-sections. When the area ratio is small (e.g., less than 2), the pressure difference generated even at the rated flow rate is small, potentially approaching the sensor's lower measurement limit, leading to increased relative error. Extensive model tests have verified that when the area ratio reaches 3 times or more, a fluid contraction pressure drop sufficient for precise measurement can still be generated even under low flow rate conditions (e.g., 30% of rated flow rate), ensuring the differential pressure transmitter's measured value is far from the lower limit of its range and guaranteeing measurement accuracy.

[0071] For example, in Embodiment 1, the cross-sectional area A of the water inlet channel 14 is 25.8 m². 2 The area of ​​section B is 6.3 m². 2The area ratio is 4.1; in Example 2, the area of ​​section A is 24.516 m². 2 The area of ​​section B is 5.73 m². 2 The area ratio is 4.28; in Example 3, the area of ​​section A is 15.915 m². 2 The area of ​​section B is 4.2m². 2 The area ratio is 3.8. The area ratios of all three embodiments satisfy the constraint that they are not less than 3.

[0072] The concrete flow channel has a rectangular or rounded rectangular cross-section, while the metal flow channel has a circular cross-section.

[0073] The choice of this cross-sectional shape has clear engineering rationale. The concrete flow channel uses a rectangular or rounded rectangle because the inlet flow channel 14 of a large inclined axial flow pump is typically wide and flat, requiring a central diaphragm 13 to meet structural stress requirements. A rectangular cross-section facilitates formwork construction and rebar tying, and fully utilizes the excavated space. The metal flow channel uses a circular cross-section because a circular cross-section provides the most uniform stress distribution under internal pressure, maximizes material utilization, and allows for a natural transition to the circular impeller chamber, reducing hydraulic losses.

[0074] From a fluid dynamics perspective, a circular cross-section also has the advantages of the largest hydraulic diameter and the smallest wetted perimeter, which can effectively reduce head loss along the pumping route and improve the overall efficiency of the pumping station.

[0075] like Figure 2 As shown, the large and medium-sized inclined axial flow pump unit 16 applicable to this invention adopts an inclined pump shaft arrangement, with the angle between the pump shaft and the horizontal plane typically ranging from 15° to 30°. The pump station flow channel system consists of three parts: the inlet flow channel 14, the pump unit 16, and the outlet flow channel 15. The two pressure measuring sections of this invention are respectively arranged in the concrete section (section A) of the inlet flow channel 14 and the metal pipe section (section B) in front of the pump unit 16.

[0076] The inlet channel 14 is a curved concrete structure, starting from the outlet of the inlet pool and smoothly transitioning to the metal pipe section of the water pump after a bend. A central diaphragm 13 is installed within the inlet channel 14, extending from the outlet of the inlet pool along the flow direction to the beginning of the bend. Its function is to share the structural load and improve the flow distribution. The pressure measurement section A1 of the concrete channel is located on the straight section before the bend, approximately 2 to 3 times the hydraulic diameter from the outlet of the inlet pool. The flow at this location is relatively stable, avoiding the contraction acceleration zone at the outlet of the inlet pool and the flow turning zone at the beginning of the bend. Section A is rectangular or rounded-corner rectangular, typically 6 to 8 meters wide and 3 to 4 meters high. Due to the presence of the central diaphragm 13, section A is effectively divided into two symmetrical sub-sections. Four static pressure measuring points are arranged at the center of the top, bottom, left side, and right side of the rectangular section, respectively. The locations of the measuring points should avoid the area where the central diaphragm 13 is located.

[0077] The pressure measurement section B of the water pump's metal flow channel is located on the metal pipe section after the elbow and before the impeller inlet, approximately 0.5 to 1.0 times the impeller diameter from the inlet. At this location, the flow channel has completed cross-sectional contraction and direction reversal, and the fluid tends to be axisymmetrically and uniformly distributed after entering the circular pipe, which is beneficial for obtaining a representative cross-sectional average static pressure. Section B is circular, with a diameter equal to or slightly larger than the impeller inlet diameter. The four static pressure measurement points are evenly distributed at 90° intervals around the circumference, located at the top (0°), right side (90°), bottom (180°), and left side (270°), respectively. The flow length between sections A and B is approximately 5 to 8 meters, depending on the elbow curvature radius and turning angle of the specific project. The turning angle of the elbow section is typically 60° to 90°, and the curvature radius is 1.5 to 2.5 times the pipe diameter.

[0078] The outlet channel 15 is located downstream of the pump unit 16 outlet. It also adopts a concrete structure and is equipped with a central diaphragm 13 to guide the pumped water flow to the outlet pool. The present invention chooses to arrange the pressure measuring section on the inlet channel 14 side rather than the outlet channel 15 side because the flow state of the inlet channel 14 is relatively stable, while the flow velocity distribution of the outlet channel 15 is more uneven due to the influence of the impeller outflow rotation component, which is not conducive to the representativeness of static pressure measurement.

[0079] The pressure difference-flow rate mathematical model is in the form of a power function: ; Where Q is the real-time flow rate of the pump unit (unit: m³ / s). 3 / s), ΔP is the static pressure difference between the concrete flow channel section and the metal flow channel section (unit: m water column or Pa), K is the flow coefficient, a is the exponent, and K and a are constants obtained by fitting the pressure difference and flow data through device model test.

[0080] The constants K and a are obtained by calibration using the following method.

[0081] The first step involved measuring the flow rate (Qm) and corresponding differential pressure data of the model pump at multiple operating points on a certified pump model test bench. According to the "Pump Station Design Standard," large axial flow pumps and mixed flow pumps should have model test data. The test bench's measurement system included high-precision electromagnetic flowmeters and differential pressure transmitters, with the electromagnetic flowmeters having a calibration accuracy of ±0.2% and the differential pressure transmitters ±0.075%. All measuring equipment on the test bench was calibrated and within its validity period. The static pressure monitoring points on the model flow channel were arranged proportionally to their corresponding positions on the prototype, ensuring that the model test data could be directly converted to the prototype.

[0082] The second step is to analyze the measured model flow rate Q. m Converting to the prototype flow rate Q' according to the similarity criterion, the conversion formula is as follows: Q'=(n×D 3 ) / (n m ×D m 3 )×Q m ; Where n is the prototype pump speed (r / min), D is the prototype pump impeller diameter (m), and n m D represents the model pump speed (r / min). m Let be the impeller diameter of the model pump (m). This conversion formula is based on the similarity law of pumps. Under the conditions of geometric and kinematic similarity, the flow rate ratio between the model and the prototype is equal to the speed ratio multiplied by the cube of the linear size ratio.

[0083] The third step involves taking the common logarithm of the converted prototype flow rate Q' and pressure difference ΔP, and then performing a linear fit to obtain log10(Q) = a × log10(ΔP) + b. For example... Figure 3 As shown, the reason for using logarithmic linear fitting lies in the mathematical properties of power functions. For Q = K × (ΔP) a Taking the common logarithm of both sides, we get: Log10(Q)=a×log10(ΔP)+log10(K); Let Y = log10(Q), X = log10(ΔP), and b = log10(K). Then the power function relationship is transformed into a linear relationship Y = aX + b. By using the least squares method to perform linear fitting on the scattered data, the slope a and intercept b can be obtained.

[0084] The fourth step is to convert the linear fitting result into a power function form to obtain the flow coefficient K and the exponent a, and then write K and a into the flow measurement processing unit.

[0085] The flow measurement processing unit optionally incorporates a deep learning-based multivariate fusion soft measurement model. This model employs a long short-term memory recurrent neural network (LSTM) architecture, with the static pressure difference signal as the main input variable and motor current, power factor, and blade adjustment angle as auxiliary correction variables.

[0086] Based on the power function flow measurement model described above, this invention further provides a series compensation soft measurement model based on deep learning to improve measurement accuracy under complex operating conditions. The flow measurement processing unit optionally incorporates a multivariate fusion soft measurement model based on a long short-term memory recurrent neural network (LSTM), using the static pressure difference signal as the main input variable and motor current, power factor, and blade adjustment angle as auxiliary correction variables.

[0087] The system also includes an electrical parameter acquisition interface and a blade angle acquisition interface. The electrical parameter acquisition interface is used to acquire motor voltage, motor current, and power factor data. The blade angle acquisition interface is used to acquire blade adjustment angle data.

[0088] The electrical parameters and blade adjustment angle data are acquired via dedicated sensors or read from the existing control system of the pumping station through a communication interface. In specific implementation, the appropriate data acquisition method can be selected based on the actual conditions of the pumping station. Method 1: Data is collected through dedicated sensors: a current transformer, a voltage transformer, and a power factor meter are installed on the pump motor side, and an angle encoder or potentiometer-type angle sensor is installed on the blade adjustment mechanism. The sensor signals are then directly connected to the flow measurement and processing unit.

[0089] Method 2: Read data from the existing control system of the pumping station via a communication interface: Large pumping stations are typically equipped with PLC control systems or SCADA systems, which have already collected parameters such as motor current, power factor, and blade angle. The flow measurement and processing unit can read the above data from the existing control system via industrial communication protocols such as Modbus, Profibus, and OPC, avoiding the need to repeatedly set up sensors and reducing system cost and complexity.

[0090] The key innovation of the LSTM model in this invention, which differs from traditional black-box neural networks, lies in the introduction of a physical information constraint mechanism. The conservation laws of fluid mechanics and the operating laws of the pump system are added as regularization terms to the loss function to ensure that the prediction results meet physical laws and eliminate erroneous predictions that may occur in purely data-driven models that violate common sense about physics.

[0091] The core idea of ​​the cascaded compensation architecture is to decompose traffic calculation into two cascaded stages: physical modeling and intelligent correction.

[0092] The first stage is the physical modeling stage. Using the aforementioned power function model, the estimated base flow rate is calculated based on the measured static differential pressure. .

[0093] The second stage is the intelligent correction stage. It utilizes an LSTM neural network to learn the flow correction amount. This correction is defined as the deviation between the actual flow rate and the base flow rate: in: Flow correction amount, in meters 3 / s, predicted output by LSTM neural network; Q: Actual flow rate, in meters (m³) 3 / s, obtained from high-precision calibration measurements during the training phase; Basic flow rate estimate, in m³ 3 / s, calculated using the power function model; I: Motor current, in A, obtained by real-time measurement from a current transformer; Power factor, dimensionless, is obtained in real time by a power quality analyzer. Blade adjustment angle, in degrees (°), is obtained in real time by an angle sensor (for fixed vane pumps, this input is a constant or omitted). T: Fluid temperature, in °C, obtained in real time by a temperature sensor; The nonlinear mapping function represented by an LSTM neural network; The model is configured to output a flow correction value through the nonlinear mapping capability of a neural network, and the sum of the flow correction value and the real-time flow rate of the pump set is used as the flow output under normal operating conditions. .

[0094] The flow correction is used to automatically compensate for flow field distortion caused by changes in blade angle and flow coefficient drift under low flow conditions.

[0095] In the power function model, the flow coefficient K and the exponent a are constants obtained through model testing and calibration under specific operating conditions (fixed blade angle, standard water temperature). However, in actual operation, the following factors can cause the pressure differential-flow rate relationship to deviate from the calibration curve: Flow field distortion caused by changes in blade angle. Large inclined axial flow pumps typically employ adjustable blade designs. Changes in blade angle alter the velocity triangle in front of the impeller, thus affecting the flow field structure within the inlet channel 14. When the blade angle deviates from the calibration condition, the actual flow rate corresponding to the same pressure difference will shift.

[0096] Flow coefficient drift under low flow conditions. Under deep partial load conditions, flow separation and backflow zones may occur within the inlet channel 14, leading to a reduction in the effective flow area and a drift in the flow coefficient K. The power function model, using a fixed K value, cannot adapt to this nonlinear change.

[0097] Density effect caused by water temperature changes. The density of water changes with temperature, although the change is small (about 1% in the range of 5℃ to 25℃), but compensation is still required for high-precision measurements.

[0098] Related information about motor operating status. Motor current and power factor indirectly reflect the pump's load status and operating efficiency, and are correlated with flow rate, thus serving as auxiliary correction information.

[0099] The task of the LSTM neural network is to learn the above factors and the flow correction amount. The complex nonlinear mapping relationship between them.

[0100] LSTM networks are deep learning architectures specifically designed for processing time series data, and their core lies in the collaborative working mechanism of three gating units.

[0101] Fluid systems typically exhibit large inertia and time delay characteristics. The current flow rate depends not only on the current blade angle and pump speed but also on the historical propagation of pressure within the pipeline. LSTM can capture the time correlations in fluid systems that can last for tens of seconds, effectively solving the problem of predicting dynamic transient processes that pure feedforward networks cannot handle.

[0102] This invention employs a dual-layer LSTM structure to predict flow correction amounts, with the specific configuration as follows: The input layer receives multidimensional feature vectors. It contains the following components: The basic flow rate output by the power function model, in meters (m). 3 / s; Normalized value of motor current, dimensionless. Rated current; Power factor, dimensionless; : Normalized value of blade adjustment angle, dimensionless. This is the maximum adjustment angle; Normalized value of fluid temperature, dimensionless. For reference temperature, Temperature variation range All input variables are fed into the LSTM layer after standardization preprocessing.

[0103] The first LSTM layer contains 64 hidden units, responsible for capturing short-term dynamic features of the input variables, such as current fluctuations and differential pressure pulsations. The second LSTM layer contains 32 hidden units, responsible for extracting long-term trend information, such as temperature drift and gradual changes in blade angle.

[0104] The fully connected output layer has a single neuron structure, and the output flow correction is... The predicted value. The activation function uses linear activation (identity mapping), allowing... Take a positive or negative value.

[0105] This invention incorporates the conservation laws of fluid mechanics and the operating principles of the pump system as regularization terms into the loss function. The total loss function is defined as follows: ; in: Total loss function value, dimensionless, used to guide network parameter optimization; Data-driven loss term, dimensionless, measures the deviation between network predictions and calibration data; The physical equations constrain the loss term, which is dimensionless, ensuring that the network output satisfies the physical laws of the pump system. Boundary condition constraint loss term, dimensionless, ensures that the output value meets the boundary constraints of the physical feasible region; Physical constraint weight coefficient, dimensionless, controls the relative strength of physical constraints and is dynamically adjusted through an adaptive strategy; Boundary constraint weight coefficient, dimensionless, controls the relative strength of boundary constraints and is dynamically adjusted through an adaptive strategy.

[0106] The data-driven loss term uses mean squared error to measure the deviation between the network's predicted correction and the actual correction. ; in: N: The total number of training samples, determined by the number of operating points obtained from device model tests or field calibration, with a typical value of 100~500; i: Sample index, =1,2,...,N; The network prediction correction for the i-th sample, in units of m. 3 / s, obtained by forward propagation calculation of the LSTM network. The true correction value for the i-th sample, in units of m. 3 / s, the calculation method is as follows ; The measured flow rate of the i-th sample, in m³ / s. 3 / s, obtained by high-precision flow meter measurement in device model test; The flow rate is calculated using a power function model for the i-th sample, in m³ / s. 3 / s.

[0107] The physical equation constraint loss term is used to ensure that the network output satisfies the physical laws of the pump system: ; in: The pump affinity law describes the proportional relationship between flow rate, head, and power at different speeds. When the training data covers multiple speed conditions, the correction amount predicted by the network should ensure that the corrected flow rate satisfies the following proportional relationship: ; The pump affinity law constraint loss term is defined as: ; in: M: The number of speed change condition pairs, determined by the number of pair combinations of different speed conditions in the training data; j: Index of operating condition pairs, j=1,2,...,M; ： The corrected flow prediction value for the two speed points in the j-th pair of operating conditions, in m³. 3 / s, the calculation method is as follows .

[0108] The measured speed values ​​at two speed points in the j-th pair of operating conditions, in r / min, are obtained by the speed sensor. Two speed points refer to the operating conditions of the same pump at two different speeds.

[0109] Since the correction amount of the series compensation model is relatively small compared to the base flow (usually 2% to 5%), its impact on the system energy balance is much smaller than the measurement uncertainty. Therefore, the physical constraints can be satisfied with the regularization requirements of model training by using only affinity constraints.

[0110] The boundary condition constraint loss term is used to ensure that the corrected flow value meets the boundary constraints of the physical feasible region, and it adopts the form of a ReLU penalty function: in: : Modified linear unit activation function, defined as When the input is negative, the output is zero; when the input is positive, the output is equal to the input. The corrected flow prediction value for the i-th sample, in meters. 3 / s, The pump's maximum physical flow rate, measured in cubic meters per second (m³). 3 / s, determined by the pump's design parameters, is typically 1.2 to 1.3 times the rated flow rate.

[0111] The physical meaning of this constraint is: when the flow forecast value is negative ( )hour, This results in a positive penalty; when the predicted traffic exceeds the maximum traffic ( )hour, This results in a positive penalty; when the traffic forecast is within the physically feasible region ( When ), both ReLU terms are zero, and there is no penalty.

[0112] This constraint ensures that the correction amount output by the neural network will not lead to the final flow rate value violating physical common sense, such as incorrect predictions like a large flow rate when the pump is stopped, a negative flow rate, or a flow rate exceeding physical limits.

[0113] To address the training imbalance problem caused by differences in the gradient magnitudes of different loss terms, this invention employs an adaptive weighting strategy based on gradient statistics. This strategy calculates the gradient norm of each loss term in each training iteration and dynamically adjusts the weight coefficients accordingly, ensuring that the gradient contributions of each loss term remain on the same order of magnitude, thereby achieving stable convergence of the training process.

[0114] The LSTM model was trained using the Adam optimizer with an initial learning rate of 0.001, a batch size of 64, and 500-700 iterations. Dropout (with a dropout rate of 0.1) was employed to prevent overfitting. The training process employed a curriculum-based learning strategy: initially focusing on data fitting with relatively small physical constraint weights; gradually increasing these weights as training progressed to converge the model to the physically feasible region; and finally, using an adaptive weighting strategy in the later stages to achieve balanced optimization of all constraints.

[0115] The training data for the series-compensated soft-sensor model primarily originates from synchronous multi-parameter acquisition during the device model test phase. According to the "Pump Station Design Standard," large axial-flow pumps and mixed-flow pumps should undergo device model tests. While conducting differential pressure-flow calibration tests on the test bench, the motor current, power factor, blade adjustment angle, and fluid temperature are simultaneously recorded at each test operating point. The difference between the calibrated flow rate measured by the high-precision electromagnetic flowmeter on the test bench and the flow rate calculated by the power function model is the actual flow correction amount corresponding to that operating point, serving as a training label. Test operating conditions are designed in combination based on two dimensions: blade angle and flow rate. The blade angle covers the full adjustment range, and the flow rate covers 30% to 110% of the rated flow rate. The resulting operating condition combinations provide no fewer than 100 independent training samples. During the commissioning phase after the pump station is put into operation, further on-site operating data can be collected using physically measured flow rates as labels to correct for scale effect deviations between the model test and the prototype. The acquisition, preprocessing, and model training processes of the above training data can all be implemented using machine learning engineering methods known in the field.

[0116] The power function model undertakes the main task of flow calculation, and its output has a clear hydrodynamic physical meaning (derived based on the Bernoulli equation). LSTM is only responsible for outputting the correction, which is typically 2% to 5% of the base flow and has a clear physical source (blade angle change, temperature drift, etc.). When the system malfunctions, maintenance personnel can check the base flow and the correction separately to quickly locate the root cause of the problem.

[0117] Pure neural networks need to learn the complete mapping from raw sensor signals to flow rate, requiring a large amount of training data and prone to overfitting when data is scarce. In the serial compensation architecture, LSTM only needs to learn the residuals, which have a small range of variation and strong regularity. Experiments show that when using the serial architecture, only about 100 calibration samples are needed to achieve ideal accuracy, while the end-to-end architecture requires more than 1000 samples.

[0118] Power function calculations require only one exponentiation operation (typically achieved through table lookup or Newton's iteration). LSTM networks are small in size (5 input dimensions, 96 hidden units), with a single inference time of less than 1 millisecond. The overall computational load is far lower than that of large-scale end-to-end neural networks, meeting the requirements of industrial real-time control (typical sampling periods of 100 milliseconds to 1 second).

[0119] When the LSTM module malfunctions (such as program errors or parameter corruption), the system can automatically switch to "basic mode," using only the power function model for flow calculation. Although the accuracy decreases (from ±0.8% to ±1.5%), it still provides reliable flow estimates, ensuring the safe operation of the pumping station. This degraded operation capability is something that pure neural network solutions cannot provide.

[0120] In summary, the power function + LSTM cascaded compensation architecture adopted in this invention fully leverages the reliability of the physical model and the adaptive capability of the neural network. Through the design of a loss function constrained by physical information and an adaptive weight strategy, it achieves high-precision, high-reliability, and interpretable soft flow measurement, providing an effective technical means for intelligent monitoring of large inclined axial flow pumps.

[0121] The pressure-tapping pipeline 4 is equipped with a signal splitting device to divide the pressure signal into two independent channels: The first channel is a static measurement channel, connected in series with a physical damper or voltage regulator 7. This damper filters out high-frequency pressure pulsations within the flow channel before connecting to the differential pressure sensor. The signal output by the differential pressure sensor is used for flow calculation. The voltage regulator 7 is a sealed pressure buffer container with a typical volume of 0.5 to 1.0 liters, which can effectively attenuate pressure pulsations with a frequency higher than 0.5 Hz.

[0122] The second channel is a dynamic diagnostic channel, employing a straight-through undamped conduit connected to a high-frequency dynamic pressure sensor independent of the differential pressure sensor. With a sampling rate of at least 100Hz, it captures turbulent pulsations within the flow channel to achieve online diagnosis of blockage in the pressure-sensing conduit 4. This dynamic pressure sensor is typically a piezoresistive or piezoelectric pressure sensor.

[0123] The two channels are independent and do not interfere with each other, enabling the system to simultaneously perform both accurate flow measurement and self-diagnosis.

[0124] The flow measurement and processing unit is configured to calculate a blockage index based on the signal from the dynamic diagnostic channel to quantify the blockage status of the pressure tapping pipeline 4.

[0125] The congestion index (BI) is defined as follows: B in To measure the variance of pressure pulsation, This represents the baseline variance for the current flow rate Q.

[0126] When fluid flows in a pipe, it is accompanied by a broad spectrum of turbulent pulsations. When the pressure tapping line 4 is unobstructed, these pulsations are transmitted to the sensor without attenuation, resulting in a high variance. When the line is blocked (similar to a low-pass filter), the high-frequency pulsations are attenuated, leading to a significant decrease in the measured variance.

[0127] When the pipeline is completely unobstructed, the measured variance is close to the reference variance, and BI is close to 0. When the pipeline is blocked, the transmission of pressure pulsation is hindered, the measured variance decreases, and BI increases. A value close to 1 indicates complete blockage. When BI is negative, it indicates that the measured variance is greater than the reference variance, which may indicate pipeline leakage or sensor malfunction.

[0128] The flow processing unit performs detrending processing on the raw signal acquired by the high-frequency dynamic pressure sensor to separate the pressure pulsation component. The detrending processing uses a moving average filter or a high-pass filter, with a cutoff frequency set to 0.1Hz, to remove the DC component and gradual trend from the signal.

[0129] The flow processing unit calculates the measured variance of the pressure pulsation component within a preset time window. The time window length is typically set to 10 to 30 seconds to include a sufficient number of turbulent pulsation cycles.

[0130] The flow measurement and processing unit incorporates a baseline variance model, which describes the relationship between turbulence intensity and flow rate under clean pipeline conditions. This baseline variance model is established through self-learning during the initial system commissioning phase; the specific steps are as follows: The first step, requirements for the number and distribution of operating points: With the pipeline clean, operate the pump unit and ensure stable operation at multiple flow rates. Establishing a baseline variance model requires collecting at least 10 independent steady-state operating points.

[0131] The second step, flow coverage requirements: sampling points must cover 30% to 110% of the rated flow range. Low flow rate range (30% to 60% of rated flow rate): Turbulent fluctuations are relatively weak in this range, so it is necessary to increase the sampling density to establish the lower limit of sensitivity. It is recommended to set at least 4 operating points in this range. Rated flow range (80% to 100% of rated flow): This is the most frequently used operating range and requires high-precision calibration. It is recommended to set 3 to 4 operating points. Overload zone (greater than 100% of rated flow): Determine the upper limit of pulsation, and it is recommended to set 1 to 2 operating points.

[0132] The third step is data acquisition: pressure pulsation data from the dynamic diagnostic channel is collected at each operating point, and the variance of the pressure pulsation is calculated. Each operating point should operate stably for at least 5 minutes, and the data acquisition length should be no less than 300 seconds.

[0133] Step 4, Blade Angle Dimension (for Fully Adjustable Pumps): For fully adjustable pumps, it is recommended to establish baseline curves for different blade angles (such as -4°, 0°, +4°), or to introduce the blade angle as a correction parameter into the model.

[0134] Step 5, Curve Fitting: Establish the correspondence between the baseline variance and the flow rate Q data points, and use a quadratic polynomial to perform curve fitting on the data points to obtain the baseline variance model: ; Where A, B, and C are fitting constants.

[0135] According to turbulence theory, the root mean square (RMS) value of pressure fluctuations is approximately equal to the dynamic head of the flow velocity (v). 2 It is directly proportional to σ, that is, σ∝Q 2 Therefore, the variance σ 2 Theoretically, there is a high-order correlation between the flow rate Q and the flow rate Q. In engineering approximations, quadratic polynomials typically provide the best goodness of fit (R²). 2 With a value >0.95 and low computational overhead, it is suitable for real-time computation in embedded units. Only at extremely low flow rates, if background electromagnetic noise is dominant and relatively constant, may the curve flatten; in this case, an exponential function can be used for correction.

[0136] The flow measurement and processing unit executes different response strategies based on the numerical range of the congestion index BI: When BI is less than the first preset blockage threshold, the pipeline is considered unobstructed and the system is operating normally. A typical value for the first preset blockage threshold is 0.3. When the BI (Blocking Interference Scale) falls between the first and second preset blockage thresholds, it is determined to be partially blocked, and a maintenance warning is issued, prompting operators to arrange cleaning and maintenance. A typical value for the second preset blockage threshold is 0.7. When BI is greater than or equal to the second preset blockage threshold, it is determined to be a serious blockage and an alarm is triggered; When BI is less than 0, it is determined that there may be a leak in the pipeline and a maintenance alarm is triggered.

[0137] The aforementioned blockage thresholds can be adjusted and set according to the specific pipeline characteristics and operation and maintenance requirements of the project.

[0138] When the flow rate is below 30% of the rated flow, turbulent fluctuations are extremely weak, background noise accounts for a large proportion, and the baseline variance model may fail, leading to missed or false alarms. Therefore, it is recommended to set a diagnostic blind zone and activate the blockage diagnosis function only when the flow rate is greater than 30% of the rated flow.

[0139] The flow measurement processing unit also has a built-in virtual flow takeover model for sensor failures, providing software-level redundancy for the flow measurement system.

[0140] The system also includes a liquid level data acquisition interface for acquiring liquid level data from the inlet and outlet pools. This liquid level data is collected via a dedicated liquid level sensor or read from the existing control system of the pump station via a communication interface. The liquid level sensor can be a pressure level gauge, ultrasonic level gauge, or radar level gauge, and its measurement accuracy should be better than ±0.5%.

[0141] To facilitate the description and differentiation of flow rates from different data sources, this invention defines the flow rate calculated by the differential pressure signal output by the differential pressure sensor and the differential pressure-flow rate mathematical model as the physical measurement flow rate.

[0142] The flow measurement and processing unit runs an independent virtual flow soft measurement model in parallel. This model does not rely on the differential pressure signal, but instead uses the motor current and power factor obtained from the electrical parameter acquisition interface, the blade adjustment angle obtained from the blade angle acquisition interface, and the inlet and outlet water level difference obtained from the liquid level data acquisition interface as inputs. It outputs virtual flow through a pre-trained neural network model. ; This is a typical "soft measurement" technique.

[0143] The virtual traffic soft measurement model also uses an LSTM neural network architecture, but the input variables and network structure are different from those of the serial compensation model.

[0144] The input layer receives a 4-dimensional feature vector, which contains the following components: : Normalized value of motor current, dimensionless; Power factor, dimensionless; : Normalized value of blade adjustment angle, dimensionless; Normalized value of the difference in water level between the inlet and outlet pools, dimensionless. .

[0145] The network architecture employs a two-layer LSTM, with the first layer containing 48 hidden units and the second layer containing 24 hidden units. The fully connected output layer is a single-neuron structure, outputting virtual traffic. .

[0146] The model employs a phased training strategy. In the first phase, data collected synchronously during the device model testing phase, including motor current, power factor, blade adjustment angle, inlet and outlet water level difference, and calibration flow rate, are converted into prototype parameters according to similarity criteria and used for pre-training of the network to obtain initial weights. In the second phase, during the pump station commissioning period, the model is fine-tuned using physically measured flow rates as labels, employing transfer learning methods known in the field. In the third phase, during the pump station's formal operation, when the finite state machine is in normal mode, the system automatically collects new samples at preset intervals using physically measured flow rates as labels and performs online incremental learning, gradually expanding the coverage of operating conditions. Through the accumulation of these three phases, the number of effective training samples can reach over 500 within a complete operating cycle after commissioning. During online incremental learning, known catastrophic forgetting prevention techniques are used to maintain the model's predictive ability for learned operating conditions.

[0147] The loss function of the virtual flow model also incorporates physical information constraints, including pump affinity law constraints and boundary condition constraints, to ensure that the virtual flow predictions satisfy physical laws.

[0148] Considering the number of parameters in the LSTM network (4-dimensional input layer, 48+24 units in hidden layer), in order to ensure generalization ability and prevent overfitting, the number of effective calibration samples for the virtual traffic model should not be less than 300 to 500.

[0149] Compared to a series compensation model that only learns the residuals (requiring approximately 100 samples), the virtual flow model needs to learn the full mapping from electrical parameters to hydraulic parameters, making it a more nonlinear black-box model and thus requiring more samples.

[0150] The specific dimensions and density of the operating condition coverage are as follows: Flow range: 0% (valve closed, start-up state) to 120% of rated flow; Blade angle: Full adjustment range, recommended step size no greater than 2°; Liquid level difference (net head): covers the pump station's historical lowest to highest head; Sampling density: It is recommended to use Latin hypercube sampling (LHS) or the maximum-minimum distance criterion to design test conditions to ensure uniform spatial distribution of samples and avoid data accumulation at rated test points.

[0151] Within the normal operating range (50% to 100% of rated flow), the relative error of the virtual flow should be controlled within ±3% to ±5%. Although this is lower than the ±1% accuracy of physical measurements, it is sufficient to meet the emergency dispatch and protection logic requirements during sensor failures.

[0152] The flow processing unit monitors the signal status of the differential pressure sensor in real time and triggers a fault flag when any of the following fault characteristics are detected: Fault characteristic one: The differential pressure sensor output signal is outside the effective measurement range. When the 4-20mA signal output by the differential pressure sensor is lower than 3.8mA or higher than 20.5mA, it is determined to be a signal abnormality.

[0153] Fault Characteristic Two: The rolling variance of the differential pressure sensor output signal remains below the preset variance threshold for a preset duration while the motor is running. During normal pump operation, the differential pressure signal will inevitably exhibit some pulsation due to turbulence in the flow channel. If the signal is too stable (rolling variance too low), it may indicate sensor malfunction or complete blockage of pressure tapping line 4. The preset variance threshold can be set according to the actual signal characteristics; a typical value is 10% of the normal pulsation variance. The typical preset duration is 60 seconds.

[0154] Fault characteristic three: The deviation between the physical measured flow rate and the virtual flow rate continuously exceeds the preset deviation percentage. When the relative deviation between the two flow rate estimates continuously exceeds the preset deviation percentage (e.g., ±10%) for a preset time (e.g., 300 seconds), it indicates that there may be an anomaly in the physical measurement system.

[0155] The flow measurement and processing unit also defines a start-stop transition window, which is the time interval from the detection of a pump start-up or stop command until the flow reaches a stable condition. During the pump start-stop transition phase, the flow state in the flow channel changes drastically, the differential pressure signal fluctuates greatly, and the reliability of physical measurement values ​​decreases.

[0156] Optimization of start-stop transition window processing logic: To avoid logic deadlock caused by the flow stability criterion relying on the differential pressure signal itself, a virtual flow mode (based on motor current and blade angle) should be forcibly used during the start-up phase until the running time exceeds a preset value (e.g., 60 seconds) and the differential pressure signal variance stabilizes, at which point the system should switch to physical measurement mode. The flow stability criterion can be a condition where the differential pressure signal rolling variance is less than the steady-state threshold for a certain period of time (e.g., 30 seconds).

[0157] The flow processing unit employs different switching strategies for two scenarios: start / stop transition window and sensor failure. Scenario 1: Direct switching during start-up / shutdown transition window activation: During pump startup, the differential pressure sensor has not yet established a valid steady-state differential pressure signal, and the physical flow measurement itself is unreliable or even unusable. Therefore, the conditions for using the physical flow measurement as a weighted starting point are not met. Thus, when the start-up / shutdown transition window is activated, the system directly forces the output of virtual flow without weighted smoothing. During shutdown, the flow state within the channel rapidly decays, and the differential pressure signal also loses its steady-state reference value. The system then directly switches to virtual flow output. Once the start-up / shutdown transition window closes and the flow state meets the stability criteria, the system switches back from virtual flow to physical flow measurement. The switchback process uses a reverse weighted smoothing algorithm, where the weighting coefficient α linearly increases from 0 to 1, achieving a seamless transition from virtual flow to physical flow measurement.

[0158] Scenario 2: Weighted Smooth Switching During Sensor Failure: When the differential pressure sensor fault flag is triggered, the system is in a steady-state operation phase. Although the physically measured flow rate gradually deviates from the true value due to sensor malfunction, it still has some reference value in the early stages of the fault. Therefore, the switching process in the sensor failure scenario uses a weighted smoothing algorithm to achieve a bumpless transition, avoiding the impact of data jumps on the downstream control system. The weighted smoothing algorithm is implemented as follows: during the switching transition period (e.g., 10 seconds), the output flow rate... The weighting coefficient α decreases linearly from 1 to 0.

[0159] The system issues a sensor fault alarm while switching to virtual flow output mode, but does not stop the unit operation, ensuring the continuity of flow data and the normal scheduling of the pumping station.

[0160] The virtual flow model relies on electrical signals such as motor current, power factor, and blade angle, all of which must be acquired via a PLC control system or communication bus. If a common-mode failure event occurs, causing a complete PLC power outage or communication bus malfunction, both the physical and virtual flow measurements will fail simultaneously, and the finite state machine's software switching logic will also fail. To address this common-mode failure scenario, this invention provides a final mechanical backup through the liquid column differential pressure display device. This liquid column differential pressure display device has an inverted U-shaped tube structure, and its working principle is entirely based on hydrostatics, allowing it to operate independently without any power supply or electronic signals. When a PLC power outage or communication bus malfunction causes a complete failure of the electronic measurement system, the liquid column differential pressure display device can still continuously indicate the pressure difference between the two measurement sections through the height difference at the oil-water interface. Operators can read the liquid column height difference Δh on-site and manually calculate the flow value using the calibrated pressure-flow formula, thus maintaining basic monitoring capabilities of the pump group's flow status even under extreme conditions where the electronic system is completely paralyzed. Therefore, the flow measurement system of this invention forms a three-level redundancy architecture: the first level is physical flow measurement based on differential pressure sensors (normal operating conditions); the second level is virtual flow soft measurement based on neural networks (sensor failure conditions); and the third level is manual reading based on a liquid column differential pressure display device (common mode failure conditions). The three-level redundancy covers the complete fault spectrum from single-point sensor failure to system-level electrical failure, ensuring that there is a usable flow monitoring method under any fault scenario.

[0161] The flow measurement processing unit adopts the operation mode and data source switching of the finite state machine management system, which enables the flow measurement system to have clear state division and standardized transition logic.

[0162] The finite state machine includes the following states and transition conditions: 1) State 0: Normal Mode: The sum of the system output physical measured flow rate and the correction amount output by the series-compensated soft-sensor model, i.e. Meanwhile, the system calculates virtual traffic in parallel and calculates the consistency index between the two at preset intervals (e.g., 60 seconds).

[0163] The consistency index is defined as the rolling average of the relative deviations between physical measured flow and virtual flow, and is calculated using the following formula: ; The typical length of the scrolling window is 600 seconds.

[0164] This consistency index also serves as an online calibration reference for differential pressure sensors: when CI remains at a low level (e.g., <3%), it indicates that the physical measurement system is working normally; when CI gradually increases, it may indicate that sensor calibration or maintenance is required.

[0165] 2) State 1: Warning Mode: This state is entered when the consistency index exceeds a first preset deviation threshold (e.g., 5%). The system still outputs physical measurement flow, but issues a calibration prompt to remind maintenance personnel to pay attention to the measurement system status and arrange calibration and maintenance.

[0166] 3) State 2: Takeover Mode: This state is entered when the aforementioned fault flag is triggered. The system automatically and smoothly switches to outputting virtual flow and issues a sensor fault alarm. In takeover mode, the system continuously monitors whether the fault condition has been eliminated.

[0167] The transitions between states are automatically executed based on real-time criteria of the fault flag and the consistency index. When the fault condition is eliminated and the physical measurement signal returns to normal, the system automatically switches back from the takeover mode to the normal mode step by step: first, it switches back from state 2 to state 1 for observation and verification, and after confirming that the physical measurement is stable, it switches back from state 1 to state 0.

[0168] When the system is equipped with both a dynamic diagnostic channel and a finite state machine, the blockage index can be incorporated into the state transition criteria of the state machine, enabling the system to comprehensively consider two failure modes: sensor malfunction and blockage of the pressure tapping pipeline 4.

[0169] When the congestion index BI is in the partial congestion range (i.e., between the first preset congestion threshold and the second preset congestion threshold), the system enters state 1 warning mode and issues a maintenance prompt, reminding maintenance personnel to arrange pipeline cleaning.

[0170] When the congestion index BI reaches the severe congestion threshold (i.e., greater than or equal to the second preset congestion threshold), the system enters state 2 takeover mode and automatically switches to output virtual traffic, while issuing a congestion alarm.

[0171] This design organically combines pipeline health status with data source switching logic, further enhancing the flow measurement system's ability to cope with complex fault scenarios.

[0172] The flow measurement and processing unit also includes a real-time device efficiency calculation module, which uses the existing flow rate, liquid level and electrical parameter data of the flow measurement system to calculate the real-time device efficiency of the pump set online.

[0173] The real-time device efficiency calculation module is configured to calculate the real-time device efficiency η of the water pump according to the following formula: ; in: ρ: Fluid density (kg / m³) 3 For room temperature clean water, take 1000 kg / m³. 3 ; g: acceleration due to gravity (m / s²) 2), take 9.81m / s 2 ; Q: The real-time flow rate (m³) output by the flow measurement and processing unit. 3 / s); : Motor input active power (W).

[0174] The motor input active power The motor voltage, motor current, and power factor obtained from the electrical parameter acquisition interface are calculated using the following formula: Where U is the motor line voltage (V), I is the motor line current (A), and cosφ is the power factor.

[0175] The net head is calculated using the following formula: in: : These are the water levels (m) of the outlet pool and the inlet pool obtained by the liquid level data acquisition interface, respectively.

[0176] The flow measurement and processing unit is configured to compare the real-time device efficiency η with the pre-stored design efficiency curve. The design efficiency curve is derived from the pump set's factory test report or device model test report, reflecting the pump set's design efficiency value at various flow rate operating points. When the real-time efficiency is lower than the preset percentage (e.g., 5% or 10%) of the design efficiency value corresponding to the current flow rate, an efficiency deviation alarm is triggered, prompting operators to pay attention to the pump set's operating status or perform necessary maintenance.

[0177] This function expands the value of the flow measurement system from simple flow measurement to the economic assessment of pump station operation, providing data support for energy saving, consumption reduction and optimized scheduling of pump stations.

[0178] The flow measurement and processing unit has a built-in adaptive noise reduction module based on signal decomposition. This module uses the adaptive noise complete ensemble empirical mode decomposition (CEEMDAN) technique to perform multi-scale decomposition on the original differential pressure signal to obtain a series of intrinsic mode function (IMF) components.

[0179] Fully Adaptive Noise Ensemble Empirical Mode Decomposition (CEEMDAN) is a cutting-edge technique for processing nonlinear, non-stationary fluid pressure signals. Compared to traditional EMD and EEMD, CEEMD effectively solves the mode aliasing problem by adding adaptive white noise at each decomposition stage, while ensuring the integrity of the reconstructed signal.

[0180] In the differential pressure signal processing of the pump station, the original signal x(t) is decomposed into a series of intrinsic mode functions (IMFs) and a residual term R(t): ; Among them, the high-frequency IMF component usually includes fluid turbulence noise, blade passing frequency (BPF) harmonics and electromagnetic interference, while the low-frequency IMF component characterizes the quasi-steady-state trend of flow rate changes.

[0181] The signal characteristics of each IMF component are automatically identified using the permutation entropy (PE) index: the noise-dominant component has high randomness and a large PE value; the signal-dominant component has regularity and a small PE value.

[0182] Permutation entropy is a dynamic metric for measuring the complexity of a time series. For a time series of length L, the phase space is reconstructed using the embedding dimension m and the delay time τ, and the probability distribution p(π) of each permutation pattern is calculated. Its normalized permutation entropy is defined as: The closer the value is to 1, the stronger the randomness of the sequence (noise characteristic). The closer the value is to 0, the stronger the regularity of the sequence (deterministic signal characteristics).

[0183] For the sampling frequency of the pump station differential pressure signal (usually 100Hz to 1000Hz), the embedding dimension m=5 to 7 and the delay time τ=1 are set.

[0184] Based on extensive hydraulic machinery signal analysis practice, the PE threshold... Set between 0.60 and 0.70: Noise-dominated group: When the permutation entropy of the i-th IMF component is greater than 0.70, the IMF component is mainly composed of random noise and needs to be strongly denoised or directly removed.

[0185] Signal dominance group: When the permutation entropy value of the i-th IMF component is <0.60, the IMF component contains the main hydrodynamic characteristics and should be retained.

[0186] Mixed mode region: Components between 0.6 and 0.7, which typically contain periodic pulsations caused by blade rotation and require further processing via SVD.

[0187] To adapt to different operating conditions at different speeds and flow rates, an adaptive threshold based on statistical distribution is adopted: ; in The mean of the PE values ​​for all IMF components. Here, is the standard deviation, and k is the adjustment coefficient (typically 0.5). This method can dynamically adapt to changes in turbulence intensity within the flow channel, avoiding misjudgments during drastic changes in operating conditions.

[0188] For the noise-dominated IMF component, a secondary purification technique, Singular Value Decomposition (SVD), is employed. A Hankel trajectory matrix H is constructed for the noise-dominated IMF component, and its SVD decomposition is as follows: Singular value sequence True signals typically correspond to the first few large singular values, while noise corresponds to a sequence of singular values ​​with flat tails and small values.

[0189] Singular value difference spectrum is defined as the difference between adjacent singular values: The peak positions of the differential spectrum reflect the abrupt change at the boundary between signal and noise.

[0190] The criteria for determining the effective rank k are as follows: When the subsequent difference value Decrease to the maximum difference When the value is below 10%, it is determined that all valid information has been extracted, and the index here is the valid rank k. That is: find k such that: .

[0191] This criterion ensures that the reconstructed signal retains the main hydraulic pulsation energy (such as the blade frequency component) while eliminating broadband random noise to the greatest extent possible.

[0192] The signal is reconstructed by retaining the first k largest singular values, thereby removing noise represented by small singular values. The dominant signal component is then superimposed with the purified effective signal to reconstruct a high signal-to-noise ratio net differential pressure signal for subsequent flow calculations.

[0193] The CEEMDAN-PE-SVD combined noise reduction method can improve the signal-to-noise ratio by more than 10dB, effectively suppress the influence of blade passage frequency, rotation frequency, fluid turbulence noise and environmental electromagnetic interference, and improve the stability of measurement readings.

[0194] The pressure sensing component has several independent pressure tapping points on the concrete flow channel cross-section and the metal flow channel cross-section, respectively. Several pressure tapping points on the same cross-section are physically connected by a ring-shaped connecting pipe to form a pressure equalization structure, which is used to output the average static pressure signal of the cross-section to the flow measurement and processing unit.

[0195] The core principle of the equalizing ring is based on the communicating vessels principle and the continuity assumption of Bernoulli's equation in fluid statics. In fluid dynamics, the static pressure distribution across a pipe cross-section is affected by the velocity field. When the velocity distribution is uneven (e.g., the velocity is high on the inner side and low on the outer side after a bend), the static pressure distribution also exhibits a gradient. If only a single-point measurement is used, the reading will fluctuate significantly with the circumferential angle at the pressure tap 3 position.

[0196] The pressure equalizing ring evenly arranges a plurality of static pressure tapping holes 3 on the circumference of the same cross-section, and connects these tapping holes 3 to each other through an external annular main pipe. According to Pascal's law, after the fluid in the connected ring reaches static equilibrium, its internal pressure will automatically tend to the arithmetic mean of the pressures at each tapping point.

[0197] This physical averaging mechanism acts as a natural low-pass spatial filter, eliminating local pressure fluctuations caused by flow field distortion or large-scale turbulent vortex structures before the signal enters the sensor, thus providing a stable static reference for subsequent differential pressure flow calculation.

[0198] The geometric parameters of the tapping hole 3 need to be strictly controlled: the ratio of the diameter d of the tapping hole 3 to the inner diameter D of the pipeline is controlled within the range of 0.01 < d / D < 0.03. If the aperture is too large, it will damage the boundary layer, resulting in a lower measured value; if the aperture is too small, it is prone to blockage and response hysteresis. The edge of the hole must be kept sharp and free of burrs to prevent the generation of parasitic eddies. The diameter of the pressure equalizing ring is at least twice the diameter of the tapping hole 3 to ensure the instantaneousness of pressure transmission.

[0199] The pressure sensing component arranged in the concrete flow channel adopts an integrated anti-blocking embedded module structure.

[0200] This module includes a panel flush with the inner wall of the flow channel, and tangential tapping holes 3 are opened on the panel. The water inlet direction of the tapping hole 3 is perpendicular to the water flow direction, forming tangential pressure tapping, which can effectively prevent the water flow from directly impacting the inside of the pressure measuring hole.

[0201] The present invention also includes a set of liquid column differential pressure display device, which is connected in parallel between the pressure guiding pipeline 4 of the concrete material flow channel section and the pressure guiding pipeline 4 of the metal material flow channel section.

[0202] As Figure 5 shown, this device is an inverted U-shaped tube structure, made of a transparent organic glass tube with an inner diameter of 15 to 20 mm, with a total height of about 800 to 1000 mm. The tube body is engraved with differential pressure scales, and the scale range is 0 to 2.0 m water column, and the graduation value is 0.01 m. The two vertical legs of the U-shaped tube are respectively connected in parallel to the pressure guiding pipelines 4 of section A and section B through three-way joints.

[0203] Its interior is filled with an oily indicating liquid with a density less than water and insoluble in water. Kerosene with a specific gravity of about 0.8 or a special indicating oil can be selected, and the filling amount is about the volume of the arc section at the top of the U-shaped tube. The top connecting part of the inverted U-shaped tube is filled with the indicating liquid, and the two vertical legs are connected to the measured water flow.

[0204] When the top medium is light oil, the height difference h at the oil-water interface directly reflects the pressure difference between the two measuring points. Due to the large density difference between oil and water, the change in liquid column height is more sensitive to changes in pressure difference, greatly improving the measurement resolution under low pressure difference.

[0205] The top of the inverted U-shaped tube is equipped with an exhaust branch pipe and a valve to expel air during initial filling, ensuring that the tube is completely filled with oil and water, and eliminating the influence of air lock on the measurement.

[0206] The liquid column differential pressure display device is used to provide a backup flow indication when the flow processing unit or the differential pressure sensor fails.

[0207] During normal operation, this device serves as a redundancy check, allowing operators to periodically compare electronic instrument readings with liquid column indications to promptly detect sensor zero drift or calibration deviations. In the event of a power outage or instrument malfunction, the device can immediately provide a qualitative assessment of the flow rate (large / medium / small flow range) to ensure the safe operation of the pumping station.

[0208] Based on the differential pressure reading and the calibrated differential pressure-flow rate formula, operators can manually calculate the flow rate. This dual-reading mechanism (automatic + manual) significantly improves system reliability and meets the redundancy requirements of critical infrastructure.

[0209] The present invention also provides an installation method for the above-mentioned mixed flow channel differential pressure measurement system. This installation method includes the installation of a metal flow channel pressure tapping structure and a concrete flow channel pressure tapping structure.

[0210] Among them, the installation of the metal flow channel pressure tapping structure is relatively simple. Because the metal pipe section has uniform material and high strength, standard machining (such as drilling, tapping, and welding) can be performed directly on it. After installing the pressure tapping nozzle, a pressure tapping point can be formed. The whole process is independently controllable and is not affected by other processes.

[0211] The installation of pressure tapping structures in concrete flow channels faces unique technological challenges. The installation of pressure taps on the concrete flow channel surface must be deeply intertwined with the complex civil engineering pouring process. Forming a pressure tapping hole perpendicular to the measuring surface, with a smooth and flat opening, and free from turbulence on the concrete surface is extremely difficult. During pre-embedding, slight displacement of the formwork and the impact of concrete pouring can cause misalignment. Later drilling carries the risk of edge chipping and cracking. The entire installation process is irreversible, making protection extremely difficult. Therefore, this invention achieves the installation of pressure tapping structures on the concrete flow channel side by pre-embedding static pressure taps during the concrete pouring stage, specifically including the following steps:

[0212] Step 1: Prepare the pressure testing head assembly. This assembly includes a pressure testing head body with internal threads (in this embodiment, it is a combination of a hydrostatic connector 18 and a hydrostatic panel 19), a positioning member (in this embodiment, an anchor bar 20) disposed on the outside of the pressure testing head body, and a positioning fastener (in this embodiment, a mounting bolt 21) adapted to the internal threads. Figure 4 As shown, the concrete static pressure testing head consists of a static pressure tap, a static pressure connector, a static pressure panel, and anchor bars. The static pressure connector 18 is welded to the static pressure panel 19, and the anchor bars 20 are welded to the four corners of the back of the static pressure panel 19. Before water is introduced into the flow channel, the mounting bolt 21 is unscrewed, and the static pressure tap 17 is screwed into its position.

[0213] like Figure 4 As shown, the component numbers and specifications of the concrete static pressure tester are as follows: Static pressure tap 17: Made of stainless steel, with M27×2 external threads, and a total length of approximately 35mm. The front face of the tap has a 4mm diameter tapping hole 3, with the hole axis perpendicular to the end face. A wrench eye is provided on the side of the tap for applying torque during installation.

[0214] Static pressure connector 18: This is a stainless steel adapter with an internal thread of M27×2 for connecting the pressure tap and an external thread of G1 for connecting the DN25 pressure tapping line 4. The connector is fixed to the static pressure panel by welding.

[0215] Static pressure panel 19: Made of 10mm thick stainless steel sheet, with dimensions of approximately 120mm × 120mm. A 32mm diameter through hole is provided in the center of the panel for installing connectors, and three M10 × 1.5 threaded holes are provided around the perimeter for securing with positioning bolts. The panel surface in contact with the concrete should be flat and smooth.

[0216] Anchor bar 20: Four HRB400 threaded steel bars, each approximately 200mm long and 12mm in diameter, are welded to the four corners of the back of the static pressure panel. The anchor bars are tied or welded to the main steel reinforcement skeleton of the flow channel to ensure that the pressure measuring head does not shift during the pouring process.

[0217] Step 2: Drill installation holes at the predetermined measuring point positions on the construction formwork 23 of the concrete flow channel. Mark the center point of the pressure measuring head on the formwork of the water inlet flow channel 14 according to the design drawings, and drill a hole with a diameter of 32 mm at that position.

[0218] Step 3: Place the pressure testing head tube body (static pressure connector 18 and static pressure panel 19) inside the template. Use the positioning fastener to pass through the mounting hole from the outside of the template and screw it into the tube body, tightly fitting and fixing the end face of the pressure testing head tube body to the inner surface of the template. Use an installation bolt instead of a static pressure tap to screw onto the static pressure panel of the pressure testing head; pass the other end of the installation bolt through a φ32 mm drilled hole in the pouring channel template and fix it to the channel template with a nut.

[0219] Drill holes from the inside of the casting template using a φ6 mm drill bit through the three M10×1.5 positioning bolt holes on the pressure head panel, then enlarge the holes to φ13 mm from the outside of the casting template. Finally, use three M10×1.5 positioning bolts to pass through these holes and secure the static pressure panel of the pressure head.

[0220] Step 4: Rigidly connect the positioning component to the steel reinforcement cage inside the concrete structure, and connect and fix the pressure-inducing pipeline to the pressure measuring head body, so that the pressure-inducing pipeline extends to the outside of the concrete structure. Specifically, weld or tie the anchor bar 20 to the main steel reinforcement cage to ensure that the pressure measuring head will not shift due to concrete impact during pouring. Screw the first section of the pressure-inducing pipeline 4 (i.e., the first DN25 stainless steel pipe directly connected to the pressure measuring head) into the external thread G1 interface of the static pressure connector 18 of the pressure measuring head, and connect it to the subsequent section of the pressure-inducing pipeline 4 that has been pre-laid along the steel reinforcement cage, so that the pressure-inducing pipeline 4 extends from the pressure measuring head to a predetermined position outside the concrete structure for subsequent connection to the differential pressure sensor and instrument system.

[0221] Step 5: Pour concrete and cure it.

[0222] Step Six: After the concrete has solidified, remove the formwork and positioning fasteners to create a pressure testing hole flush with the wall surface of the concrete flow channel. Install the pressure tap to complete the sealing process. Specifically, when removing the formwork of the poured block, unscrew the three M10×1.5 positioning bolts and simultaneously unscrew the nut 24 from the mounting bolt 21. After removing the flow channel formwork, leave the mounting bolt 21 as a sealing bolt. Fill the three M10×1.5 bolt holes on the panel with lead and polish. Before water flows through the flow channel, unscrew the mounting bolt 21 again and screw the static pressure tap 17 into its position. After tightening, fill the wrench hole of the static pressure tap 17 with lead and polish. The inlet of the static pressure tap 17 should be perpendicular to the water flow direction to form tangential pressure tapping.

[0223] The present invention also includes a measurement point optimization step before step two.

[0224] First, a three-dimensional hydrodynamic numerical model of the pump station's inlet channel 14 is established. This model should include the complete inlet pool, inlet channel 14, central diaphragm 13, elbow section, and metal pipe section. Structured or unstructured meshing is used, with a total number of meshes not less than 1 million to ensure calculation accuracy.

[0225] Secondly, the flow field under rated operating conditions is simulated, and the three-dimensional Reynolds time-averaged Navier-Stokes equations are solved using the k-ε or SSTk-ω turbulence model. Pressure distribution contour maps of pre-selected cross sections are extracted to display the time-averaged static pressure distribution at each point on the cross section.

[0226] Finally, the coefficient of variation of the cross-sectional pressure distribution is calculated, and the coordinate point with the smallest pressure fluctuation amplitude and avoiding the local vortex zone is selected as the exact location of the drilling of the construction template in step two.

[0227] The technical benefits of measuring point optimization are significant: it avoids placing measuring points in areas with drastic pressure fluctuations, such as flow separation zones, recirculation zones, or vortex core zones, thereby reducing random measurement errors; it selects locations with relatively uniform pressure distribution, enabling single-point measurements to better represent the average cross-sectional pressure; and it reduces on-site trial-and-error costs through numerical simulation prediction.

[0228] The positioning fastener is a long bolt with a length greater than the thickness of the template. In step three, the depth to which the long bolt is screwed into the pressure testing head tube is configured to completely fill the threaded section at the front end of the tube, so as to prevent concrete slurry from seeping into the inside of the pressure testing head during the pouring process and to maintain the cleanliness of the internal threads after demolding.

[0229] The use of long bolts has a dual function: first, to provide reliable tension during the template fixing stage, ensuring that the end face of the pressure probe fits tightly against the inner surface of the template; second, to act as a temporary seal during the pouring stage, protecting the inside of the pressure probe from contamination by cement slurry.

[0230] The following examples further illustrate the application scenarios of the present invention: Example 1: A large inclined axial flow pump has a prototype impeller diameter of 2.9 m and a rotational speed of 125 r / min. A model pump with a test impeller diameter of 0.3 m and a rotational speed of 1208 r / min has an inlet flow channel (section 14A) with a cross-sectional area of ​​25.8 m². 2 The area of ​​section B is 6.3 m². 2 The area ratio is 4.1. The pressure difference of the model section at the corresponding cross-sectional position of the prototype was measured in the device model test, as shown in Table 1. The relationship between the linearly fitted log10(Q) data and the log10(ΔP) data is log10(Q) = 0.4938log10(ΔP) + 1.3836. Figure 6 Therefore, we get Q = 24.1936 × ΔP 0.4938 By writing this relational function into the software of the local flow measurement unit 12, long-term flow monitoring can be performed simply by measuring the pressure difference ΔP between sections A and B of the inlet flow channel 14. The flow rate Q can also be easily calculated by manually reading the pressure difference using an inverted U-tube differential pressure gauge 9.

[0231] Table 1. Device Model Tests and Conversions of the Invention in Example 1 Table 1. Example 2: A large inclined axial flow pump has a prototype impeller diameter of 2.7m and a rotational speed of 141 r / min. A model pump with an impeller diameter of 0.3m and a rotational speed of 1269 r / min was tested. The inlet channel 14A has a cross-sectional area of ​​24.516 m². 2 The area of ​​section B is 5.73 m². 2 The area ratio is 4.28. The pressure difference of the model section at the corresponding cross-sectional location of the prototype was measured in the device model test, as shown in Table 1. The relationship between the linearly fitted log10(Q) data and the log10(ΔP) data is log10(Q) = 0.5062log10(ΔP) + 1.4213. Figure 7 Therefore, we get Q = 26.38153128 × ΔP 0.5062 By writing this relational function into the software of the local flow measurement unit 12, long-term flow monitoring can be performed simply by measuring the pressure difference ΔP between sections A and B of the inlet flow channel 14. The flow rate Q can also be easily calculated by manually reading the pressure difference using an inverted U-tube differential pressure gauge 9.

[0232] Table 2. Device Model Tests and Conversions of the Invention in Example 2 Example 3: A large inclined axial flow pump has a prototype impeller diameter of 2.2m and a rotational speed of 172 r / min. A model pump with an impeller diameter of 0.3m and a rotational speed of 1261 r / min was tested. The inlet channel (section 14A) has a cross-sectional area of ​​15.915 m². 2 The area of ​​section B is 4.2m². 2 The area ratio is 3.8. The pressure difference of the model section at the corresponding cross-sectional position of the prototype was measured in the device model test, as shown in Table 3. The relationship between the linearly fitted log10(Q) data and the log10(ΔP) data is log10(Q) = 0.4916log10(ΔP) + 1.3411. Figure 8 Therefore, we get Q = 21.93309905 × ΔP 0.4916 By writing this relational function into the software of the local flow measurement unit 12, long-term flow monitoring can be performed simply by measuring the pressure difference ΔP between sections A and B of the inlet flow channel 14. The flow rate Q can also be easily calculated by manually reading the pressure difference using an inverted U-tube differential pressure gauge 9.

[0233] Table 3. Device Model Tests and Conversions of the Invention in Example 3. The three embodiments cover large-scale inclined axial flow pump units 16 of varying sizes (impeller diameters from 2.2m to 2.9m), and their common features include: The exponent 'a' ranges from 0.4916 to 0.5062, all very close to the theoretical value of 0.5, proving the physical rationality of the power function model. The flow coefficient K ranges from 21.93 to 26.38, depending on factors such as the channel geometry and surface roughness of the specific project. The area ratios all satisfy the constraint of not less than 3, ensuring measurement accuracy. The goodness of fit R0 is... 2 All values ​​are greater than 0.999, indicating that the pressure difference-flow rate relationship has extremely high certainty.

[0234] These results demonstrate that the flow measurement method of the present invention has universal applicability and can be extended to various large-scale inclined axial flow pump stations.

[0235] In summary, the large-scale inclined axial flow pump mixing channel differential pressure measurement system and the matching concrete flow channel static pressure measuring head installation method provided by this invention solve the following key technical problems and achieve significant technical effects: Regarding the flow measurement system: First, the measurement principle is direct and reliable. Based on the definite physical relationship between static pressure difference and flow rate, it avoids the accumulation of errors that can be indirectly calculated based on water level and performance curves, and the measurement accuracy can reach ±1% to ±2%.

[0236] Second, it is not sensitive to flow field distortion. It utilizes the inherent area ratio difference of the concrete-metal transition channel to generate pressure difference, eliminating the need for equal-diameter straight pipe sections, and is suitable for the characteristics of short and highly curved inlet channels of large inclined axial flow pumps.

[0237] Third, costs are significantly reduced. No additional flow measurement device structure is required; the sensor and installation cost per unit is only one-tenth that of an ultrasonic flow meter.

[0238] Regarding installation methods: First, precise pre-embedding ensures one-time molding. Through techniques such as template positioning and long bolt protection, the pressure probe is ensured to be precisely positioned during concrete pouring, with the orifice flush with the wall surface, eliminating the need for subsequent drilling and repair.

[0239] Second, CFD optimization reduces errors. Optimal measurement point locations are determined in advance through numerical simulation, avoiding areas of severe pressure fluctuations and improving measurement representativeness.

[0240] This invention fills the gap in dedicated flow measurement technology for inclined axial flow pumps, and has significant engineering value and promotional significance for promoting intelligent operation and maintenance and refined management of large pumping stations.

[0241] The undescribed parts of this invention are the same as or implemented using existing technology. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this invention, and these improvements and modifications should also be considered within the scope of protection of this invention. The above are merely specific embodiments of this invention, but the scope of protection of this invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this invention should be included within the scope of protection of this invention.

Claims

1. A differential pressure flow measurement system for the mixing channel of a large inclined axial flow pump, characterized in that, include: The pump station's inlet channel is characterized by a smooth transition from a concrete channel at the front to a metal channel at the rear. The pressure sensing component includes a pressure tapping structure disposed on the concrete flow channel section and the metal flow channel section respectively, a pressure tapping pipe connected to the pressure tapping structure, and a differential pressure sensor connected between the two pressure tapping pipes, for acquiring the fluid pressure difference signal between the two different material sections. The flow measurement and processing unit is communicatively connected to the differential pressure sensor. The flow processing unit is configured to receive the differential pressure signal output by the differential pressure sensor and calculate the real-time flow rate of the pump group based on the differential pressure-flow rate mathematical model calibrated in advance through device model testing.

2. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, The geometry of the water inlet channel satisfies the following constraints: The ratio of the flow area of ​​the concrete flow channel cross-section to the flow area of ​​the metal flow channel cross-section is set to a predetermined threshold or higher, so as to generate a fluid contraction pressure drop between the two cross-sections that can be precisely measured under low flow rate conditions; the shape of the concrete flow channel cross-section is rectangular or rounded rectangle, and the shape of the metal flow channel cross-section is circular.

3. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, The flow measurement processing unit has a built-in adaptive noise reduction module based on signal decomposition: This module is configured to process the raw differential pressure signal using an adaptive noise-complete ensemble empirical mode decomposition technique; This module decomposes the acquired time-series signal into several intrinsic mode function components, and automatically identifies and removes components containing channel turbulence noise and high-frequency pulsations of dynamic-static interference by calculating the complexity statistics of each component and the original signal. This module reconstructs the remaining low-frequency effective components into a high signal-to-noise ratio net differential pressure signal for subsequent flow calculation, thereby improving the stability of the measurement readings.

4. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, The pressure tapping structure has several independent pressure tapping holes on the concrete flow channel cross-section and the metal flow channel cross-section, respectively. Several pressure taps on the same cross-section are physically connected by a ring-shaped connecting pipe to form a pressure equalization structure, which is used to output the average static pressure of the cross-section to the differential pressure sensor.

5. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, It also includes a liquid column differential pressure display device: The device is connected in parallel between the pressure-inducing pipe of the concrete material flow channel section and the pressure-inducing pipe of the metal material flow channel section. The device is an inverted U-shaped tube structure, filled with an oily indicator liquid with a density less than water and insoluble in water. It is used to provide a backup flow indication when the flow processing unit or differential pressure sensor fails, or to visually verify the accuracy of the differential pressure sensor output on site.

6. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, The pressure difference-flow rate mathematical model is in the form of a power function: Q=K×(ΔP) a Where Q is the real-time flow rate of the pump group, ΔP is the differential pressure signal value output by the differential pressure sensor, K is the flow coefficient, a is the exponent, and K and a are constants obtained by fitting the differential pressure and flow rate data through device model tests.

7. The differential pressure flow measurement system for a mixed flow channel according to claim 6, characterized in that, The flow measurement processing unit has a built-in deep learning-based series compensation soft measurement model: The system also includes an electrical parameter acquisition interface and a blade angle acquisition interface; the electrical parameter acquisition interface is used to acquire motor current and power factor data, and the blade angle acquisition interface is used to acquire blade adjustment angle data; the electrical parameters and blade adjustment angle data are acquired through dedicated sensors or read from the existing control system of the pump station through a communication interface; The series compensation soft measurement model adopts a long short-term memory recurrent neural network architecture, receives the real-time flow rate of the pump group calculated by the differential pressure-flow mathematical model as a reference input, and uses motor current, power factor and blade adjustment angle as auxiliary correction variables. The model is configured to output a flow correction through the nonlinear mapping capability of a neural network, and the sum of the flow correction and the real-time flow of the pump set is used as the final flow output. The flow correction is used to automatically compensate for flow field distortion caused by changes in blade angle and flow coefficient drift under low flow conditions.

8. The differential pressure flow measurement system for a mixed flow channel according to claim 6, characterized in that, The constants K and a are obtained by calibration using the following method: On the calibrated water pump model test bench, the flow rate Q of the model pump was measured at multiple operating points. m and the corresponding pressure difference data of the model flow channel cross section; For the measured model flow Q m Converted to the prototype flow rate Q' according to the similarity criterion, the conversion formula is: Q'=(n×D) 3 ) / (n m ×D m 3 )×Q m Where n is the prototype pump speed, D is the prototype pump impeller diameter, and n m D represents the pump speed in the model. m The impeller diameter of the model pump; Taking the common logarithm of the converted prototype flow rate Q' and pressure difference ΔP respectively, and performing linear fitting, we obtain log10(Q)=a×log10(ΔP)+b; The linear fitting result is converted into a power function form to obtain the flow coefficient K=10b and the exponent a, and K and a are written into the flow measurement processing unit.

9. The differential pressure flow measurement system for a mixed flow channel according to claim 1, characterized in that, The pressure-sensing pipeline is equipped with a signal splitter to divide the pressure signal into two independent channels: The first channel is a static measurement channel, connected in series with a physical damper or voltage regulator to filter out high-frequency pressure pulsations in the flow channel before being connected to the differential pressure sensor. The signal output by the differential pressure sensor is used for flow calculation. The second channel is a dynamic diagnostic channel, which uses a straight-through undamped pipeline and connects to a high-frequency dynamic pressure sensor independent of the differential pressure sensor. This sensor is used to capture the turbulent pulsation characteristics in the flow channel to achieve online diagnosis of the blockage status of the pressure-sensing pipeline.

10. The differential pressure flow measurement system for a mixed flow channel according to claim 9, characterized in that, The flow measurement and processing unit is configured to calculate a blockage index based on the signal from the dynamic diagnostic channel to quantify the blockage status of the pressure tapping pipeline. The flow processing unit performs detrending processing on the raw signal acquired by the high-frequency dynamic pressure sensor to separate the pressure pulsation component. The flow measurement processing unit calculates the measured variance of the pressure pulsation component within a preset time window; The flow measurement and processing unit has a built-in benchmark variance model, which describes the relationship between the intensity of turbulent fluctuations and the flow rate under clean pipeline conditions. The benchmark variance model is established through self-learning in the early stage of system debugging. The formula for calculating the congestion index BI is as follows: ； This represents the measured variance; This represents the baseline variance for the current flow rate Q. The flow measurement and processing unit executes different response strategies based on the range of the blockage index BI: when BI is less than the first preset blockage threshold, the pipeline is determined to be unobstructed; When BI is between the first preset blockage threshold and the second preset blockage threshold, it is determined to be a partial blockage and a maintenance warning is issued; when BI is greater than or equal to the second preset blockage threshold, it is determined to be a serious blockage and an alarm is triggered; when BI is less than 0, it is determined that there may be a leak in the pipeline and a maintenance alarm is triggered.

11. The differential pressure flow measurement system for a mixed flow channel according to claim 7, characterized in that, The flow measurement and processing unit also has a built-in virtual flow takeover model for sensor failures. The system also includes a liquid level data acquisition interface for acquiring liquid level data of the inlet and outlet pools; the liquid level data is acquired through a dedicated liquid level sensor or read from the existing control system of the pump station through a communication interface. The flow processing unit defines the flow rate calculated by the differential pressure signal output by the differential pressure sensor and the differential pressure-flow mathematical model as the physical measurement flow rate. The flow measurement and processing unit runs an independent virtual flow soft measurement model in parallel. This model does not rely on the differential pressure signal, but takes the motor current and power factor collected by the electrical parameter acquisition interface, the blade adjustment angle collected by the blade angle acquisition interface, and the liquid level difference between the inlet and outlet pools measured by the liquid level sensor as inputs, and outputs the virtual flow through a pre-trained neural network model. The flow processing unit monitors the signal status of the differential pressure sensor in real time. When any of the following fault characteristics are detected, a fault flag is triggered: the sensor output signal exceeds the effective range; the rolling variance of the sensor output signal is continuously lower than the preset variance threshold for a preset duration while the motor is running; or the deviation between the physical measured flow rate and the virtual flow rate continuously exceeds the preset deviation percentage. The flow measurement processing unit also defines a start-stop transition window, which is the time interval from the detection of the pump start or stop command to the flow reaching the stability criterion; when the start-stop transition window is activated, the system automatically switches to the virtual flow output mode, and switches back to the physical measurement mode after the flow stabilizes. When the fault flag is triggered or the start-stop transition window is activated, the flow processing unit automatically switches the flow data source output by the system from physical flow measurement to virtual flow. The switching process uses a weighted smoothing algorithm to achieve a smooth transition. At the same time, a sensor fault alarm is issued but the unit operation is not stopped.

12. The differential pressure flow measurement system for a mixed flow channel according to claim 11, characterized in that, The flow measurement processing unit adopts the operation mode and data source switching of the finite state machine management system: The finite state machine includes the following states and transition conditions: State 0 is the normal mode. The system outputs the sum of the physical measured flow rate and the correction amount output by the series compensation soft measurement model. At the same time, the virtual flow rate is calculated in parallel and the consistency index between the two is calculated according to a preset period. The consistency index is defined as the rolling average of the relative deviation between the physical measured flow rate and the virtual flow rate. This consistency index also serves as an online calibration reference for the differential pressure sensor. State 1 is the warning mode. When the consistency index exceeds the first preset deviation threshold, the system enters this state. The system still outputs the physical measurement flow rate but issues a calibration prompt. State 2 is the takeover mode. When the aforementioned fault flag is triggered, the system enters this state, automatically and smoothly switches to outputting virtual flow and issuing a sensor fault alarm. The transitions between states are automatically executed based on real-time criteria of the fault flag and the consistency index. When the fault condition is eliminated and the physical measurement signal returns to normal, the system automatically switches back from the takeover mode to the normal mode step by step.

13. The differential pressure flow measurement system for a mixed flow channel according to claim 12, characterized in that, The pressure-tapping pipeline is also equipped with a dynamic diagnostic channel, which is connected to a high-frequency dynamic pressure sensor to collect pressure pulsation signals; the state transition conditions of the finite state machine also include a blockage index criterion. The blockage index is calculated based on the ratio of the measured variance of the pressure pulsation signal to the reference variance, and is used to characterize the degree of blockage in the pressure tapping pipeline. When the congestion index is in the partial congestion range, the system enters state 1 warning mode and issues a maintenance prompt; when the congestion index BI reaches the severe congestion threshold, the system enters state 2 takeover mode and automatically switches to output virtual traffic.

14. The differential pressure flow measurement system for a mixed flow channel according to claim 11, characterized in that, The flow measurement and processing unit also includes a real-time device efficiency calculation module: The real-time device efficiency calculation module is configured to calculate the real-time device efficiency η of the water pump according to the following formula: Where ρ is the fluid density, g is the gravitational acceleration, and Q is the real-time flow rate output by the flow measurement and processing unit; Input active power to the motor; For net headway; The flow measurement processing unit is configured to compare the real-time device efficiency with the pre-stored design efficiency curve, and trigger an efficiency deviation alarm when the real-time efficiency is more than a preset percentage lower than the design efficiency value corresponding to the current flow.

15. A method for installing a differential pressure flow measurement system for a mixing channel of a large inclined axial flow pump according to any one of claims 1 to 14, characterized in that, The pressure tapping structure of the metal flow channel section is achieved by setting pressure taps on the metal pipe section; the pressure tapping structure of the concrete flow channel section is achieved by pre-embedding a static pressure measuring head during the concrete pouring stage, specifically including the following steps: Step 1: Prepare a pressure head assembly, which includes a pressure head body with internal threads, a positioning member disposed on the outside of the pressure head body, and a positioning fastener adapted to the internal threads; Step 2: Drill installation holes at the predetermined measuring point positions on the construction formwork of the concrete flow channel; Step 3: Place the pressure testing head body on one side of the inner surface of the construction template, and use the positioning fastener to pass through the mounting hole from one side of the outer surface of the construction template and screw it into the internal thread of the pressure testing head body, so that the pressure tapping end face of the pressure testing head body is tightly fitted and fixed to the inner surface of the construction template. Step 4: Rigidly connect the positioning component to the steel reinforcement skeleton inside the concrete structure, and connect and fix the pressure tapping pipe to the pressure measuring head body, so that the pressure tapping pipe is led out to the outside of the concrete structure. Step 5: Pour concrete and cure it; Step 6: After the concrete has solidified, remove the construction template and the positioning fasteners to form a pre-embedded pressure testing hole flush with the wall surface on the concrete flow channel surface; screw the pressure tap into the pre-embedded pressure testing hole and seal it.

16. The installation method according to claim 15, characterized in that, It also includes the measurement point optimization step before step two: Establish a three-dimensional numerical model of the pump station's inlet flow channel; Simulation of the flow field under rated operating conditions is performed to extract the pressure distribution cloud map of the pre-selected section. Calculate the coefficient of variation of the cross-sectional pressure distribution, and select the coordinate point with the smallest pressure fluctuation amplitude that avoids the local vortex zone as the exact location of the drilling of the construction template in step two.

17. The installation method according to claim 15, characterized in that: The positioning fastener is a long bolt with a length greater than the template thickness; In step three, the depth to which the long bolt is screwed into the pressure testing head body is configured to completely fill the threaded section at the front end of the body, so as to prevent concrete slurry from seeping into the inside of the pressure testing head during the pouring process and to maintain the cleanliness of the internal threads after demolding.

Images