Centrifugal pump flow dynamic detection system based on Internet of Things

Abstract

The invention discloses a centrifugal pump flow dynamic detection system based on the Internet of Things, and relates to the technical field of intelligent sensing systems. Comprising a data acquisition module used for acquiring multi-source signals in real time and obtaining a sensing feature data set after preprocessing; the physical property parameter module is used for calculating the influence parameters of the physical property change of the medium on the pump performance based on the sensing characteristic data set to obtain a medium physical property parameter set; the flow prediction module is used for correcting a preset flow prediction model through the medium physical property parameter set to obtain a corrected flow prediction model; inputting a sensing characteristic data set acquired in real time into the corrected flow prediction model, and outputting to obtain a flow prediction value; the traffic detection module is used for judging whether the current traffic is abnormal or not based on the traffic predicted value, the historical traffic data and a preset operation threshold parameter set, and giving a traffic detection result; the influence of medium physical property changes on flow detection can be accurately captured, and the accuracy and stability of flow detection are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing system technology, and more specifically, to an Internet of Things-based centrifugal pump flow dynamic detection system. Background Technology

[0002] Centrifugal pumps are widely deployed as core fluid transport equipment in practical applications such as water supply, industrial circulating water, and petrochemical fluid transportation. Accurate detection of their flow parameters is crucial for ensuring stable system operation, optimizing energy consumption control, and guaranteeing production process compliance. Especially in urban water supply networks, centrifugal pumps need to adapt to complex operating conditions such as seasonal water temperature fluctuations, changes in impurity content in the network, and instantaneous gas content fluctuations. In industrial circulating water systems, the viscosity of the medium dynamically changes with process temperature adjustments and impurity accumulation. In the petrochemical field, changes in the composition of the transported medium further exacerbate the instability of physical properties. In these scenarios, IoT-based non-invasive centrifugal pump flow dynamic detection technology, due to its advantages such as convenient installation, no need to damage the pipeline structure, and low operation and maintenance costs, has gradually replaced some traditional invasive flow detection solutions and become one of the mainstream application technologies.

[0003] Existing IoT-based non-invasive centrifugal pump flow dynamic detection technology works by collecting multi-source signals during operation using vibration sensors, current sensors, and temperature sensors deployed on the pump body or motor. Data-driven algorithms or semi-physical models based on pump characteristic curves are then used to establish a mapping relationship between sensor signal characteristics and actual flow rate, enabling indirect flow rate estimation. This technology uses an edge gateway for signal preprocessing and preliminary calculations, then uploads the data to a cloud platform via an IoT transmission module for final flow rate output and operational status monitoring. Signal feature extraction and mapping model construction are the core technologies. Existing technologies typically train models by offline collection of sample data under specific operating conditions, determining the correspondence between signal characteristics and flow rate, and then embedding the model for online detection.

[0004] Existing technologies are prone to detection deviations caused by changes in the physical properties of the medium in practical applications. The core reason is that existing technologies generally implicitly assume that the medium properties remain stable, failing to fully consider the dynamic changes in the medium's physical parameters in real-world scenarios. During actual operation, changes in water temperature alter the density and viscosity of the fluid, thereby affecting the flow state of the fluid within the pump and the pump's operating efficiency. Increased gas content in the medium disrupts the continuous flow characteristics of the fluid, altering the pressure distribution and energy transfer efficiency within the pump. Changes in impurity content not only change the viscosity of the medium but also exacerbate impeller wear within the pump, indirectly affecting the mapping relationship between signal characteristics and flow rate.

[0005] The aforementioned changes in the physical properties of the medium directly cause systematic distortions in the statistical characteristics of vibration, current, and other signals collected by sensors. Existing fixed mapping models cannot distinguish whether this distortion is caused by changes in physical properties or changes in flow rate. Consequently, signal variations resulting from changes in physical properties are misinterpreted as flow rate fluctuations, or the adaptability of the mapping model decreases due to changes in physical properties, ultimately leading to flow rate detection errors. This fails to provide reliable data support for production process control. In industrial production scenarios, this may cause process parameters to deviate from set thresholds, affecting product quality. In water supply scenarios, flow rate estimation errors may lead to a mismatch between water supply and actual demand, resulting in water waste or insufficient supply. Secondly, frequent detection errors can trigger false alarms or missed alarms. False alarms lead to ineffective inspections by maintenance personnel, increasing maintenance costs; missed alarms may cause pumps to continue operating under abnormal conditions, exacerbating impeller wear, cavitation, and other faults, shortening equipment lifespan, and even causing serious production accidents such as equipment shutdowns.

[0006] In view of this, the present invention proposes an Internet of Things-based centrifugal pump flow dynamic detection system to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a centrifugal pump flow dynamic detection system based on the Internet of Things, comprising:

[0008] The data acquisition module is used to acquire multi-source signals in real time and obtain sensor feature datasets after preprocessing.

[0009] The physical property parameter module, based on the sensor feature dataset, calculates the parameters that affect the pump performance by the changes in the physical properties of the medium, and obtains the set of physical property parameters of the medium;

[0010] The flow prediction module modifies the preset flow prediction model using a set of medium property parameters to obtain a modified flow prediction model; it inputs the real-time collected sensor feature dataset into the modified flow prediction model and outputs the flow prediction value.

[0011] The traffic detection module determines whether the current traffic is abnormal based on traffic prediction values, historical traffic data, and a preset set of operating threshold parameters, and provides the traffic detection results.

[0012] Furthermore, the sensing feature dataset includes vibration signal feature values, average motor current, medium temperature, medium gas content, medium impurity concentration, and pump speed.

[0013] Furthermore, the set of medium physical property parameters includes the current medium density, current medium dynamic viscosity, current medium effective viscosity, and current medium gas holdup;

[0014] The density and dynamic viscosity of the current medium are calculated by reading the density and dynamic viscosity calculation rules corresponding to the current medium temperature from the preset medium reference physical property data.

[0015] Determine the gas phase ratio based on the gas content of the medium and record it as the current gas content of the medium;

[0016] The cleanliness of the medium is determined based on the concentration of impurities in the medium, and the additional viscosity increment is calculated accordingly. The current dynamic viscosity of the medium is added to the additional viscosity increment to obtain the current effective viscosity of the medium.

[0017] Furthermore, methods for calculating the additional viscosity increment include:

[0018] Set a cleanliness coefficient, which is determined by the impurity concentration of the current medium and the reference impurity concentration given in the medium's reference physical property data;

[0019] When the concentration of impurities in the medium does not exceed the reference impurity concentration, the cleanliness coefficient is set to 1 and the additional viscosity increment is set to 0; when the concentration of impurities in the medium exceeds the reference impurity concentration, the cleanliness coefficient is calculated based on the ratio of the concentration of impurities in the medium to the reference impurity concentration, and the additional viscosity increment is determined based on the difference between the product of the dynamic viscosity of the medium and the cleanliness coefficient.

[0020] Furthermore, a density correction factor is determined based on the current medium density and the reference density. The density correction factor is equal to the reference density divided by the current medium density.

[0021] Furthermore, the viscosity correction function is determined according to the viscosity change ratio;

[0022] The viscosity change ratio is equal to the effective viscosity of the current medium divided by the reference dynamic viscosity. The viscosity correction function is defined as the negative exponential form of the viscosity change ratio.

[0023] Furthermore, the gas content correction coefficient is determined as a piecewise continuous function based on the current gas content of the medium;

[0024] When the current gas content of the medium does not exceed the preset gas content stability threshold, the gas content correction coefficient is calculated by linear decay; when the current gas content of the medium exceeds the gas content stability threshold, the gas content correction coefficient is calculated by exponential decay.

[0025] Furthermore, the density correction coefficient, viscosity correction function, and gas content correction coefficient are combined to form a comprehensive correction coefficient, which is then used as a property correction factor in the output of the flow prediction model.

[0026] The corrected predicted flow rate is obtained by multiplying the predicted flow rate by a comprehensive correction factor.

[0027] Furthermore, the traffic forecast is compared with the historical normal range. When the traffic forecast is within the historical normal range, the traffic detection result is judged as normal.

[0028] When the traffic prediction value exceeds the historical normal range, the anomaly detection process is initiated and a graded judgment is made in combination with the set of operating threshold parameters.

[0029] Furthermore, the methods for determining the classification include:

[0030] The absolute value of the difference between the predicted flow rate and the baseline value is calculated to obtain the flow rate deviation. If the flow rate deviation does not exceed the preset alarm threshold, the flow rate detection result is judged as an acceptable deviation and monitoring continues. If the flow rate deviation exceeds the alarm threshold, a physical property interpretation judgment is performed.

[0031] The comprehensive correction coefficient is compared with the property change threshold in the set of operating threshold parameters; when the comprehensive correction coefficient is within the interpretable range of the property change threshold and the rate of change of the comprehensive correction coefficient does not exceed the preset rate of change threshold, the flow detection result is determined to be a flow deviation caused by property change and the alarm is suppressed.

[0032] When the comprehensive correction coefficient exceeds the interpretable range of the property change threshold, or when the rate of change of the comprehensive correction coefficient exceeds the rate of change threshold, a consistency check of the vibration signal characteristic value and the motor power is introduced, and the consistency residual is calculated. The consistency residual is determined by the physical consistency relationship between the predicted flow change direction and the vibration signal characteristic value change direction and the motor power change direction. If the consistency residual continuously exceeds the residual threshold in the set of operating threshold parameters and meets the preset duration threshold, the flow detection result is judged as an abnormal mode and an alarm is triggered.

[0033] Compared with existing technologies, the technical effects and advantages of the IoT-based centrifugal pump flow dynamic detection system of the present invention are as follows:

[0034] The present invention relates to an IoT-based centrifugal pump flow dynamic detection system, which acquires multi-source signals in real time through a data acquisition module and preprocesses them to obtain a sensor feature dataset. The system then uses a physical property parameter module to calculate a set of medium physical property parameters. After correcting the preset flow prediction model using this parameter set, the system inputs the sensor feature dataset to obtain the flow prediction value. Finally, the flow detection module combines historical flow data and a set of operating threshold parameters to classify whether the flow is abnormal.

[0035] This invention effectively solves the problems of existing technologies that, due to the implicit assumption of stable media properties, fail to fully consider the dynamic changes in media properties such as water temperature, gas content, and impurity concentration, resulting in deviations, false alarms, or missed alarms in flow detection, which in turn affect production process control, cause resource waste, or equipment damage. It can accurately capture the impact of changes in media properties on flow detection, improve the accuracy and stability of flow detection, reduce ineffective operation and maintenance costs, and ensure the stable operation of centrifugal pump systems. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of a centrifugal pump flow dynamic detection system based on the Internet of Things according to an embodiment of the present invention;

[0037] Figure 2 This is an execution flowchart of an IoT-based centrifugal pump flow dynamic detection system according to an embodiment of the present invention.

[0038] Figure 3 This is a flowchart illustrating the method for obtaining a set of medium property parameters in an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be described in detail, clearly, and completely below with reference to the accompanying drawings. It should be particularly noted that the specific embodiments described below are only for better illustrating and explaining the technical solutions of the present invention, and are intended to enable those skilled in the art to better understand and implement the present invention, and should not be construed as limiting the scope of protection of the present invention. Without departing from the spirit and substance of the present invention, those skilled in the art can modify, adjust, or make equivalent substitutions based on the content disclosed in the present invention, and these should all be considered within the scope of protection of the present invention.

[0040] Example 1:

[0041] Please see Figure 1 As shown in the figure, this embodiment discloses a centrifugal pump flow dynamic detection system based on the Internet of Things, including a data acquisition module, a physical property parameter module, a flow prediction module and a flow detection module. Each module is connected by wires and / or wirelessly to realize data transmission.

[0042] The data acquisition module is used to acquire multi-source signals in real time and obtain sensor feature datasets after preprocessing.

[0043] Various deployed sensors synchronously collect multi-source signal data during pump operation. For example, accelerometers collect vibration sensor signals, current transformers collect current sensor signals, temperature probes acquire temperature sensor signals, acoustic sensors acquire bubble monitoring sensor signals, turbidity meters acquire water turbidity sensor signals, and speed sensors acquire pump speed signals. All sensor signals are transmitted in real-time to the edge computing unit via an IoT gateway. Specifically, vibration sensor signals are acquired by using an accelerometer mounted on the surface of the centrifugal pump body to collect equipment vibration data, reflecting the mechanical vibration characteristics of the pump during operation. Current sensor signals are acquired by installing current sensors, such as Hall effect sensors or current transformers, on the pump motor power supply circuit to collect the motor operating current in real time, reflecting changes in pump shaft power. Temperature sensor signals are acquired by installing temperature sensors on the outer wall of the pump outlet pipe or at the medium contact point to collect the real-time temperature of the transported medium, used to calculate changes in the medium's density and viscosity. Bubble monitoring sensor signals are acquired by installing ultrasonic or acoustic sensors on the pump body or pipeline to monitor changes in bubbles or gas content in the medium, such as detecting cavitation noise or ultrasonic attenuation, to obtain the medium's gas content index. The water turbidity sensor signal acquisition method involves installing an optical turbidity sensor in the pipeline bypass or using medium sampling to detect impurity content, obtaining the concentration of suspended impurities or turbidity index in the medium, reflecting changes in medium purity. The pump speed signal acquisition method involves using a speed sensor installed at the motor shaft end, such as a magnetoelectric sensor or built-in detection in the frequency converter, to obtain the real-time speed of the centrifugal pump, reflecting changes in pump operating conditions.

[0044] After completing the synchronous acquisition of multi-source signals, the edge computing device performs windowing processing on the vibration sensor signal, current sensor signal, temperature sensor signal, bubble monitoring sensor signal, water quality turbidity sensor signal, and pump speed signal using a unified timestamp. The length of each time window is given by a pre-configured set of operating threshold parameters and remains fixed, ensuring that the feature calculation of each signal is completed within the same statistical interval and guaranteeing the time consistency of subsequent model inputs. The logic for setting the time window length is to cover more than one stable pump speed cycle and not exceed the typical operating condition adjustment response time, thereby suppressing transient noise while maintaining the ability to track dynamic changes. At the same time, transition regions are set at both ends of each time window to eliminate filter boundary effects. The length of the transition region is determined by the filter order and cutoff frequency and is fixedly configured.

[0045] When denoising and filtering the vibration sensor signal, the edge computing device first performs DC bias removal processing on the vibration sensor signal, then determines the rotational frequency based on the pump speed signal, and uses the frequency band containing the rotational frequency and its harmonics as the reserved bandwidth. A bandpass digital filter is used to suppress low-frequency drift and high-frequency random noise. The filter cutoff frequency and order are given by the operating threshold parameter set. The setting logic is to ensure that the reserved bandwidth covers the main energy range of the pump structural response and eliminates environmental interference unrelated to rotation. After filtering, the root mean square amplitude of the filtered vibration sensor signal is calculated within the same time window as the overall vibration amplitude index. A fast Fourier transform is then performed on the vibration sensor signal according to the preset frequency band division to calculate the spectral energy of each preset frequency band and normalize it with the total energy, obtaining the frequency band energy index used to characterize the strength of hydraulic excitation and structural response. The overall vibration amplitude index and the frequency band energy index are combined to form the vibration signal characteristic value, and the vibration signal characteristic value is written into the sensor feature dataset.

[0046] When smoothing the current sensor signal, the edge computing device first performs anti-aliasing low-pass filtering and synchronous sampling alignment on the current sensor signal. Then, it calculates the effective current value of the current sensor signal within the same time window as the basis for calculating the average motor current or power. Subsequently, it calculates the average motor power by combining the motor's rated parameters and power supply parameters. The average motor power is calculated in the form that power equals the product of the effective voltage and current values, multiplied by the power factor. The power factor is given by the motor's rated parameters or configured as a fixed value by the operating threshold parameter set. When the conditions for calculating the average motor power are not met, the effective current value is used as a substitute output quantity and the dimension is kept consistent, so that the subsequent flow prediction model can receive the average motor current or power based on the same input interface. The reason for smoothing is that the current sensor signal is easily affected by power supply ripple and instantaneous load disturbances. Low-pass filtering and in-window statistics can stably reflect changes in pump shaft load and reduce the interference of transient spikes on power estimation.

[0047] When reading the temperature sensor signal, the edge computing device performs calibration conversion on the temperature sensor signal, converting the raw voltage or resistance value of the temperature sensor signal into the current medium temperature value through a pre-stored calibration curve. The calibration curve is established during the factory or field calibration stage and forms a mapping table using a multi-point calibration method. The edge computing device uses piecewise linear interpolation for the current sampling point to obtain the current medium temperature value. To ensure the stability of the temperature conversion, a moving average smoothing is performed on the current medium temperature value. The length of the moving average window is given by the operating threshold parameter set, and its setting logic is to cover the smallest significant time scale of temperature change and be significantly larger than the sampling period of the temperature sensor, thereby filtering out sampling noise without weakening the true temperature trend.

[0048] When converting the bubble monitoring sensor signal into a medium gas content index, the edge computing device first performs bandpass filtering on the bubble monitoring sensor signal to retain the effective frequency band related to bubble scattering and cavitation noise while suppressing background noise other than pump structure resonance. The effective frequency band is configured by the operating threshold parameter set and determined during the installation and commissioning phase through comparison between no-load and loaded conditions. Subsequently, within the same time window, the energy index of the bubble monitoring sensor signal is calculated and its amplitude is normalized. The normalization reference is taken from the reference energy of the reference gas-free state corresponding to the medium reference physical property data. The edge computing device substitutes the normalized energy index into a pre-calibrated gas content conversion curve to obtain the current medium gas content. The gas content conversion curve is established from field calibration data and a monotonically increasing constraint ensures a unique correspondence between gas content and signal energy. To avoid sudden changes in gas content caused by occasional impact noise, the edge computing device performs deburring processing on the current medium gas content. The deburring threshold is given by the operating threshold parameter set. Its setting logic is to ensure that the single window change of gas content does not exceed the upper limit of the maximum measurement fluctuation of the sensor under stable medium conditions. When it exceeds the upper limit, it is clipped according to the upper limit and recorded as an abnormal sampling point for subsequent diagnosis.

[0049] When converting the water turbidity sensor signal into a medium impurity content index, the edge computing device reads the water turbidity sensor signal and converts the original output into the current medium impurity concentration based on the calibration coefficient of the water turbidity sensor. The calibration coefficient is obtained from factory calibration or on-site calibration and stored in the operating threshold parameter set. Range clipping is performed on the current medium impurity concentration to ensure physical rationality. The upper limit of clipping is determined by the range of the water turbidity sensor, and the lower limit is set to zero to avoid negative values ​​from being included in subsequent calculations. To reduce the impact of spikes caused by short-term sediment disturbances, exponential smoothing is performed on the current medium impurity concentration. The exponential smoothing coefficient is configured by the operating threshold parameter set, and its setting logic is to suppress short-term jumps caused by transient particle aggregation without masking the slow trend of impurity accumulation.

[0050] When reading the pump speed signal and converting it into a current pump speed value, the edge computing device calculates the average speed within the same time window based on the pulse count or frequency output of the pump speed signal, and converts the average speed into the current pump speed value. To ensure the synchronization of speed calculation with other features, the speed statistics window is consistent with the feature windows of the aforementioned vibration sensor signal and current sensor signal. When a speed pulse is lost, the previous window holding strategy is used to output the current pump speed value, and a lost pulse marker is recorded for cloud-based quality control.

[0051] The preprocessed sensor feature dataset includes vibration signal feature values, average motor current or power, current medium temperature, current medium gas content, current medium impurity concentration, and current pump speed. All data are aligned with the same time window and can be directly used as input for subsequent construction of medium property parameter sets and flow prediction model calculations.

[0052] The flow prediction module modifies the preset flow prediction model using a set of medium property parameters to obtain a modified flow prediction model. The real-time collected sensor feature dataset is then input into the modified flow prediction model, and the flow prediction value is output.

[0053] Please see Figure 3 As shown, methods for obtaining the set of medium property parameters include:

[0054] By calculating the influence coefficient of medium property changes on pump performance, the systematic impact of medium property changes on the hydraulic process and electromechanical response of centrifugal pumps is explicitly quantified and introduced into the subsequent flow estimation link. This avoids misjudging the statistical characteristic deformation of signals such as vibration and current caused by medium property changes as flow fluctuations. The energy transfer and hydraulic loss of centrifugal pumps are jointly affected by medium density, dynamic viscosity, and gas phase ratio. Changes in medium density will change the torque demand and current response amplitude at the same speed. Changes in medium dynamic viscosity will change the boundary layer and internal friction loss, leading to shifts in efficiency and characteristic frequency band energy distribution. Changes in medium gas content will disrupt the continuous medium assumption and cause pressure pulsation and changes in vibration modes. By calculating the influence coefficient of medium property changes on pump performance, the predictable offset caused by properties can be separated from the flow mapping relationship without changing the existing sensor deployment. This allows subsequent correction steps to compensate for the mapping model in a targeted manner, reducing systematic bias and minimizing false alarms and missed alarms. This quantification step plays a key supporting role in solving the technical problem of flow detection deviation caused by dynamic changes in medium properties.

[0055] The density and dynamic viscosity calculation rules corresponding to the current medium temperature are retrieved from the medium reference physical property data. The medium reference physical property data pre-stores the corresponding data sequences and interpolation methods for temperature and density, and temperature and dynamic viscosity. After trimming the current medium temperature at the edges, piecewise linear interpolation or cubic spline interpolation is used to calculate the current medium density and dynamic viscosity. The adjacent temperature nodes used for interpolation are determined by locating the current medium temperature within the temperature sequence of the medium reference physical property data. The current medium temperature is then substituted into the parameterized functions for temperature-to-density and temperature-to-dynamic viscosity defined in the medium reference physical property data to obtain the current medium density and dynamic viscosity. The applicable temperature range given by the medium reference physical property data is used as the validity boundary; when the temperature exceeds the boundary, the boundary value is fixed to prevent extrapolation from causing numerical distortion.

[0056] The gas phase ratio is determined based on the gas content of the medium and recorded as the gas content value. When the gas content of the medium is output as a percentage from the previous signal processing stage, the gas content is first divided by 100 to obtain the gas phase ratio, and then the gas phase ratio is multiplied by 100 to write back the gas content value to maintain a consistent dimensional expression with the gas content of the medium. To ensure the physical rationality of the gas phase ratio, lower and upper thresholds are set to trim the gas content of the medium. The lower threshold is set to zero to avoid negative gas content, and the upper threshold is set to the maximum calibrable value of the gas content measurement link to avoid abnormal spikes entering the model. The threshold setting logic is based on the sensor range and the coverage of calibration data. During operation, a short window de-straining process is performed on the gas content of the medium to remove isolated pulses. The de-straining window length is set to cover a complete statistical window for acoustic signal feature extraction, thereby ensuring the time alignment consistency between the gas content value and the acoustic features.

[0057] The cleanliness of the medium is determined based on the concentration of impurities, and the additional viscosity increment is calculated accordingly. Medium cleanliness is characterized by a dimensionless cleanliness coefficient, which is jointly determined by the medium impurity concentration and a reference impurity concentration given in the medium's reference physical property data: when the medium impurity concentration does not exceed the reference impurity concentration, the cleanliness coefficient is set to 1, and the additional viscosity increment is set to 0; when the medium impurity concentration exceeds the reference impurity concentration, the cleanliness coefficient is calculated as the ratio of the medium impurity concentration to the reference impurity concentration, and the additional viscosity increment is determined as the difference between the product of the medium's dynamic viscosity and the cleanliness coefficient, i.e., the additional viscosity increment equals the product of the current medium dynamic viscosity and the cleanliness coefficient minus the current medium dynamic viscosity. To avoid excessive amplification of the additional viscosity increment during high-impurity stages, an upper limit threshold for the cleanliness coefficient is set. The logic for setting the upper limit threshold is based on the calibrated upper bound of the maximum achievable medium impurity concentration on-site, ensuring that the additional viscosity increment calculated from this upper bound does not exceed the maximum compensation range within which the flow estimation model can stably converge. The current dynamic viscosity of the medium is then added to the additional viscosity increment to obtain the current effective viscosity of the medium. This incorporates the influence of the medium impurity concentration on the equivalent viscosity loss into the same viscosity variable, facilitating subsequent unified correction.

[0058] In summary, a set of medium physical property parameters is formed, which includes the current medium density, current medium dynamic viscosity, current medium effective viscosity, and current medium gas holdup. The set of medium physical property parameters is used to quantitatively correct the flow prediction model and provide interpretable physical property basis for alarm suppression criteria.

[0059] The flow prediction model is used to establish a quantitative correlation between input and output data. The input data consists of a sensor feature dataset and a synchronized set of medium property parameters, while the output data is the actual flow rate corresponding to the same time window. Vibration signal eigenvalues ​​characterize the mechanical response of the pump body and pipeline under specific operating conditions, reflecting the influence of hydraulic excitation, internal backflow, and pressure pulsation on structural vibration. Motor power characterizes the change in pump shaft input energy, reflecting the influence of changes in medium density, load torque, and hydraulic losses on energy consumption. Pump speed is a key variable in the pump similarity law, directly determining the theoretical flow rate level and the fundamental frequency scale of vibration and power characteristics. The current medium density, current effective viscosity, and current gas holdup in the medium property parameter set are used to characterize the dynamic changes in medium properties; these are the root causes of the systematic deviation between vibration signal eigenvalues ​​and motor power under the same actual flow rate. The aforementioned input data can simultaneously characterize the pump's energy input and dynamic response from both the electromechanical and structural perspectives. Furthermore, it can use the medium property parameter set to explain the morphological drift of the input data when the properties change, thereby solving the problem that existing solidified mapping models cannot distinguish between property changes and flow rate changes and are prone to flow rate detection errors.

[0060] During training, the flow prediction model should be trained using data collected under the condition that the target centrifugal pump and installation structure remain unchanged. The pump model, impeller geometry, sensor installation location and method, edge sampling frequency, and feature extraction window length should all be consistent. The training data should cover the usable operating range of the pump speed and the flow variation range formed by valve adjustments under the same pipe network or equivalent hydraulic resistance conditions. During the training phase, actual flow rate must be obtained simultaneously as a supervisory output. This actual flow rate is obtained from a standard flow meter temporarily installed or permanently maintained on the pipeline. The standard flow meter's range must cover the training flow rate range, and the sampling timestamp must be aligned with the vibration signal characteristic value, motor power, and pump speed to ensure that the output data of each training sample truly corresponds to the same statistical window of its input data. Simultaneously, the current medium temperature, gas content, and impurity concentration are collected during the training phase to generate a set of medium physical property parameters. This ensures that the training samples cover different combinations of current medium density, effective viscosity, and gas content, allowing the flow prediction model to learn the influence of input data on output data under varying physical property conditions, avoiding the flow prediction model only fitting a single medium condition.

[0061] The flow prediction model specifically employs a physically constrained regression model, which takes the form of a regularized gradient boosting regression tree. The inputs are vibration signal feature values, motor power, pump speed, current medium density, current effective viscosity, and current gas holdup. The output is the predicted flow rate. The physically constrained regression model effectively characterizes the nonlinear coupling relationships between input variables and exhibits strong robustness to multi-source features. Furthermore, interpretability can be enhanced by constraining the monotonicity of the training samples, ensuring that the predicted flow rate remains monotonically increasing with pump speed, monotonically decreasing with current effective viscosity, and monotonically decreasing with current gas holdup, thus conforming to the pump similarity law and the fundamental laws governing two-phase flow. The training method employs supervised learning. First, training samples are divided into sample pairs according to time windows. Each sample pair consists of an input data vector and the corresponding actual flow rate. Then, the samples are stratified according to operating conditions to divide them into training and validation sets. Cross-validation is used to determine the number of trees, maximum depth, and learning rate. The mean square error and maximum absolute error on the validation set are used as optimization objectives to finally obtain a fixed flow prediction model. At the same time, the reference physical property parameters of the training phase are saved as the reference physical properties of the flow prediction model. The reference physical properties of the flow prediction model are determined by the reference density and reference dynamic viscosity of the medium at the reference temperature. The reference density and reference dynamic viscosity are used as references for subsequent online correction.

[0062] When dynamically adjusting the flow prediction model to adapt to changes in the medium, the current medium density, current effective viscosity, and current gas holdup in the medium property parameter set are first compared with the baseline properties of the flow prediction model. The baseline properties of the flow prediction model are the baseline density, baseline dynamic viscosity, and baseline gas holdup. The baseline gas holdup is set to zero to represent the baseline operating condition during the training phase as a gas-free state. To quantify the impact of changes in medium density on energy input characteristics and suppress the resulting flow prediction deviation, a density correction coefficient is determined based on the current medium density and the baseline density. The density correction coefficient is equal to the baseline density divided by the current medium density.

[0063] The density correction coefficient is designed based on the positive correlation between the shaft power of a centrifugal pump and the density of the medium. When the pump speed and hydraulic state are approximately constant, the change in motor power contains a proportional component directly introduced by the change in medium density. An increase in medium density will increase the motor power under the same hydraulic state, and a decrease in medium density will decrease the motor power under the same hydraulic state. Since the flow prediction model uses motor power as one of its key inputs, without compensation for changes in medium density, the flow prediction model is prone to misidentifying the increase in motor power caused by medium density as an increase in motor power caused by an increase in flow rate, thus producing a systematic deviation in the flow prediction value. By introducing a density correction coefficient to compensate for the flow prediction value in reverse, the proportional influence of the change in medium density on the motor power is canceled out on the output side. This makes the flow prediction value more concentrated in reflecting the flow-dominated trend, bringing the flow prediction value back to a scale more consistent with the hydraulic state, thereby improving the prediction stability and transferability under cross-seasonal and cross-medium conditions.

[0064] To ensure the physical rationality and numerical stability of the density correction coefficient, a lower and upper threshold are set for the density correction coefficient and a pruning process is performed. The lower and upper thresholds are jointly determined by the current density statistical range of the medium physical property parameter set during the training phase and the available density range of the medium reference physical property data. This ensures that the density correction coefficient does not exceed the confidence interval already covered by the flow prediction model, and avoids excessive amplification or reduction of the flow prediction value under extreme medium conditions that exceed the calibration range.

[0065] When introducing a viscosity correction function based on the current effective viscosity of the medium, the influence of medium impurity concentration on equivalent viscous loss is already incorporated into the same viscosity variable. This directly corresponds to the influence path of pump internal friction loss and efficiency changes on vibration signal characteristic values ​​and motor power. The viscosity correction function is determined by the viscosity change ratio, which is equal to the current effective viscosity of the medium divided by the reference dynamic viscosity. The viscosity correction function is defined as the negative exponential form of the viscosity change ratio, i.e., the viscosity correction function is equal to the negative exponent of the viscosity change ratio. The exponent is the correction sensitivity parameter determined by the validation set during the training phase. The correction sensitivity parameter is obtained by minimizing the error between the corrected flow prediction value and the actual flow on the training data. Specifically, the correction method is to multiply the flow prediction value output by the flow prediction model by the viscosity correction function. When the current effective viscosity of the medium is higher than the reference dynamic viscosity, the viscosity change ratio is greater than 1, and the viscosity correction function is less than 1, thus implementing a downward correction for the flow prediction value; when the current effective viscosity of the medium is lower than the reference dynamic viscosity, the viscosity change ratio is less than 1, and the viscosity correction function is greater than 1, thus implementing an upward correction for the flow prediction value. The advantage of this correction is that it maps the hydraulic efficiency reduction and power shift caused by viscosity changes into the compensation term of the flow prediction in a controllable continuous function form, avoiding overcompensation in the low viscosity region or undercompensation in the high viscosity region caused by using a fixed ratio reduction, thereby reducing systematic deviation and improving the portability between different operating conditions.

[0066] When considering the influence of gas holdup, the flow prediction model introduces a gas holdup correction coefficient. This correction coefficient is determined as a piecewise continuous function based on the current gas holdup: when the current gas holdup does not exceed the gas holdup stability threshold, the correction coefficient is calculated with linear decay, for example, gas holdup correction coefficient = (1 - current gas holdup / gas holdup stability threshold) × maximum gas holdup decay coefficient; when the current gas holdup exceeds the gas holdup stability threshold, the correction coefficient is calculated with exponential decay, causing it to decrease rapidly with increasing gas holdup, thus reflecting the fact that two-phase flow leads to a significant decrease in energy transfer efficiency and increased flow instability. The gas holdup stability threshold is taken as the gas holdup inflection point obtained through experimental calibration during the training phase. Its setting logic is to ensure that the linear segment of the gas holdup correction coefficient covers the approximately linear range of the gas holdup's influence on flow, and that after exceeding this inflection point, it enters a rapid decay region to suppress overly optimistic predictions under high gas holdup conditions. The maximum gas holdup decay coefficient is taken as the minimum conservative proportion at the gas holdup stability threshold that still allows the prediction results to be used for control and alarm judgment, and is determined through the maximum absolute error constraint on the validation set. The specific correction method is to multiply the predicted flow rate by the gas content correction factor and reduce the effective flow rate according to the proportion of the current medium gas content. This allows the model to adopt a conservative output for the predicted flow rate when cavitation or two-phase flow is present, reducing the risk of misjudging power and vibration anomalies caused by gas content as an increase in flow rate. At the same time, it provides an interpretable quantitative basis for the gas content impact on subsequent alarm logic.

[0067] A comprehensive correction coefficient is formed by combining the density correction factor, viscosity correction function, and gas content correction factor. This comprehensive correction coefficient equals the density correction factor multiplied by the viscosity correction function, and then multiplied by the gas content correction factor. This comprehensive correction coefficient is used as a property correction factor in the flow prediction model output. The correction method involves first using a fixed flow prediction model at the edge, based on vibration signal characteristic values, motor power, pump speed, and a set of medium property parameters, to output a flow prediction value. Then, this flow prediction value is multiplied by the comprehensive correction coefficient to obtain the corrected flow prediction value. Simultaneously, the comprehensive correction coefficient and its components, along with the corrected flow prediction value, are output for cloud monitoring and alarm suppression criterion calculation. This ensures the system can trace the source of prediction deviations and maintain the stability of flow estimation when property changes occur.

[0068] After acquiring vibration signal characteristic values, motor power, and pump speed within the same statistical time window, the edge computing device writes these values ​​as input to the input interface of the modified real-time flow prediction model, triggering the model to perform a flow prediction operation. The modified real-time flow prediction model first calls the fixed flow prediction model to perform a nonlinear mapping calculation on the input, obtaining a preliminary flow prediction value corresponding to the statistical time window. This preliminary flow prediction value is the prediction result directly output by the fixed flow prediction model without applying property compensation. Subsequently, the modified real-time flow prediction model reads the comprehensive correction coefficient generated synchronously with the statistical time window and performs property correction operations on the preliminary flow prediction value to obtain the final flow prediction value. The property correction operation uses a multiplicative correction method, determining the flow prediction value as the product of the preliminary flow prediction value and the comprehensive correction coefficient. This compensates for the systematic deviation caused by changes in medium properties on pump hydraulic efficiency and electromechanical response with a unified scale in the flow prediction result, ensuring that the flow prediction value maintains a consistent trend and order of magnitude with the actual medium flow even under conditions of changing medium properties.

[0069] The numerical meaning of the comprehensive correction coefficient is used to explain the correction direction and correction magnitude. When the comprehensive correction coefficient is less than 1, it means that the combined effect of the current medium density, current medium effective viscosity, and current medium air holdup corresponding to the medium property parameter set leads to an increase in pump hydraulic loss or a decrease in energy transfer efficiency. Under the same input characteristics, the actual medium flow rate tends to decrease relative to the reference physical property conditions. Therefore, the flow prediction value is adjusted downward relative to the initial flow prediction value through multiplicative correction. When the comprehensive correction coefficient is greater than 1, it means that the combined effect of the current medium density, current medium effective viscosity, and current medium air holdup corresponding to the medium property parameter set leads to a relative decrease in hydraulic loss or a relative increase in energy transfer efficiency. Under the same input characteristics, the actual medium flow rate tends to increase relative to the reference physical property conditions. Therefore, the flow prediction value is adjusted upward relative to the initial flow prediction value through multiplicative correction. To ensure the physical rationality and numerical stability of the calculation results, upper and lower thresholds are set for the comprehensive correction coefficient and a pruning process is performed. The upper and lower thresholds are determined by the statistical range of the medium property parameter set during the training phase, so that the comprehensive correction coefficient varies within the confidence interval covered by the training data, avoiding excessive amplification or reduction of the initial flow prediction value when it exceeds the calibration range.

[0070] The traffic detection module determines whether the current traffic is abnormal based on traffic prediction values, historical traffic data, and a preset set of operating threshold parameters, and provides the traffic detection results.

[0071] After obtaining the predicted flow rate, the edge computing device transmits the predicted flow rate along with the corresponding timestamp, pump speed, motor power, vibration signal characteristic value, and a comprehensive correction coefficient synchronized with the time window to the cloud monitoring platform via the Internet of Things. The cloud monitoring platform stores the received data and generates the flow rate detection result for that moment. The flow rate detection result includes the predicted flow rate, the comprehensive correction coefficient, and the correlation characteristics used to explain the source of flow rate changes, thereby providing traceable data for real-time operating condition monitoring and feedback control.

[0072] The cloud-based monitoring platform establishes a historical normal range based on historical flow data. This range is calculated using a sliding time window, which covers multiple stable pump operating cycles and spans the duration of short-term disturbances to reduce the impact of transient noise on the estimation of the normal range. Within each sliding time window, the median of the historical flow data is calculated as the baseline value, and the median of the absolute deviation is calculated as the fluctuation scale. The upper and lower thresholds of the historical normal range are then set as the baseline value plus the fluctuation scale multiplied by the range coefficient, and the baseline value minus the fluctuation scale multiplied by the range coefficient, respectively. The range coefficient is provided by the operating threshold parameter set, and its setting logic is to ensure coverage of most fluctuations under historical normal operating conditions while maintaining sufficient sensitivity to abnormal deviations. The range coefficient can be offline tuned and fixed into the operating threshold parameter set based on the alarm error tolerance and false alarm rate targets in historical data. The reason for using the median and the median of the absolute deviation to construct the historical normal range is that it is robust to occasional spikes and short-term disturbances, maintaining threshold stability even when outliers occur due to water supply scheduling, valve actions, etc., thereby improving the consistency of subsequent judgments.

[0073] The cloud-based monitoring platform compares the current traffic prediction value with the historical normal range. When the current traffic prediction value is within the historical normal range, the current traffic detection result is directly judged as normal. When the current traffic prediction value exceeds the historical normal range, an anomaly judgment process is initiated, and a graded judgment is performed based on the operating threshold parameter set. First, the traffic deviation is calculated, which is the absolute value of the difference between the current traffic prediction value and the baseline value. The traffic deviation is then compared with an alarm threshold, which is given by the operating threshold parameter set and set proportionally to the fluctuation scale. The logic for setting the alarm threshold is to trigger further diagnosis only when the deviation exceeds a multiple of the normal fluctuation scale, so as to avoid misjudging normal scheduling fluctuations as fault symptoms. If the traffic deviation does not exceed the alarm threshold, the traffic detection result at that moment is judged as an acceptable deviation and monitoring continues. If the traffic deviation exceeds the alarm threshold, the process enters the physical property interpretation judgment stage to distinguish between interpretable deviations caused by changes in media physical properties and uninterpretable deviations caused by equipment or pipeline anomalies.

[0074] The interpretation of physical properties uses a comprehensive correction coefficient as a quantitative indicator of the impact of physical properties. The cloud-based monitoring platform reads the comprehensive correction coefficient and its rate of change within the same time window and compares it with the physical property change threshold in the operating threshold parameter set. The physical property change threshold is jointly determined by the coverage range of the medium physical property parameter set during the training phase and the acceptable range of physical property fluctuations on-site. Its setting logic is that changes in the comprehensive correction coefficient within the confidence interval covered by the training data are considered interpretable changes, while changes exceeding this interval or sudden changes exceeding the rate of change threshold within a short period are considered uninterpretable changes. When the comprehensive correction coefficient is within the interpretable range and the rate of change does not exceed the rate of change threshold, the cloud-based monitoring platform determines that the current flow deviation is mainly driven by changes in medium physical properties, outputs the flow detection result as the flow deviation caused by the physical property change, and suppresses the alarm. This ensures that the system can maintain stable control without generating invalid inspections even when there are fluctuations in water temperature, impurity accumulation, or gas content. When the comprehensive correction coefficient exceeds the interpretable range or the rate of change of the comprehensive correction coefficient exceeds the rate of change threshold, the cloud monitoring platform determines that the current flow deviation cannot be fully explained by changes in the physical properties of the medium. It further introduces a consistency check between the vibration signal characteristic value and the motor power, calculating the consistency residual. The consistency residual is determined by the physical consistency relationship between the predicted flow change direction and the changes in the vibration signal characteristic value and the motor power. If the consistency residual continuously exceeds the residual threshold in the operating threshold parameter set and meets the duration threshold, the flow detection result at that moment is judged as an abnormal mode and an alarm is triggered. The logic for setting the duration threshold is to cover more than one stable pump operating cycle and be greater than the duration of short-term hydraulic disturbances, ensuring that the anomaly corresponding to the alarm is persistent rather than instantaneous noise. After triggering the alarm, the cloud monitoring platform outputs alarm information and the corresponding flow detection result, and sends the alarm information to the operation and maintenance terminal. Simultaneously, it marks the flow detection result as an abnormal sample requiring manual review for subsequent maintenance and model iteration.

[0075] The cloud-based monitoring platform also performs long-term correlation analysis on the concentration of media impurities and the comprehensive correction coefficient to assess the trend of pump impeller wear or efficiency decline. The long-term analysis uses daily or weekly statistical windows to calculate the mean and slope of the media impurity concentration and the comprehensive correction coefficient, and determines whether the comprehensive correction coefficient shows a continuous decreasing trend or a continuously increasing fluctuation trend under similar media impurity concentration conditions. When the long-term slope of the comprehensive correction coefficient exceeds the maintenance prompt threshold in the operating threshold parameter set and the duration reaches the maintenance prompt duration threshold, the cloud-based monitoring platform outputs maintenance prompt information. The logic for setting the maintenance prompt threshold is to statistically analyze the long-term drift upper bound of the comprehensive correction coefficient during historical stable operation phases, and set the maintenance prompt threshold to a conservative level exceeding this drift upper bound, thus making the maintenance prompt more likely to correspond to actual efficiency degradation rather than normal fluctuations.

[0076] The flow detection results are generated by the cloud monitoring platform based on the predicted flow value, historical flow data, and operating threshold parameter set. They include real-time monitoring conclusions of the predicted flow value, property interpretation judgment conclusions, alarm suppression or alarm triggering results, and necessary maintenance prompts. This satisfies the production process's need for accurate flow data and reduces the risk of water supply imbalance or process parameter out-of-control due to prediction deviations. Furthermore, the property interpretation judgment and continuous criteria reduce false alarms and missed alarms, improving operation and maintenance efficiency and equipment safety. Finally, the corrected flow data and health status information are provided to users through the Internet of Things platform to achieve refined management of centrifugal pump operation.

[0077] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0078] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A centrifugal pump flow dynamic detection system based on Internet of Things, characterized in that, The method comprises the following steps: A data acquisition module is used to collect multi-source signals in real time, and a sensing feature dataset is obtained after preprocessing; A physical property parameter module is used to calculate the influence parameters of the physical property changes of the medium on the pump performance based on the sensing feature dataset, and a medium physical property parameter set is obtained; A flow prediction module is used to correct a preset flow prediction model based on the medium physical property parameter set, and a corrected flow prediction model is obtained; The real-time collected sensing feature dataset is input into the corrected flow prediction model, and a flow prediction value is output; A flow detection module is used to determine whether the current flow is abnormal based on the flow prediction value, historical flow data, and a preset operating threshold parameter set, and a flow detection result is given.

2. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 1, characterized in that, The sensing feature dataset includes vibration signal characteristic values, average motor current, medium temperature, medium gas content, medium impurity concentration, and pump speed.

3. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 2, characterized in that, The medium physical property parameter set includes current medium density, current medium dynamic viscosity, current medium effective viscosity, and current medium gas content; The density and dynamic viscosity calculation rules corresponding to the current medium temperature are read from the preset medium reference physical property data, and the current medium density and the current medium dynamic viscosity are calculated; The gas phase ratio is determined according to the medium gas content and recorded as the current medium gas content; The medium cleanliness is determined according to the medium impurity concentration, and the viscosity additional increment is calculated accordingly. The current medium effective viscosity is obtained by adding the current medium dynamic viscosity and the viscosity additional increment.

4. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 3, characterized in that, The method for calculating the viscosity additional increment comprises: A cleanliness coefficient is set, which is determined by the impurity concentration of the current medium and the reference impurity concentration given in the medium reference physical property data; When the medium impurity concentration does not exceed the reference impurity concentration, the cleanliness coefficient is 1 and the viscosity additional increment is 0. When the medium impurity concentration exceeds the reference impurity concentration, the cleanliness coefficient is calculated according to the ratio of the medium impurity concentration to the reference impurity concentration, and the viscosity additional increment is determined according to the product difference of the medium dynamic viscosity and the cleanliness coefficient.

5. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 1, characterized in that, A density correction coefficient is determined according to the current medium density and the reference density, and the density correction coefficient is equal to the reference density divided by the current medium density.

6. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 5, characterized in that, The viscosity correction function is determined according to the viscosity change ratio; The viscosity change ratio is equal to the current medium effective viscosity divided by the reference dynamic viscosity, and the viscosity correction function is defined as the negative exponential form of the viscosity change ratio.

7. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 6, characterized in that, The gas content correction coefficient is determined as a piecewise continuous function according to the current medium gas content; When the current medium gas content does not exceed the preset gas content stability threshold, the gas content correction coefficient is calculated according to linear decay. When the current medium gas content exceeds the gas content stability threshold, the gas content correction coefficient is calculated according to exponential decay.

8. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 7, characterized in that, The comprehensive correction coefficient is formed by combining the density correction coefficient, the viscosity correction function, and the gas content correction coefficient, and the comprehensive correction coefficient is used as the physical property correction on the flow prediction model output; The corrected predicted flow is obtained by multiplying the flow prediction value by the comprehensive correction coefficient.

9. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 8, characterized in that, The flow prediction value is compared with the historical normal range. When the flow prediction value is within the historical normal range, the flow detection result is determined to be normal; When the flow prediction value exceeds the historical normal range, the abnormal determination process is started and the operating threshold parameter set is used for hierarchical determination.

10. The centrifugal pump flow dynamic detection system based on the Internet of Things according to claim 9, characterized in that, The method for hierarchical determination comprises: An absolute value of a difference between the flow prediction value and the baseline value is calculated to obtain a flow deviation amount; if the flow deviation amount does not exceed a preset alarm threshold, the flow detection result is determined as an acceptable deviation and continues to be monitored; if the flow deviation amount exceeds the alarm threshold, a physical property interpretation determination is performed; The comprehensive correction coefficient is compared with a physical property change threshold in the operating threshold parameter set; when the comprehensive correction coefficient is in an interpretable interval of the physical property change threshold and a change rate of the comprehensive correction coefficient does not exceed a preset change rate threshold, the flow detection result is determined as a flow deviation caused by a physical property change and an alarm is inhibited; When the comprehensive correction coefficient exceeds the interpretable interval of the physical property change threshold or the change rate of the comprehensive correction coefficient exceeds the change rate threshold, a cooperative consistency check of a vibration signal feature value and motor power is introduced, a consistency residual is calculated, the consistency residual is determined by a physical consistency relationship of a prediction flow change direction and a vibration signal feature value change direction and a motor power change direction, if the consistency residual continuously exceeds a residual threshold in the operating threshold parameter set and meets a preset continuous time length threshold, the flow detection result is determined as an abnormal mode and an alarm is triggered.

Images