Flowmeter predictive maintenance method and system based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a flowmeter predictive maintenance method and system based on artificial intelligence, and the method comprises the steps: firstly carrying out the learning of multi-dimensional operation parameters through an artificial intelligence model, and preliminarily predicting the type of an electrode fault; and then, introducing a physical verification link based on equivalent capacitance, comparing an AI prediction result with an ideal capacitance value determined by a fault type so as to verify and correct the preliminary diagnosis, and finally outputting an accurate target fault type and confidence. Then, fault evolution prediction is carried out based on the target fault type and the corresponding target confidence coefficient, and fault remaining maintenance time is calculated; and finally, generating a personalized maintenance strategy according to the residual maintenance time of the fault and the target fault type. Therefore, different fault types of the electrode can be predicted and identified in advance, and a targeted maintenance strategy is provided.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and specifically to a predictive maintenance method and system for flow meters based on artificial intelligence. Background Technology

[0002] In industrial production and various fluid monitoring scenarios, flow meters are crucial equipment. Their stable operation and accurate measurement play a key role in optimizing production processes, rationally utilizing resources, and ensuring product quality. Traditional flow meter maintenance methods mostly consist of periodic maintenance and reactive repair after a failure. Periodic maintenance involves checking, cleaning, and calibrating the flow meter according to fixed time cycles. For example, monthly or quarterly inspections of the flow meter's appearance are conducted to check for obvious damage, corrosion, or leaks; annual calibration is performed using a standard flow meter for comparison to ensure the accuracy of readings. While this approach can ensure the normal operation of the flow meter to a certain extent, it suffers from significant blindness. Within the fixed cycle, the flow meter may not have any actual problems, yet unnecessary maintenance is performed, wasting a lot of manpower, material resources, and time. Reactive repair after a failure, on the other hand, only addresses the issue when the flow meter has already malfunctioned and is affecting production or monitoring work. This often leads to production interruptions and serious economic losses. For example, in the petrochemical industry, flow meter failures can cause inaccurate material delivery, affecting product quality and even causing safety accidents.

[0003] In related technologies, the limitations of traditional maintenance methods for electromagnetic flowmeters are becoming increasingly apparent, making it difficult to meet the high reliability and stability requirements of modern industry for flowmeters. Summary of the Invention

[0004] The main objective of this invention is to provide a predictive maintenance method and system for flow meters based on artificial intelligence, aiming to solve the technical problem that traditional maintenance methods in the prior art cannot meet the high reliability and stability requirements of modern industry for flow meters.

[0005] To achieve the above objectives, in a first aspect, this application provides an artificial intelligence-based predictive maintenance method for flow meters, applied to a flow meter maintenance system, the method comprising:

[0006] The flow meter operates by acquiring multi-dimensional parameters, which include at least one or more of the following: flow parameters, conductivity parameters, vibration parameters, environmental parameters, instrument power supply parameters, and medium characteristic parameters.

[0007] Feature extraction is performed on the collected multi-dimensional parameters to obtain static features, dynamic features, and coupling features;

[0008] The extracted features are input into a pre-trained fault prediction model, which outputs the electrode fault prediction type and corresponding confidence level. The electrode fault prediction type includes electrode water film coverage fault, electrode scale coverage fault, electrode vibration loosening fault, and electrode corrosion damage fault.

[0009] Based on the ideal capacitance value and the actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type, capacitance verification is performed on the electrode fault prediction type to obtain the target fault type and target confidence level of the electrode.

[0010] Based on the target fault type and the corresponding target confidence level, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual working conditions and calculate the remaining maintenance time of the fault.

[0011] Based on the remaining maintenance time of the fault and the target fault type, a targeted maintenance strategy is matched and generated from the preset strategy library.

[0012] In one possible implementation, the static feature is the degree to which the parameter deviates from a preset normal range, including the degree of deviation of the parameter, and the formula for calculating the degree of deviation is:

[0013] ;

[0014] Where δ is the parameter deviation, x is the real-time collected value of the parameter, x0 is the preset standard value of the parameter, x_max is the maximum allowed value of the parameter, and x_min is the minimum allowed value of the parameter; when δ≥20%, it is marked as a high deviation feature.

[0015] In one possible implementation, the dynamic feature is the temporal change trend of the parameter, including the rate of change of the parameter. The rate of change is obtained by calculating the slope of parameter change within adjacent sliding time windows using a linear fitting algorithm. The length of the sliding time window is adaptively adjusted according to the parameter change period.

[0016] In one possible implementation, the coupling feature is the synchronous change correlation between multiple parameters, including the correlation coefficient of parameter pairs. The step of inputting the extracted features into a pre-trained fault prediction model and outputting the electrode fault prediction type and corresponding confidence level includes:

[0017] The fault type is initially predicted by calling static and dynamic features through the feature selection algorithm, and the candidate electrode fault prediction type is obtained.

[0018] The target coupling features are obtained by matching the corresponding coupling features according to the candidate electrode fault prediction type; wherein, the target coupling features include the correlation coefficients of the target parameter pairs;

[0019] By using parallel attention mechanism network layers to assign weights and focus key information on static features, dynamic features and target coupling features respectively, the model inference outputs the fault prediction type and corresponding confidence level of the flow meter electrode.

[0020] In one possible implementation, when there are multiple candidate electrode fault prediction types, the parallel attention mechanism network layer performs weight allocation and key information focusing on static features, dynamic features, and target coupling features respectively, and then outputs the fault prediction type and corresponding confidence level of the flowmeter electrode through model inference, including:

[0021] Extract the correlation coefficient r of the coupling feature corresponding to each candidate fault prediction type, and compare it with the preset correlation coefficient threshold r0 of the coupling feature of the fault prediction type. Retain the target coupling feature and the corresponding candidate fault prediction type with r≥r0 to form an effective candidate set.

[0022] For each candidate fault prediction type in the effective candidate set, its corresponding static features, dynamic features and target coupling features are input into a parallel attention mechanism network layer. The network layer assigns weights to the three types of features and outputs the preliminary confidence level for each candidate type.

[0023] The candidate fault prediction type with the highest initial confidence level and not lower than the preset value is selected as the final fault prediction type. If the initial confidence level of all candidate types in the effective candidate set is lower than the preset value, the correlation coefficient details of the top two candidate types with the highest confidence level and their target coupling features are output, triggering the manual review process, and the final fault type is confirmed by human.

[0024] In one possible implementation, based on the ideal capacitance value and the actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type, a capacitance verification is performed on the electrode fault prediction type to obtain the electrode target fault type and target confidence level, including:

[0025] Calculate the percentage relative deviation between the actual equivalent capacitance value and the ideal capacitance value;

[0026] If the relative deviation percentage is less than the first threshold, the capacitor verification is deemed to have passed, and the electrode fault prediction type is directly determined as the electrode target fault type.

[0027] If the relative deviation percentage is determined to be greater than or equal to the first threshold and less than the second threshold, it is determined to be an operating condition disturbance. The original prediction type is maintained as the target fault type, and its corresponding confidence level is reduced.

[0028] If the relative deviation percentage is determined to be greater than or equal to the second threshold, the capacitance verification is deemed to have failed, its corresponding confidence level is reduced to zero, and a manual maintenance mechanism is triggered.

[0029] In one possible implementation, based on the target fault type and the corresponding target confidence level, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual operating conditions, and the remaining maintenance time of the fault is calculated, including:

[0030] Based on the target fault type, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual working conditions and obtain an initial assessment value of the remaining maintenance time.

[0031] The remaining maintenance time is obtained by correcting the initial assessment value of the remaining maintenance time with a confidence correction factor. The formula for calculating the corrected remaining maintenance time is as follows:

[0032] ;

[0033] Where T is the corrected remaining maintenance time, T0 is the remaining maintenance time initially calculated by the model, α is the confidence correction coefficient, which ranges from 0.3 to 0.8, and γ is the target confidence level.

[0034] In one possible implementation, based on the target fault type, a preset fault deterioration model is invoked to simulate the fault's evolution path under the current actual operating conditions to obtain an initial assessment of the remaining maintenance time, including:

[0035] The fault deterioration model incorporates real-time dynamic adjustment factors when simulating the evolution path. These dynamic adjustment factors include the medium temperature fluctuation coefficient, pressure change coefficient, and environmental humidity influence coefficient. The target evolution curve is obtained by weighted calculation and correction of the evolution curve.

[0036] The initial assessment value of the remaining maintenance time is determined based on the target evolution curve.

[0037] In one possible implementation, the target evolution curve is obtained by weighted calculation of the modified evolution curve, including:

[0038] Preset weights are assigned to the medium temperature fluctuation coefficient k1, pressure change coefficient k2, and environmental humidity influence coefficient k3, respectively.

[0039] Calculate the overall adjustment factor K using the following formula: ,in They are respectively The corresponding preset weights;

[0040] The time axis parameter t of the initial fault evolution curve is corrected by the comprehensive adjustment factor K. The correction formula is t' = t×(1 + K). If K is negative, then t' = t×(1 - |K|). The corrected target evolution curve is obtained.

[0041] Secondly, embodiments of this application also provide a flow meter maintenance system, including: a memory and a processor, wherein the memory is used to store program code; and the processor is used to call the program code to execute the method as described in the first aspect.

[0042] Unlike existing technologies, the AI-based predictive maintenance method for flow meters provided in this application first learns multi-dimensional operating parameters through an AI model to preliminarily predict electrode fault types. Then, a physical verification step based on equivalent capacitance is introduced, comparing the AI ​​prediction results with the ideal capacitance value determined by the fault type to verify and correct the preliminary diagnosis, ultimately outputting an accurate target fault type and confidence level. Next, based on the target fault type and corresponding target confidence level, fault evolution prediction is performed to calculate the remaining maintenance time. Finally, a personalized maintenance strategy is generated based on the remaining maintenance time and the target fault type. Thus, this application, through the dual guarantee of AI preliminary screening and secondary verification using physical mechanisms, effectively overcomes the false alarm and false negative problems that may exist in single data-driven models, significantly improving the ability to distinguish and identify different fault types such as water film coverage, scaling, loosening, and corrosion. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the flow meter structure in some embodiments of this application;

[0045] Figure 2 This is a flowchart illustrating the predictive maintenance method for flow meters in some embodiments of this application;

[0046] Figure 3 This is a flowchart illustrating step S300 of the predictive maintenance method for flow meters in some embodiments of this application;

[0047] Figure 4 This is a flowchart illustrating step S400 of the predictive maintenance method for a flow meter in some embodiments of this application;

[0048] Figure 5 This is a flowchart illustrating step S500 of the predictive maintenance method for a flow meter in some embodiments of this application;

[0049] Figure 6This is a schematic diagram of the hardware structure of the flow meter maintenance system in some embodiments of this application.

[0050] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0052] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0053] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the term "and / or" throughout the text includes three solutions; taking A and / or B as an example, it includes technical solution A, technical solution B, and a technical solution that simultaneously satisfies A and B. Furthermore, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of a person skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0054] In industrial production and various fluid monitoring scenarios, flow meters are crucial equipment. Their stable operation and accurate measurement play a key role in optimizing production processes, rationally utilizing resources, and ensuring product quality. Traditional flow meter maintenance methods mostly consist of periodic maintenance and reactive repair after a failure. Periodic maintenance involves checking, cleaning, and calibrating the flow meter according to fixed time cycles. For example, monthly or quarterly inspections of the flow meter's appearance are conducted to check for obvious damage, corrosion, or leaks; annual calibration is performed using a standard flow meter for comparison to ensure the accuracy of readings. While this approach can ensure the normal operation of the flow meter to a certain extent, it suffers from significant blindness. Within the fixed cycle, the flow meter may not have any actual problems, yet unnecessary maintenance is performed, wasting a lot of manpower, material resources, and time. Reactive repair after a failure, on the other hand, only addresses the issue when the flow meter has already malfunctioned and is affecting production or monitoring work. This often leads to production interruptions and serious economic losses. For example, in the petrochemical industry, flow meter failures can cause inaccurate material delivery, affecting product quality and even causing safety accidents.

[0055] In related technologies, the limitations of traditional maintenance methods for electromagnetic flowmeters are becoming increasingly apparent, making it difficult to meet the high reliability and stability requirements of modern industry for flowmeters.

[0056] To address the aforementioned technical issues, this application provides an artificial intelligence-based predictive maintenance method for flow meters, applicable to flow meter maintenance systems. The flow meter maintenance system includes an electromagnetic flow meter, various flow meter parameter monitoring sensors, and a data processing cloud platform. These sensors continuously collect operational status data from the flow meter and transmit the data to the cloud via a network. On the data processing cloud platform, this massive amount of real-time monitoring data is input into a pre-trained fault prediction artificial intelligence model. This model, through deep analysis and learning of multi-dimensional parameter sequences, can accurately identify potential fault symptoms and fault development trends, ultimately achieving early warning and predictive maintenance of electromagnetic flow meter faults.

[0057] like Figure 1 As shown, the electromagnetic flowmeter of this application includes a fluid pipe 100, an electromagnetic coil 200, two electrode plates 300 disposed inside the fluid pipe 100, and a converter 400 connected to the electrode plates 300, wherein the diameter of the fluid pipe 100 is D.

[0058] The fluid pipe 100, serving as the carrier for the conductive fluid, has an inner wall coated with an insulating lining made of materials such as polytetrafluoroethylene (PTFE). This not only prevents interference from the pipe's conductivity with the measurement signal but also effectively prevents corrosion of the pipe's inner wall by the fluid, providing a stable physical environment for flow measurement. The electromagnetic coils 200 are symmetrically installed on both sides of the fluid pipe 100, typically employing a ring-wound structure. When energized, they generate a uniform alternating magnetic field perpendicular to the fluid flow direction inside the pipe, with stable and controllable magnetic field strength, serving as the core magnetic field source for generating the induced electromotive force. The two electrode plates 300 are made of corrosion-resistant metals such as stainless steel and Hastelloy, and are laid flat. The electrode plate 300 is symmetrically fixed in the insulating lining of the inner wall of the fluid pipe 100 in an embedded manner. The sensing surface of the electrode plate 300 is flush with the inner wall of the pipe, which does not obstruct the fluid flow and can directly contact the conductive fluid being measured, so as to realize the real-time capture of the weak induced electromotive force signal generated by the fluid cutting magnetic field lines. The converter 400, as the core component for signal processing and output, integrates a signal amplification module, a filtering circuit, an A / D conversion unit and a microprocessor. It can process the raw signal transmitted by the electrode plate 300 and convert it into flow data. At the same time, it has data storage and communication functions, providing raw monitoring data support for subsequent predictive maintenance work.

[0059] The following explanation uses a flow meter maintenance system as an example to illustrate the predictive maintenance method for flow meters. It should be noted that although the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order. Please refer to the appendix. Figure 2 The method includes the following steps S100-S600:

[0060] Step S100: Obtain multi-dimensional parameters of the flow meter operation, wherein the multi-dimensional parameters include at least one or more of the following: flow parameters, conductivity parameters, vibration parameters, environmental parameters, instrument power supply parameters, and medium characteristic parameters;

[0061] Specifically, based on the specific application requirements and monitoring objectives, it can be determined which types of parameters need to be acquired, including flow rate parameters, conductivity parameters, vibration parameters, environmental parameters, instrument power supply parameters, and media characteristic parameters. For example, to monitor the flow rate of industrial wastewater and whether there is scaling fault on the flowmeter electrodes, flow rate parameters, conductivity parameters, and media characteristic parameters can be acquired to determine whether scaling fault has occurred.

[0062] Among them, flow parameters can be measured by electromagnetic flowmeters, including instantaneous flow or average flow; conductivity parameters can be measured by conductivity sensors. Conductivity parameters are data related to the measured medium and electrode state, and can help determine whether there are abnormalities on the electrode surface. When the purity of the medium is determined, conductivity parameters can be used to determine whether there is a water film or scaling on the electrode surface; vibration parameters refer to the vibration of the pipeline, which can be detected by multi-axis accelerometers and can be used to help determine whether the electrodes are vibrating and loose; environmental parameters can include ambient temperature, humidity, air pressure, etc., and can be detected by corresponding sensors; instrument power supply parameters can include the flowmeter's power supply voltage, current, etc.; and medium characteristic parameters can include the viscosity, density, corrosiveness, etc. of the measured fluid medium, which can be determined according to the type of fluid medium.

[0063] For example, the cloud platform can acquire one or more of the following parameters in real time: flow rate, conductivity, vibration, environmental, power supply, and media characteristics of the flow meter, in order to provide comprehensive and reliable data support for subsequent fault diagnosis and prediction.

[0064] Step S200: Extract features from the collected multi-dimensional parameters to obtain static features, dynamic features, and coupling features;

[0065] Static characteristics refer to the degree to which a parameter deviates from a preset normal range, including the degree of deviation, such as the deviation of the current flow rate value from the normal range. The degree of deviation can be calculated using the following formula:

[0066] ;

[0067] In the formula, δ is the parameter deviation, x is the real-time collected value of the parameter, x0 is the preset standard value of the parameter, x_max is the maximum allowed value of the parameter, and x_min is the minimum allowed value of the parameter; when δ≥20%, it is marked as a high deviation feature.

[0068] Among them, the dynamic feature is the temporal change trend of the parameter, including the rate of change of the parameter, such as the rate of change of conductivity. The rate of change can be obtained by calculating the slope of parameter change within adjacent sliding time windows using a linear fitting algorithm. Furthermore, the length of the sliding time window can be adaptively adjusted according to the parameter change period to accurately capture the temporal trend at different scales.

[0069] The coupling feature refers to the synchronous change correlation between multiple parameters, including the correlation coefficient of parameter pairs, such as the correlation coefficient between the change in flow rate and the change in conductivity, and the correlation coefficient between the change in flow rate and the change in vibration intensity. The correlation coefficient can be calculated using the Pearson correlation coefficient formula to determine the degree of linear correlation between two parameters in the same time series, with a value range of [-1, 1]. When the absolute value of the correlation coefficient is ≥ 0.8, it is marked as strong coupling, and when the absolute value of the correlation coefficient is ≤ 0.3, it is marked as weak coupling.

[0070] It should be noted that the static, dynamic and coupling characteristics of different parameters can be used individually or in combination, and the different characteristics can be used to reflect different types of electrode faults.

[0071] For example, the cloud platform performs feature extraction on the collected multi-dimensional parameters to obtain static features, dynamic features, and coupling features.

[0072] Step S300: Input the extracted features into the pre-trained fault prediction model and output the electrode fault prediction type and corresponding confidence level. The electrode fault prediction type includes electrode water film coverage fault, electrode scale coverage fault, electrode vibration loosening fault and electrode corrosion damage fault.

[0073] After obtaining the static features, dynamic features, and coupling features between the parameters in step S200, the cloud platform can input these three types of features into the pre-trained fault prediction model to output the electrode fault prediction type and corresponding confidence score. The confidence score is used to provide feedback on the reliability of the model's fault judgment. The output predicted fault types include electrode water film coverage fault, electrode scale coverage fault, electrode vibration loosening fault, and electrode corrosion damage fault.

[0074] For example, the static characteristics are as follows: the deviation δ of the conductivity parameter gradually increases from 10% to 30% over time (showing a continuously increasing high deviation trend). This is due to the accumulation of scale thickness on the electrode surface, which reduces the effective contact area between the electrode and the medium, thus weakening the signal transmission capability. Simultaneously, the deviation δ of the flow parameter occasionally exceeds 20% (because scale alters the flow field distribution near the electrode, interfering with the accuracy of flow measurement). The dynamic characteristics show that the slope of the conductivity change rate is -0.05 S / cm·h (showing a continuously decreasing trend), and within a 2-hour sliding window, the absolute value of the slope gradually increases with operating time (e.g., from 0.03 to 0.07), reflecting the gradual accumulation of scale thickness over time—the thicker the scale, the faster the conductivity signal attenuates. The coupling characteristics are manifested as: the correlation coefficient |r| between the conductivity change value and the flow change value is ≥0.8 (strong coupling). This is because scale not only directly affects the conductivity measurement of the electrode but also indirectly affects the flow detection results by changing the fluid resistance in the pipeline, resulting in a significant synchronicity between the abnormal changes of the two. Model output: Considering the sustained high deviation trend of conductivity, the rate of decrease with the accumulation of scale, and the strong coupling characteristics of conductivity and flow rate changes (consistent with the synergistic effect of scale on both), the model is judged as "electrode scale coverage failure" with a confidence level of 92%.

[0075] Step S400: Based on the ideal capacitance value and the actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type, perform capacitance verification on the electrode fault prediction type to obtain the target fault type and target confidence level of the electrode.

[0076] After obtaining the electrode fault prediction type and corresponding confidence level in step S300, the cloud platform can further perform capacitance verification on the electrode fault prediction type based on the ideal capacitance value and actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type to obtain the target fault type and target confidence level, thereby further improving the accuracy of fault prediction.

[0077] Step S500: Based on the target fault type and the corresponding target confidence level, call the preset fault deterioration model to simulate the evolution path of the fault under the current actual working conditions and calculate the remaining maintenance time of the fault.

[0078] After obtaining the target fault type and corresponding target confidence level of the electrode through capacitance verification in step S400, the cloud platform can automatically call a preset fault deterioration model based on the target fault type and its corresponding target confidence level. This model pre-integrates the deterioration patterns of similar faults under different operating parameters (such as medium temperature, flow rate, conductivity, pressure, etc.) in historical fault data. By inputting the current actual operating parameters (real-time collected operating data) into the model, it can dynamically simulate the complete evolution path of the fault from the current state to the critical state that requires maintenance. Based on the above simulated evolution path, the model further calculates the time span required from the current moment to the fault reaching the critical state, i.e., the remaining maintenance time. For example, if the target fault type is "electrode scaling" and the target confidence level is 90%, and the current operating conditions have a low medium flow rate and a high temperature (accelerating scaling), the model will combine the time taken for the scaling thickness to increase from the current value to the critical thickness affecting the metering accuracy under similar operating conditions in historical data, and output the specific remaining maintenance time (e.g., 72 hours).

[0079] Step S600: Based on the remaining maintenance time of the fault and the target fault type, match and generate a targeted maintenance strategy from the preset strategy library.

[0080] After calculating the remaining maintenance time for the fault in step S500, the cloud platform can match and generate a targeted maintenance strategy from the preset strategy library based on the remaining maintenance time and the target fault type. The strategy library pre-stores standard handling procedures for different fault types under different remaining times, including operation steps, required tools, safety specifications, etc. For example, for the case of "electrode scaling" with a remaining maintenance time of 48 hours, the strategy of "online chemical cleaning + secondary testing after 24 hours" will be matched. At the same time, the cleaning agent model and circulating cleaning pressure and other parameters are specified to ensure that the generated maintenance strategy can not only accurately solve the fault, but also adapt to the time urgency and reduce the impact on the normal operation of the equipment.

[0081] Based on this, the AI-based predictive maintenance method for flow meters provided in this application first learns multi-dimensional operating parameters through an AI model to preliminarily predict electrode fault types. Then, a physical verification step based on equivalent capacitance is introduced, comparing the AI ​​prediction results with the ideal capacitance value determined by the fault type to verify and correct the preliminary diagnosis, ultimately outputting an accurate target fault type and confidence level. Next, based on the target fault type and corresponding target confidence level, fault evolution prediction is performed to calculate the remaining maintenance time. Finally, a personalized maintenance strategy is generated based on the remaining maintenance time and the target fault type. Thus, this application, through the dual guarantee of AI preliminary screening and secondary verification of physical mechanisms, effectively overcomes the false alarm and false negative problems that may exist in a single data-driven model, significantly improving the ability to distinguish and identify different fault types such as water film coverage, scaling, loosening, and corrosion.

[0082] like Figure 3 As shown, in one embodiment, step S300: inputting the extracted features into the pre-trained fault prediction model, and outputting the electrode fault prediction type and corresponding confidence level, includes:

[0083] S310. The static and dynamic features are called through the feature selection algorithm to make an initial prediction of the fault type and obtain the candidate electrode fault prediction type.

[0084] S320. Match the corresponding coupling features according to the candidate electrode fault prediction type to obtain the target coupling features; wherein, the target coupling features include the correlation coefficient of the target parameter pair;

[0085] S330: After using parallel attention mechanism network layers to assign weights and focus key information on static features, dynamic features and target coupling features respectively, the model inference outputs the fault prediction type and corresponding confidence level of the flow meter electrode.

[0086] Specifically, firstly, a feature selection algorithm is used to perform initial fault type prediction by calling static and dynamic features, resulting in candidate electrode fault prediction types. The feature selection algorithm can quickly filter out abnormal static and dynamic features, and based on these features, the range of fault types is initially narrowed down, yielding several candidate fault prediction types, laying the foundation for subsequent accurate prediction. Then, the corresponding coupling features are matched according to the candidate electrode fault prediction types to obtain target coupling features, which include the correlation coefficients of target parameter pairs. Different fault types often correspond to specific correlation patterns between parameters. By directionally matching candidate fault types with related parameter pairs and calculating the correlation coefficients of these parameter pairs, deep features reflecting the inherent correlation of the fault can be obtained, supplementing the prediction limitations of single static or dynamic features. Finally, a parallel attention mechanism network layer is used to assign weights and focus key information on static features, dynamic features, and target coupling features respectively. The parallel structure ensures that the effective information of each of the three types of features is not obscured, and the attention mechanism automatically strengthens the weights of features that play a key role in fault prediction and weakens interference from irrelevant information. After model inference, the final output is the fault prediction type and corresponding confidence level of the flowmeter electrode.

[0087] Thus, the embodiments of this application, through the steps of first screening candidate fault types, then matching deep coupling features, and finally accurately focusing on key information through an attention mechanism, can efficiently utilize multi-dimensional features to achieve accurate prediction of electrode faults, while outputting confidence scores to provide a reliable reference for fault judgment, thereby improving the accuracy and practicality of fault prediction.

[0088] In one embodiment, when there are multiple candidate electrode fault prediction types, the parallel attention mechanism network layer performs weight allocation and key information focusing on static features, dynamic features, and target coupling features respectively, and then outputs the fault prediction type and corresponding confidence level of the flowmeter electrode through model inference, including:

[0089] Extract the correlation coefficient r of the coupling feature corresponding to each candidate fault prediction type, and compare it with the preset correlation coefficient threshold r0 of the coupling feature of the fault prediction type. Retain the target coupling feature and the corresponding candidate fault prediction type with r≥r0 to form an effective candidate set.

[0090] For each candidate fault prediction type in the effective candidate set, its corresponding static features, dynamic features and target coupling features are input into a parallel attention mechanism network layer. The network layer assigns weights to the three types of features and outputs the preliminary confidence level for each candidate type.

[0091] The candidate fault prediction type with the highest initial confidence level and not lower than the preset value is selected as the final fault prediction type. If the initial confidence level of all candidate types in the effective candidate set is lower than the preset value, the correlation coefficient details of the top two candidate types with the highest confidence level and their target coupling features are output, triggering the manual review process, and the final fault type is confirmed by human.

[0092] Specifically, when there are multiple candidate electrode fault prediction types, the correlation coefficient r of the coupling feature corresponding to each candidate fault prediction type is first extracted and compared with the preset correlation coefficient threshold r0 of the coupling feature for that fault prediction type. Target coupling features with r ≥ r0 and their corresponding candidate fault prediction types are retained to form an effective candidate set. The correlation coefficient r reflects the strength of the association between the target parameter and the fault type. The preset threshold r0 is set based on historical fault data and expert experience. This screening process can eliminate candidate types with weak correlation, reduce invalid calculations, and improve the targeting of subsequent predictions. For example, when the candidate type is "electrode scaling," its corresponding coupling feature might be the "correlation coefficient between conductivity fluctuations and flow rate changes." If this coefficient is lower than the preset threshold, it indicates that the current coupling feature is insufficient to support the judgment of this fault type and needs to be excluded.

[0093] Subsequently, for each candidate fault prediction type in the valid candidate set, its corresponding static features, dynamic features, and target coupling features are input into a parallel attention mechanism network layer. The network layer assigns weights to the three types of features and outputs the preliminary confidence score for each candidate type. The parallel attention mechanism network layer can independently process the features of different candidate types: static features, dynamic features, and target coupling features. By assigning weights, it strengthens key features (e.g., in the "electrode corrosion" fault, the weight of "pH change rate" in the dynamic features may be increased) and weakens secondary information. The final output preliminary confidence score can quantify the probability of each candidate type.

[0094] Finally, the candidate fault prediction type with the highest preliminary confidence level, which is not lower than the preset value, is selected as the final fault prediction type to ensure the reliability of the output results. If the preliminary confidence level of all candidate types in the effective candidate set is lower than the preset value, the correlation coefficient details of the top two candidate types with the highest confidence levels and their target coupling features are output, triggering a manual review process where the final fault type is confirmed manually. This mechanism ensures automatic and accurate output under high confidence conditions, while introducing manual intervention when features are ambiguous and the model is difficult to judge. It provides data support for manual decision-making by combining the correlation coefficient details, avoiding misjudgment.

[0095] Thus, when there are many candidate fault types, this embodiment of the application can effectively narrow the prediction range and improve the accuracy of fault type judgment by using coupled feature correlation screening, attention mechanism for precise weighting, and confidence level decision-making. It can also make up for the limitations of the model through manual review mechanism, ensuring the rigor and practicality of fault prediction.

[0096] like Figure 4 As shown, in one embodiment, step S400: based on the ideal capacitance value and actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type, a capacitance verification is performed on the electrode fault prediction type to obtain the electrode target fault type and target confidence level, including:

[0097] S410. Calculate the percentage of relative deviation between the actual equivalent capacitance value and the ideal capacitance value;

[0098] S420. If the relative deviation percentage is less than the first threshold, the capacitor verification is deemed to have passed, and the electrode fault prediction type is directly determined as the electrode target fault type.

[0099] S430. If the relative deviation percentage is determined to be greater than or equal to the first threshold and less than the second threshold, it is determined to be an operating condition disturbance. The original prediction type is maintained as the target fault type, and its corresponding confidence level is reduced.

[0100] S440. If the relative deviation percentage is determined to be greater than or equal to the second threshold, the capacitor verification is deemed to have failed, its corresponding confidence level is reduced to zero, and the manual maintenance mechanism is triggered.

[0101] The actual equivalent capacitance value refers to the equivalent capacitance measurement value between the two electrode plates of the flowmeter under the current operating conditions, which is collected in real time by the capacitance detection circuit and directly reflects the actual electrical characteristics of the electrode at present.

[0102] The ideal capacitance value of an electrode refers to a benchmark capacitance reference value that is matched with a specific fault prediction type and is formed based on the annotation of historical fault case data. This value is determined by statistical analysis of capacitance data when similar faults occur (such as taking the mean or range). It represents the capacitance characteristics that the fault type should exhibit under typical operating conditions and provides a benchmark for comparison and verification of actual measured values.

[0103] Specifically, firstly, the relative percentage deviation between the actual equivalent capacitance value and the ideal capacitance value is calculated. The relative percentage deviation can be calculated using the formula "(actual equivalent capacitance value - ideal capacitance value) / ideal capacitance value × 100%", quantifying the degree of difference between the two and providing a quantitative basis for subsequent verification. If the relative percentage deviation is less than a first threshold (e.g., 5%), the capacitance verification is considered passed, and the electrode fault prediction type is directly determined as the target electrode fault type. This indicates that the actual capacitance state highly matches the typical characteristics of this fault type, further verifying the reliability of the prediction result, and the fault type can be confirmed without adjustment. If the relative percentage deviation is greater than or equal to the first threshold and less than the second threshold (e.g., 5% ≤ deviation < 15%), it is determined to be an operating condition disturbance (e.g., a sudden change in medium temperature), and the original prediction type is maintained as the target fault type, but its corresponding confidence level is reduced (e.g., from 80% to 60%). At this time, the capacitance deviation does not exceed the reasonable fluctuation range, and the fault type still has a high probability, but the uncertainty caused by the disturbance needs to be reflected by reducing the confidence level. If the relative deviation percentage is determined to be greater than or equal to the second threshold (e.g., ≥15%), the capacitor verification is deemed to have failed, its corresponding confidence level is lowered to zero, and a manual maintenance mechanism is triggered. This indicates that the actual capacitor condition is significantly inconsistent with the typical characteristics of the predicted fault type, and the original prediction result has extremely low reliability. Manual intervention is required to detect the actual condition of the electrodes (e.g., on-site inspection of whether the electrodes are damaged, the degree of scaling, etc.) to avoid misjudgment leading to equipment damage.

[0104] Thus, this application embodiment uses relative deviation quantitative analysis of capacitance values, combined with a dual-threshold hierarchical verification mechanism, to not only enhance the accuracy of fault prediction through capacitance characteristic verification, but also to distinguish between operating condition interference and prediction errors. While ensuring the efficiency of automatic judgment, it compensates for model limitations through a manual maintenance mechanism, and finally outputs reliable electrode target fault types and target confidence levels.

[0105] like Figure 5 As shown, in one embodiment, step S500: based on the target fault type and the corresponding target confidence level, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual operating conditions, and the remaining maintenance time of the fault is calculated, including:

[0106] S510. Based on the target fault type, call the preset fault deterioration model to simulate the evolution path of the fault under the current actual working conditions to obtain the initial assessment value of the remaining maintenance time.

[0107] S520. A confidence correction factor is introduced to correct the initial assessment value of the remaining maintenance time to obtain the remaining maintenance time for the fault. The formula for calculating the corrected remaining maintenance time is as follows:

[0108] ;

[0109] Where T is the corrected remaining maintenance time, T0 is the remaining maintenance time initially calculated by the model, α is the confidence correction coefficient, which ranges from 0.3 to 0.8, and γ is the target confidence level.

[0110] Specifically, firstly, based on the target fault type, a preset fault deterioration model is invoked to simulate the fault's evolution path under the current actual operating conditions, obtaining an initial assessment of the remaining maintenance time. Then, a confidence correction factor is introduced to correct the initial assessment of the remaining maintenance time, yielding the remaining maintenance time for the fault. The correction formula is as follows: Among them, the target confidence level γ reflects the degree of certainty of the target fault type (range 0-1), and the confidence level correction coefficient α (0.3-0.8) is an empirical coefficient calibrated based on historical operation and maintenance data, used to adjust the weight of the confidence level on the time assessment. For example, if the initial assessment value T0 = 120 hours, the target confidence level γ = 0.9 (high certainty), and α = 0.5, then the corrected T = 120 × (1 - 0.5 × (1 - 0.9)) = 120 × 0.95 = 114 hours. Because the confidence level is high, the correction is small. If γ = 0.6 (medium certainty), then T = 120 × (1 - 0.5 × 0.4) = 120 × 0.8 = 96 hours, and the uncertainty is addressed by shortening the assessment time. If γ = 0.3 (low certainty), then T = 120 × (1 - 0.5 × 0.7) = 120 × 0.65 = 78 hours, and the time is further compressed to reserve a maintenance buffer.

[0111] Thus, this embodiment of the application obtains the initial assessment time based on the fault type and real-time operating conditions, and then dynamically corrects it by combining the target confidence level. This not only ensures the basic accuracy of the time assessment by using the model, but also reflects the impact of the uncertainty of fault judgment on the maintenance timing through the confidence level factor. This makes the output remaining maintenance time of the fault more in line with the actual operation and maintenance needs, avoids maintenance too early or too late due to assessment deviation, and improves the rationality of operation and maintenance decisions.

[0112] In one embodiment, step S510: Based on the target fault type, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual operating conditions to obtain an initial assessment value of the remaining maintenance time, including:

[0113] The fault deterioration model incorporates real-time dynamic adjustment factors when simulating the evolution path. These dynamic adjustment factors include the medium temperature fluctuation coefficient, pressure change coefficient, and environmental humidity influence coefficient. The target evolution curve is obtained by weighted calculation and correction of the evolution curve.

[0114] The initial assessment value of the remaining maintenance time is determined based on the target evolution curve.

[0115] Specifically, firstly, the fault deterioration model incorporates real-time dynamic adjustment factors when simulating the fault evolution path. These dynamic adjustment factors include the medium temperature fluctuation coefficient, the pressure change coefficient, and the environmental humidity influence coefficient. The medium temperature fluctuation coefficient reflects the impact of the deviation between the current medium temperature and historical typical operating temperature on the fault deterioration rate (e.g., high temperatures may accelerate electrode corrosion, corresponding to a coefficient greater than 1). The pressure change coefficient quantifies the effect of pressure fluctuations within the pipeline on fault development (e.g., high-pressure environments may exacerbate the aging of electrode seals, with the coefficient increasing with the degree of pressure exceeding the limit). The environmental humidity influence coefficient addresses the external working environment of the flowmeter; a humid environment may reduce electrode insulation performance, causing this coefficient to increase with rising humidity. These coefficients are calculated using data collected in real-time by sensors.

[0116] Subsequently, the aforementioned dynamic adjustment factors are integrated into the model's base evolution curve through weighted calculations, thus obtaining the target evolution curve. The weights of each factor can be pre-set for different fault types—for example, pre-set weights are assigned to the medium temperature fluctuation coefficient k1, pressure change coefficient k2, and environmental humidity influence coefficient k3. Through the weighted formula "correction coefficient" "Calculate the comprehensive adjustment value, and then correct the time axis parameter t of the initial fault evolution curve according to the comprehensive adjustment coefficient K: if K is positive, the correction formula is t' = t × (1 + K); if K is negative, then t' = t × (1 - |K|). Through this correction, the initial curve is adjusted to a target evolution curve that fits the current real-time operating conditions, and can accurately reflect the dynamic change law of fault characteristic parameters over time."

[0117] Finally, the initial assessment value of the remaining maintenance time is determined based on the target evolution curve. This is achieved by identifying the time points in the curve where fault characteristic parameters reach preset critical thresholds (e.g., capacitance values ​​falling below the critical value required for metering accuracy), and calculating the difference between these time points and the current time. This difference is the initial assessment value of the remaining maintenance time. For example, if the target evolution curve shows that capacitance decay caused by electrode scaling reaches a critical value after 150 hours, the difference between the current time and that time point is 150 hours, which is used as the initial assessment result.

[0118] It should be noted that when K>0 (accelerated failure), t'=t×(1+K) extends the time axis to simulate accelerated deterioration, corresponding to a shorter remaining maintenance time; when K<0 (decelerated failure), t'=t×(1-|K|) compresses the time axis to simulate decelerated deterioration, resulting in a longer remaining maintenance time.

[0119] Thus, by incorporating multi-dimensional dynamic adjustment factors of real-time operating conditions into the fault deterioration model, the fault evolution curve that accurately reflects the current operating conditions is obtained through weighted correction. The initial assessment value of the remaining maintenance time calculated accordingly can fully adapt to the impact of the actual operating environment, avoid the assessment deviation caused by ignoring the fluctuation of operating conditions due to the fixed model, provide more reliable basic data for subsequent time correction, and further improve the rationality of the operation and maintenance plan.

[0120] like Figure 6 As shown, Figure 6 The diagram below shows the hardware structure of a flow meter maintenance system in some embodiments of this application. The flow meter maintenance system provided in this application includes a memory 1000 and a processor 2000. The memory 1000 is used to store computer-readable instructions, and the processor 2000 is used to call the computer-readable instructions to execute the flow meter predictive maintenance method as described above.

[0121] The processor 2000 provides computing and control capabilities to control the flowmeter maintenance system to perform corresponding tasks. For example, it controls the flowmeter maintenance system to perform the flowmeter predictive maintenance method in any of the above method embodiments. The method includes: acquiring multi-dimensional parameters of the flowmeter operation, wherein the multi-dimensional parameters include at least flow parameters, conductivity parameters, vibration parameters, environmental parameters, instrument power supply parameters, and medium characteristic parameters; extracting features from the acquired multi-dimensional parameters to obtain static features, dynamic features, and coupling features; inputting the extracted features into a pre-trained fault prediction model, and outputting the electrode fault prediction type and corresponding confidence level. The predicted fault types include electrode water film coverage faults, electrode scaling coverage faults, electrode vibration loosening faults, and electrode corrosion damage faults. Based on the ideal capacitance value and actual equivalent capacitance value of the electrode corresponding to each predicted fault type, capacitance verification is performed on the predicted fault type to obtain the target fault type and target confidence level. Based on the target fault type and corresponding target confidence level, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual working conditions, and the remaining maintenance time is calculated. Based on the remaining maintenance time and target fault type, a targeted maintenance strategy is matched and generated from a preset strategy library.

[0122] The processor 2000 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a hardware chip, or any combination thereof; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The aforementioned PLD can be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0123] The memory 1000, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the flow meter predictive maintenance method in the embodiments of this application. The processor 2000 can implement the flow meter predictive maintenance method in any of the above method embodiments by running the non-transitory software programs, instructions, and modules stored in the memory 1000.

[0124] Specifically, memory 1000 may include volatile memory (VM), such as random access memory (RAM); memory 1000 may also include non-volatile memory (NVM), such as read-only memory (ROM), flash memory, hard disk drive (HDD), solid-state drive (SSD), or other non-transitory solid-state storage devices; memory 1000 may also include combinations of the above types of memory.

[0125] In summary, the flow meter maintenance system of this application adopts the technical solution of any of the above-described flow meter predictive maintenance method embodiments. Therefore, it has at least the beneficial effects brought about by the technical solutions of the above embodiments, which will not be elaborated here.

[0126] This application also provides a computer-readable storage medium, such as a memory including program code, which can be executed by a processor to complete the flow meter predictive maintenance method described in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CDROM), magnetic tape, a floppy disk, or an optical data storage device, etc.

[0127] This application also provides a computer program product comprising one or more lines of program code stored in a computer-readable storage medium. The processor of the early warning system reads the program code from the computer-readable storage medium and executes the program code to complete the steps of the flow meter predictive maintenance method provided in the above embodiments.

[0128] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program or program code related to hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0129] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0130] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0131] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

Claims

1. A predictive maintenance method for flow meters based on artificial intelligence, applied to a flow meter maintenance system, characterized in that, The method includes: The flow meter operates by acquiring multi-dimensional parameters, which include at least one or more of the following: flow parameters, conductivity parameters, vibration parameters, environmental parameters, instrument power supply parameters, and medium characteristic parameters. Feature extraction is performed on the collected multi-dimensional parameters to obtain static features, dynamic features, and coupling features; The extracted features are input into a pre-trained fault prediction model, which outputs the electrode fault prediction type and corresponding confidence level. The electrode fault prediction type includes electrode water film coverage fault, electrode scale coverage fault, electrode vibration loosening fault, and electrode corrosion damage fault. Based on the ideal capacitance value and the actual equivalent capacitance value of the electrode corresponding to the electrode fault prediction type, capacitance verification is performed on the electrode fault prediction type to obtain the target fault type and target confidence level of the electrode. Based on the target fault type and the corresponding target confidence level, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual working conditions and calculate the remaining maintenance time of the fault. Based on the remaining maintenance time of the fault and the target fault type, a targeted maintenance strategy is matched and generated from the preset strategy library.

2. The predictive maintenance method for flow meters as described in claim 1, characterized in that, The static feature refers to the degree to which the parameter deviates from a preset normal range, including the degree of deviation of the parameter. The formula for calculating the degree of deviation is: ; Where δ is the parameter deviation, x is the real-time collected value of the parameter, x0 is the preset standard value of the parameter, x_max is the maximum allowed value of the parameter, and x_min is the minimum allowed value of the parameter; when δ≥20%, it is marked as a high deviation feature.

3. The predictive maintenance method for flow meters as described in claim 1, characterized in that, The dynamic feature is the temporal change trend of the parameter, including the rate of change of the parameter. The rate of change is obtained by calculating the slope of parameter change within adjacent sliding time windows using a linear fitting algorithm. The length of the sliding time window is adaptively adjusted according to the parameter change period.

4. The predictive maintenance method for flow meters as described in claim 1, characterized in that, The coupling feature is the synchronous change correlation between multiple parameters, including the correlation coefficient of parameter pairs. The extracted features are input into a pre-trained fault prediction model, which outputs the electrode fault prediction type and corresponding confidence level, including: The fault type is initially predicted by calling static and dynamic features through the feature selection algorithm, and the candidate electrode fault prediction type is obtained. The target coupling features are obtained by matching the corresponding coupling features according to the candidate electrode fault prediction type; wherein, the target coupling features include the correlation coefficients of the target parameter pairs; By using parallel attention mechanism network layers to assign weights and focus key information on static features, dynamic features and target coupling features respectively, the model inference outputs the fault prediction type and corresponding confidence level of the flow meter electrode.

5. The predictive maintenance method for flow meters as described in claim 4, characterized in that, When there are multiple candidate electrode fault prediction types, the parallel attention mechanism network layer performs weight allocation and key information focusing on static features, dynamic features, and target coupling features, respectively, and then outputs the fault prediction type and corresponding confidence level of the flowmeter electrode through model inference, including: Extract the correlation coefficient r of the coupling feature corresponding to each candidate fault prediction type, and compare it with the preset correlation coefficient threshold r0 of the coupling feature of the fault prediction type. Retain the target coupling feature and the corresponding candidate fault prediction type with r≥r0 to form an effective candidate set. For each candidate fault prediction type in the effective candidate set, its corresponding static features, dynamic features and target coupling features are input into a parallel attention mechanism network layer. The network layer assigns weights to the three types of features and outputs the preliminary confidence level for each candidate type. The candidate fault prediction type with the highest initial confidence level and not lower than the preset value is selected as the final fault prediction type. If the initial confidence level of all candidate types in the effective candidate set is lower than the preset value, the correlation coefficient details of the top two candidate types with the highest confidence level and their target coupling features are output, triggering the manual review process, and the final fault type is confirmed by human.

6. The predictive maintenance method for flow meters as described in claim 1, characterized in that, Based on the ideal capacitance value and actual equivalent capacitance value of the electrode corresponding to the predicted electrode fault type, capacitance verification is performed on the predicted electrode fault type to obtain the target fault type and target confidence level, including: Calculate the percentage relative deviation between the actual equivalent capacitance value and the ideal capacitance value; If the relative deviation percentage is less than the first threshold, the capacitor verification is deemed to have passed, and the electrode fault prediction type is directly determined as the electrode target fault type. If the relative deviation percentage is determined to be greater than or equal to the first threshold and less than the second threshold, it is determined to be an operating condition disturbance. The original prediction type is maintained as the target fault type, and its corresponding confidence level is reduced. If the relative deviation percentage is determined to be greater than or equal to the second threshold, the capacitance verification is deemed to have failed, its corresponding confidence level is reduced to zero, and a manual maintenance mechanism is triggered.

7. The predictive maintenance method for flow meters as described in claim 1, characterized in that, The step of calling a preset fault deterioration model to simulate the evolution path of the fault under the current actual operating conditions based on the target fault type and the corresponding target confidence level, and calculating the remaining maintenance time of the fault, includes: Based on the target fault type, a preset fault deterioration model is invoked to simulate the evolution path of the fault under the current actual working conditions and obtain an initial assessment value of the remaining maintenance time. The remaining maintenance time is obtained by correcting the initial assessment value of the remaining maintenance time with a confidence correction factor. The formula for calculating the corrected remaining maintenance time is as follows: ; Where T is the corrected remaining maintenance time, T0 is the remaining maintenance time initially calculated by the model, α is the confidence correction coefficient, which ranges from 0.3 to 0.8, and γ is the target confidence level.

8. The predictive maintenance method for flow meters as described in claim 7, characterized in that, Based on the target fault type, a preset fault deterioration model is invoked to simulate the fault's evolution path under the current actual operating conditions, obtaining an initial assessment of the remaining maintenance time, including: The fault deterioration model incorporates real-time dynamic adjustment factors when simulating the evolution path. These dynamic adjustment factors include the medium temperature fluctuation coefficient, pressure change coefficient, and environmental humidity influence coefficient. The target evolution curve is obtained by weighted calculation and correction of the evolution curve. The initial assessment value of the remaining maintenance time is determined based on the target evolution curve.

9. The predictive maintenance method for flow meters as described in claim 8, characterized in that, The target evolution curve is obtained by weighted calculation and correction of the evolution curve, including: Preset weights are assigned to the medium temperature fluctuation coefficient k1, pressure change coefficient k2, and environmental humidity influence coefficient k3, respectively. Calculate the overall adjustment factor K using the following formula: ,in They are respectively The corresponding preset weights; The time axis parameter t of the initial fault evolution curve is corrected by the comprehensive adjustment factor K. The correction formula is t' = t × (1 + K). If K is negative, then t' = t × (1 - |K|). The corrected target evolution curve is obtained.

10. A flow meter maintenance system, characterized in that, include: Memory and processor, wherein the memory is used to store program code; The processor is used to call the program code to perform the method as described in any one of claims 1 to 9.

Images