Flowmeter self-diagnosis method and device based on digital twinning

Abstract

This invention discloses a flow meter self-diagnosis method and device based on digital twins. The method involves acquiring the flow meter's original measurement signal and current operating parameters, obtaining the actual measured flow value and intermediate signal characteristics based on signal processing, inputting the operating parameters into an embedded digital twin model to generate simulated flow values ​​and expected signal characteristics, extracting a first deviation sequence and a second deviation sequence by the deviation between the simulated and actual values, performing trend separation to generate a systematic drift component, cross-validating the two deviation sequences to generate effective drift markers, bidirectionally matching the systematic drift component with a fault mode library to obtain suspected fault types, performing rate of change analysis on the drift component to obtain drift acceleration, and combining the fault type to perform health quantification to generate a flow meter health index and parameter drift amount, and adaptively compensating based on the parameter drift amount to output the compensated flow value and form a monitoring result, thus realizing online identification and adaptive compensation of flow meter parameter drift.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology for industrial instruments, and in particular to a self-diagnostic method and device for flow meters based on digital twins. Background Technology

[0002] As a core instrument in industrial process control and trade measurement, the accuracy of flow meters directly impacts production efficiency and economic benefits. During long-term operation, flow meters are affected by factors such as fluid erosion, temperature changes, and vibration, causing sensor performance and circuit parameters to gradually drift, leading to increased measurement errors. Timely detection and quantification of this parameter drift are crucial for ensuring measurement accuracy and developing appropriate maintenance plans.

[0003] Existing flowmeter condition monitoring methods primarily rely on periodic offline calibration or simple threshold alarms. Periodic calibration requires disassembling the flowmeter from the field and transporting it to a laboratory for calibration, which is not only time-consuming and labor-intensive but also makes it impossible to monitor the instrument's real-time status during the calibration interval. Threshold alarm methods only issue alarms when the measurement deviation exceeds a set range, lacking the ability to analyze drift trends and identify fault types, making it difficult to distinguish between different fault mechanisms such as sensor degradation and circuit aging. Furthermore, traditional methods fail to consider the individual differences in flowmeter characteristics developed during manufacturing and use, and using uniform judgment standards may lead to misjudgments or omissions. Summary of the Invention

[0004] This invention discloses a flow meter self-diagnosis method and device based on digital twins. By embedding a digital twin model in the flow meter to generate a simulation benchmark in real time, the actual measurement and simulation results are analyzed for deviation and trend separation in multiple dimensions. Systematic drift is identified and fault types are matched, realizing quantitative assessment of parameter drift and adaptive compensation for measurement error, providing technical support for intelligent operation and maintenance of flow meters.

[0005] The first aspect of this invention proposes a self-diagnostic method for flow meters based on digital twins, comprising the following steps: The system collects the original measurement signal and current operating parameters of the flow meter, performs signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and inputs the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics. A first deviation sequence is obtained by extracting the flow deviation between the simulated flow value and the actual measured flow value. A second deviation sequence is obtained by extracting the feature deviation between the expected signal feature and the intermediate signal feature. A systematic drift component is generated by performing trend separation based on the first deviation sequence and the second deviation sequence. Cross-validation is performed on the first deviation sequence and the second deviation sequence to generate effective drift markers. Based on the effective drift markers, the systematic drift components are bidirectionally verified and matched with a preset fault mode library to obtain suspected fault types. A rate of change analysis is performed on the systematic drift component to obtain the drift acceleration. The flow meter health index and parameter drift amount are generated by quantifying the suspected fault type and the drift acceleration. Based on the parameter drift, the actual measured flow rate value is adaptively and dynamically compensated to output the compensated flow rate value. The flow meter health index and the compensated flow rate value are used to form the flow meter monitoring result.

[0006] A second aspect of the present invention provides a flow meter self-diagnostic device based on digital twins, comprising: The signal acquisition module is used to acquire the original measurement signal and current operating parameters of the flow meter, perform signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and input the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics. The deviation extraction module is used to extract a first deviation sequence by comparing the simulated flow rate value with the actual measured flow rate value, extract a second deviation sequence by performing feature deviation extraction on the expected signal features and the intermediate signal features, and perform trend separation based on the first deviation sequence and the second deviation sequence to generate a systematic drift component. The fault diagnosis module is used to perform cross-validation on the first deviation sequence and the second deviation sequence to generate a valid drift marker, and to perform bidirectional verification and matching on the systematic drift component with a preset fault mode library based on the valid drift marker to obtain the suspected fault type. The health assessment module is used to perform a rate of change analysis on the systematic drift component to obtain the drift acceleration, and to quantify the health of the flow meter and the parameter drift by using the suspected fault type and the drift acceleration. The compensation output module is used to adaptively and dynamically compensate the actual measured flow rate value according to the parameter drift and output the compensated flow rate value. The flow meter health index and the compensated flow rate value are used to form the flow meter monitoring result.

[0007] The beneficial effects of this invention are reflected in the following points: 1. By inputting the current operating condition parameters into the embedded digital twin model to generate simulated flow values ​​and expected signal characteristics, an ideal response benchmark for the flowmeter under the current operating condition is established. The simulation results and actual measurement results are compared to extract flow deviation and multi-dimensional signal feature deviation, respectively, to obtain the first deviation sequence and the second deviation sequence. Through filtering and drift monotonicity verification, the systematic drift component is separated from the deviation sequence, thus achieving effective differentiation between random fluctuation components and trend drift components in the measurement deviation. 2. By performing synchronous change detection and directional consistency analysis on the first and second deviation sequences, effective drift markers are generated. Based on the effective drift markers, the systematic drift components are compared with a preset fault mode library for forward similarity matching and reverse compatibility verification to obtain suspected fault types. Forward matching is sorted by similarity based on drift direction, rate, and curve shape, while reverse verification is performed for compatibility screening based on physical feature constraints and fault mutual exclusion rules. The bidirectional verification mechanism improves the accuracy of fault type identification. 3. By extracting adjacent-period increments and identifying drift direction inflection points of systematic drift components, a degradation stage marker is established and drift acceleration is output. Combined with suspected fault types, fault-specific attenuation curves are queried and degradation risk levels are analyzed to generate a flowmeter health index and parameter drift. Based on the parameter drift, the actual measured flow value is adaptively and dynamically compensated to output the compensated flow value. The health index combined with the compensated flow value forms a flowmeter monitoring result that includes measurement data, health assessment, and maintenance recommendations, thus realizing online compensation of parameter drift and continuous monitoring of health status.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0010] Unless otherwise specified, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0011] Figure 1 This is a flowchart illustrating a flow meter self-diagnosis method based on digital twins according to the present invention.

[0012] Figure 2 This is a schematic diagram of the structure of a Coriolis mass flow meter according to the present invention.

[0013] Figure 3 This is a structural block diagram of a flow meter self-diagnostic device based on digital twins according to the present invention.

[0014] Wherein: 1-Flowmeter body; 2-Vibration tube; 3-Drive coil; 4-Detection coil; 5-Temperature sensor; 6-Pressure sensor; 7-Signal processing unit; 8-Operating condition monitoring device; 9-Inlet flange; 10-Inlet pipe; 11-Outlet pipe; 12-Measured fluid; 13-Outlet flange. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0016] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0017] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0018] The technical solutions of the embodiments of this application will be described below.

[0019] like Figure 1 As shown, this embodiment of the invention provides a flow meter self-diagnosis method based on digital twins, including the following steps S110-S150: Step S110: Collect the original measurement signal and current operating parameters of the flow meter, perform signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and input the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics.

[0020] Specifically, it collects the flow meter's raw measurement signal and current operating parameters. For example... Figure 2 As shown, the flowmeter body 1 is connected to the inlet pipe 10 and the outlet pipe 11 via the inlet flange 9 and the outlet flange 13, respectively. When the measured fluid 12 flows through the vibrating tube 2 inside the flowmeter body 1, it generates a Coriolis force effect. The original measurement signal is acquired in real time by the built-in sensor of the flowmeter. The drive coil 3 is installed in the middle of the vibrating tube 2 to excite the vibrating tube 2 to vibrate at its natural frequency. The detection coils 4 are installed on the inlet and outlet sides of the vibrating tube 2 to detect the motion response of the vibrating tube 2. The Coriolis force generated when the measured fluid 12 flows through the vibrating tube 2 causes a phase difference between the two sides of the vibrating tube 2. The magnitude of the phase difference is proportional to the mass flow rate. The phase difference signal output by the detection coil 4 and the vibration frequency signal constitute the main components of the original measurement signal. The original measurement signal is transmitted to the signal processing unit 7 for subsequent analysis after analog-to-digital conversion. The sampling frequency is set according to the flowmeter type and measurement accuracy requirements. Current operating parameters are read from the operating condition monitoring device 8. Temperature sensor 5, installed on the flowmeter body 1, collects the temperature of the fluid 12 being measured, and pressure sensor 6, also installed on the flowmeter body 1, collects the pipeline pressure. Density and viscosity are calculated based on the fluid type and current temperature and pressure conditions. The operating condition monitoring device 8 collects these physical property parameters, including temperature, pressure, density, and viscosity. The quality of the original measurement signal is directly affected by the current operating parameters. Changes in fluid temperature alter the elastic modulus of the vibrating tube 2, causing resonant frequency drift, while pressure fluctuations lead to signal zero-point drift. The acquisition cycle of the current operating parameters is synchronized with the original measurement signal to ensure accurate time correspondence between the signal and the operating conditions.

[0021] Signal processing is performed on the raw measurement signal to obtain the actual measured flow rate and intermediate signal characteristics. The raw measurement signal first undergoes a preprocessing stage to eliminate noise interference introduced during acquisition. Preprocessing methods include low-pass filtering to remove high-frequency noise, baseline correction to eliminate DC offset, and outlier removal to eliminate transient interference. The filter cutoff frequency is set according to the flowmeter's operating frequency band. The preprocessed raw measurement signal then enters the feature extraction stage of signal processing unit 7, extracting key feature parameters reflecting flow information from both the time and frequency domains. Time-domain features include signal amplitude, period, phase difference, and waveform distortion; frequency-domain features include fundamental frequency amplitude, harmonic distribution, and spectral energy distribution. The intermediate signal characteristics are derived by combining these time-domain and frequency-domain feature parameters. The dimension of the feature vector is determined according to the flowmeter type; for Coriolis flowmeters, the intermediate signal characteristics focus on phase and frequency parameters. The actual measured flow rate is calculated from the core characteristic parameters of the original measurement signal based on the flow meter's measurement principle. The Coriolis flow meter calculates the mass flow rate Qm using the calibration relationship between phase difference and flow rate. The calculation formula is Qm = K × Δφ, where K is the flow meter calibration coefficient, and Δφ is the phase difference detected by the detection coil 4. The calibration coefficient K is determined during the flow meter's factory calibration. In addition to the core flow calculation parameters, the intermediate signal characteristics also include auxiliary diagnostic parameters. These auxiliary diagnostic parameters are extracted by the signal processing unit 7 from the spectrum analysis results and reflect the sensor's operating status and signal quality, including indicators such as signal-to-noise ratio, waveform symmetry, and harmonic suppression ratio.

[0022] The current operating parameters are input into the embedded digital twin model to generate simulated flow values ​​and expected signal characteristics. The embedded digital twin model is a mathematical mapping of the flowmeter's physical structure and measurement principle. Built into the flowmeter's signal processing unit 7, it enables edge computing. Compared to cloud computing solutions, edge computing offers lower response latency and is unaffected by network conditions. The embedded digital twin model's architecture comprises three parts: a sensor response submodule, a fluid dynamics submodule, and a signal generation submodule. When the fluid temperature parameter from the current operating conditions is input into the embedded digital twin model, it triggers temperature compensation calculations in the sensor response submodule, calculating the temperature drift correction coefficient for sensor sensitivity. Pressure parameters trigger pressure compensation calculations in the sensor response submodule, calculating the correction amount for the impact of pipe deformation on measurement. Density and viscosity parameters trigger property calculations in the fluid dynamics submodule, calculating the influence of fluid characteristics on the signal response. The embedded digital twin model solves for the flowmeter's ideal response under the current operating conditions based on the parameters. Internally, the model uses the finite element method to calculate the sensor's dynamic response and computational fluid dynamics to calculate the flow field distribution. The two methods are coupled to obtain the theoretical value of the sensor's output signal. The simulated flow rate value represents the measurement result that the flow meter should provide under current operating parameters in a fault-free state. This value serves as a reference benchmark for actual measurements. The expected signal characteristics are synchronously generated by the embedded digital twin model, maintaining consistency with the intermediate signal characteristics in terms of feature dimensions. This includes signal amplitude, phase, frequency, and waveform parameters under ideal operating conditions. This dimensional consistency ensures that the two sets of characteristics can be compared dimension-by-dimensionally. Any change in the current operating parameters will cause corresponding adjustments to the simulated flow rate value and the expected signal characteristics. The output of the embedded digital twin model tracks changes in operating conditions in real time, with a response delay controlled within one sampling period. The difference between the simulated flow rate value and the actual measured flow rate value reflects the flow meter's measurement deviation, while the difference between the expected signal characteristics and the intermediate signal characteristics reflects changes in the state of the sensor and signal channel.

[0023] Step S120: Extract the flow deviation by comparing the simulated flow value with the actual measured flow value to obtain the first deviation sequence; extract the feature deviation by comparing the expected signal features with the intermediate signal features to obtain the second deviation sequence; and perform trend separation based on the first deviation sequence and the second deviation sequence to generate a systematic drift component.

[0024] Specifically, the first deviation sequence is obtained by extracting the flow deviation between simulated flow rate values ​​and actual measured flow rate values. The simulated flow rate value represents the theoretical output of the flow meter under current operating conditions, while the actual measured flow rate value represents the actual output of the flow meter. The difference between the two reflects the degree to which the flow meter deviates from the ideal state. The formula for calculating the first deviation sequence is ΔQ_i = Q_measured_i - Q_simulated_i, where Q_measured_i is the actual measured flow rate value in the i-th sampling period, Q_simulated_i is the simulated flow rate value in the same period, and ΔQ_i is the flow deviation value in that period. The simulated flow rate value and the actual measured flow rate value are paired periodically according to timestamps. After pairing, the deviation value for each period is calculated sequentially, and the deviation values ​​are arranged in chronological order to form the first deviation sequence. In the first deviation sequence, a positive deviation value indicates that the actual measurement is too large, that is, the reading given by the flow meter is higher than the actual flow rate; a negative deviation value indicates that the actual measurement is too small; and a deviation value close to zero indicates that the measurement is accurate. Actual measured flow rates are subject to short-term fluctuations due to random noise and operating condition variations, resulting in high-frequency oscillations in the first deviation sequence. Simulated flow rates, however, already account for changes in operating conditions; therefore, the low-frequency trend components in the first deviation sequence primarily reflect the flowmeter's own drift characteristics. The length of the first deviation sequence continuously increases with monitoring time, with new deviation values ​​calculated for each sampling period added to the end of the sequence. The sequence comprehensively records the deviation evolution of the flowmeter from its commissioning to the current moment.

[0025] In some embodiments, the step of extracting the feature deviation between the expected signal features and the intermediate signal features to obtain a second deviation sequence includes: performing multi-dimensional feature decomposition on the expected signal features and the intermediate signal features respectively to obtain feature components of each dimension; performing multi-dimensional deviation extraction on the feature components of each dimension to generate a multi-dimensional deviation set; performing inter-dimensional consistency analysis based on the multi-dimensional deviation set to identify the combination of deviation mismatch dimensions; and locating fault-related links and constructing a second deviation sequence based on the combination of deviation mismatch dimensions.

[0026] The expected signal features and intermediate signal features are decomposed into feature components of each dimension using multi-dimensional feature decomposition. The expected signal features are split into multi-dimensional feature vectors according to their dimension indices. The feature values ​​of each dimension are independently extracted to form the feature components of that dimension. After decomposition, independent feature sequences such as amplitude components, phase components, frequency components, harmonic components, and waveform components are obtained. The intermediate signal features are processed using the same decomposition method. The number and definition of the dimensions after decomposition correspond completely with those of the expected signal features, ensuring comparability between components of the same dimension. Each feature component of each dimension is stored in a pair of component names and component values. The expected signal features and intermediate signal features each generate a set of feature components of each dimension. The dimension indices of the two sets of components correspond one-to-one, facilitating the calculation of deviations. Some dimensions of the feature components have physical coupling relationships. These coupling relationships are determined by the physical principles of the embedded digital twin model and are pre-stored in the diagnostic knowledge base. For example, there is an energy conservation constraint between the vibration frequency and vibration amplitude of a Coriolis flowmeter. Changes in the mass load of the vibrating tube will simultaneously affect both the frequency and amplitude dimensions. This coupling relationship plays a key role in the consistency analysis between dimensions. The decomposition process is performed independently for each sampling period of the expected signal features and intermediate signal features, forming a time-ordered component sequence with a sequence length equal to the number of monitoring periods.

[0027] Dimensional deviation extraction is performed on the feature components of each dimension to generate a multidimensional deviation set. The difference between the j-th dimension component of the expected signal feature and the j-th dimension component of the intermediate signal feature is calculated, and the difference result is stored as the deviation component for that dimension in the multidimensional deviation set. The difference calculation uses algebraic subtraction to subtract the expected value from the actual value. Dimensional deviation extraction is performed sequentially on all dimensions, with each dimension calculating its deviation value independently without interference. The number of deviation components in the multidimensional deviation set corresponds to the number of dimensions in each feature component, maintaining consistency in the number of dimensions throughout the processing flow. The multidimensional deviation set organizes data using a mapping structure with dimension indices as keys and deviation values ​​as values. Within the same sampling period, the deviation values ​​of each dimension constitute a complete deviation record. The relative magnitude and sign of the deviation values ​​in each dimension contain fault diagnosis information. The sign and magnitude of the deviation value reflect the direction and degree of deviation between the actual and expected signals in that dimension. A deviation value of zero indicates a perfect match between the actual and expected signals in that dimension, while a large deviation value indicates a significant anomaly in that dimension requiring close attention. In a multidimensional deviation set, the dimensions and orders of magnitude of the deviation values ​​differ across dimensions. Amplitude deviations are measured in voltage or displacement, phase deviations in angles, and frequency deviations in Hertz. Directly comparing the deviation values ​​across different dimensions is meaningless. Before conducting inter-dimensional consistency analysis, it is necessary to normalize the deviations of each dimension to eliminate the influence of dimensional differences. The normalization method involves dividing the deviation of each dimension by the nominal value of that dimension's characteristic under healthy conditions to obtain a dimensionless relative percentage deviation, which facilitates cross-dimensional comparisons.

[0028] Inter-dimensional consistency analysis is performed based on a multi-dimensional deviation set to identify mismatched dimension combinations. This analysis examines whether the changes in deviations across dimensions within the multi-dimensional deviation set conform to physical constraints. During normal degradation, deviations in related dimensions should change in a coordinated manner. For example, a decrease in sensor sensitivity will simultaneously lead to an increase in both amplitude and signal-to-noise ratio deviations. A single-dimensional anomaly while other related dimensions remain normal violates physical laws. Dimension pairs with physical coupling relationships within the multi-dimensional deviation set constitute the objects of consistency testing. The testing method involves calculating the correlation coefficient of the deviations between the dimension pairs. A strong correlation indicates good consistency, meaning the deviations of the two dimensions change synchronously; a weak correlation or a negative correlation indicates poor consistency, meaning the deviations of the two dimensions change uncoordinatedly. Mismatched dimension combinations are included, representing dimension pairs that fail the consistency test. The deviation changes in these dimension pairs violate the coordinated relationships expected during normal degradation, suggesting a specific type of fault or anomaly rather than overall performance degradation. In a multidimensional deviation set, an increase in amplitude dimension deviation while the frequency dimension deviation remains unchanged indicates a deviation mismatch. This combination suggests a circuit fault in the drive coil or detection coil, rather than a fault in the vibrating tube itself. An increase in phase dimension deviation while the amplitude dimension deviation remains unchanged indicates another type of deviation mismatch, suggesting an abnormal signal channel delay rather than a sensor fault. Deviation mismatch dimension combinations are recorded in a pairwise format of dimension pair numbers and mismatch degrees. The degree of mismatch is quantified by the magnitude of the deviation of the correlation coefficient from the expected value; a larger mismatch degree indicates a more severe anomaly in that dimension pair. Consistency analysis is performed separately for each sampling period of the multidimensional deviation set. The deviation mismatch dimension combinations identified in different periods may differ, and the temporal evolution of the combinations reflects the development process of the fault.

[0029] A second deviation sequence is constructed by locating fault-related links based on the deviation mismatch dimension combination. Each dimension pair in the deviation mismatch dimension combination corresponds to a specific fault-related link hypothesis. The mapping relationship between the dimension pair and the fault-related link is determined by the physical structure and signal flow of the flowmeter, and the mapping relationship is pre-stored in the diagnostic module in the form of a knowledge base. Amplitude-frequency dimension combination mismatch points to a fault-related link in the sensor excitation circuit or detection circuit. A decrease in the efficiency of the drive coil will lead to a decrease in vibration amplitude but will not affect the resonant frequency of the vibrating tube. Phase-amplitude dimension combination mismatch points to a fault-related link in the signal conditioning circuit or analog-to-digital conversion channel. This link fault will introduce phase delay but will not change the signal amplitude. Frequency-harmonic dimension combination mismatch points to a fault-related link in the mechanical structure or fluid coupling of the vibrating tube. Scale or corrosion on the inner wall of the vibrating tube will affect both the resonant frequency and harmonic distribution. The fault-related link location results are attached as diagnostic tags to the corresponding records of the multi-dimensional deviation set. The tag content includes the name of the suspected fault-related link and the location confidence level. The confidence level is determined by the degree of mismatch in the deviation mismatch dimension combination and the number of dimension pairs. The second deviation sequence is constructed by integrating fault-related link location information based on the multidimensional deviation set. Each element in the sequence contains the multidimensional deviation value and associated fault label for that sampling period. The multidimensional deviation value and fault label together constitute a complete characteristic deviation description. When the deviation mismatch dimension combination is empty, it indicates that the deviation changes in each dimension are coordinated and conform to the normal degradation law. The fault label for that period in the second deviation sequence is set to normal degradation. The second deviation sequence organizes data in chronological order, with a sequence length equal to the number of monitoring periods, and completely records the multidimensional evolution process of the characteristic deviation and the fault evolution trajectory.

[0030] In some embodiments, the step of performing trend separation to generate a systematic drift component based on the first deviation sequence and the second deviation sequence includes: filtering the first deviation sequence and the second deviation sequence to obtain a smoothed deviation sequence; identifying drift direction features from the smoothed deviation sequence to generate a drift direction identifier; performing drift monotonicity verification on the drift direction identifier to generate a pseudo-drift marker; and separating the smoothed deviation sequence based on the pseudo-drift marker to obtain a systematic drift component.

[0031] The first and second deviation sequences are filtered to obtain a smoothed deviation sequence. High-frequency fluctuations in the first deviation sequence are suppressed using a low-pass filter. The filter employs either moving average or exponential smoothing methods. The window width is determined based on the fluctuation period and sampling interval; a window that is too narrow will result in insufficient filtering and residual noise, while a window that is too wide may obscure the details of the true trend. The filtered first deviation sequence retains low-frequency trend components while smoothing out high-frequency noise, resulting in a smooth curve that facilitates identification of the overall drift direction. The second deviation sequence, as a multidimensional sequence, is filtered separately for each dimension. The same filtering parameters are used for each dimension to maintain temporal synchronization. The smoothed sequences of each dimension are then recombined into a smoothed multidimensional sequence, maintaining the original multidimensional structure. The smoothed deviation sequence comprises two parts: the filtering results of the first and second deviation sequences. The former is a one-dimensional smoothed sequence representing the smoothing trend of the flow deviation, while the latter is a multidimensional smoothed sequence representing the smoothing trend of the deviations in each feature dimension. Both are organized using the same time index for convenient synchronous analysis.

[0032] Drift direction identifiers are generated by identifying drift direction features from a smoothed deviation sequence. These drift direction features describe the monotonic trend of the deviation value over time in the smoothed deviation sequence. A continuously increasing deviation value corresponds to positive drift, meaning the flowmeter measurement value gradually increases; a continuously decreasing deviation value corresponds to negative drift, meaning the flowmeter measurement value gradually decreases; and fluctuations around zero value correspond to no drift, meaning the flowmeter is stable. The smoothing result of the first deviation sequence in the smoothed deviation sequence is fitted with a trend line using linear regression. The sign and magnitude of the trend line slope characterize the drift direction and rate; a positive slope indicates that the deviation increases over time, and a negative slope indicates that the deviation decreases over time. Drift direction identifiers are generated based on the trend line slope. A positive slope exceeding a threshold indicates positive drift; a negative slope exceeding a threshold in absolute value indicates negative drift; and a slope below a threshold in absolute value indicates stable, drift-free flow. The threshold setting considers the range of measurement noise and normal fluctuations. In the smoothed deviation sequence, the direction of each dimension of the second deviation sequence is identified separately. Each dimension generates an independent drift direction identifier reflecting the direction of change in the characteristic deviation of that dimension. When the directions of most dimensions are consistent, the overall drift direction identifier taking the majority direction indicates a reliable drift judgment. When the dimensional directions are dispersed, the overall drift direction identifier is marked as uncertain and requires further analysis. In addition to the direction category, the drift direction identifier also includes a direction confidence level. The confidence level is determined by combining the results of the trend line goodness of fit and slope significance test. A high confidence level indicates a reliable drift direction judgment when the goodness of fit is high and the slope is significant, while a low confidence level indicates uncertainty in the drift direction when the goodness of fit is low or the slope is not significant.

[0033] A drift monotonicity check is performed on the drift direction indicator to generate a pseudo-drift marker. The normal degradation process of a flowmeter exhibits monotonicity. Fatigue aging of the vibrating tube material leads to a gradual decrease in the elastic modulus, a process that is irreversible. Sensor sensitivity typically decreases monotonically without spontaneous recovery, and circuit component parameters typically drift monotonically without reverse correction. This monotonicity is a crucial basis for distinguishing true degradation from external interference. The drift monotonicity check analyzes the temporal evolution of the drift direction indicator to verify whether the drift direction remains consistent within the monitoring period. The entire monitoring period is divided into several sub-periods, and the drift direction of each sub-period is determined and the consistency of the direction sequence is checked. When the drift direction indicator changes from positive to negative or vice versa during a certain period, it indicates that the drift during that period does not conform to the monotonicity rule. This may be a pseudo-drift caused by sudden changes in operating conditions, measurement anomalies, or external interference, rather than true degradation within the flowmeter. False drift markers record the time periods when the drift direction indicator reverses. The markers include the start and end times of the reversal, the direction before the reversal, and the direction after the reversal. Deviation data during the reversal period requires special processing when extracting systematic drift components to avoid false drift contaminating the estimation of true drift. Data corresponding to the false drift marker period in the smoothed deviation sequence is marked as suspicious data; deviation changes during this period are not directly included in the cumulative drift amount but are corrected or removed. When the drift direction indicator remains in the same direction throughout the monitoring period, the false drift markers are empty, indicating that the drift within the entire monitoring period follows a monotonic pattern. In this case, all drift data can be used for systematic drift component extraction, and the drift estimation has high reliability.

[0034] Systematic drift components are obtained by separating the smoothed deviation sequence based on pseudo-drift markers. Data from time periods not covered by pseudo-drift markers in the smoothed deviation sequence are retained as valid drift data; this data conforms to the degenerate monotonicity rule and can be directly used for drift component extraction. Data from time periods covered by markers are removed or corrected to prevent pseudo-drift components from being mixed into the systematic drift components. The correction method involves subtracting the deviation change during the reversal period from the cumulative drift, retaining only the drift increment conforming to the monotonicity rule. The subtraction operation is achieved by calculating the deviation difference between the start and end points of the reversal period and subtracting it from the total drift. Systematic drift components are extracted from the corrected smoothed deviation sequence by accumulating and summing or trend fitting the valid drift data. The accumulated result represents the total drift from the monitoring start point to each time point. The starting point of the drift is set to zero based on the initial state of the flowmeter when it is put into operation. The smoothed deviation sequence of the first deviation sequence generates a flow drift component reflecting the cumulative offset of the flow measurement value. The smoothed deviation sequence of the second deviation sequence generates characteristic drift components for each dimension, reflecting the cumulative offset of each signal feature. The systematic drift component, with the flow drift component as the main component and supplemented with characteristic drift information of key dimensions, provides a more comprehensive drift description. The existence period of pseudo-drift markers is used to fill data gaps in the systematic drift component through interpolation. The interpolation result is obtained by linear estimation based on the effective drift data before and after the marked period, maintaining the continuity of the drift curve. The final form of the systematic drift component is a drift time series, which records the cumulative drift of the flowmeter relative to its initial state at each moment.

[0035] Step S130: Cross-validate the first deviation sequence and the second deviation sequence to generate a valid drift marker. Based on the valid drift marker, perform bidirectional verification and matching between the systematic drift component and the preset fault mode library to obtain the suspected fault type.

[0036] In some embodiments, the step of cross-validating the first deviation sequence and the second deviation sequence to generate a valid drift marker includes: performing synchronization change detection on the first deviation sequence and the second deviation sequence to generate synchronization features; performing directional consistency analysis on the synchronization features to generate a drift source marker; determining the degradation source based on the drift source marker to obtain an internal degradation marker; and generating a valid drift marker based on the internal degradation marker.

[0037] Synchronization features are generated by detecting synchronous changes in the first and second deviation sequences. The synchronous change detection analyzes whether the changes in the first and second deviation sequences occur synchronously over time. The detection method calculates the ratio between the time difference and the magnitude of the change in the two sequences. A small time difference and a stable ratio indicate good synchronization, while a large time difference or fluctuating ratio indicates poor synchronization. The first deviation sequence, as a one-dimensional sequence, directly extracts its change time and magnitude. The change time is defined as the point where the slope of the deviation value exceeds a threshold, and the magnitude of the change is defined as the difference between the deviation values ​​before and after the change. The second deviation sequence, as a multi-dimensional sequence, selects the dimension with the strongest correlation to flow measurement for synchronization detection. The correlation is determined by statistical correlation coefficients of historical data; typically, the amplitude and phase dimensions have high correlations with flow measurement. The synchronization features record quantitative indicators of the synchronization degree between the first and second deviation sequences, including the average time difference of the change time and the correlation coefficient of the magnitude of the change. The average time difference reflects the time lag of the responses of the two sequences; a smaller time difference indicates better synchronization. The correlation coefficient of the magnitude of the change reflects the degree of coordination of the intensity of the changes in the two sequences; a coefficient closer to 1 indicates better coordination. Synchronization characteristics can be used to evaluate the entire monitoring period as a whole, or time-varying synchronization characteristic sequences can be calculated in time periods. Time-varying sequences can reveal the evolution of synchronicity over time.

[0038] Drift source markers are generated through directional consistency analysis using synchronization features. Directional consistency analysis examines whether the first and second deviation sequences change in the same direction. A positive correlation coefficient in the synchronization features indicates that the directions are consistent (both sequences increase or decrease simultaneously), a negative correlation coefficient indicates that the directions are opposite (one increases and the other decreases), and a correlation coefficient close to zero indicates that the directions are unrelated (the changes are independent). Directional consistency is a necessary condition for determining that drift originates from internal degradation of the flowmeter. A decrease in flowmeter sensor sensitivity will simultaneously cause flow measurement deviation and signal characteristic deviation to change in the same direction. Opposite or unrelated changes suggest that drift may originate from external interference rather than internal degradation. For example, electromagnetic interference or sudden changes in ambient temperature can cause short-term fluctuations in the first deviation sequence, but the mechanical characteristic dimension of the second deviation sequence remains unaffected. Drift source markers are generated based on the results of directional consistency analysis. When the directions are consistent and the time difference in the synchronization features is small, it is marked as internal drift; when the directions are inconsistent or the time difference is large, it is marked as external drift or an uncertain source. Drift source markers are applied separately to each monitoring period. Different periods may have different source markers. The period division is determined based on the time-varying characteristics of the synchronization features; continuous intervals with stable synchronization features are classified as the same period.

[0039] Internal degradation markers are obtained by identifying degradation sources based on drift source markers. The degradation sources for periods marked as having internal drift are further determined. These sources are categorized into three types: sensor degradation, circuit degradation, and mechanical structure degradation. Different degradation sources exhibit different deviation distribution patterns across the dimensions of the second deviation sequence. Sensor degradation primarily affects the amplitude and sensitivity-related dimensions; insulation aging of the detection coil leads to a gradual decrease in the output signal amplitude. Circuit degradation primarily affects the phase and noise-related dimensions; component parameter drift in the signal conditioning circuit introduces additional phase delay. Mechanical structure degradation primarily affects the frequency and harmonic-related dimensions; scaling on the inner wall of the vibrating tube increases the added mass, causing a decrease in the resonant frequency. Degradation source determination is based on matching the most prominent deviation dimension combinations in the second deviation sequence with the aforementioned patterns. Internal degradation markers record the degradation source determination results for each period, including the degradation source category and determination confidence level. The confidence level is determined by the degree of matching between the deviation distribution pattern and the standard pattern; a higher matching degree results in a higher confidence level. Periods marked as having external drift sources are not subject to degradation source determination; these periods are labeled as non-internal degradation in the internal degradation marker. Periods marked as having uncertain drift sources are treated conservatively; these periods are marked as pending in the internal degradation marker, and will be determined after more data is acquired.

[0040] Valid drift markers are generated based on internal degradation markers. Periods marked by internal degradation (sensor degradation, circuit degradation, or mechanical structure degradation) are considered valid drift periods. The systematic drift component data within these periods reflects the true performance degradation of the flowmeter and can be used for fault diagnosis. Valid drift markers assign validity labels to each period. Validity label values ​​include valid, invalid, and pending. Periods with clearly defined internal degradation markers are labeled valid, those without internal degradation markers are labeled invalid, and those with pending markers are labeled pending. In addition to validity labels, valid drift markers also include a validity score. The score is calculated by combining the synchronization degree of synchronization features, the degree of directional consistency, and the confidence level of the internal degradation marker. The score ranges from 0 to 1, with higher scores indicating more reliable drift validity for that period. Periods with valid validity labels and high scores are prioritized for fault mode matching. Periods with valid validity labels but low scores are used as auxiliary references. Periods with invalid or pending validity labels are excluded or downweighted during fault matching. Valid drift markers are organized chronologically to form a marker sequence. The sequence length is equal to the number of monitoring periods. The validity labels and scores for each period constitute the complete marker information for that period.

[0041] In some embodiments, the step of performing bidirectional verification matching between the systematic drift component and a preset fault mode library based on the effective drift marker to obtain a suspected fault type includes: filtering systematic drift components to be matched according to the effective drift marker; performing forward similarity matching between the systematic drift component and the preset fault mode library to generate a candidate fault ranking; performing reverse compatibility verification on the candidate fault ranking to establish a filtered fault set; and selecting the one with the highest similarity from the filtered fault set to determine it as a suspected fault type.

[0042] Systematic drift components to be matched are filtered based on valid drift markers. Data for systematic drift components with valid validity labels are extracted and used as matching data; data for time periods with invalid validity labels are removed from the matching set to avoid spurious drifts interfering with the matching results. The validity score in the valid drift markers is used to weight the matching data; time periods with higher scores are given greater weight in the matching calculation, while those with lower scores are given less weight. This weighting mechanism ensures that the matching results rely more on high-confidence drift data. The drift amount time series of the systematic drift components is truncated to the valid time period to form the drift curve to be matched. The drift direction and drift rate are recalculated from the drift curve to ensure that the feature values ​​only reflect the characteristics of valid drifts. The systematic drift components to be matched are compiled into a data packet containing the drift curve, drift direction, drift rate, and a summary of associated second deviation sequence features. This data packet serves as input for fault mode matching. The amount of data to be matched after filtering may be less than the amount of data of the original systematic drift components. When the amount of data is too small, the reliability of fault matching decreases. In this case, the filtering conditions of effective drift markers can be relaxed or the matching results can be labeled as low confidence.

[0043] Systematic drift components are positively similar to a pre-defined fault mode library to generate a candidate fault ranking. Positive similarity matching calculates the similarity between the systematic drift component to be matched and each fault mode in the fault mode library. The similarity is calculated using a multi-dimensional distance metric, including drift direction, drift rate, drift curve shape, and associated feature anomaly combinations. The drift direction dimension checks whether the positive or negative direction of the drift to be matched matches the direction specified by the fault mode; matching directions earn full marks, while opposite directions earn zero marks. The drift rate dimension calculates the degree to which the drift rate to be matched conforms to the rate range specified by the fault mode; rates falling within the range earn full marks, and the greater the deviation, the lower the score. The drift curve shape dimension uses a dynamic time warping algorithm to calculate the shape similarity between the curve to be matched and the standard curve of the fault mode. The dynamic time warping algorithm uses non-linear time alignment to handle local differences in drift rate, thus accurately assessing the similarity of the curve shape. The associated feature anomaly combination dimension checks whether the deviation patterns in the second deviation sequence match the feature anomaly combinations specified by the fault mode; the more matching dimensions, the higher the score. The comprehensive similarity score is obtained by weighted summation of scores from each dimension. The weights are set according to the importance of each dimension to fault diagnosis. Drift direction and abnormal correlation features are usually given higher weights. The weights are determined by statistical analysis of historical fault cases. Candidate faults are ranked according to the comprehensive similarity score from high to low. Fault types with scores exceeding the threshold are included in the candidate list, while fault types with scores below the threshold are excluded from the candidate list.

[0044] For example, the step of performing reverse compatibility verification on the candidate fault ranking to establish a filtered fault set includes: extracting physical feature constraints corresponding to each candidate fault from the candidate fault ranking; performing fault mutual exclusion rule matching on the physical feature constraints to generate mutually exclusive fault pairs; identifying coexistence conflicts between candidate faults based on the mutually exclusive fault pairs to generate conflict markers; and performing compatibility screening on the candidate fault ranking according to the conflict markers to form a filtered fault set.

[0045] Physical characteristic constraints corresponding to each candidate fault are extracted from the candidate fault ranking. Each candidate fault in the ranking has a corresponding physical characteristic constraint record in the fault mode library. The constraint record describes the physical conditions and signal characteristic requirements that must be met when the fault occurs. The content of the constraints varies depending on the fault type. Physical characteristic constraints for sensor sensitivity degradation faults include negative drift direction, negative amplitude dimension deviation, and frequency dimension deviation close to zero. Aging of the detection coil insulation causes output signal attenuation but does not affect the mechanical characteristics of the vibrating tube. Physical characteristic constraints for circuit gain drift faults include drift rate being temperature-dependent and small changes in phase dimension deviation. Physical characteristic constraints for vibrating tube corrosion faults include negative frequency dimension deviation, increased harmonic dimension deviation, and irreversible drift. Corrosion causes thinning of the tube wall, leading to a decrease in the stiffness of the vibrating tube and a reduction in the resonant frequency. The physical characteristic constraints of each candidate fault are retrieved from the fault mode library. The extraction results are organized in the form of a constraint list. Each entry in the list describes a specific constraint and its allowed value range. For example, the constraint list for sensor sensitivity degradation faults includes three entries: "drift direction = negative", "amplitude deviation range = [-15%, -3%]", and "frequency deviation range = [-0.5%, +0.5%]". The completeness and accuracy of the physical characteristic constraints depend on the quality of the fault mode library. The library should cover all types of faults that the flowmeter may experience and accurately describe the physical constraints of each fault.

[0046] The system generates mutually exclusive fault pairs by matching fault mutual exclusion rules to physical feature constraints. A mutually exclusive fault pair is a combination of two faults that cannot occur simultaneously due to contradictory physical feature constraints. When one fault requires a certain feature to be positive while another fault requires the same feature to be negative, they are logically mutually exclusive. When contradictory constraints exist in the physical feature constraints, the corresponding two fault types constitute a mutually exclusive fault pair. For example, a sensor sensitivity decrease fault requires a negative drift direction, while a zero-point positive drift fault requires a positive drift direction. Since their constraints on the drift direction are opposite, sensor sensitivity decrease and zero-point positive drift constitute a mutually exclusive fault pair. Similarly, a vibrating tube corrosion fault requires a negative frequency dimension deviation (i.e., a decrease in resonant frequency), while a vibrating tube scaling fault also requires a negative frequency dimension deviation. Since their constraints on frequency deviation are consistent, they are not mutually exclusive. The fault mutual exclusion rule base predefines the mutual exclusion relationships between various types of faults. The rule base is organized in the form of fault pairs and mutually exclusive causes. Physical feature constraints are matched with the mutual exclusion rule base to identify mutually exclusive pairs among candidate faults. Mutually exclusive fault pairs are paired with fault types that are mutually exclusive in the candidate fault ranking. The pairing information includes the type numbers of the two faults and a description of the mutual exclusion reason, such as recording "sensor sensitivity decrease - zero point positive drift, mutual exclusion reason: drift direction constraint conflict". When there is no mutual exclusion relationship in the candidate fault ranking, the mutually exclusive fault pair is an empty set, indicating that the candidate faults can coexist or occur independently physically; when there is a mutual exclusion relationship, the mutually exclusive fault pair is not empty, and further analysis is needed to determine which fault to keep and which to exclude.

[0047] Conflict markers are generated based on the coexistence conflict identification of candidate faults using mutually exclusive fault pairs. The two faults in a mutually exclusive fault pair cannot both be output as diagnostic conclusions simultaneously; one must be selected and retained based on similarity scores and the degree of satisfaction of physical constraints. Coexistence conflict identification analyzes each pair of mutually exclusive fault pairs, comparing the degree of agreement between the two faults and the observed data. The fault with a higher degree of agreement is identified as the dominant fault and retained, while the fault with a lower degree of agreement is identified as the inferior fault and marked. Conflict markers record the conflict analysis results for each mutually exclusive fault pair, including the mutually exclusive pair number, dominant fault, inferior fault, and the basis for judgment. The comparison of agreement is based on the comprehensive similarity score in the candidate fault ranking; the fault with the higher score is the dominant fault. For example, in the mutually exclusive fault pair "sensor sensitivity decrease - positive zero-point drift," if the similarity score for sensor sensitivity decrease is 0.82 and the score for positive zero-point drift is 0.65, then sensor sensitivity decrease is identified as the dominant fault, and positive zero-point drift is identified as the inferior fault. When the scores are close, the degree of satisfaction of physical constraints is further compared; the fault with the higher degree of constraint satisfaction is identified as the dominant fault. Candidate faults marked as inferior in the conflict flag will be excluded in subsequent screening, while candidate faults marked as superior will remain in the candidate list. When there are multiple mutually exclusive fault pairs, chain conflicts may occur, i.e., A and B are mutually exclusive and B and C are mutually exclusive. The handling of chain conflicts follows the transitivity principle: if A is superior to B and B is superior to C, then A is ultimately retained and B and C are excluded.

[0048] Based on conflict markers, candidate fault ranking is performed for compatibility screening to form a filtered fault set. Candidate faults marked as inferior in the conflict markers are removed from the candidate fault ranking. This removal operation is performed one by one according to the inferior fault field of the conflict markers, ensuring that there are no mutually exclusive conflicts in the removed candidate list. Candidate faults not involved by any conflict markers are directly retained in the list; these faults do not have a mutual exclusion relationship with other candidate faults and can be considered in parallel. After compatibility screening, a filtered fault set is formed. Fault types in the set do not have mutual exclusion conflicts and can physically coexist or occur independently. For example, if the original candidate fault ranking included three candidates—sensor sensitivity decrease, positive zero-point drift, and circuit gain drift—positive zero-point drift is identified as an inferior fault by the conflict markers and removed. The filtered fault set retains only sensor sensitivity decrease and circuit gain drift. The ranking order of fault types in the filtered fault set inherits from the original order of the candidate fault ranking; only fault types marked as inferior due to conflicts are removed, while the relative ranking of the faults remains unchanged. When the set of mutually exclusive fault pairs is empty, compatibility screening does not eliminate any candidate faults. After screening, the fault set is the same as the candidate list of ranked candidate faults. In this case, reverse verification mainly relies on the direct verification of physical feature constraints rather than screening by mutual exclusion rules.

[0049] The fault type with the highest similarity score from the filtered fault set is identified as the suspected fault type. All fault types in the filtered fault set have passed reverse compatibility verification, and each can potentially explain the observed drift phenomenon from a physical constraint perspective. Further optimization based on similarity scores is required. The similarity score is inherited from the comprehensive similarity calculation result of the forward matching stage. The fault type with the highest similarity score in the filtered fault set is determined as the most likely fault type, i.e., the suspected fault type. For example, if the filtered fault set includes two fault types, sensor sensitivity decrease and circuit gain drift, with similarity scores of 0.82 and 0.71 respectively, then sensor sensitivity decrease is identified as the suspected fault type due to its highest score. The suspected fault type represents the optimal fault judgment made by the current diagnostic system based on existing data, and this judgment has the highest data support and physical rationality. When multiple fault types with similarity scores are found in the filtered fault set, the fault types with the highest scores are all listed as suspected fault types. The diagnostic conclusion is output as a list of fault types rather than a single type. For example, if a sensor sensitivity decrease score is 0.82 and a circuit gain drift score is 0.79, and the difference between the two is less than the proximity threshold of 0.05, then the suspected fault type output will be a list containing both. The result of determining the suspected fault type includes the fault category name, which is used to query the corresponding fault-specific decay curve for health measurement. When the filtered fault set is empty, it means that all candidate faults failed the reverse verification, and the suspected fault types are marked as unknown fault types, indicating that there may be new faults not covered by the pattern library or data quality issues that caused the matching failure.

[0050] Step S140: Perform a rate of change analysis on the systematic drift component to obtain the drift acceleration, and perform health quantification based on the suspected fault type and drift acceleration to generate the flow meter health index and parameter drift amount.

[0051] In some embodiments, performing rate of change analysis on the systematic drift component to obtain drift acceleration includes: extracting adjacent period increments of the systematic drift component along the time axis to obtain a drift increment sequence; performing sign monitoring on the drift increment sequence to identify drift direction inflection points; identifying degradation stage boundaries based on the drift direction inflection points to establish degradation stage markers; and combining the degradation stage markers with the drift increment sequence to output drift acceleration.

[0052] The drift increment sequence is obtained by extracting the incremental values ​​of the systematic drift components along the time axis from adjacent periods. The systematic drift components are stored in discrete time series form. Each element in the sequence corresponds to the cumulative drift value of one sampling period. The difference in drift values ​​between adjacent periods is the drift increment within that period, and the increment value reflects the change in drift value within a single period. The formula for calculating the drift increment sequence is ΔD_i = D_i - D_(i-1), where D_i is the cumulative drift value of the i-th period, D_(i-1) is the cumulative drift value of the (i-1)-th period, and ΔD_i is the drift increment of the i-th period. The sign of the increment value reflects the direction of change of drift value within that period; a positive value indicates an increase in drift value, and a negative value indicates a decrease in drift value. The increment extraction is performed sequentially along the time axis from front to back. The first sampling period has no previous period reference, and its increment value is set to zero or filled with the initial drift rate estimate. The increment values ​​of subsequent periods are calculated sequentially to form a complete drift increment sequence. The length of the drift increment sequence is one less than the length of the systematic drift component. The elements in the sequence are arranged chronologically to form a discrete sampled representation of the drift velocity. The absolute value of the increment in the drift increment sequence reflects the severity of the drift within that period; a large absolute value indicates a rapid drift change and active degradation process, while a small absolute value indicates a gradual drift change and slow degradation process. In addition to the increment value, the drift increment sequence also records the corresponding timestamp information. The timestamps are stored in pairs with the increment values ​​for time alignment and correlation analysis with degradation stage markers.

[0053] For example, the step of detecting and identifying the drift direction inflection point by symbol monitoring of the drift increment sequence includes: extracting symbols from the drift increment sequence to generate an incremental symbol sequence; statistically analyzing consecutive identical sign segments in the incremental symbol sequence to generate a degenerate monotonic interval; locating symbol reversal points based on the boundaries of the degenerate monotonic interval to establish a candidate set of inflection points; and selecting effective transition points from the candidate set of inflection points to determine them as drift direction inflection points.

[0054] Symbol extraction is performed on the drift increment sequence to generate an increment symbol sequence. Each increment value in the drift increment sequence is classified into three categories: positive, negative, and zero. A positive value is marked as +1, indicating an increase in drift during that period; a negative value is marked as -1, indicating a decrease in drift during that period; and a value with an absolute value below the dead zone threshold is marked as 0, considered as uncertain and in a zero-drift state. Symbol extraction is performed on each element of the drift increment sequence sequentially, with the determination made by comparing it to the dead zone threshold. An increment value greater than the positive threshold is determined as +1, an increment value less than the negative threshold is determined as -1, and an increment value between the positive and negative thresholds is determined as 0. The determination results are arranged in the original sequence order to form the increment symbol sequence, with a sequence length equal to the drift increment sequence. Each element takes three discrete values: {-1, 0, +1}. The dead zone threshold is set by comprehensively considering the measurement noise level and drift detection sensitivity requirements; the threshold is usually set as a certain proportion of the standard deviation of the drift increment sequence to balance noise suppression and signal preservation. Incremental symbol sequences transform continuous incremental numerical sequences into discrete symbol sequences. The pattern structure of symbol sequences is easier to analyze and identify than that of numerical sequences. Segments with consecutive identical symbols correspond to monotonically changing time periods, and the positions where symbols change correspond to candidate points where the drift direction turns.

[0055] Degenerate monotonic intervals are generated by statistically analyzing consecutive identical-sign segments in the incremental symbol sequence. A consecutive identical-sign segment is defined as a sequence of consecutive elements in the incremental symbol sequence whose signs remain consistent. All elements within the segment have signs of +1, -1, or 0. The segment boundary is defined by the position where the sign changes, and the segment length is the number of consecutive elements contained within the segment. Segment statistics traverse the incremental symbol sequence to identify all consecutive identical-sign segments. Starting from the beginning of the sequence, each element is scanned element by element. If the sign of the current element is different from the previous element, the segment boundary is marked. After scanning, the start and end positions and sign type information of all segments are obtained. Degenerate monotonic intervals consist of consecutive identical-sign segments with signs of +1 or -1. Segments with signs of +1 correspond to time intervals where the drift increases monotonically, i.e., intervals where the bias continues to worsen. Segments with signs of -1 correspond to time intervals where the drift decreases monotonically, i.e., intervals where the bias continues to improve. Continuous segments with a sign of 0 are marked as stagnant segments in the degenerate monotonic intervals. Within the stagnant segments, the drift remains essentially constant, and the flowmeter's state is relatively stable. Stagnant segments typically appear during the transition period of the degradation phase or during the stable period when degradation tends to stabilize. After recording the segment length, start and end times, and sign type of each degenerate monotonic interval, the interval boundary positions are extracted to form a boundary position set for locating the sign reversal point.

[0056] A candidate set of inflection points is established based on the boundaries of degenerate monotonic intervals to locate sign reversal points. When adjacent intervals within a degenerate monotonic interval have different sign types, the boundary between the two intervals is the sign reversal point. At the reversal point, the sign of the incremental sign sequence undergoes a transpolar change, either from positive to negative or from negative to positive. The sign reversal point location process traverses the list of degenerate monotonic intervals, checking the sign type pairings of adjacent intervals. A reversal point is identified as a positive-to-negative reversal point when the preceding interval's sign is +1 and the following interval's sign is -1, indicating a shift from deterioration to improvement. Conversely, a reversal point is identified as a negative-to-positive reversal point when the preceding interval's sign is -1 and the following interval's sign is +1, indicating a shift from improvement to deterioration. The candidate set of inflection points includes all identified sign reversal points. Each element in the candidate set records four pieces of information: the time position of the reversal point, the reversal type, the length of the interval before the reversal, and the length of the interval after the reversal. The interval length information is used to subsequently determine the validity of the reversal point. Adjacent intervals with the same sign do not constitute a reversal point. For example, if two intervals with +1 signs are separated by an interval with 0 signs, the separating position is not a turning point in the drift direction, but merely a point of stagnation or discontinuity; the drift direction has not fundamentally changed. Some reversal points in the candidate set of turning points may be caused by measurement noise rather than a true change in drift direction, especially when the intervals before and after the reversal point are very short. These short intervals are often formed by random fluctuations, and the corresponding reversal points are false reversal points that need to be eliminated during the screening process.

[0057] Valid transition points are selected from the candidate set of transition points to determine the drift direction transition points. The selection of valid transition points is based on a comprehensive judgment of the length characteristics of the interval before and after the transition point and the amplitude characteristics of the increment value. A true drift direction transition point should be accompanied by a relatively long monotonic interval and a significant change in the increment amplitude. These two conditions together ensure that the transition point reflects a true degradation direction transition rather than noise fluctuations. A transition point is considered valid when the length of both the interval before and after the transition point exceeds the minimum interval threshold. If the length of either interval is below the threshold, it is considered a suspicious transition point requiring further verification. The minimum interval threshold is set based on the sampling period and typical degradation time constant to ensure that the interval length is sufficient to reflect the true degradation trend. Further verification of suspicious transition points examines the amplitude change characteristics of the drift increment sequence at the transition point. If the absolute values ​​of the increments before and after the transition are both large and in opposite directions, it is confirmed as a valid transition point, indicating a true reversal of the drift direction. If the absolute values ​​of the increments before and after the transition are very small and close to the dead zone, it is excluded as an invalid transition point, indicating that the sign change is merely random fluctuation of noise near zero. Drift direction transition points consist of the selected valid transition points; invalid transition points are removed from the candidate set of transition points and not included in the final result. The drift direction turning points are arranged in chronological order. The time position, flip type, and confidence information of each turning point constitute a complete description of the turning point. The confidence level is determined by a comprehensive score of the length of the preceding and following intervals and the increment magnitude. The longer the interval and the larger the magnitude, the higher the confidence level, indicating that the turning point is more reliable.

[0058] Degradation stage boundary identification is performed based on drift direction inflection points to establish degradation stage markers. Drift direction inflection points divide the entire monitoring period into several monotonically changing segments. Within each segment, the drift amount continuously changes in the same direction, exhibiting monotonically increasing or decreasing characteristics. The start and end boundaries of each segment are determined by the time position of adjacent transition points; the starting point of the first segment is the monitoring start time, and the ending point of the last segment is the current monitoring time. Degradation stage boundary identification analyzes the drift increment characteristics of each segment to determine the corresponding degradation stage type. S130 classifies degradation into three categories based on the source of degradation: sensor degradation, circuit degradation, and mechanical structure degradation. This step classifies the degradation process into three stages based on the degradation rate change characteristics: accelerated degradation, uniform degradation, and decelerated degradation. These two classifications describe degradation characteristics from different perspectives. Segments with positive drift increments and gradually increasing values ​​are classified as accelerated degradation stages, indicating that the degradation process is intensifying. Segments with positive drift increments but relatively stable values ​​are classified as uniform degradation stages, indicating that the degradation process is maintaining a constant rate. Segments where the drift increment changes from positive to negative or gradually decreases are classified as decelerating degradation stages, indicating that the degradation process is slowing down. Degradation stage markers record the boundary times and stage types of each degradation stage. The marker content includes three pieces of information: stage number, start and end times, and stage type. The stages are arranged in chronological order to form a complete degradation stage sequence. The boundary moments where adjacent stage types differ are stage transition points. Degradation stage markers highlight the location of these transition points and the change in stage type before and after the transition. Stage transition points are key nodes in the degradation process, reflecting important turning points in the degradation dynamics.

[0059] The drift acceleration is output by combining the degradation stage markers and the drift increment sequence. The drift acceleration is calculated by performing a quadratic difference operation on the drift increment sequence, with the formula a_i = (ΔD_i - ΔD_(i-1)) / Δt, where ΔD_i and ΔD_(i-1) are the drift increments of adjacent periods, Δt is the sampling period interval, and a_i is the drift acceleration of the i-th period. A positive acceleration value indicates that the drift speed is accelerating and the degradation trend is strengthening, while a negative value indicates that the drift speed is slowing down and the degradation trend is weakening. The degradation stage markers are used to segment and analyze the drift acceleration calculation results. The acceleration within different degradation stages is statistically analyzed for its mean and extreme value range. A positive mean acceleration in the accelerating degradation stage reflects the continued intensification of degradation, a negative mean acceleration in the decelerating degradation stage reflects the gradual slowing down of degradation, and a mean acceleration close to zero in the uniform degradation stage reflects that the degradation speed is basically constant. The drift acceleration output includes an acceleration time series, which records the instantaneous acceleration values ​​for each sampling period to form an acceleration evolution curve. Drift acceleration at the stage boundaries in the degradation stage markers often exhibits extreme values ​​or abrupt changes. These characteristic points reflect key turning points in the degradation process and are the focus of degradation dynamic analysis.

[0060] In some embodiments, the step of generating a flowmeter health index and parameter drift by quantifying health based on the suspected fault type and the drift acceleration includes: querying the corresponding fault-specific decay curve according to the suspected fault type; extracting the acceleration fluctuation characteristics by continuously periodically fluctuating the drift acceleration; performing degradation stability analysis based on the acceleration fluctuation characteristics and the drift acceleration to generate a degradation risk level; and outputting the flowmeter health index and parameter drift by combining the fault-specific decay curve and the degradation risk level.

[0061] Based on the suspected fault type, the corresponding fault-specific decay curve is queried. Different types of faults lead to different patterns of flowmeter health status decay. Fault-specific decay curves are pre-established in the fault mode library, describing the typical decay trajectory of the health index as a function of drift or time under each fault type. The decay curves are based on statistical analysis of historical fault cases and have empirical basis. The fault-specific decay curve corresponding to sensor sensitivity reduction faults shows an S-shaped characteristic of slow initial decay followed by accelerated decay. In the early stages of aging of the detection coil insulation material, the impact on the signal is small, and the health index decays slowly. After aging accumulates to a certain extent, the insulation performance deteriorates rapidly, leading to accelerated decay of the health index. The fault-specific decay curve corresponding to circuit gain drift faults shows an approximately linear decay characteristic. The drift of electronic component parameters is basically proportional to time, and the health index decreases at a uniform rate. The fault-specific decay curve corresponding to vibrating tube mechanical damage faults shows a step decay characteristic. When sudden damage such as cracks occurs in the vibrating tube, the health index drops sharply, forming a cliff-like decline. After the damage stabilizes, the health index remains at a low level and no longer decreases further. The suspected fault type is used as the query key to retrieve the corresponding fault-specific attenuation curve from the fault mode library. If the suspected fault type output by S130 is a single type, the corresponding curve is directly queried; if there are multiple candidate types, the curve corresponding to the type with the highest similarity is selected. The query results are returned in the form of mathematical function expressions or discrete data points. The function parameters or data points are obtained by fitting historical monitoring data of similar faults. When the suspected fault type is unknown, a general attenuation curve is used for conservative estimation. The general curve is taken as the envelope of the attenuation curves of various faults to ensure that the estimation result is not too optimistic.

[0062] Acceleration fluctuation characteristics are obtained by extracting the amplitude of continuous periodic fluctuations in drift acceleration. Drift acceleration, as a time series, exhibits different values ​​in each sampling period. The amplitude of change between adjacent periods reflects the stability of acceleration; a small amplitude indicates a stable and predictable acceleration degradation process, while a large amplitude indicates an unstable and unpredictable acceleration degradation process. The amplitude extraction calculates the absolute value of the difference between adjacent elements in the drift acceleration sequence. The formula for the difference is Δa_i = |a_i - a_(i-1)|, where a_i is the drift acceleration in the i-th period, a_(i-1) is the drift acceleration in the (i-1)-th period, and Δa_i is the amplitude of acceleration fluctuation in the i-th period. The absolute value difference sequence reflects the periodic fluctuation of acceleration, and its length is one less than that of the drift acceleration sequence. Acceleration fluctuation characteristics utilize statistical analysis of the absolute value sequence of differences to obtain an overall description of the fluctuation characteristics. Statistical indicators include the mean fluctuation amplitude reflecting the average fluctuation level, the maximum value reflecting the maximum instantaneous fluctuation, and the standard deviation reflecting the dispersion of the fluctuation. The calculation of the mean and standard deviation covers the entire monitoring period to obtain a comprehensive assessment of the fluctuation. The core output of acceleration fluctuation characteristics is the fluctuation degree judgment result. The fluctuation degree is determined based on a comprehensive evaluation of statistical indicators. When the mean fluctuation amplitude is lower than a preset mean threshold and the standard deviation is lower than a preset dispersion threshold, the fluctuation degree is judged as low. When the mean fluctuation amplitude exceeds the preset mean threshold or the standard deviation exceeds the preset dispersion threshold, the fluctuation degree is judged as high. The preset thresholds are determined statistically based on historical operating data of similar flowmeters to reflect individualized characteristics. The fluctuation degree judgment result is output in a high / low binary label format. A high label indicates an unstable degradation process requiring close monitoring, while a low label indicates a stable degradation process that can be monitored routinely.

[0063] Degradation stability analysis is performed based on acceleration fluctuation characteristics combined with drift acceleration to generate a degradation risk level. Degradation stability reflects the predictability and controllability of the flowmeter degradation process. High stability indicates a clear and predictable degradation process, allowing for planned maintenance strategies. Low stability indicates a volatile and unpredictable degradation process, requiring enhanced real-time monitoring and emergency response. Degradation stability analysis combines the mean sign of drift acceleration and the degree of fluctuation in acceleration fluctuation characteristics. A negative or near-zero mean drift acceleration indicates a slowing or stable degradation trend, a benign state. Low fluctuation in acceleration fluctuation characteristics indicates a smooth and clear degradation process; both conditions are met for high stability. A positive mean drift acceleration indicates an increasing degradation trend, a deteriorating state. Low fluctuation in acceleration fluctuation characteristics indicates deterioration, but the change pattern is clear and predictable; the combination of both conditions indicates medium stability. A positive mean drift acceleration and high volatility indicate accelerated degradation and instability. This is common in flow meters operating under harsh conditions for extended periods, representing the worst-case scenario characterized by both deterioration and unpredictability, and is classified as low stability. Degradation risk levels are graded based on degradation stability: High stability corresponds to degradation risk level A, indicating controllable degradation with low risk, allowing for routine maintenance; medium stability corresponds to degradation risk level B, indicating generally controllable degradation but requiring monitoring of trends and appropriate strengthening of monitoring; and low stability corresponds to degradation risk level C, indicating uncontrollable degradation requiring close monitoring and emergency preparedness. The degradation risk level is also dynamically adjusted based on the current value and recent trend of drift acceleration. A larger current value and an upward trend indicate a lower degradation risk level, reflecting increased risk; conversely, a smaller current value and a downward trend indicate a higher degradation risk level, reflecting decreased risk.

[0064] The flowmeter health index and parameter drift are output by combining the fault-specific decay curve and the degradation risk level. The base score of the flowmeter health index is obtained from the fault-specific decay curve based on the current drift. The fault-specific decay curve uses drift as the independent variable and the health index as the dependent variable. The current cumulative value of the systematic drift component is input as the independent variable to query and output the corresponding base value of the health index. When the flowmeter is put into operation, the drift is zero, corresponding to an initial health index value of 100, indicating a brand-new state. The base score is adjusted according to the degradation risk level. At degradation risk level A, the base score remains unchanged, reflecting a good state. At degradation risk level B, the base score is appropriately lowered, reflecting a certain risk that needs attention. At degradation risk level C, the base score is further lowered, reflecting a high risk that requires close attention. The adjustment range is determined by combining the level and the drift acceleration value, reflecting a dynamic adjustment mechanism. The adjusted score, after rounding and boundary constraint processing, is the final output value of the flowmeter health index, ranging from 0 to 100. The higher the value, the better the health status and the lower the maintenance urgency. The parameter drift is taken as the latest value of the systematic drift component time series. This value represents the systematic offset of the flowmeter's current measurement value relative to the initial calibration value. The offset can be positive or negative, indicating either an overestimation or underestimation of the measurement. The sign of the parameter drift indicates the direction of the offset: a positive value indicates an overestimation (actual flow rate is less than the measured value), and a negative value indicates an underestimation (actual flow rate is greater than the measured value). The absolute value of the offset indicates the magnitude of the offset. The flowmeter health index and parameter drift together constitute the core output of health measurement. The flowmeter health index is used for health status assessment, maintenance decisions, and risk warnings, while the parameter drift is used for online compensation and correction of measurement results and accuracy improvement.

[0065] Step S150: Based on the parameter drift, adaptive dynamic compensation is performed on the actual measured flow rate value, and the compensated flow rate value is output. The flow rate monitoring result is formed based on the flow rate health index and the compensated flow rate value.

[0066] Specifically, the system adaptively and dynamically compensates for the actual measured flow rate based on the parameter drift, outputting the compensated flow rate value. The parameter drift reflects the systematic measurement deviation currently present in the flowmeter. Adaptive dynamic compensation improves the accuracy of the measurement results by correcting this deviation. For applications requiring high measurement accuracy, such as trade settlement, the compensation function can effectively reduce measurement errors caused by flowmeter drift. The basic formula for compensation calculation is Q_compensated = Q_measured - D_drift, where Q_measured is the actual measured flow rate value for the current period, D_drift is the current parameter drift, and the sign of the parameter drift is consistent with the deviation direction defined in S120 (a positive value indicates that the actual measurement is too large). Q_compensated is the compensated flow rate value. The adaptive characteristic of adaptive dynamic compensation is reflected in the dynamic update of the parameter drift during the monitoring process. The compensation amount for each sampling period is determined based on the latest estimated drift amount. When the drift amount changes, the compensation amount is adjusted synchronously without manual intervention, tracking drift changes and achieving automatic calibration. Adaptive dynamic compensation also considers the time characteristics of drift changes. When the drift acceleration is large, the compensation amount adopts a predictive correction mechanism. The predicted correction amount is the sum of the product of the current drift acceleration and the predicted time step, added to the current drift amount, improving the tracking ability for rapidly changing drifts. After the compensation calculation is completed, the flow rate value undergoes a reasonableness check. The compensated value should fall within the physically reasonable flow range, and the compensation amount should not exceed a certain proportion of the actual measured value. Compensation results exceeding the reasonable range are judged as abnormal, triggering an alarm and reverting to the uncompensated value output. When the flow meter health index is too low, the estimation of parameter drift may be inaccurate. When the health index is below the set threshold, a low confidence mark is added to the compensated flow rate value to prompt the user to use it with caution.

[0067] The flowmeter monitoring results are generated based on the flowmeter health index and the compensated flow rate. The flowmeter monitoring results summarize three aspects: flow measurement, health assessment, and maintenance recommendations, providing users with complete information about the flowmeter's current status. The flow measurement section records the compensated flow rate and its compensation reliability, while retaining the actual measured flow rate as raw data for reference. The difference between the two is the compensation amount executed in this cycle. The compensation reliability is determined by a combination of the flowmeter health index and the relative magnitude of the compensation amount; a high health index and a small compensation amount result in high compensation reliability, while a low health index or a large compensation amount results in lower compensation reliability. The health assessment section records the flowmeter health index and its changing trends. The suspected fault types output by S130 and the degradation risk levels output by S140 are also included in the health assessment section. The suspected fault types indicate the possible types of faults, and the degradation risk levels indicate the predictability and risk level of the degradation process. The maintenance recommendations section generates differentiated maintenance operation guidelines based on the flowmeter's health index values. When the health index is above 80, it is recommended to continue normal use and monitor according to the regular cycle. When the health index is between 60 and 80, it is recommended to shorten the monitoring cycle, monitor degradation trends, and assess whether early maintenance is necessary. When the health index is between 40 and 60, it is recommended to schedule planned maintenance or calibration to restore measurement accuracy. When the health index is below 40, it is recommended to replace or send the flowmeter for repair as soon as possible to avoid serious measurement inaccuracies. These thresholds can be adjusted according to the accuracy requirements of the flowmeter type and application scenario. Flowmeter monitoring results are output in a structured report format, including four standard columns: timestamp, measurement data, health assessment, and maintenance recommendations. Flowmeter monitoring results support both local display and remote transmission. Flowmeter monitoring results are continuously output according to the monitoring cycle to form a monitoring record sequence. The record sequence is stored in a database to support historical queries and statistical analysis. Long-term accumulated monitoring records can be used to analyze the degradation patterns of the flowmeter and predict its remaining service life, providing data support for the formulation of predictive maintenance strategies.

[0068] To implement the digital twin-based flow meter self-diagnosis method corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects. See also Figure 3 , Figure 3 This diagram illustrates a structural block diagram of a flow meter self-diagnostic device 300 based on a digital twin, according to an embodiment of this application. For ease of explanation, only the parts relevant to this embodiment are shown. The flow meter self-diagnostic device 300 based on a digital twin provided in this application includes: The signal acquisition module 301 is used to acquire the original measurement signal and current operating parameters of the flow meter, perform signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and input the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics. The deviation extraction module 302 is used to extract a first deviation sequence by comparing the simulated flow rate value with the actual measured flow rate value, extract a second deviation sequence by performing feature deviation extraction on the expected signal features and the intermediate signal features, and perform trend separation based on the first deviation sequence and the second deviation sequence to generate a systematic drift component. The fault diagnosis module 303 is used to perform cross-validation on the first deviation sequence and the second deviation sequence to generate a valid drift marker, and to perform bidirectional verification and matching on the systematic drift component with a preset fault mode library based on the valid drift marker to obtain the suspected fault type. The health assessment module 304 is used to perform a rate of change analysis on the systematic drift component to obtain the drift acceleration, and to perform health quantification by the suspected fault type and the drift acceleration to generate the flow meter health index and parameter drift amount; The compensation output module 305 is used to adaptively and dynamically compensate the actual measured flow rate value according to the parameter drift and output the compensated flow rate value, and form the flow meter monitoring result based on the flow meter health index and the compensated flow rate value.

[0069] The aforementioned digital twin-based flow meter self-diagnostic device 300 can implement the digital twin-based flow meter self-diagnostic method of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application's embodiments can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0070] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.

[0071] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A flow meter self-diagnosis method based on digital twin, characterized in that, include: The system collects the original measurement signal and current operating parameters of the flow meter, performs signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and inputs the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics. A first deviation sequence is obtained by extracting the flow deviation between the simulated flow value and the actual measured flow value. A second deviation sequence is obtained by extracting the feature deviation between the expected signal feature and the intermediate signal feature. A systematic drift component is generated by performing trend separation based on the first deviation sequence and the second deviation sequence. Cross-validation is performed on the first deviation sequence and the second deviation sequence to generate effective drift markers. Based on the effective drift markers, the systematic drift components are bidirectionally verified and matched with a preset fault mode library to obtain suspected fault types. A rate of change analysis is performed on the systematic drift component to obtain the drift acceleration. The flow meter health index and parameter drift amount are generated by quantifying the suspected fault type and the drift acceleration. Based on the parameter drift, the actual measured flow rate value is adaptively and dynamically compensated to output the compensated flow rate value. The flow meter health index and the compensated flow rate value are used to form the flow meter monitoring result.

2. The method according to claim 1, characterized in that, The step of extracting the feature deviation between the expected signal features and the intermediate signal features to obtain the second deviation sequence includes: The expected signal features and the intermediate signal features are respectively subjected to multi-dimensional feature decomposition to obtain feature components of each dimension; A multidimensional deviation set is generated by extracting dimensional deviations from the feature components of each dimension. Based on the multidimensional deviation set, inter-dimensional consistency analysis is performed to identify combinations of mismatched dimensions. A second deviation sequence is constructed by locating the fault-related links based on the deviation mismatch dimension combination.

3. The method according to claim 1, characterized in that, The step of performing trend separation based on the first deviation sequence and the second deviation sequence to generate a systematic drift component includes: The first deviation sequence and the second deviation sequence are filtered to obtain a smoothed deviation sequence; Drift direction identifiers are generated by identifying drift direction features from the smoothed deviation sequence; The drift direction identifier is subjected to drift monotonicity verification to generate a pseudo drift marker; The systematic drift component is obtained by separating the smoothed deviation sequence based on the pseudo-drift marker.

4. The method according to claim 1, characterized in that, The step of cross-validating the first deviation sequence and the second deviation sequence to generate a valid drift marker includes: Synchronization features are generated by detecting synchronous changes between the first deviation sequence and the second deviation sequence. The drift source marker is generated by performing directional consistency analysis based on the aforementioned synchronization features. The internal degradation marker is obtained by determining the degradation source based on the drift source marker; A valid drift marker is generated based on the internal degradation marker.

5. The method according to claim 1, characterized in that, The step of obtaining suspected fault types by bidirectionally verifying and matching the systematic drift components with a preset fault mode library based on the effective drift markers includes: Based on the effective drift markers, filter the systematic drift components to be matched; The systematic drift components are matched with a preset fault mode library using positive similarity to generate a candidate fault ranking. A filtered fault set is established by performing reverse compatibility verification on the candidate fault ranking. The faults with the highest similarity from the filtered fault set are identified as suspected fault types.

6. The method according to claim 1, characterized in that, The process of performing rate-of-change analysis on the systematic drift components to obtain drift acceleration includes: The drift increment sequence is obtained by extracting adjacent period increments of the systematic drift component along the time axis; Sign monitoring is performed on the drift increment sequence to identify the drift direction inflection point; Degradation stage boundary identification and degradation stage markers are established based on the drift direction inflection points. The drift acceleration is output by combining the degradation stage marker with the drift increment sequence.

7. The method according to claim 1, characterized in that, The step of generating a flow meter health index and parameter drift by quantifying the health of the suspected fault type and the drift acceleration includes: Based on the suspected fault type, query the corresponding fault-specific attenuation curve; The acceleration fluctuation characteristics are obtained by extracting the amplitude of continuous periodic fluctuations of the drift acceleration. Degradation risk level is generated by performing degradation stability analysis based on the acceleration fluctuation characteristics and the drift acceleration. The fault-specific decay curve is combined with the degradation risk level to output the flow meter health index and parameter drift.

8. The method according to claim 5, characterized in that, The step of performing reverse compatibility verification on the candidate fault ranking to establish a filtered fault set includes: Extract the physical feature constraints corresponding to each candidate fault from the candidate fault ranking; The physical feature constraints are matched using fault mutual exclusion rules to generate mutually exclusive fault pairs; Based on the mutual exclusion fault pair, a conflict marker is generated to identify the coexistence conflict between candidate faults. Based on the conflict markers, the candidate faults are sorted and compatibility-based screening is performed to form a filtered fault set.

9. The method according to claim 6, characterized in that, The step of identifying drift direction inflection points by sign monitoring of the drift increment sequence includes: The drift increment sequence is subjected to symbol extraction to generate an increment symbol sequence; For the incremental symbol sequence, statistical analysis of consecutive identical sign segments is performed to generate degenerate monotonic intervals; Based on the boundaries of the degenerate monotonic interval, sign reversal points are located to establish a candidate set of inflection points; Valid transition points are selected from the candidate set of turning points to determine the drift direction turning points.

10. A flow meter self-diagnostic device based on digital twin, characterized in that, include: The signal acquisition module is used to acquire the original measurement signal and current operating parameters of the flow meter, perform signal processing based on the original measurement signal to obtain the actual measured flow value and intermediate signal characteristics, and input the current operating parameters into the embedded digital twin model to generate the simulated flow value and expected signal characteristics. The deviation extraction module is used to extract a first deviation sequence by comparing the simulated flow rate value with the actual measured flow rate value, extract a second deviation sequence by performing feature deviation extraction on the expected signal features and the intermediate signal features, and perform trend separation based on the first deviation sequence and the second deviation sequence to generate a systematic drift component. The fault diagnosis module is used to perform cross-validation on the first deviation sequence and the second deviation sequence to generate a valid drift marker, and to perform bidirectional verification and matching on the systematic drift component with a preset fault mode library based on the valid drift marker to obtain the suspected fault type. The health assessment module is used to perform a rate of change analysis on the systematic drift component to obtain the drift acceleration, and to quantify the health of the flow meter and the parameter drift by using the suspected fault type and the drift acceleration. The compensation output module is used to adaptively and dynamically compensate the actual measured flow rate value according to the parameter drift and output the compensated flow rate value. The flow meter health index and the compensated flow rate value are used to form the flow meter monitoring result.

Images