An anti-clogging measurement and control system for wedge flow meters

Abstract

This invention discloses an anti-clogging measurement and control system for wedge flowmeters, specifically relating to the field of industrial flow monitoring technology. It addresses the limitations of existing technologies in effectively identifying entanglement-type localized blockages caused by flexible impurities such as fibers and viscous substances in fluids, leading to flow measurement distortion and maintenance delays. The system extracts the proportion of low-frequency energy and the abrupt changes in high-frequency noise of flow field distortion through frequency domain energy analysis and time domain multi-scale entropy attenuation fitting of differential pressure signals. This is combined with dynamic fusion of fluid characteristics to generate an entanglement risk coefficient, precisely triggering directional pulse cleaning and closed-loop calibration of the signal zero point. The technical solution covers the entire chain of signal detection, feature analysis, risk assessment, and control execution, effectively distinguishing between real blockages and transient interference, avoiding the limitations of traditional threshold alarms or periodic cleaning, and significantly improving measurement accuracy and maintenance efficiency under complex operating conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial flow monitoring, and more particularly, to a kind of anti-blocking measurement and control system for wedge flowmeter. BACKGROUND

[0002] Wedge flowmeter is widely used in industrial flow monitoring scenarios of impurity-containing fluid, such as sewage treatment, papermaking, food processing and other fields, due to its simple structure, wear resistance and other characteristics. In the prior art, to solve the problem of fluid impurity blockage, the conventional method is to use mechanical filtration, regular cleaning or add protective cover, etc. to reduce the direct impact or deposition of solid particles on the measuring element. However, for fibers, viscous substances or flexible impurities (such as paper pulp fibers, hair, etc.) in the fluid, the above methods are difficult to effectively cope with. Such impurities are easy to adhere to the edge of the wedge structure or the surface of the flow passage, and form winding and sticking phenomenon with the fluid flow, resulting in local deformation of the flow passage cross-sectional area, but it has not yet reached the critical state of complete blockage.

[0003] The existing anti-blocking technology has the problem that when fibers or soft impurities in the fluid form winding type local blockage in the wedge area, the system cannot timely identify the abnormality because the degree of flow passage deformation does not trigger the preset differential pressure alarm threshold. At the same time, such local winding changes the flow field distribution characteristics, making the flow measurement value continuously deviate from the true value, and the traditional control logic based on fixed threshold or periodic cleaning is difficult to distinguish the cause of signal abnormality, eventually leading to cumulative measurement error and maintenance lag. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an anti-blocking measurement and control system for wedge flowmeter to solve the problems raised in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] An anti-blocking measurement and control system for wedge flowmeter, comprising:

[0007] A differential pressure detection module for detecting the differential pressure signal between the inlet and outlet of the wedge flowmeter and generating a corresponding frequency spectrum distribution graph;

[0008] A feature extraction module for extracting the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points from the frequency spectrum distribution graph;

[0009] A trigger determination module for determining whether to trigger winding type blockage analysis according to whether the low-frequency energy proportion exceeds a first threshold and the number of high-frequency noise mutation points is less than a second threshold;

[0010] an entropy analysis module configured to perform multi-scale sample entropy analysis on the differential pressure signal when the analysis of the winding type blockage is triggered, fit the entropy value decay slope of each scale, and generate a complexity decay rate;

[0011] a risk fusion module configured to generate a winding risk coefficient based on the complexity decay rate and the low-frequency energy proportion;

[0012] a control execution module configured to control the cleaning device to perform directional pulse flushing on the wedge-shaped region when the winding risk coefficient exceeds a third threshold value, and recalibrate the differential pressure signal zero point after the flushing.

[0013] In a preferred embodiment, detecting the differential pressure signal between the inlet and outlet of the wedge flowmeter and generating a corresponding frequency spectrum distribution chart comprises:

[0014] continuously collecting real-time differential pressure signals between the inlet and outlet of the wedge flowmeter through a pressure sensor;

[0015] preprocessing the differential pressure signal to remove high-frequency noise and baseline drift, and generating a preprocessed differential pressure signal;

[0016] based on the preprocessed differential pressure signal, generating a frequency spectrum distribution chart covering a preset frequency band by fast Fourier transform.

[0017] In a preferred embodiment, the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points are extracted from the frequency spectrum distribution chart, comprising:

[0018] integrating and calculating the normalized amplitude energy in the preset low-frequency band of the frequency spectrum distribution chart to obtain a low-frequency energy value;

[0019] integrating and calculating the normalized amplitude energy in the preset high-frequency band of the frequency spectrum distribution chart to obtain a high-frequency energy value;

[0020] taking the ratio of the low-frequency energy value to the sum of the low-frequency energy value and the high-frequency energy value as the low-frequency energy proportion;

[0021] detecting the number of amplitude energy differences between adjacent frequency points in the high-frequency band that exceed a preset mutation threshold, and counting as the number of high-frequency noise mutation points.

[0022] In a preferred embodiment, whether to trigger the analysis of the winding type blockage is determined according to whether the low-frequency energy proportion exceeds a first threshold value and the number of high-frequency noise mutation points is less than a second threshold value, comprising:

[0023] comparing the low-frequency energy proportion with the preset first threshold value, and simultaneously comparing the number of high-frequency noise mutation points with the preset second threshold value;

[0024] When the low-frequency energy proportion exceeds the first threshold value and the number of high-frequency noise mutation points is less than the second threshold value, the entanglement type blockage analysis is triggered, otherwise it is not triggered.

[0025] In a preferred embodiment, the first threshold value is dynamically adjusted according to a preset multiple range of the average low-frequency energy proportion under historical normal conditions corresponding to the fluid type; and the second threshold value is set according to a high quantile point of the statistical distribution of the number of high-frequency noise mutation points under the non-blockage condition.

[0026] In a preferred embodiment, when the entanglement type blockage analysis is triggered, a multi-scale sample entropy analysis is performed on the differential pressure signal, the entropy values at each scale are fitted to generate a complexity decay rate, including:

[0027] Based on the dynamic fluctuation characteristics of the differential pressure signal, a coarse-grained time scale range is divided according to the fluid turbulence intensity;

[0028] According to the coarse-grained time scale range, a corresponding coarse-grained sequence is generated;

[0029] The sample entropy value of each coarse-grained sequence is calculated, and the non-stationary decay trend with the increase of the scale is detected;

[0030] For the non-stationary decay trend, the decay slopes of the initial segment and the stable segment are fitted respectively by using piecewise linear regression; and the absolute value of the difference between the decay slopes of the initial segment and the stable segment is taken as the complexity decay rate.

[0031] In a preferred embodiment, the low turbulence intensity corresponds to a long scale sequence, and the high turbulence intensity corresponds to a short scale sequence.

[0032] In a preferred embodiment, based on the complexity decay rate and the low-frequency energy proportion, an entanglement risk coefficient is generated, including:

[0033] According to the viscosity characteristics corresponding to the fluid type, the fusion weight of the complexity decay rate and the low-frequency energy proportion is dynamically allocated;

[0034] The complexity decay rate and the low-frequency energy proportion are normalized respectively to eliminate the dimensional difference;

[0035] The normalized complexity decay rate and the low-frequency energy proportion are linearly weighted and summed according to the allocated weight to generate an entanglement risk coefficient;

[0036] When the sum of the fusion weight exceeds a preset range, the weight is reallocated to a set value in proportion to the weight.

[0037] In a preferred embodiment, the weight of the complexity decay rate corresponding to the high viscosity fluid is increased.

[0038] In a preferred embodiment, when the winding risk coefficient exceeds a third threshold value, the control washing device performs a directional pulse flush on the wedge-shaped area, and after the flush, recalibrates the differential pressure signal zero point, including:

[0039] When the winding risk coefficient exceeds the third threshold value, dynamically adjust the intensity and duration of the flush pulse according to the mapping relationship between the fluid type and the winding risk coefficient;

[0040] After the flush is completed, delay for a preset time window to wait for the fluid flow to stabilize, and then collect the differential pressure signal in the stable state to calculate the zero point offset;

[0041] If the zero point offset exceeds the preset tolerance range, repeat the flush and calibration process until the offset meets the standard;

[0042] After calibration, update the zero point reference value in the historical database for subsequent detection cycles.

[0043] Compared with the prior art, the present application has the following beneficial effects:

[0044] 1、The present application effectively solves the problem of traditional anti-clogging technology that cannot identify winding type local clogging and misjudge signal abnormalities through multi-dimensional signal feature extraction and dynamic fusion mechanism. The system uses dual analysis of frequency energy distribution and time domain complexity decay, combined with adaptive weight distribution strategy based on fluid characteristics, significantly improving detection sensitivity and working condition adaptability. By capturing the energy concentration effect of flow field distortion through low-frequency energy proportion, filtering particle interference through high-frequency noise mutation points, and quantifying dynamic instability rate through multi-scale sample entropy, the system can distinguish the difference between fiber winding and conventional impurity flushing, avoiding false triggering caused by single threshold or fixed cleaning logic. At the same time, the closed-loop calibration mechanism automatically corrects the differential pressure signal zero point after cleaning, eliminating the residual error of flow field disturbance and ensuring the continuous reliability of measurement data.

[0045] 2、By dynamically adjusting the cleaning pulse parameters to adapt to different fluid viscosity and clogging degree, it not only avoids the accumulation of errors caused by insufficient cleaning, but also prevents energy waste caused by over-cleaning. Through nonlinear fusion of winding risk coefficient, the system can trigger accurate cleaning at the early stage of flow field distortion, solving the maintenance delay problem caused by flow passage deformation lag in traditional methods, and greatly improving the long-term stability and measurement accuracy of wedge flowmeter in complex impurity environment. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 The present application is a structure diagram of a kind of anti-clogging detection and control system for wedge flowmeter. DETAILED DESCRIPTION

[0047] With reference to the accompanying drawings: clear and complete description of the technical solutions in the embodiments of the present application will be described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of the present application.

[0048] Embodiments: Figure 1 The structural diagram of the anti-blocking measurement and control system for the wedge flowmeter is given, and the anti-blocking measurement and control system for the wedge flowmeter comprises the following modules:

[0049] The differential pressure detection module is used for detecting the differential pressure signal between the inlet and outlet of the wedge flowmeter and generating a corresponding frequency spectrum distribution graph.

[0050] The feature extraction module is used for extracting the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points from the frequency spectrum distribution graph.

[0051] The trigger determination module is used for determining whether to trigger the analysis of the winding type blockage according to whether the low-frequency energy proportion exceeds a first threshold value and whether the number of high-frequency noise mutation points is less than a second threshold value.

[0052] The entropy value analysis module is used for performing multi-scale sample entropy analysis on the differential pressure signal when the analysis of the winding type blockage is triggered, fitting the entropy value decay slope of each scale, and generating a complexity decay rate.

[0053] The risk fusion module is used for generating a winding risk coefficient based on the complexity decay rate and the low-frequency energy proportion.

[0054] The control execution module is used for controlling the cleaning device to perform directional pulse flushing on the wedge region when the winding risk coefficient exceeds a third threshold value, and recalibrating the differential pressure signal zero point after flushing.

[0055] The differential pressure detection module collects the differential pressure signals of the inlet and outlet in real time through the sensor, generates a frequency spectrum distribution graph after preprocessing, and transmits the frequency spectrum data to the feature extraction module. The feature extraction module extracts the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points from the frequency spectrum distribution graph and outputs them to the trigger judgment module. The trigger judgment module judges whether to start the subsequent analysis process based on the double conditions that whether the low-frequency energy proportion exceeds the first threshold value and whether the number of high-frequency noise mutation points is less than the second threshold value. If it is determined to trigger, the entropy analysis module performs multi-scale sample entropy analysis on the original differential pressure signal, fits the entropy value attenuation slope of each scale to generate the complexity attenuation rate, and inputs the result into the risk fusion module. The risk fusion module dynamically allocates weights according to the fluid type, nonlinearly fuses the normalized complexity attenuation rate and low-frequency energy proportion, and generates the winding risk coefficient. The control execution module monitors the risk coefficient in real time, and when it exceeds the third threshold value, controls the cleaning device to perform directional flushing according to the pulse parameters mapped by the fluid type, delays for a specific time window after flushing to stabilize the flow field, then re-collects the differential pressure signal to calibrate the zero point, and completes the closed-loop control.

[0056] Each module forms a cooperative mechanism through the series connection of data flow and logic chain: differential pressure detection and feature extraction realize multi-dimensional analysis of signals, trigger judgment avoids the risk of single parameter misjudgment, entropy analysis captures time domain dynamic instability characteristics, risk fusion enhances the robustness of complex working conditions, and control execution solves the defects of traditional methods such as cleaning overload or deficiency by combining dynamic parameters and closed-loop calibration. Through frequency domain-time domain joint analysis, multi-threshold cooperative judgment and fluid adaptive dynamic control, this connection breaks through the limitations of conventional threshold alarm or fixed cleaning strategy, especially for the concealment, slow change and fluid dependence of winding type blockage, and realizes the whole link closed-loop management from signal perception to precise execution.

[0057] Detect the differential pressure signal between the inlet and outlet of the wedge flowmeter and generate the corresponding frequency spectrum distribution graph, which is implemented as follows:

[0058] When the pressure sensor continuously collects the real-time differential pressure signal between the inlet and outlet of the wedge flowmeter, the specific implementation is as follows: differential pressure type pressure sensors are installed on the outer walls of the inlet and outlet pipes of the wedge flowmeter, and the installation distance between the two sensors is 1 to 3 times the diameter of the pipe to avoid interference of fluid disturbance. The sampling frequency of the pressure sensor is set to 1000 Hz to 5000 Hz, covering the main frequency range of fluid flow fluctuation. During the collection process, the pressure sensor outputs continuous voltage analog signals, which are converted into digital signals by an analog-to-digital converter with a conversion accuracy of 12 to 16 bits to ensure that the dynamic range of the differential pressure signal meets the measurement requirements.

[0059] In the preprocessing of the differential pressure signal to remove high-frequency noise and baseline drift, the specific implementation is as follows: first, the original differential pressure signal is filtered by a Butterworth low-pass filter, and the cutoff frequency of the filter is set to one-tenth to one-fifth of the sampling frequency, for example, when the sampling frequency is 2000 Hz, the cutoff frequency is 200 Hz to 400 Hz, to filter out sensor circuit noise and high-frequency fluid turbulence noise. Second, the filtered signal is corrected for baseline drift, and the baseline drift correction method is as follows: the signal segment of the differential pressure signal when the fluid is stationary or stably flowing is taken as the reference, the mean value of the segment is calculated as the baseline offset, and the offset is subtracted from the entire signal. If the fluid cannot be completely stationary, a sliding window polynomial fitting method is used, with a window length of 10 to 30 seconds, to perform a quadratic polynomial fitting on the signal in the window, and the low-frequency trend of the fitting curve is subtracted as the baseline drift to generate the preprocessed differential pressure signal.

[0060] Based on the preprocessed differential pressure signal, the specific implementation of generating a frequency spectrum distribution graph covering a preset frequency band by fast Fourier transform is as follows: the preprocessed differential pressure signal is divided into multiple continuous time windows according to the time sequence, and the length of each time window is 1 to 5 seconds, and the overlap rate between adjacent windows is 20% to 50%. The signal in each time window is applied with a Hanning window function to reduce spectral leakage, and then the windowed signal is subjected to fast Fourier transform to calculate its amplitude spectrum. The preset frequency band is set according to the fluid type and pipe size, for example, for water-based fluid and pipe diameter of 50 to 200 mm, the preset frequency band covers 0 to 100 Hz, of which 0 to 10 Hz is defined as the low frequency band and 10 to 100 Hz is defined as the high frequency band. The horizontal axis of the frequency spectrum distribution graph is the frequency value, and the vertical axis is the normalized amplitude energy, and the amplitude energy value of each frequency point is the square of the modulus value of the Fourier transform result at that frequency divided by the signal length.

[0061] In the preprocessing process, the order of the Butterworth low-pass filter is set to 4 to 8 to achieve a steep transition between the passband and the stopband, avoiding signal distortion. For example, when the cutoff frequency is 300 Hz, a 6th order Butterworth filter is used, with a passband fluctuation of less than 0.1 dB and a stopband attenuation of more than 60 dB. The sliding window polynomial fitting method used in baseline drift correction adjusts the polynomial order according to the stability of fluid flow: for stable flow conditions, a 1st to 2nd order polynomial is used; for fluctuating flow conditions, a 3rd to 4th order polynomial is used. If there is still residual offset after baseline drift correction, further mean zero normalization processing is performed, i.e. the mean value of the entire signal sequence is calculated and subtracted.

[0062] The specific parameter settings of the fast Fourier transform are as follows: the number of signal points in each time window is determined according to the sampling frequency and the window length, for example, when the sampling frequency is 2000 Hz and the window length is 2 seconds, the number of signal points is 4000 points, and the number of points of the fast Fourier transform is expanded to 4096 points to match an integer power of 2. The frequency resolution of the spectral distribution map is the sampling frequency divided by the number of transform points, for example, when the sampling frequency is 2000 Hz and the number of transform points is 4096, the frequency resolution is 0.488 Hz. The normalization method of spectral energy is to divide the amplitude value of each frequency point by the maximum possible amplitude value of the signal, which is determined according to the range of the analog-to-digital converter, for example, when the analog-to-digital converter range is ±5 volts, the maximum possible amplitude value is 5 volts.

[0063] After generating the spectral distribution map, the system automatically records the spectral data of each time window and stores it in matrix form in chronological order, with the rows corresponding to the time window index, the columns corresponding to the frequency point index, and the matrix element value being the normalized amplitude energy. The division of the preset frequency band is based on the fluid flow characteristics and the common interference frequency range, for example, in the sewage treatment scenario, the upper limit of the high frequency band of the preset frequency band is set to 80 Hz to avoid the influence of water pump vibration noise. If the type of fluid or the size of the pipeline changes in actual application, the preset frequency band can be manually adjusted through the user interface, and the adjusted frequency band data is saved to the system configuration file for subsequent analysis and call.

[0064] The low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points are extracted from the spectral distribution map, and the specific implementation is as follows:

[0065] When extracting the low-frequency energy proportion of flow field distortion from the spectral distribution map, the specific implementation is as follows: integrate the normalized amplitude energy in the preset low-frequency band, and the frequency range of the preset low-frequency band is dynamically adjusted according to the type of fluid, for example, set to 0 Hz to 10 Hz for water-based fluid and 0 Hz to 5 Hz for oil-based high-viscosity fluid. The integral calculation adopts the trapezoidal integral method to adapt to the case where the frequency interval in the spectral distribution map is not uniform, and the specific steps are as follows: add the normalized amplitude energy values of the adjacent two frequency points, multiply by the frequency interval, and then divide by 2, and accumulate point by point to obtain the low-frequency energy value. The frequency interval is determined by the spectral resolution of the fast Fourier transform, for example, when the spectral resolution is 0.488 Hz, the range of 0 Hz to 10 Hz contains 20 intervals, and the integral contribution value of each interval is (energy value 1 + energy value 2) x 0.488 / 2. The source of the normalized amplitude energy value is the preprocessed differential pressure signal generated by the fast Fourier transform, and its value is determined by the ratio of the 16-bit analog-to-digital converter range ±5V to the digital range 32767, i.e. 1 corresponds to 5V.

[0066] The specific implementation method for extracting the number of high-frequency noise abrupt change points is as follows: Within a preset high-frequency band, the number of adjacent frequency points whose amplitude energy difference exceeds a preset abrupt change threshold is detected. The preset high-frequency band is defined to cover the main noise interference frequency bands; for example, in industrial wastewater containing sand, it is set to 20Hz to 150Hz to avoid pump fundamental frequency interference. The difference between adjacent frequency points is calculated using the absolute value method, i.e., |energy value n+1 - energy value n|, where n is the frequency point index. The preset abrupt change threshold is set in two modes: static threshold and dynamic threshold. In static mode, a fixed value is preset according to the type of fluid impurities; for example, 0.03 for fiber impurities and 0.08 for particulate impurities. In dynamic mode, 10% of the maximum energy value within the high-frequency band is taken as the adaptive threshold. For example, when the maximum energy value of the high-frequency band is 0.5, the dynamic threshold is 0.05. During the detection process, if a difference exceeds the threshold, the counter is incremented by 1, and the final statistical result is the number of abrupt change points.

[0067] The specific implementation method for calculating the proportion of low-frequency energy is as follows: the ratio of the low-frequency energy value to the total energy value (low-frequency energy value + high-frequency energy value) is converted into a percentage. The calculation method for the high-frequency energy value is the same as that for the low-frequency energy value, with the integration interval being a preset high-frequency band. For example, in a paper mill pulp pipeline, if the integrated result of the low-frequency energy value is 0.8 and the high-frequency energy value is 1.2, then the proportion of low-frequency energy is 0.8 / (0.8+1.2)×100%=40%. The calculation of the total energy value excludes interference components outside the preset frequency band, such as ultrasonic noise above 150Hz. If the total energy value is zero (theoretically impossible, but possibly due to signal loss), the proportion is forcibly reset to zero and a sensor fault alarm is triggered.

[0068] Dynamic adjustment of preset frequency bands is achieved through the user interface. After the user inputs the fluid type (such as "sewage" or "crude oil"), the system automatically matches the built-in frequency band parameter library. For example, selecting "sewage" loads the 0-10Hz low-frequency band and the 10-100Hz high-frequency band; selecting "crude oil" loads the 0-5Hz low-frequency band and the 5-50Hz high-frequency band. The frequency band parameter library is established based on multi-condition experimental data. The experimental methods include: collecting pressure difference signals under clean flow fields and simulated clogging flow fields, comparing the differences in spectral energy distribution, and determining the energy concentration frequency band. For example, fiber entanglement leads to an energy increase of more than 30% in the 0-10Hz range, while particle accumulation causes a 50% increase in the number of abrupt change points in the 10-100Hz range.

[0069] When detecting abrupt changes in high-frequency noise, the fault-tolerance mechanism for instantaneous signal anomalies (such as sensor momentary disconnection) is as follows: if the number of abrupt changes within a single time window exceeds 50% of the total number of frequency points in the high-frequency band, it is determined to be a signal anomaly. The data in the current window is discarded and replaced with interpolated data from the previous window. The interpolation method is linear extrapolation. For example, if the number of abrupt changes in the previous window was 20, and the current window detects 100 abrupt changes due to an anomaly, it is replaced with 20.

[0070] The hardware dependencies for integral calculation and mutation point detection include: a processor supporting floating-point operations (such as an ARM Cortex-M7) and at least 100KB of memory for storing the spectrum matrix data. Software dependencies are open-source mathematical libraries (such as the `trapz` function in NumPy 1.21.5). During calculation, if the processor load exceeds 80%, the time window overlap of the spectrum distribution plot is automatically reduced to 20% to ensure real-time performance.

[0071] The optimization and validation of the preset mutation threshold were conducted using cross-validation: 100 sets of historical data were collected (50 sets under normal operating conditions and 50 sets under congested operating conditions), and the false alarm rate and false negative rate were tested under different thresholds. The threshold that resulted in the highest F1 score was selected. For example, in the fiber impurity scenario, when the threshold was 0.04, the F1 score reached 0.92 (95% precision and 89% recall). The threshold data is stored in non-volatile memory and is automatically loaded when the system starts.

[0072] Whether to trigger entanglement-type blockage analysis is determined based on whether the proportion of low-frequency energy exceeds the first threshold and the number of high-frequency noise abrupt changes is less than the second threshold. The specific implementation is as follows:

[0073] When determining whether to trigger entanglement-type blockage analysis based on whether the proportion of low-frequency energy exceeds a first threshold and the number of high-frequency noise abrupt changes is less than a second threshold, the specific implementation method is as follows: The first threshold is dynamically adjusted based on a preset multiple range of the average proportion of low-frequency energy under historical normal operating conditions corresponding to the fluid type. Historical normal operating condition data is obtained by long-term monitoring of flow signals under non-blockage conditions and stored in an embedded database (such as SQLite), with a data retention period of 3 months to 1 year. The preset multiple range is set according to the fluid viscosity characteristics: for low-viscosity fluids (such as water-based solutions), the multiple range is 1.2 to 1.5 times; for high-viscosity fluids (such as crude oil), the multiple range is 1.1 to 1.3 times. For example, when the average proportion of low-frequency energy under historical normal operating conditions is 30%, the first threshold for low-viscosity fluids is set to 36% to 45%. The adjusted first threshold is written to the system configuration file and automatically loaded when the fluid type changes.

[0074] The second threshold is set based on the high percentile of the statistical distribution of the number of high-frequency noise mutation points under non-congestion conditions. The high percentile is determined by statistically analyzing the distribution of mutation points during non-congestion periods in historical data. Specifically, at least 100 sets of non-congestion data are collected, sorted in ascending order, and the 90th percentile value is taken as the second threshold. For example, if the number of mutation points at the 90th percentile in 100 sets of data is 50 times / second, then the second threshold is set to 50 times / second. If there are fewer than 100 sets of historical data, a linear interpolation method is used to estimate the high percentile: when there are N sets of data, the 90th percentile position is 0.9×(N+1), and the weighted average of the two data points adjacent to this position is taken.

[0075] The comparison logic between the low-frequency energy percentage and the number of high-frequency noise mutation points is as follows: the real-time calculated low-frequency energy percentage is compared with the dynamically adjusted first threshold, and the real-time number of high-frequency noise mutation points is compared with the second threshold. If and only if the low-frequency energy percentage exceeds the first threshold and the number of high-frequency noise mutation points is less than the second threshold, it is determined to meet the characteristics of entanglement-type blockage, triggering subsequent analysis. For example, if the real-time low-frequency energy percentage is 42% (the first threshold is 40%) and the real-time number of high-frequency mutation points is 30 times / second (the second threshold is 50 times / second), analysis is triggered if both conditions are met.

[0076] The abnormal handling mechanism for dynamic threshold adjustment includes: if there are fewer than 10 sets of historical normal operating condition data, the first threshold is set to the default multiple of 1.3, and a data shortage alarm is issued; if the historical data for the number of high-frequency noise mutation points is all zero (theoretically impossible, possibly due to sensor failure), a preset empirical value (e.g., 40 times / second) is temporarily used as the second threshold, and the sensor calibration program is initiated. The threshold update cycle is set to weekly or monthly, adjusted according to the stability of the system operating environment: weekly updates in highly polluted scenarios such as chemical plants, and monthly updates in stable scenarios such as water treatment.

[0077] The hardware dependencies for the decision-making logic include: a microprocessor supporting floating-point operations (such as an ARM Cortex-M4), and at least 50KB of memory for storing threshold parameters and historical data indexes. Software dependencies include open-source statistical libraries (such as the percentile function in SciPy 1.7.3) and linear interpolation algorithms. During the decision-making process, if real-time data is abnormal (e.g., the proportion of low-frequency energy is greater than 100% or the number of high-frequency abrupt change points is negative), the current data is discarded and re-acquired. After three consecutive abnormal occurrences, a system self-check is triggered.

[0078] When triggering the entanglement-type blockage analysis, multi-scale sample entropy analysis is performed on the differential pressure signal. The attenuation slope of the entropy value at each scale is fitted to generate the complexity attenuation rate. The specific implementation is as follows:

[0079] When triggering entanglement-type blockage analysis, multi-scale sample entropy analysis is performed on the pressure difference signal to fit the attenuation slope of the entropy value at each scale, generating a complexity attenuation rate. The specific implementation method is as follows: First, the fluid turbulence intensity is calculated based on the dynamic fluctuation characteristics of the pressure difference signal. The fluid turbulence intensity is quantified by the variance of the pressure difference signal within a preset time window; a larger variance indicates higher turbulence intensity. For example, when the variance of the pressure difference signal is in the range of 0.1 to 0.5, it is determined to be low turbulence intensity; when the variance is in the range of 0.5 to 2.0, it is determined to be high turbulence intensity. A coarse-grained time scale range is defined according to the turbulence intensity. The time window length for long-scale sequences corresponding to low turbulence intensity is 2 to 5 seconds, and the time window length for short-scale sequences corresponding to high turbulence intensity is 0.5 to 1 second. The time window length for turbulence intensity calculation is fixed at 10 seconds and is independent of the window in the subsequent coarse-graining step.

[0080] The specific method for generating the coarse-grained sequence is as follows: the original differential pressure signal is segmented into non-overlapping segments according to the length of the defined time window. The average value of the data points within each segment is used to generate the coarse-grained sequence. For example, when the time window length is 2 seconds and the sampling frequency is 1000Hz, each segment contains 2000 data points, and the coarse-grained sequence is the average sequence of each 2000 points. Long-scale sequences with low turbulence intensity retain low-frequency trend characteristics, while short-scale sequences with high turbulence intensity retain high-frequency detail characteristics. The standard deviation of the coarse-grained sequence is calculated as an index of the dispersion of each data segment, which is used for subsequent adjustment of the sample entropy parameter.

[0081] When calculating the sample entropy value for each coarse-grained sequence, the calculation parameters are set as follows: embedding dimension m = 2 (according to the general standard of "Nonlinear Time Series Analysis"), and tolerance r = 0.2 times the standard deviation of the coarse-grained sequence. The sample entropy value reflects the complexity of the sequence; the lower the value, the more regular the sequence. For example, the sample entropy value of a long-scale coarse-grained sequence might be 1.2, while the sample entropy value of a short-scale sequence might be 1.8. When detecting the non-stationary decay trend of the sample entropy value with increasing scale, non-stationarity is determined by the fluctuation amplitude of the entropy value with scale: if the difference between the entropy values ​​of adjacent scales exceeds 1.5 times the average difference of the previous three scales, it is determined to be a non-stationary decay. This judgment rule has been validated with historical data, and the accuracy rate exceeds 90%.

[0082] The method for calculating sample entropy is as follows:

[0083]

[0084] In the formula: a represents the coarse-grained sequence X = [x1, x2, ..., x N ], containing N data points; b = m represents the embedding dimension (typically 2), used to define the vector length; c = r represents the tolerance parameter, which takes the value of 0.2 times the standard deviation of the coarse-grained sequence; E m(d) represents the proportion of all vector pairs with a distance less than r in dimension m.

[0085] The specific calculation steps are as follows:

[0086] Construct a vector of dimension m:

[0087]

[0088] Where, in the formula This represents a vector consisting of m consecutive data points starting from the i-th point in sequence x.

[0089] Probability calculation:

[0090]

[0091] Among them, E m (r) represents the average similarity ratio of all vector pairs in dimension m; Let E represent the similarity ratio of the i-th vector to all other vectors in dimension m; similarly, E is calculated. m+1 (r) The post-adoption formula yields the sample entropy value; E m+1 (r) represents the average similarity ratio of all vector pairs in dimension m+1, calculated in the same way as E. m (r) Same.

[0092] When using piecewise linear regression to fit the decay slopes of the initial and stable segments for non-stationary decay trends, the segmentation method is as follows: the sample entropy value sequence of the first one-third scale is divided into the initial segment, and the latter two-thirds scale is divided into the stable segment. For example, when the total number of scales is 6, the initial segment contains the first 2 scales, and the stable segment contains the last 4 scales. The fitting objective of linear regression is the linear relationship between the scale index and the sample entropy value. The slope of the initial segment reflects the rapid rate of decrease in complexity, and the slope of the stable segment reflects the sustained rate of decrease. The absolute value of the difference in decay slopes is calculated by subtracting the absolute value of the slope of the stable segment from the slope of the initial segment. For example, if the slope of the initial segment is -0.3 and the slope of the stable segment is -0.1, the difference is 0.2, and the absolute value is the complexity decay rate.

[0093] The anomaly handling mechanism includes: if the standard deviation of the coarse-grained sequence is zero (theoretically impossible, possibly due to sensor malfunction), the calculation of the sample entropy for the current sequence is skipped and it is marked as invalid data; if the sample entropy value shows an increasing trend with increasing scale (contradicting theoretical decay), it is determined as data anomaly and a re-acquisition is triggered. Parameter adjustment is based on: the turbulence intensity threshold is determined statistically through the variance distribution of normal and blocked operating conditions in historical data; for example, the variance for normal operating conditions is concentrated between 0.1 and 0.8, and the variance for blocked operating conditions is concentrated between 0.8 and 2.0. Threshold data is stored in an embedded database and supports dynamic loading according to fluid type.

[0094] Hardware dependencies include: a microcontroller supporting multithreaded processing (such as an STM32H743, 400MHz clock speed, 512KB memory) for parallel computation of sample entropy values ​​at different scales; software dependencies include an open-source entropy computation library (such as EntropyLibrary 2.3). During computation, if the microcontroller's memory usage exceeds 80%, the number of coarse-grained scales is automatically reduced to a preset minimum (e.g., 4 scales) to ensure real-time performance.

[0095] The specific rules for defining the coarse-grained timescale range are adaptively adjusted based on the fluid type: For water-based fluids, the turbulence intensity threshold is set to a variance of 0.5, meaning that a long scale (2 to 5 seconds) is used when the variance is below 0.5, and a short scale (0.5 to 1 second) is used when the variance is above 0.5; for oil-based fluids, the threshold is set to a variance of 0.3, because the turbulence intensity of high-viscosity fluids is generally lower. For example, a long-scale analysis is still used when the variance of oil-based fluids is 0.4 to capture slow flow field distortions. The adaptive adjustment logic is implemented through a configuration file, and users can select the fluid type through the human-machine interface.

[0096] The tolerance parameter *r* in the sample entropy calculation is adjusted according to the dynamic range of the coarse-grained sequence: when the standard deviation of the sequence is less than 0.1, *r* = 0.15; when the standard deviation is greater than 0.1, *r* = 0.2. This rule avoids excessively large tolerances for low-fluctuation sequences, which could distort the entropy value. For example, when the standard deviation of the coarse-grained sequence is 0.08, *r* = 0.15; when the standard deviation is 0.12, *r* = 0.2. The adjusted *r* value is written to a temporary cache for subsequent calculations.

[0097] The ratio of the initial segment to the stable segment in piecewise linear regression is optimized based on historical data: Analysis of 100 sets of normal and blocked data revealed that the entropy decrease rate in the initial stage of entanglement blockage (initial segment) is 2 to 3 times that of the stable segment. Therefore, the initial segment is set to account for one-third of the total scale. If the total number of scales is 9, the initial segment consists of the first 3 scales, and the stable segment consists of the last 6 scales. The ratio parameter is stored in read-only memory to prevent tampering during runtime.

[0098] The application logic for the complexity decay rate is as follows: when the decay rate exceeds a dynamic threshold (e.g., 0.25), it is judged as a high-risk entanglement blockage. The dynamic threshold is set according to the fluid type and pipe diameter; for example, the threshold is 0.2 for oil-based fluids in a DN100 pipe and 0.3 for sewage fluids in a DN200 pipe. Threshold data is stored in non-volatile memory, allowing on-site engineers to calibrate it via a human-machine interface. During calibration, the system automatically records historical decay rate data for the past 30 days and generates recommended threshold values ​​for engineer confirmation.

[0099] Boundary condition handling includes: if the variance of the differential pressure signal exceeds the sensor's range (e.g., variance > 5.0), it is considered a hardware fault, the multi-scale analysis function is disabled, and an alarm is triggered; if the sample entropy value cannot be calculated due to data loss, the entropy value from the previous time period is used as a substitute. The interpolation method is linear extrapolation; for example, if the entropy value of the previous time period is 1.5 and the current time period is missing, then the interpolation value is 1.5. The interpolated data is marked as an estimated value for reference in subsequent steps.

[0100] Based on the complexity decay rate and the proportion of low-frequency energy, an entanglement risk coefficient is generated, specifically implemented as follows:

[0101] When generating an entanglement risk coefficient by weighted fusion of complexity attenuation rate and low-frequency energy proportion, the specific implementation method is as follows: Fusion weights are dynamically allocated based on the viscosity characteristics corresponding to the fluid type. The fluid type is obtained through a viscosity sensor or a preset fluid parameter library, and viscosity characteristics are divided into low viscosity (less than 100 cP), medium viscosity (100 cP to 1000 cP), and high viscosity (greater than 1000 cP). The weight allocation rule is as follows: for low-viscosity fluids, the weight of low-frequency energy proportion is 60% to 70%, and the weight of complexity attenuation rate is 30% to 40%; for high-viscosity fluids, the weight of low-frequency energy proportion is adjusted to 30% to 40%, and the weight of complexity attenuation rate is 60% to 70%. For example, when the detected fluid is high-viscosity crude oil (viscosity 1200 cP), the weight of complexity attenuation rate is set to 65%, and the weight of low-frequency energy proportion is set to 35%. The weight data is stored in an embedded database and supports dynamic loading based on real-time viscosity detection results.

[0102] When normalizing the complexity decay rate and the proportion of low-frequency energy, the normalization method is as follows: calculate the maximum and minimum values ​​of the complexity decay rate and the proportion of low-frequency energy under the current operating conditions, and map the real-time values ​​to the range of 0 to 1. For example, if the maximum value of the complexity decay rate in historical data is 0.8, the minimum value is 0.1, and the current value is 0.5, then the normalization result is (0.5-0.1) / (0.8-0.1) = 0.57; the maximum value of the low-frequency energy proportion is 50%, the minimum value is 10%, and the current value is 30%, then the normalization result is (30-10) / (50-10) = 0.5. The normalization parameters are updated every 24 hours, based on the extreme values ​​statistically analyzed from the historical data of the last 7 days.

[0103] When the normalized complexity attenuation rate and the proportion of low-frequency energy are linearly weighted and summed according to the assigned weights, the calculation formula is as follows:

[0104] crx = fv*Az + sf*Bz

[0105] Where crx is the entanglement risk coefficient, fv is the normalized complexity decay rate, sf is the normalized low-frequency energy proportion, Az is the weight A, and Bz is the weight B.

[0106] The sum of weights A and B is forced to be 100%. For example, if weight A is 65%, weight B is 35%, the normalized complexity decay rate is 0.57, and the normalized low-frequency energy proportion is 0.5, then the risk coefficient = 0.57 × 0.65 + 0.5 × 0.35 = 0.555. If the sum of the input weights is not 100% due to a configuration error (e.g., weight A = 70%, weight B = 40%), the system will automatically scale proportionally: weight A = 70 / (70+40) = 63.6%, weight B = 40 / (70+40) = 36.4%, ensuring the sum is 100%.

[0107] The dynamic weight allocation is adjusted based on the following: through experimental testing of the sensitivity of various parameters under different viscosity fluids, the proportion of low-frequency energy in low viscosity fluids is more sensitive to blockage, and the complexity attenuation rate in high viscosity fluids is more significant due to flow instability.

[0108] The exception handling mechanism includes: if the maximum value of the normalized parameter equals the minimum value (no data fluctuation), the normalization step is skipped, and the original value is directly multiplied by the weight; if there is no matching item for the current fluid type in the weight configuration table, the default weight (50%:50%) is used and an alarm is triggered. Hardware dependencies include: a microcontroller supporting floating-point operations (such as STM32F746, 216MHz clock speed, 320KB memory) for real-time calculation of the weighted sum; software dependencies are open-source linear algebra libraries (such as Arm CMSIS-DSP 5.7.0).

[0109] Viscosity characteristics can be obtained in two ways: one is through real-time detection using an online viscosity sensor (such as a rotary viscometer, with a measurement range of 1 cP to 10000 cP and an accuracy of ±2% FS); the other is by retrieving historical viscosity data from a preset parameter library based on the fluid type. The preset parameter library stores typical viscosity values ​​for common fluids, such as water-based fluids (1 cP to 10 cP), crude oil (800 cP to 1200 cP), and polymer solutions (5000 cP to 10000 cP). When using the parameter library mode, the user selects the fluid name (such as "water" or "crude oil") through the human-machine interface, and the system automatically matches the viscosity value.

[0110] The extreme value statistics rules for normalization are as follows: the maximum and minimum values ​​are taken from the rolling extreme values ​​of the same time period over the most recent 7 days. For example, data at 8:00 AM each day is counted separately to avoid data confusion between different working conditions and time periods. If there is insufficient data within 7 days (e.g., less than 10 sets), the statistical period is extended to 30 days to ensure the reliability of extreme values. The normalization result is retained to two decimal places; values ​​exceeding the range of 0 to 1 are forcibly truncated to 0 or 1, and anomalies are logged.

[0111] The dynamic adjustment of weight allocation is achieved through a lookup table: the system has a built-in weight mapping table, where each row corresponds to a fluid type and its viscosity range, and each column corresponds to weight A (complexity attenuation rate) and weight B (low-frequency energy proportion). For example, when the fluid type is "water" with a viscosity range of 1-10 cP, weight A = 30% and weight B = 70%; when the fluid type is "crude oil" with a viscosity range of 800-1200 cP, weight A = 65% and weight B = 35%. The lookup data is generated through factory calibration experiments. The calibration method is as follows: under simulated blockage scenarios, the detection accuracy of different weight combinations is tested, and the combination with the highest accuracy is selected and written into the table.

[0112] In real-time calculation of linear weighted summation, if the input parameters are abnormal (e.g., the normalized value is greater than 1 or less than 0), the normalized value from the previous time period is used as an interpolation substitute. The interpolation method is as follows: if the current data is abnormal, the moving average of the data from the previous 3 time periods is used as the substitute value. For example, if the current normalization complexity decay rate is 1.2 (abnormal), and the data from the previous 3 time periods are 0.6, 0.7, and 0.8, then the substitute value is (0.6 + 0.7 + 0.8) / 3 = 0.7.

[0113] Boundary condition handling includes: when the fluid viscosity exceeds the sensor's range (e.g., >10000 cP), switching to parameter library mode and employing a nearest neighbor matching strategy. For example, if a viscosity of 15000 cP is detected, the weight of the fluid type with the highest viscosity in the parameter library (e.g., "asphalt" with a viscosity of 10000 cP) is used. If there is no matching item in the parameter library, manual intervention mode is triggered, pausing automatic detection until an engineer manually configures the settings.

[0114] When the entanglement risk factor exceeds the third threshold, the control cleaning device performs directional pulse flushing on the wedge-shaped area, and recalibrates the zero point of the differential pressure signal after flushing. The specific implementation is as follows:

[0115] When the entanglement risk factor exceeds the third threshold, the specific implementation method for controlling the cleaning device to perform directional pulse flushing on the wedge-shaped area is as follows: The intensity and duration of the flushing pulse are dynamically adjusted according to the mapping relationship between fluid type and entanglement risk factor. The fluid type is obtained through a viscosity sensor or a preset parameter library. The flushing pulse intensity ranges from 0.5 MPa to 5 MPa, and the duration ranges from 1 second to 10 seconds. For example, if the detected fluid is high-viscosity crude oil and the entanglement risk factor is 0.8 (the third threshold is 0.6), the pulse intensity is set to 4 MPa and the duration to 8 seconds; if the fluid is a low-viscosity water-based solution and the risk factor is 0.7, the intensity is adjusted to 2 MPa and the duration to 5 seconds. The parameter mapping relationship is calibrated experimentally: under simulated blockage scenarios, the pulse intensity is gradually increased until the entanglement is cleared, and the minimum effective intensity and corresponding time are recorded to form an optimal parameter table for different fluid types.

[0116] The third threshold is set based on the statistical distribution of entanglement risk coefficients under normal and entanglement / clogging conditions in historical data, selecting a critical value to distinguish between the two types of conditions. Specifically, at least 100 sets of historical data (50 sets for normal conditions and 50 sets for clogging conditions) are collected, a histogram of risk coefficient distribution is plotted, and the 95th percentile of the normal condition distribution is taken as the third threshold. For example, if the maximum risk coefficient for normal conditions is 0.55 and the minimum for clogging conditions is 0.62, then the third threshold is set to 0.6. The threshold is dynamically adjusted for different fluid types: for high-viscosity fluids (>1000 cP), the threshold is increased by 0.05 to 0.1; for low-viscosity fluids (<100 cP), the threshold is decreased by 0.05 to 0.1. Threshold data is stored in a parameter library and can be retrieved based on real-time fluid type. If no matching item is found in the parameter library, the default threshold of 0.6 is used and an alarm is triggered.

[0117] After flushing, a preset time window is allowed to allow the fluid flow to stabilize. This preset time window is set based on the pipe diameter and fluid viscosity: 5 to 10 seconds for diameters smaller than DN100 and viscosities less than 100 cP; and 20 to 30 seconds for diameters larger than DN200 or viscosities greater than 1000 cP. For example, when transporting a medium-viscosity fluid (500 cP) through a DN150 pipe, the delay time is set to 15 seconds. Differential pressure signal acquisition is disabled during the delay period to avoid fluid disturbance interference.

[0118] When calculating the zero-point offset using differential pressure signals acquired under steady-state conditions, the zero-point offset is calculated as the absolute difference between the current average differential pressure signal and the zero-point reference value stored in the historical database. For example, if the historical reference value is -0.1 kPa and the current average signal value is 0.3 kPa, then the zero-point offset is 0.4 kPa. The preset tolerance range is set based on the sensor accuracy: if the sensor accuracy is ±0.05% FS (full scale ±10 kPa), then the tolerance range is set to ±0.1 kPa. If the offset exceeds the tolerance, the system automatically marks the current calibration as failed.

[0119] The rinsing and calibration process is repeated until the offset reaches the target value. The maximum number of repetitions is set to 3, with an increasing delay time of 10 seconds after each rinse to avoid over-rinsing. For example, the delay is 15 seconds after the first rinse, 25 seconds after the second, and 35 seconds after the third. If the target value is still not reached after 3 rinses, a manual intervention alarm is triggered and automatic rinsing is paused. The zero-point reference value is updated according to the following rule after calibration: the average of the 3 most recent successful zero-point calibrations is written to the historical database, overwriting the old reference value. The historical database uses a circular buffer to store reference data for the most recent 30 days for retrieval.

[0120] The hardware dependencies for dynamically adjusting flushing parameters include: a high-pressure pulse pump (such as a plunger pump, with a pressure range of 0.1 MPa to 10 MPa) and a flow control valve (with a response time of less than 0.5 seconds); the software dependencies are a real-time control system (such as FreeRTOS 10.4.3) and a calibration algorithm library (such as the sliding mean function of NumPy 1.21.5). During flushing, if the pulse pump pressure sensor detects an abnormal pressure (exceeding the set value by 10%), flushing is immediately terminated and fault diagnosis is triggered.

[0121] The mapping table between fluid type and flushing parameters is generated through the following steps: In a laboratory setting, clogging scenarios involving different fluids (water, crude oil, polymer solution) are simulated. The pulse intensity is gradually increased until the blockage is completely cleared, and the minimum effective intensity and corresponding time are recorded. For example, for crude oil at 25°C and a viscosity of 800 cP, the minimum effective intensity is 3.8 MPa for 7 seconds; for polymer solution (5000 cP), an intensity of 5.2 MPa is required to maintain the pressure for 12 seconds. The mapping table is stored in JSON format and includes four columns: fluid type, viscosity range, pulse intensity, and duration, and supports online updates.

[0122] The delay time window is set based on fluid dynamics simulation results: by calculating the Reynolds number and flow recovery time of the fluid in the pipe, the quantitative relationship between delay time and viscosity and pipe diameter is determined. For example, when the Reynolds number is less than 2000 (laminar flow), delay time = pipe diameter (m) × 10 + viscosity (cP) × 0.02; when the Reynolds number is greater than 4000 (turbulent flow), delay time = pipe diameter (m) × 5 + viscosity (cP) × 0.01. The simulation model is validated using ANSYS Fluent, with an error rate of less than 5%.

[0123] In the zero-point offset calculation, the average pressure difference signal is obtained using the sliding window method: the window length is set to the data within 10 seconds after the delay time ends, and the average is calculated every second, removing outliers within ±3 standard deviations. For example, if 1000 pressure difference points are collected within 10 seconds, 950 valid points are retained after removal, with an average of 0.05 kPa. If the percentage of valid data is less than 80%, the flow is considered unstable, and the delay time is extended by 10 seconds.

[0124] The historical database's circular buffer management strategy is as follows: the earliest data is automatically deleted at 00:00 every day, while the latest 30 days' records are retained. When the baseline value is updated, the system verifies the reasonableness of the new value: if the deviation between the new baseline value and the previous value exceeds 3 times the sensor's accuracy (e.g., ±0.15 kPa), a manual confirmation process is triggered to prevent abnormal data from polluting the database.

[0125] The hardware fault diagnosis logic includes: after the pulse pump pressure sensor detects three consecutive out-of-limit readings, switching to the backup pump and recording the fault code; when the flow control valve response times out (failure to reach the set opening degree within 0.5 seconds), activating the backup valve and calibrating the opening position. Fault data is uploaded to the monitoring center via the Modbus protocol, supporting remote diagnosis.

[0126] The technological improvements in this embodiment are reflected in the collaborative design of signal feature extraction, multi-dimensional fusion, and dynamic control. Traditional anti-clogging methods typically rely on a single signal dimension (such as differential pressure threshold) or fixed cleaning logic, making it difficult to distinguish between entanglement-type clogging and particulate impurity interference, and lacking adaptability to fluid dynamic characteristics. This embodiment constructs a dual criterion for entanglement risk through cross-dimensional coupling of frequency domain energy distribution (low-frequency energy proportion) and time domain complexity attenuation (sample entropy multi-scale analysis). Frequency domain features capture the energy concentration effect of flow field distortion, while time domain features quantify the dynamic evolution rate of flow instability. The two are nonlinearly fused through fluid viscosity-adaptive weight allocation, overcoming the limitations of traditional linear weighting or independent criteria. Furthermore, the cleaning control logic does not simply respond to threshold exceedances, but dynamically adjusts pulse parameters according to fluid type and risk level, and combines delay stabilization and closed-loop calibration mechanisms to solve the problem of false detection caused by flow field disturbances after a single flush. The synergistic effect of the above steps enables the system to distinguish the differential signal characteristics of fiber entanglement and particulate scouring, while overcoming the inherent interference of signal hysteresis in high-viscosity fluids.

[0127] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0128] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0129] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0130] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0131] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0132] The above description is merely a specific embodiment of this application, but the scope of protection of this application 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 application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0133] 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. An anti-clogging measurement and control system for a wedge flowmeter, comprising: The method comprises the following steps: a differential pressure detection module for detecting the differential pressure signal between the inlet and outlet of the wedge flowmeter and generating a corresponding frequency spectrum distribution; a feature extraction module for extracting the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points from the frequency spectrum distribution; a trigger determination module for determining whether to trigger the analysis of the winding type blockage according to whether the low-frequency energy proportion exceeds the first threshold value and the number of high-frequency noise mutation points is less than the second threshold value; an entropy value analysis module for performing multi-scale sample entropy analysis on the differential pressure signal when the analysis of the winding type blockage is triggered, fitting the entropy value decay slope of each scale, and generating a complexity decay rate, including: based on the dynamic fluctuation characteristics of the differential pressure signal, the fluid turbulence intensity is divided into a coarse-grained time scale range; generate the corresponding coarse-grained sequence according to the coarse-grained time scale range; calculate the sample entropy value of each coarse-grained sequence and detect its non-stationary decay trend with the increase of the scale; non-stationarity is determined by the fluctuation amplitude of the entropy value with the change of the scale: if the difference between adjacent scale entropy values exceeds 1.5 times the average difference of the first three scales, it is determined to be non-stationary decay; for the non-stationary decay trend, the decay slope of the initial segment and the stable segment is fitted respectively by using piecewise linear regression; the segmentation method is: the sample entropy value sequence of the first third of the scale is divided into the initial segment, and the last two-thirds of the scale is divided into the stable segment; the absolute value of the difference between the decay slopes of the initial segment and the stable segment is taken as the complexity decay rate; a risk fusion module for generating a winding risk coefficient based on the complexity decay rate and the low-frequency energy proportion, including: dynamically allocate the fusion weight of the complexity decay rate and the low-frequency energy proportion according to the viscosity characteristics of the fluid type; normalize the complexity decay rate and the low-frequency energy proportion respectively to eliminate the dimensional difference; linearly weight and sum the normalized complexity decay rate and low-frequency energy proportion according to the allocated weight to generate the winding risk coefficient; when the sum of the fusion weight exceeds the preset range, the weight is redistributed to the sum of the set value in proportion to the weight; a control execution module for controlling the cleaning device to perform directional pulse flushing on the wedge-shaped area when the winding risk coefficient exceeds the third threshold value, and recalibrating the differential pressure signal zero point after flushing.

2. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, Detecting the differential pressure signal between the inlet and outlet of the wedge flowmeter and generating a corresponding frequency spectrum distribution includes: continuously collecting real-time differential pressure signals between the inlet and outlet of the wedge flowmeter through pressure sensors; preprocessing the differential pressure signal to remove high-frequency noise and baseline drift to generate a preprocessed differential pressure signal; based on the preprocessed differential pressure signal, generate a frequency spectrum distribution covering a preset frequency band through fast Fourier transform.

3. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, Extracting the low-frequency energy proportion of flow field distortion and the number of high-frequency noise mutation points from the frequency spectrum distribution includes: integrate the normalized amplitude energy in the preset low-frequency band of the frequency spectrum distribution to obtain the low-frequency energy value; integrate the normalized amplitude energy in the preset high-frequency band of the frequency spectrum distribution to obtain the high-frequency energy value; take the ratio of the low-frequency energy value to the sum of the low-frequency energy value and the high-frequency energy value as the low-frequency energy proportion; The number of amplitude energy differences between adjacent frequency points in the high frequency band exceeding a preset mutation threshold is detected, and the number of high frequency noise mutation points is counted.

4. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, According to whether the low frequency energy proportion exceeds the first threshold and the number of high frequency noise mutation points is less than the second threshold, it is determined whether to trigger the winding type blockage analysis, including: Comparing the low frequency energy proportion with the preset first threshold, and comparing the number of high frequency noise mutation points with the preset second threshold; When the low frequency energy proportion exceeds the first threshold and the number of high frequency noise mutation points is less than the second threshold, the winding type blockage analysis is triggered, otherwise it is not triggered.

5. The anti-clogging measurement and control system for a wedge flowmeter of claim 4, wherein, The first threshold is dynamically adjusted according to the preset multiple range of the average low frequency energy proportion under the historical normal working condition corresponding to the fluid type; the second threshold is set according to the high quantile point of the statistical distribution of the number of high frequency noise mutation points under the non-blockage working condition.

6. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, Low turbulence intensity corresponds to long scale sequence, and high turbulence intensity corresponds to short scale sequence.

7. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, The weight of the complexity decay rate corresponding to the high viscosity fluid is increased.

8. The anti-clogging measurement and control system for a wedge flowmeter of claim 1, wherein, When the winding risk coefficient exceeds the third threshold, the cleaning device is controlled to perform directional pulse flushing on the wedge-shaped area, and the pressure difference signal zero point is recalibrated after flushing, including: When the winding risk coefficient exceeds the third threshold, the intensity and duration of the flushing pulse are dynamically adjusted according to the mapping relationship between the fluid type and the winding risk coefficient; After flushing is completed, a preset time window is delayed to wait for the fluid flow to stabilize, and then the pressure difference signal in the stable state is collected to calculate the zero point offset; If the zero point offset exceeds the preset tolerance range, the flushing and calibration process is repeated until the offset meets the standard; After calibration is completed, the zero point reference value in the historical database is updated for subsequent detection cycles.

Images