Vortex shedding flowmeter monitoring method and device based on deep learning, medium and product

Abstract

The invention discloses a vortex shedding flowmeter monitoring method and device based on deep learning, a medium and a product, and relates to the field of electrical digital data processing.The method comprises the steps that historical time sequence data of a vortex shedding flowmeter are obtained and preprocessed to obtain standardized data; dividing the standardized data into a training set and a verification set in a preset proportion; according to the flow characteristics of the vortex shedding flowmeter, segmenting the training set according to a preset flow threshold interval, and respectively constructing deep learning models to obtain a flow model set; verifying a plurality of deep learning models in the traffic model set by using the verification set, and optimizing the traffic model set; acquiring real-time data of the vortex shedding flowmeter, and performing model calculation according to the flow threshold interval to which the real-time data belongs to obtain a flow correction value; when the flow meter performance obtained through calculation based on the flow correction value is lower than a preset performance threshold value, abnormal early warning information is generated. According to the vortex shedding flowmeter, the monitoring timeliness of the vortex shedding flowmeter can be improved, and the maintenance cost of the vortex shedding flowmeter is reduced.

Description

Technical Field

[0001] The present application relates to the field of electrical digital data processing, and in particular to a vortex flowmeter monitoring method, device, medium and product based on deep learning. Background Art

[0002] In the industrial production process, the operation status monitoring and performance evaluation of the flowmeter are important links to ensure production safety and measurement accuracy. With the development of industrial automation and intelligent manufacturing, flowmeter monitoring has evolved from traditional periodic inspections to continuous monitoring, which can not only detect measurement anomalies in a timely manner, but also predict equipment failures, which is of great significance to improving production efficiency and reducing maintenance costs. Especially in the fields of chemical industry, oil and gas, the monitoring quality of vortex flowmeters is directly related to the accuracy of process control and material statistics.

[0003] In the related technology, the monitoring method of vortex flowmeter mainly adopts the method of regular calibration and comparison. The staff conducts on-site calibration of the flowmeter in use through a portable standard device, or performs comparative measurement by installing a standard flowmeter in series online. These methods evaluate the performance status of the flowmeter by comparing the deviation between the measured value and the standard value. At the same time, by recording the output signal characteristics of the flowmeter, such as signal amplitude, signal frequency and other parameters, it is judged whether the equipment is in normal working condition.

[0004] However, this periodic detection method has the problem of monitoring lag. Due to the lack of effective monitoring during the detection interval, the flow meter performance degradation or failure is often discovered only after it has affected production quality. This passive way of discovering problems will cause great economic losses. Summary of the invention

[0005] The present application provides a vortex flowmeter monitoring method, device, medium and product based on deep learning, which are used to improve the timeliness of vortex flowmeter monitoring and reduce its maintenance cost.

[0006] In the first aspect, the present application provides a vortex flowmeter monitoring method based on deep learning, which is applied to a monitoring device, and the method includes: obtaining historical time series data of the vortex flowmeter, and preprocessing the historical time series data to obtain standardized data; based on a preset random number seed, using Bernoulli distribution to divide the standardized data into a training set and a verification set of preset proportions; according to the flow characteristics of the vortex flowmeter, the training set is segmented according to a preset flow threshold interval, and a deep learning model is constructed based on the data in each flow threshold interval to obtain a flow model set; multiple deep learning models in the flow model set are verified using the verification set, and the flow model set is optimized based on the verification error results; real-time data of the vortex flowmeter is collected, and according to the flow threshold interval to which the real-time data belongs, the corresponding deep learning model in the flow model set is called for calculation to obtain a flow correction value; when the flow meter performance calculated based on the flow correction value is lower than the preset performance threshold, an abnormal warning information is generated.

[0007] In the above embodiment, the monitoring device realizes intelligent real-time monitoring of the vortex flowmeter through a complete process of acquiring historical data and preprocessing, dividing the data set, building a flow model set, verifying the optimization model, real-time monitoring and early warning. It models different flow intervals based on the deep learning model, can accurately capture the performance characteristics of the vortex flowmeter under different working conditions, and promptly detect abnormal situations, thereby improving the accuracy and timeliness of monitoring.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the steps of obtaining historical time series data of the vortex flowmeter and preprocessing the historical time series data to obtain standardized data specifically include: collecting the original data of the vortex flowmeter within a preset time period, the original data including flow value, temperature value, pressure value and signal frequency; performing outlier processing on the original data, and calculating the local mean, filling the identified outliers based on the local mean to obtain initialization data; segmenting the initialization data according to a preset time window length to obtain multiple data segments; the preset time window length is determined based on the inherent frequency characteristics of the vortex flowmeter; performing standardization processing on each data segment to obtain standardized data; the standardization processing includes maximum and minimum value normalization and Z-score standardization.

[0009] In the above embodiment, the monitoring device performs outlier processing, local mean filling and standardization on the original data, and determines the time window in combination with the inherent frequency characteristics of the vortex flowmeter, thereby ensuring the quality and feature expression capability of the input data.

[0010] In combination with some embodiments of the first aspect, in some embodiments, according to the flow characteristics of the vortex flowmeter, the training set is segmented according to preset flow threshold intervals, and deep learning models are constructed based on the data in each flow threshold interval to obtain the steps of the flow model set, specifically including: based on fluid mechanics theory, the working range of the vortex flowmeter is divided into laminar flow interval, transition interval and turbulent flow interval; in each interval, according to the correspondence between the Reynolds number and the Strouhal number, multiple flow threshold points are determined; the data of the training set is segmented according to the determined flow threshold points to obtain multiple sub-data sets; for each sub-data set, a deep neural network model including a convolutional layer, a recurrent layer and a fully connected layer is constructed to obtain a flow model set; the neural network structure of each interval is set with different convolution kernel sizes and numbers of recurrent units according to different fluid characteristics.

[0011] In the above embodiment, the monitoring device divides the working range into different flow regimes and designs a corresponding neural network structure according to the fluid mechanics characteristics, thereby achieving accurate modeling of the vortex flowmeter under different flow regimes.

[0012] In combination with some embodiments of the first aspect, in some embodiments, before the step of constructing a deep neural network model including a convolutional layer, a recurrent layer and a fully connected layer for each sub-data set to obtain a flow model set, the method also includes: extracting the characteristic matrix of the vortex street signal within each flow threshold interval; performing singular value decomposition on the characteristic matrix to obtain the main eigenvectors; and determining the network structure parameters of each interval based on the distribution characteristics of the main eigenvectors.

[0013] In the above embodiment, the monitoring device extracts the characteristic matrix of the vortex signal and performs singular value decomposition to obtain the main characteristic vectors to guide the network structure design, thereby achieving the optimal configuration of the network parameters.

[0014] In combination with some embodiments of the first aspect, in some embodiments, before using the validation set to validate multiple deep learning models in the flow model set and optimizing the flow model set based on the validation error results, the method also includes: constructing a loss function based on the physical characteristics of the vortex flowmeter; the loss function includes a flow prediction error term, a signal frequency consistency constraint term and a physical law constraint term; the flow prediction error term characterizes the deviation between the model predicted value and the actual value, the signal frequency consistency constraint term characterizes the correlation between the prediction result and the vortex signal frequency, and the physical law constraint term is constructed based on the Bernoulli equation and the continuity equation in fluid mechanics; a cross-validation method is used to evaluate the generalization performance of the flow model set under different data segmentation schemes to obtain a generalization evaluation result; based on the generalization evaluation result, the network structure and hyperparameters of each deep learning model in the flow model set are adjusted and optimized.

[0015] In the above embodiment, the monitoring device optimizes the deep learning model by constructing a loss function containing multiple constraints and using a cross-validation method to evaluate the generalization performance of the model.

[0016] In combination with some embodiments of the first aspect, in some embodiments, after collecting real-time data of the vortex flowmeter and calling the corresponding deep learning model in the flow model set to calculate according to the flow threshold interval to which the real-time data belongs to obtain the flow correction value, the method also includes: constructing a time series-based state assessment model, using the flow correction value, temperature value and pressure value in multiple consecutive time windows as input to determine the equipment state score; probabilistically predicting the potential fault type of the equipment state score based on the historical fault case library to obtain a fault prediction result; when it is determined that the vortex flowmeter has a performance degradation trend according to the equipment state score, generating maintenance time and maintenance recommendations according to the fault prediction result.

[0017] In the above embodiment, the monitoring device realizes dynamic evaluation of equipment status and fault prediction by constructing a status evaluation model and combining it with a historical fault case library.

[0018] In combination with some embodiments of the first aspect, in some embodiments, before collecting real-time data of the vortex flowmeter and calling the corresponding deep learning model in the flow model set to calculate according to the flow threshold interval to which the real-time data belongs to obtain the flow correction value, the method also includes: obtaining the vortex signal frequency sequence and fluid parameter sequence within a preset time window; using Fourier transform to calculate the power spectrum of the vortex signal frequency sequence, and extracting the main frequency component; based on the correspondence between the main frequency component and the fluid parameter sequence, calculating the Strouhal number under the current operating conditions; and dynamically adjusting the flow threshold interval according to the changing trend of the Strouhal number.

[0019] In the above embodiment, the monitoring device realizes dynamic adjustment of the flow threshold range by analyzing the power spectrum characteristics of the vortex signal frequency and the change trend of the Strouhal number.

[0020] In a second aspect, an embodiment of the present application provides a monitoring device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the monitoring device to execute the method described in the first aspect and any possible implementation method of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when the computer program product is run on a monitoring device, enables the monitoring device to execute the method described in the first aspect and any possible implementation of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, comprising instructions, which, when executed on a monitoring device, enables the monitoring device to execute the method described in the first aspect and any possible implementation of the first aspect.

[0023] It is understandable that the monitoring device provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the method provided in the embodiment of the present application. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Due to the use of segmented modeling and real-time monitoring methods based on deep learning, the historical data of the vortex flowmeter is preprocessed and divided into training sets, and a special deep learning model set is constructed according to the flow characteristics. Therefore, accurate modeling and real-time monitoring can be performed for the characteristics of different flow intervals, which effectively solves the problem that a single model in the existing technology is difficult to adapt to all working conditions and the monitoring accuracy is insufficient, thereby realizing intelligent real-time monitoring of the vortex flowmeter performance, improving monitoring accuracy and timeliness, and reducing maintenance costs.

[0025] 2. Due to the adoption of a partition modeling strategy based on fluid mechanics theory, the working range is divided into laminar, transitional and turbulent intervals, and the flow threshold point is determined according to the correspondence between the Reynolds number and the Strouhal number. A special neural network structure is designed for different intervals. Therefore, the differences in fluid characteristics under different flow states can be fully considered, which effectively solves the problem that the existing model is difficult to accurately describe the characteristics of different flow states, thereby realizing the accurate modeling of the vortex flowmeter in the full range of operating conditions, and improving the adaptability and prediction accuracy of the model.

[0026] 3. Due to the use of an adaptive partitioning method based on vortex signal frequency analysis, the main frequency characteristics are extracted through Fourier transform, and the flow threshold interval is dynamically adjusted in combination with the changing trend of the Strouhal number. Therefore, it is possible to adapt to changes in operating conditions in real time and optimize the model partitioning, effectively solving the problem that fixed partitioning in the existing technology is difficult to cope with changes in operating conditions, thereby realizing the adaptive optimization of the flow meter monitoring model and improving the monitoring performance of the model under dynamic conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of a vortex flowmeter monitoring method based on deep learning in an embodiment of the present application; Figure 2 is another flow chart of the vortex flowmeter monitoring method based on deep learning in an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of a physical device of the monitoring device in the embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification of the present application, the singular expressions "one", "a kind of", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more of the listed items.

[0029] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0030] For ease of understanding, the application scenarios of the embodiments of the present application are introduced below.

[0031] The ethylene production line of a large chemical plant uses vortex flowmeters to monitor the flow of raw materials and products. Due to strict process requirements, the accuracy of flow measurement directly affects product quality and production safety. Traditional vortex flowmeters will have problems such as zero drift and range change during long-term operation, and these performance degradations are often gradual and difficult to detect. For example, in the first quarter of a certain year, the factory found that the quality of some products was abnormal. After investigation, it was found that the raw material ratio was imbalanced due to the decrease in flowmeter accuracy, resulting in an economic loss of about one million yuan. In addition, the production stoppage losses and labor costs caused by regular calibration and maintenance are also huge. Therefore, how to achieve real-time monitoring and predictive maintenance of vortex flowmeters has become an urgent problem to be solved in the process industry.

[0032] In the related art, the performance monitoring of vortex flowmeter can be achieved by combining periodic calibration and manual inspection. Specifically, regular calibration is performed through a portable standard device, or the flowmeter performance is evaluated through online comparison, while the operating parameters of the equipment are recorded by manual inspection to determine the working status. The following introduces the scenario of using the vortex flowmeter monitoring method based on deep learning in the related art.

[0033] At present, the chemical plant uses a combination of regular calibration and manual inspection to monitor vortex flowmeters. Every quarter, metering personnel are arranged to carry portable standard devices to conduct on-site calibration of key flowmeters, and manual inspections are carried out twice a week to record the output signal characteristics of the flowmeters. However, this method has obvious defects: first, there is a lack of effective monitoring during the calibration interval, and abnormal performance of the flowmeter is often discovered only after it has affected the process quality; second, manual inspections mainly rely on experience and judgment, and it is difficult to detect minor abnormalities in time; third, the fixed-cycle calibration and maintenance method is neither economical nor flexible enough, and it is impossible to optimize the maintenance strategy according to the actual status of the equipment. For example, during a routine calibration in a certain month of a certain year, it was found that the accuracy of a certain flowmeter was out of tolerance, but through tracing back the production data, it was found that the problem had occurred a month ago.

[0034] By adopting the deep learning-based vortex flowmeter monitoring method in the embodiment of the present application, by constructing a deep learning model set and a multi-objective loss function between partitions, real-time evaluation of flowmeter performance and fault prediction are achieved, which not only improves the timeliness and accuracy of monitoring, but also can predict potential faults and give maintenance suggestions. The following introduces the scenarios in which the deep learning-based vortex flowmeter monitoring method in the present application is used.

[0035] After adopting the deep learning monitoring method of the present application, the monitoring device can collect and analyze the operating data of the vortex flowmeter in real time. The system automatically selects the most suitable deep learning model for calculation according to the working range of the flowmeter to accurately evaluate the status of the equipment. For example, in a certain month of a certain year, the system detected an abnormal fluctuation in the correction coefficient of a flowmeter. By analyzing historical data and physical model predictions, it was determined that the sensor sensitive element may have begun to deteriorate. The system promptly issued an early warning message and recommended maintenance inspections within two weeks. The inspection results confirmed that there was indeed a problem of wear of the sensitive components, and timely replacement avoided possible measurement inaccuracies. In this way, the chemical plant significantly improved the reliability of the flowmeter and reduced measurement-related quality accidents.

[0036] It can be seen that the deep learning-based vortex flowmeter monitoring method in the embodiment of the present application can not only realize real-time monitoring of the flowmeter, but also effectively solve the problems of monitoring lag, high false alarm rate, high maintenance cost and so on in the traditional method, thereby realizing accurate evaluation of equipment status and predictive maintenance.

[0037] For ease of understanding, the following describes the process of the method provided by this implementation in combination with the above scenario. Figure 1 , which is a flow chart of a vortex flowmeter monitoring method based on deep learning in an embodiment of the present application.

[0038] S101, obtaining historical time series data of a vortex flowmeter, and preprocessing the historical time series data to obtain standardized data.

[0039] Among them, historical time series data refers to the data set continuously recorded by the vortex flowmeter at a fixed sampling interval in the past specific time period, including parameters such as flow value, temperature value, pressure value and signal frequency; preprocessing refers to the process of cleaning, denoising and formatting the original data to improve data quality; standardized data refers to the data sequence after normalization, so that parameters of different dimensions can be calculated uniformly.

[0040] Before the monitoring device starts to build a deep learning model, it needs to first obtain sufficient historical data as the basis for model training. Specifically, the monitoring device first establishes a connection with the vortex flowmeter through the data acquisition interface, sets an appropriate sampling time interval (such as 1 second, 5 seconds, etc.), and continuously collects operating data for a preset time period (such as 3 months); then performs outlier detection on the collected raw data, and uses the 3σ criterion to identify and mark abnormal data points; then calculates the mean in the local time window to fill in the outliers; then the processed data is segmented according to the preset time window length (determined based on the inherent frequency characteristics of the vortex flowmeter) to obtain multiple data segments; finally, each data segment is standardized, including maximum and minimum value normalization and Z-score standardization, so that the data distribution is more suitable for the subsequent deep learning model training; the sampling interval refers to the time interval between two data collections, which is usually determined according to the response characteristics of the flowmeter; data quality indicators are used to measure the integrity, accuracy and consistency of the data.

[0041] In some embodiments, data preprocessing and standardization can be achieved in a variety of ways: Optionally, the monitoring device can use a statistical-based method to process data. First, the raw data is smoothed using a moving average method to eliminate high-frequency noise; then the mean μ and standard deviation σ of the data are calculated, and data points outside the range of [μ-3σ, μ+3σ] are marked as outliers; finally, the outliers are corrected using a local linear regression method. Optionally, the monitoring device can also use a method based on wavelet transform. First, the original signal is decomposed at multiple scales to separate different frequency components; then the coefficients at each scale are thresholded to remove noise components; finally, the cleaned signal is obtained through wavelet reconstruction. It is understandable that other data preprocessing methods can also be used, such as principal component analysis, independent component analysis and other signal processing techniques, which are not limited here.

[0042] S102. Based on a preset random number seed, the standardized data is divided into a training set and a validation set of preset proportions using Bernoulli distribution.

[0043] Among them, the preset random number seed represents a fixed value used to initialize the random number generator to ensure the repeatability of data partitioning; Bernoulli distribution refers to a probability distribution with only two possible results, which is used here to randomly determine the distribution of data samples; training set refers to the data subset used for model training; validation set refers to an independent data subset used to evaluate model performance.

[0044] After completing data standardization, the monitoring device needs to divide the data set into two parts: training and verification to ensure the objectivity of model evaluation. Specifically, the monitoring device first sets a fixed random number seed (such as 9191972) to ensure experimental repeatability; then applies Bernoulli distribution to each data sample to generate a binary random variable (1 represents the training set, probability p=0.7; -1 represents the verification set, probability q=0.3); finally, the data samples are assigned to the corresponding sets according to the value of the random variable, and a training set and a verification set with a ratio of 7:3 are obtained.

[0045] In some embodiments, data set division can be achieved in a variety of ways: Optionally, the monitoring device can use a stratified sampling strategy, first divide the data into multiple layers according to the flow value, and sample in each layer according to the Bernoulli distribution to ensure that the divided training set and verification set can contain samples from each flow interval; then check the data distribution similarity of the two sets, and adjust the sampling ratio if necessary; finally, rearrange the data index to obtain the final training set and verification set. Optionally, a time series division method can also be used, using data from an earlier time period for training and data from a later time period for verification, and setting a buffer period between the two time periods to avoid data leakage. It is understandable that other data division methods can also be used, such as random K-fold cross-validation, etc., which are not limited here.

[0046] S103. According to the flow characteristics of the vortex flowmeter, the training set is segmented according to the preset flow threshold intervals, and a deep learning model is constructed based on the data in each flow threshold interval to obtain a flow model set.

[0047] Among them, flow characteristics refer to the measurement characteristics of the vortex flowmeter in different flow ranges; flow threshold interval refers to the different flow state ranges divided according to fluid mechanics characteristics; deep learning model refers to a mathematical model containing multi-layer neural networks; flow model set refers to a collection of multiple special models built for different flow intervals.

[0048] The monitoring device needs to build a special model for the characteristic differences of different flow intervals. Specifically, the monitoring device first divides the working range into laminar interval, transition interval and turbulent interval based on fluid mechanics theory; then, in each interval, the specific flow threshold point is determined according to the correspondence between the Reynolds number and the Strouhal number; then the training set data is segmented according to these threshold points to obtain multiple sub-data sets; finally, a deep neural network model containing convolutional layers, recurrent layers and fully connected layers is constructed for each sub-data set, in which the network structure of each interval sets different convolution kernel sizes and the number of recurrent units according to the different fluid characteristics.

[0049] In some embodiments, model construction can be achieved in a variety of ways: Optionally, the monitoring device can adopt an adaptive modeling strategy, first perform feature analysis on the data of each flow interval, extract key features such as signal frequency features, temperature and pressure features; then automatically adjust the number of network layers and the number of neurons according to the complexity of the features; finally, determine the optimal network structure through iterative optimization. Optionally, an integrated learning method can also be used to construct multiple deep learning models with different structures in each flow interval, and form the final prediction model through weighted combination. It is understandable that other modeling methods can also be used, such as transfer learning, multi-task learning, etc., which are not limited here.

[0050] S104. Use the validation set to validate multiple deep learning models in the traffic model set, and optimize the traffic model set based on the validation error results.

[0051] Among them, the validation error represents the deviation between the model prediction value and the actual value; model optimization refers to adjusting the model structure and parameters according to the validation results; and traffic model set optimization refers to improving the performance of the entire model set.

[0052] The monitoring device needs to evaluate and improve the performance of the constructed model. Specifically, the monitoring device first constructs a loss function based on the physical characteristics of the vortex flowmeter, which includes flow prediction error terms, signal frequency consistency constraints, and physical law constraints; then uses the validation set data to evaluate the performance of the model in each flow interval, and calculates the prediction error and generalization error; then, based on the evaluation results, adjusts the model's network structure and hyperparameters, including the number of layers, number of neurons, activation function, etc.; finally, through multiple rounds of iterative optimization, the model set with the best performance is obtained.

[0053] It should be noted that in the construction of the loss function, the total loss function L consists of three terms: L=α1L1+α2L2+α3L3. Among them, the flow prediction error term L1 uses the improved Huber loss to calculate the deviation between the predicted flow and the actual flow. When the deviation is small, the mean square error form is used to maintain sensitivity, and when the deviation is large, the absolute error form is used to reduce the influence of outliers. The signal frequency consistency constraint term L2 calculates the mutual correlation coefficient between the theoretical frequency corresponding to the predicted flow and the measured vortex signal frequency to ensure that the prediction result is consistent with the actual physical signal. The physical law constraint term L3 is constructed based on the Bernoulli equation and the continuity equation, and the basic laws of fluid dynamics are integrated into the model training process. The weight coefficients α1, α2, and α3 are determined by the grid search method on the validation set to determine the optimal combination. For example, in the vortex flowmeter monitoring of a chemical plant, the optimized weight ratio is 0.5:0.3:0.2. This configuration effectively improves the physical interpretability of the model while ensuring the prediction accuracy.

[0054] S105. Collect real-time data from the vortex flowmeter, and call the corresponding deep learning model in the flow model set to perform calculations based on the flow threshold interval to which the real-time data belongs, to obtain a flow correction value.

[0055] Among them, real-time data refers to the flow, temperature, pressure and other parameter values ​​collected by the vortex flowmeter at the current moment; flow threshold interval judgment refers to determining the working interval to which it belongs based on the real-time flow value; flow correction value refers to the flow value after calibration by the deep learning model.

[0056] After completing model training and optimization, the monitoring device needs to perform online real-time calculations. Specifically, the monitoring device first obtains the measurement data of the vortex flowmeter in real time through the data acquisition interface, including the original flow value, temperature value, pressure value and signal frequency; then preprocesses and standardizes the collected data to meet the model input requirements; then determines the threshold interval to which the current flow value belongs based on the current flow value, and calls the corresponding deep learning model; finally, inputs the preprocessed data into the model, calculates the corrected flow value, and outputs the result to the monitoring system.

[0057] In some embodiments, real-time computing can be achieved in a variety of ways: Optionally, the monitoring device can adopt a sliding window strategy, maintain a certain length of historical data cache, and conduct comprehensive analysis in combination with the current collected values ​​to improve the stability of the calculation results; then optimize the computing efficiency through model quantization and compression technology; finally achieve a millisecond-level response speed. Optionally, a distributed computing framework can also be used to deploy models of different traffic intervals on multiple processing units for parallel computing. It is understandable that other computing methods can also be used, such as edge computing, stream computing, etc., which are not limited here.

[0058] S106. When the flow meter performance calculated based on the flow correction value is lower than a preset performance threshold, generate abnormal warning information.

[0059] Among them, the flowmeter performance index shows the comprehensive evaluation value of the vortex flowmeter measurement accuracy and reliability; the preset performance threshold represents the minimum acceptable performance standard; the abnormal warning information includes the equipment status assessment results and maintenance recommendations.

[0060] The monitoring device needs to evaluate the equipment status in real time and issue warnings in time. Specifically, the monitoring device first builds a status assessment model based on time series, using the flow correction value, temperature value and pressure value in multiple consecutive time windows as input; then calculates various performance indicators, including measurement error, signal stability, response characteristics, etc.; then compares the calculation results with the preset threshold to determine whether there is a performance degradation trend; finally, when an abnormality is detected, the fault type is identified based on the historical fault case library, and warning information containing fault prediction results, maintenance time and maintenance suggestions is generated.

[0061] In some embodiments, the status warning judgment adopts a multi-index comprehensive evaluation mechanism. The system calculates multiple performance indicators based on the flow correction value, including: the root mean square of relative error (RMSE), signal stability index (SSI) and dynamic response characteristics (DRC). After these indicators are normalized, the comprehensive performance index CPI is obtained by weighted summation. The preset performance thresholds are divided into three levels: warning level (CPI < 0.85), alarm level (CPI < 0.75) and emergency level (CPI < 0.65). At the same time, the system will also track the rate of change of CPI. When the rate of change of three consecutive evaluations is negative and the cumulative decline exceeds the preset threshold, even if the CPI has not reached the warning level, it will trigger an early warning. For example, the CPI of a flow meter gradually dropped from 0.92 to 0.87 within a week. Although it has not reached the warning level, due to the obvious continuous downward trend, the system still generates an early warning message to remind the operation and maintenance personnel to pay attention to the status of the equipment.

[0062] It should be noted that the status assessment model adopts a multi-level time series feature extraction architecture. First, the flow correction value, temperature value and pressure value in multiple consecutive time windows are decomposed at multiple scales to obtain the change characteristics at different time scales. Then, the local pattern features are extracted through the convolutional neural network, and the long-term dependencies are captured using the bidirectional long short-term memory network (Bi-LSTM). Finally, the equipment status score is obtained through full connection layer mapping. When training the model, known fault cases in historical operation data are used as labeled samples, and the model parameters are optimized by minimizing the cross entropy loss function. The status score is normalized from 0 to 1, where 1 indicates that the equipment is in the best state and 0 indicates that the equipment has completely failed. For example, when the status score shows a downward trend in three consecutive evaluation cycles and the decline exceeds 20%, the system will trigger a performance degradation warning.

[0063] In the above embodiment, the monitoring device processes the data of different flow intervals through multiple deep learning models respectively, and optimizes them in combination with physical constraints. In practical applications, this interval modeling method can better adapt to the nonlinear characteristics of the vortex flowmeter under different working conditions and improve the prediction accuracy of the model. The following supplements the scenario of this embodiment.

[0064] As the system continues to operate and data accumulates, the performance of the monitoring method is further improved. By analyzing a large number of historical cases, the system has established a complete fault feature library and prediction model. For example, in a certain month, based on historical data analysis, the system found that a certain type of flow meter was prone to zero drift under specific working conditions, so it automatically adjusted the monitoring parameters and warning thresholds under these conditions. At the same time, the system can automatically optimize the maintenance plan according to the equipment status, realizing true predictive maintenance. Half a year after the implementation of this solution, the chemical plant's flow meter unplanned downtime was reduced by 85%, maintenance costs were reduced by 30%, and product quality stability was improved by 25%. In addition, the data and experience accumulated by the system also helped manufacturers optimize product design and further improve the reliability of new products.

[0065] After combining the above scenarios, the following is a more detailed description of the process of the method provided by this implementation. Figure 2 , is another flow chart of the vortex flowmeter monitoring method based on deep learning in an embodiment of the present application.

[0066] S201, obtaining historical time series data of the vortex flowmeter, and preprocessing the historical time series data to obtain standardized data.

[0067] Referring to step S101, the monitoring device acquires the historical operation data of the vortex flowmeter through the data acquisition interface, processes and standardizes the data for outliers, and ensures that the data quality meets the modeling requirements.

[0068] In some embodiments, the monitoring device will perform maximum and minimum value normalization and Z-score standardization on the data, that is, the monitoring device will collect the original data of the vortex flowmeter within a preset time period, the original data including flow value, temperature value, pressure value and signal frequency; perform outlier processing on the original data, and calculate the local mean, fill the identified outliers based on the local mean to obtain initialization data; segment the initialization data according to a preset time window length to obtain multiple data segments; the preset time window length is determined based on the inherent frequency characteristics of the vortex flowmeter; perform standardization processing on each data segment to obtain standardized data; the standardization processing includes maximum and minimum value normalization and Z-score standardization.

[0069] Among them, raw data refers to the unprocessed real-time measurement data collected from the vortex flowmeter, including the flow value, fluid temperature value, pressure value and vortex oscillation signal frequency generated by the fluid flow. The local mean refers to the arithmetic mean of the data within the preset time window. An outlier refers to a data point whose deviation from the local mean exceeds a preset threshold. A data segment refers to a continuous data sequence divided by a fixed time window length. Standardization refers to the mathematical transformation process of converting data to a uniform scale.

[0070] During the normal operation of the vortex flowmeter, the monitoring device needs to continuously collect and process measurement data to evaluate the status of the equipment. Specifically, the monitoring device first continuously collects raw data for a period of time at a preset sampling frequency, and then uses a sliding time window to calculate the local mean, identify and correct abnormal data points. Then, according to the natural oscillation frequency of the vortex flowmeter, the appropriate time window length is determined, and the processed data is divided into segments of equal length. Finally, each data segment is normalized by maximum and minimum values ​​and Z-score standardization to obtain standardized data with a mean of 0 and a variance of 1.

[0071] In some embodiments, data preprocessing can be achieved in the following manner: the monitoring device first uses the median absolute deviation method to detect outliers, calculates the median and deviation distribution of the data, and marks the points that deviate from the median by more than 3 times the standard deviation as outliers. Then, local polynomial regression is used to interpolate and correct the outliers to maintain the temporal continuity of the data. Finally, the main frequency components of the signal are determined based on Fourier analysis, and a time window of not less than 3 vortex oscillation cycles is selected for data segmentation and standardization. Optionally, an anomaly detection method based on kernel density estimation can also be used, combined with a multi-scale analysis method of wavelet transform for data preprocessing. The specific processing parameters need to be optimized and adjusted according to the actual working characteristics of the vortex flowmeter. It is understandable that other data preprocessing methods can also be used to achieve the cleaning and standardization of raw data, which is not limited here.

[0072] S202: Based on a preset random number seed, the standardized data is divided into a training set and a validation set of preset proportions using Bernoulli distribution.

[0073] Referring to step S102, the monitoring device will use a fixed random number seed to initialize the data division process, and divide the standardized data into a training set and a validation set in a ratio of 7:3 through Bernoulli distribution.

[0074] S203. Based on the theory of fluid mechanics, the working range of the vortex flowmeter is divided into laminar flow range, transition range and turbulent flow range.

[0075] Among them, fluid mechanics theory represents the basic physical principles that describe the laws of fluid motion and is used to guide flow state analysis; the working range refers to the Reynolds number range that the vortex flowmeter can effectively measure; the laminar range refers to the state range in which the fluid flows in a layered and orderly manner; the transition range refers to the intermediate state range in which the fluid changes from laminar flow to turbulent flow; the turbulent range refers to the state range in which the fluid exhibits irregular mixed motion; the Reynolds number is a dimensionless parameter that characterizes the ratio of the inertial force to the viscous force of the fluid.

[0076] Before constructing the model in sections, the monitoring device needs to first scientifically divide the flow state based on fluid mechanics theory. Specifically, the monitoring device first calculates the critical Reynolds number based on parameters such as pipe diameter, fluid viscosity and density; then defines the range of Reynolds numbers less than 2300 as the laminar flow interval, at which time the fluid motion presents regular laminar flow; then defines the range of Reynolds numbers between 2300 and 4000 as the transition interval, at which time the fluid motion state is unstable and has intermittent characteristics; then defines the range of Reynolds numbers greater than 4000 as the turbulent flow interval, at which time the fluid motion presents strong three-dimensional random fluctuations; finally, establishes a characteristic parameter database for each interval for subsequent model construction.

[0077] In some embodiments, flow interval division can be achieved in a variety of ways: Optionally, the monitoring device can adopt a multi-parameter comprehensive judgment method, first combining dimensionless parameters such as Reynolds number, Froude number and Weber number to construct a flow state judgment criterion; then refine the interval boundaries by analyzing the velocity pulsation characteristics and pressure fluctuation characteristics of the fluid; finally, establish fuzzy logic rules to achieve a smooth transition of the interval. Optionally, the monitoring device can also adopt a data-driven method, first identifying different characteristic modes of fluid motion through cluster analysis; then using time-frequency analysis methods to extract the main frequency components of each interval; finally, establishing an adaptive interval division standard based on statistical characteristics. It is understandable that other flow state division methods can also be used, such as phase space reconstruction, entropy analysis and other technologies, which are not limited here.

[0078] S204. In each interval, a plurality of flow threshold points are determined according to the corresponding relationship between the Reynolds number and the Strouhal number.

[0079] Among them, the correspondence between the Reynolds number and the Strouhal number represents a mathematical model that describes the relationship between the vortex generation frequency and the fluid motion state; the flow threshold point refers to the key numerical point that distinguishes different flow metering characteristics; the Strouhal number represents the dimensionless ratio of the vortex frequency to the flow velocity; the corresponding relationship curve is used to characterize the changing trend of the vortex generation law under different flow states.

[0080] The monitoring device needs to determine the specific flow threshold point within the divided flow interval to achieve more accurate segmented modeling. Specifically, the monitoring device first collects a large amount of experimental data to establish a database of the corresponding relationship between the Reynolds number and the Strouhal number; then analyzes the change trend of the Strouhal number in each flow interval and identifies the inflection point and mutation point of the curve; then calculates the flow value corresponding to these characteristic points in combination with fluid mechanics theory; then considers the measurement accuracy requirements and adds transition threshold points near the characteristic points; finally, sorts all threshold points and creates an index for subsequent data segmentation.

[0081] It should be noted that the calculation process of the Strouhal number (St) comprehensively considers the influence of temperature and pressure on fluid properties. First, the actual density and viscosity of the fluid are calculated using the state equation based on the temperature and pressure values ​​of the current operating conditions. Then, based on the main frequency component f and the characteristic length d (usually the characteristic size of the flowmeter detection body), the initial Strouhal number St_0 = fd / v is calculated in combination with the flow velocity v. Then, the temperature correction coefficient K_t and the pressure correction coefficient K_p are introduced. These two coefficients are obtained by polynomial fitting, and finally the corrected Strouhal number St = St_0·K_t·K_p is obtained. For example, when the temperature rises, the vortex shedding characteristics will change due to the decrease in fluid viscosity. At this time, the temperature correction coefficient K_t will be less than 1, thereby compensating for the temperature effect.

[0082] S205 , segmenting the data of the training set according to the determined flow threshold points to obtain multiple sub-data sets.

[0083] Among them, the segmentation operation refers to the process of dividing the data set into multiple continuous intervals according to the flow threshold point; the sub-dataset refers to the set of data samples in each flow interval; the flow interval boundary represents the dividing point between adjacent sub-datasets; the data distribution characteristics are used to describe the statistical laws of samples within each sub-dataset; and the data density refers to the distribution density of sample points in each sub-dataset.

[0084] After determining the flow threshold point, the monitoring device needs to scientifically segment the training data for subsequent modeling. Specifically, the monitoring device first sorts the training set data according to the flow value; then divides the data into multiple intervals according to the determined flow threshold point; then checks whether the amount of data in each interval meets the minimum sample number requirement, and adjusts the position of adjacent threshold points if necessary; then analyzes the statistical characteristics of the data in each interval, including the mean, variance, and distribution form; finally, numbers and marks each sub-data set, and establishes a data index for quick access.

[0085] In some embodiments, data segmentation can be achieved in a variety of ways: Optionally, the monitoring device can adopt an adaptive segmentation strategy, first calculate the gradient change of the data near each threshold point; then dynamically adjust the width of the segmentation interval according to the gradient size; finally, ensure that the data volume of each sub-data set is relatively balanced through sample density analysis. Optionally, the monitoring device can also adopt an overlapping segmentation method, first set an overlapping area between adjacent intervals; then use a weighted average method to process the data in the overlapping area; finally, ensure the continuity of the model at the interval boundary through a smooth transition. It is understandable that other data segmentation methods can also be used, such as quantile segmentation, cluster segmentation and other techniques, which are not limited here.

[0086] S206. For each sub-data set, a deep neural network model including a convolutional layer, a recurrent layer, and a fully connected layer is constructed to obtain a traffic model set.

[0087] Among them, the neural network structure of each interval is set with different convolution kernel sizes and numbers of recurrent units according to different fluid characteristics. The deep neural network model represents a machine learning model composed of multiple processing layers; the convolution layer is used to extract the local features and time-frequency characteristics of the signal; the recurrent layer is used to capture the dynamic change law of time series data; the fully connected layer is used to realize the nonlinear mapping of features; the flow model set refers to a collection of multiple special models built for different flow intervals.

[0088] The monitoring device needs to build a dedicated deep learning model for each sub-dataset. Specifically, the monitoring device first analyzes the feature dimensions and time series characteristics of each sub-dataset; then designs the network structure, including determining the convolution kernel size, the type of recurrent unit, and the number of hidden layer nodes; then initializes the model parameters using the Xavier or He initialization method; then selects an appropriate loss function and optimization algorithm, such as mean square error loss and Adam optimizer; finally, it iteratively optimizes the model parameters through the back-propagation algorithm until convergence or the preset training rounds are reached.

[0089] In some embodiments, model construction can be achieved in a variety of ways: Optionally, the monitoring device can adopt a multi-scale modeling strategy, first construct a feature extraction network containing multiple scale convolutional layers; then use a bidirectional LSTM layer to process dependencies at different time scales; finally, enhance the expressiveness of the model through residual connections and attention mechanisms. Optionally, the monitoring device can also adopt an integrated learning method, first construct multiple basic models with different structures for each sub-dataset; then use Bagging or Boosting methods to combine the prediction results of multiple models; finally, determine the optimal integration strategy through cross-validation. It is understandable that other modeling methods, such as transfer learning, multi-task learning and other technologies, can also be used, which are not limited here.

[0090] In some embodiments, the monitoring device will first perform network optimization, that is, the monitoring device will extract the characteristic matrix of the vortex signal within each flow threshold interval; perform singular value decomposition on the characteristic matrix to obtain the main eigenvectors; and determine the network structure parameters of each interval based on the distribution characteristics of the main eigenvectors.

[0091] Among them, the characteristic matrix represents the multidimensional characteristic representation of the vortex signal in the time domain and frequency domain. Singular value decomposition is a mathematical method to decompose a matrix into the product of three matrices. The main eigenvector refers to the eigenvector group corresponding to the largest singular value. The network structure parameters include the number of layers, number of nodes, activation function and other configuration items of the neural network.

[0092] Before building a deep learning model, the monitoring device needs to analyze the data characteristics of different flow intervals to determine the appropriate network structure. Specifically, the monitoring device first extracts the time domain statistical characteristics (mean, variance, skewness, etc.) and frequency domain characteristics (power spectrum, phase spectrum, etc.) of the vortex signal to form a multidimensional feature matrix. Then, through singular value decomposition, dimensionality reduction processing is performed to retain the main eigenvectors with an energy share of more than 95%. Finally, based on the distribution characteristics of the eigenvectors, the structural parameters of the neural network for each flow interval are determined, including the size of the convolution kernel, the number of cyclic units, etc.

[0093] In some embodiments, feature extraction and network design can be achieved in the following ways: first, the short-time Fourier transform of the vortex signal is calculated using the sliding window method to obtain a time-frequency spectrum. Then, the time-frequency spectrum is subjected to singular value decomposition to analyze the variation pattern of the eigenvector with frequency. According to the main distribution range of the eigenvector, the receptive field size and the number of filters of the convolutional layer are determined, and the number of memory units of the recurrent layer is determined according to the time series correlation. Optionally, wavelet transform can be used to extract multi-scale features, and principal component analysis can be combined to perform feature dimensionality reduction and network structure optimization. The optimal network configuration is determined by repeated experimental verification. It is understandable that other feature engineering methods can also be used to guide the structural design of the neural network, which is not limited here.

[0094] S207. Use the validation set to validate multiple deep learning models in the traffic model set, and optimize the traffic model set based on the validation error results.

[0095] Referring to step S104, the monitoring device will use the validation set data to evaluate the prediction performance of each deep learning model, optimize and adjust the model structure and parameters according to the validation error results, and improve the generalization ability of the model.

[0096] In some embodiments, the monitoring device will dynamically evaluate and adjust the flow threshold interval before calling the deep learning model to ensure that the model is suitable for the current operating conditions, that is, the monitoring device will construct a loss function based on the physical characteristics of the vortex flowmeter; the loss function includes a flow prediction error term, a signal frequency consistency constraint term, and a physical law constraint term; the flow prediction error term characterizes the deviation between the model predicted value and the actual value, the signal frequency consistency constraint term characterizes the correlation between the prediction result and the vortex signal frequency, and the physical law constraint term is constructed based on the Bernoulli equation and the continuity equation in fluid mechanics; a cross-validation method is used to evaluate the generalization performance of the flow model set under different data segmentation schemes to obtain a generalization evaluation result; based on the generalization evaluation result, the network structure and hyperparameters of each deep learning model in the flow model set are adjusted and optimized.

[0097] Among them, the loss function represents a mathematical expression to measure the prediction effect of the model. The flow prediction error term is used to quantify the degree of deviation between the predicted value and the actual value. The consistency constraint term reflects the degree of correlation between the prediction result and the physical signal. The physical law constraint term ensures that the model prediction result conforms to the basic laws of fluid mechanics. Generalization performance refers to the model's ability to predict unseen data.

[0098] During the model training and optimization phase, the monitoring device needs to construct a suitable loss function and evaluate the model performance. Specifically, the monitoring device first constructs the flow prediction error term based on the mean square error, and then introduces the signal frequency correlation constraint and the fluid mechanics equation constraint to form a multi-objective loss function. Then, the K-fold cross-validation method is used to evaluate the generalization ability of the model using different data partitioning schemes. Finally, according to the verification results, the network's structural parameters and training hyperparameters, such as learning rate and batch size, are adjusted.

[0099] In some embodiments, model optimization can be achieved in the following manner: first, a weighted multi-objective loss function is constructed, the prediction error term uses Huber loss to reduce the impact of outliers, the consistency constraint uses the correlation coefficient, and the physical constraint uses the dimensionless Bernoulli equation residual. Then, the grid search method is used to optimize the weight coefficient of the loss function, and strategies such as learning rate decay and early stopping are used to avoid overfitting. Based on the performance indicators of the model on the validation set, such as mean relative error, R square value, etc., the optimal model configuration is selected. It is understandable that other model optimization methods can also be used to improve prediction accuracy, which is not limited here.

[0100] S208, obtaining a vortex signal frequency sequence and a fluid parameter sequence within a preset time window.

[0101] Among them, the preset time window represents the length of a fixed time period for signal analysis; the vortex signal frequency sequence refers to the vortex shedding frequency data collected continuously over a period of time; the fluid parameter sequence includes the time series of physical quantities such as temperature and pressure; the sampling frequency refers to the time interval for signal collection; and data synchronization represents the temporal correspondence between different parameter sequences.

[0102] The monitoring device needs to obtain the equipment operation data in real time for online analysis. Specifically, the monitoring device first determines the appropriate time window length according to the dynamic characteristics of the vortex flowmeter; then synchronously collects the vortex signal and fluid parameters through the high-speed data acquisition module; then performs time alignment and synchronization processing on the collected raw data; then checks the integrity and continuity of the data, and performs interpolation and completion when necessary; finally, the processed data sequence is organized and stored in a predetermined format for subsequent analysis.

[0103] In some embodiments, data collection and preprocessing can be achieved in a variety of ways: Optionally, the monitoring device can adopt an adaptive sampling strategy, first monitor the rate of change of the signal; then dynamically adjust the sampling frequency according to the rate of change; and finally ensure data quality through digital filtering. Optionally, the monitoring device can also adopt a multi-cache mechanism, first set up a multi-level data buffer; then implement pipeline processing of the data; and finally ensure data accuracy through real-time verification. It is understandable that other data collection methods can also be used, such as distributed collection, edge computing and other technologies, which are not limited here.

[0104] S209, using Fourier transform to calculate the power spectrum of the vortex signal frequency sequence, and extracting the main frequency component.

[0105] Among them, Fourier transform refers to the mathematical tool for converting time domain signals into frequency domain; power spectrum refers to the representation of signal energy distribution in the frequency domain; the main frequency component represents the frequency component with the strongest energy in the power spectrum; frequency resolution refers to the minimum frequency interval that can be distinguished; and the spectral leakage effect refers to the phenomenon of spectrum energy diffusion caused by limited sampling length.

[0106] The monitoring device needs to perform spectrum analysis on the vortex signal to obtain the characteristic frequency. Specifically, the monitoring device first performs windowing on the signal sequence and selects a suitable window function (such as Hanning window) to reduce spectral leakage; then performs fast Fourier transform on the windowed signal to obtain a complex spectrum; then calculates the modulus square of the spectrum to obtain the power spectrum density function; then identifies the significant peak in the power spectrum through the peak detection algorithm; finally, determines the main frequency component that characterizes the vortex shedding characteristics based on the energy size and frequency position of the peak.

[0107] In some embodiments, spectrum analysis can be implemented in a variety of ways: Optionally, the monitoring device can use a short-time Fourier transform method, first dividing the signal into multiple overlapping short-time windows; then performing Fourier transform on each window; and finally analyzing the time-varying characteristics of the main frequency through a time-frequency diagram. Optionally, the monitoring device can also use a wavelet analysis method, first selecting an appropriate wavelet basis function; then performing multi-scale wavelet decomposition; and finally extracting characteristic frequencies through wavelet energy distribution. It is understandable that other spectrum analysis methods can also be used, such as Hilbert transform, empirical mode decomposition and other techniques, which are not limited here.

[0108] S210. Calculate the Strouhal number under the current working condition based on the correspondence between the main frequency component and the fluid parameter sequence.

[0109] Among them, the operating conditions represent the actual operating state of the fluid, including parameters such as temperature and pressure; the Strouhal number is a dimensionless parameter that characterizes the vortex frequency characteristics; the main frequency-flow velocity relationship refers to the functional relationship between the vortex frequency and the fluid velocity; the parameter correction coefficient is used to compensate for the influence of temperature and pressure on the measurement; the calculation accuracy represents the measurement accuracy of the Strouhal number.

[0110] The monitoring device needs to calculate the Strouhal number in real time to characterize the vortex street characteristics. Specifically, the monitoring device first obtains the fluid physical parameters under the current working conditions, including density, viscosity, etc.; then calculates the characteristic velocity based on the main frequency component and fluid parameters; then considers the influence of temperature and pressure on the fluid physical properties and performs parameter correction; then calculates the Strouhal number using the corrected parameters and main frequency component; finally, performs uncertainty analysis on the calculation results to evaluate the measurement accuracy.

[0111] In some embodiments, the Strouhal number calculation can be implemented in a variety of ways: Optionally, the monitoring device can use a physical property compensation method to first establish a mapping relationship between temperature pressure and fluid physical properties; then update the correction coefficient in real time; and finally obtain an accurate Strouhal number through iterative calculation. Optionally, the monitoring device can also use a neural network method to first train a prediction model that includes the influence of physical properties; then input the current operating parameters; and finally directly output the compensated Strouhal number. It is understandable that other calculation methods can also be used, such as table lookup interpolation, polynomial fitting and other techniques, which are not limited here.

[0112] S211. Dynamically adjust the flow threshold interval according to the changing trend of the Strouhal number.

[0113] Among them, the change trend represents the evolution law of the Strouhal number over time; dynamic adjustment refers to the adaptive update of the threshold interval according to real-time data; the adjustment amplitude refers to the change range of the threshold point position; the stability criterion is used to evaluate the necessity of adjustment; the threshold update cycle represents the time interval for performing the adjustment operation.

[0114] The monitoring device needs to optimize the flow interval division according to the actual operating status. Specifically, the monitoring device first calculates the rate of change of the Strouhal number within a certain time window; then analyzes whether the change trend exceeds the preset fluctuation range; then determines whether the current threshold interval is still applicable; then determines whether the threshold point position needs to be moved based on the judgment result; finally, when it is confirmed that adjustment is needed, the new threshold point coordinates are calculated and the interval division is updated.

[0115] In some embodiments, dynamic adjustment can be achieved in a variety of ways: Optionally, the monitoring device can use a sliding window analysis method, first maintain a fixed-length historical data window; then calculate the statistical features within the window; and finally determine the adjustment strategy based on the feature changes. Optionally, the monitoring device can also use a fuzzy control method, first build a fuzzy rule base based on the rate of change; then evaluate the necessity and magnitude of the adjustment in real time; and finally obtain the specific adjustment amount through defuzzification. It is understandable that other adjustment methods can also be used, such as adaptive control, state observer and other technologies, which are not limited here.

[0116] It should be noted that the dynamic adjustment of the flow threshold interval is based on the real-time monitoring results of the Strouhal number. The system maintains a sliding time window (usually 24 hours) and calculates the mean μ and standard deviation σ of the Strouhal number within the window. When the trend of the Strouhal number exceeds the range of μ±2σ, the threshold adjustment mechanism is triggered. The adjustment process uses an adaptive step size algorithm, and the step size is proportional to the degree of deviation, but there is a maximum adjustment limit to ensure system stability. The new threshold position is determined by an exponential smoothing method to avoid the impact of sudden changes on model predictions. For example, when the temperature of a flow meter gradually increases, the system detects an overall shift in the Strouhal number, and then adjusts the threshold point of the corresponding flow interval upward by 5%, so that the model always works within the most suitable range.

[0117] S212. Collect real-time data from the vortex flowmeter, and call the corresponding deep learning model in the flow model set to perform calculations based on the flow threshold interval to which the real-time data belongs, to obtain a flow correction value.

[0118] Referring to step S105, the monitoring device will collect the operating data of the vortex flowmeter in real time, determine the flow threshold interval to which the data belongs, and call the corresponding deep learning model to perform flow calculation and correction.

[0119] In some embodiments, the monitoring device will establish an equipment status assessment model based on the continuous monitoring data, perform fault prediction in combination with historical fault cases, and generate corresponding maintenance recommendations. That is, the monitoring device will construct a status assessment model based on time series, using the flow correction value, temperature value and pressure value in multiple consecutive time windows as input to determine the equipment status score; perform probability prediction on the potential fault type of the equipment status score based on the historical fault case library to obtain a fault prediction result; when it is determined that the vortex flowmeter has a performance degradation trend based on the equipment status score, generate maintenance time and maintenance recommendations based on the fault prediction result.

[0120] Among them, the state assessment model refers to the mathematical model used to evaluate the operating state of the equipment. The state score is a numerical indicator that quantifies the performance level of the equipment. The fault prediction results include potential fault types and their probability of occurrence. The performance degradation trend refers to the changing pattern of the equipment performance gradually decreasing over time.

[0121] In the real-time monitoring stage, the monitoring device needs to continuously evaluate the equipment status and predict potential failures. Specifically, the monitoring device first builds a status assessment model based on time series, using flow correction values, temperature values, and pressure values ​​in multiple time windows as input features. Then, based on the characteristic patterns of historical failure cases, a probabilistic statistical method is used to predict the type of failure that may occur in the current state. Finally, based on the changing trend of the status score and combined with the failure prediction results, targeted maintenance recommendations and optimal maintenance time are generated.

[0122] In some embodiments, state assessment and fault prediction can be achieved in the following manner: first, a state assessment model based on a long short-term memory network is constructed, multi-scale time series features are extracted, and state scores are calculated. Then, a Gaussian mixture model is used to perform cluster analysis on historical fault cases and establish a fault feature library. Based on the similarity between the current state score and the fault feature, the probability of occurrence of various types of faults is calculated and the maintenance strategy is determined. Optionally, other machine learning methods such as support vector machines can also be used for state assessment and fault diagnosis. It is understandable that other state monitoring methods can also be used to achieve equipment health management, which is not limited here.

[0123] S213: When the flow meter performance calculated based on the flow correction value is lower than a preset performance threshold, generate abnormal warning information.

[0124] Referring to step S106, the monitoring device evaluates the performance status of the flow meter based on the flow correction value, and generates abnormal warning information including the fault type, severity and processing suggestions when the performance index is lower than the preset threshold.

[0125] In the embodiments of the present application, due to the adoption of a partitioned interval modeling strategy based on fluid mechanics characteristics, a special deep learning model is constructed according to the different characteristics of laminar, transitional and turbulent intervals; a multi-objective loss function design integrating physical constraints is used, and the laws of fluid mechanics are used as constraints for model training; a state evaluation and fault prediction method based on time series analysis is used to achieve accurate quantification of equipment status and early warning of faults. Therefore, it is possible to maintain a high monitoring accuracy under different working conditions, effectively solving the problems of monitoring lag, high false alarm rate, and low maintenance efficiency in traditional methods. Four important goals are achieved: the monitoring timeliness of the vortex flowmeter is improved, the equipment maintenance cost is reduced, the metering reliability is improved, and predictive maintenance is achieved.

[0126] The monitoring device in the embodiment of the present invention is described below from the perspective of hardware processing. Figure 3 , is a schematic diagram of a physical device structure of a monitoring device in an embodiment of the present application.

[0127] It should be noted that Figure 3 The structure of the monitoring device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0128] like Figure 3 As shown, the monitoring device includes a CPU 301, which can perform various appropriate actions and processes according to the program stored in the ROM 302 or the program loaded from the storage part 308 into the RAM 303, such as executing the method described in the above embodiment. In the RAM 303, various programs and data required for system operation are also stored. The CPU 301, the ROM 302 and the RAM 303 are connected to each other via a bus 304. The I / O interface 305 is also connected to the bus 304.

[0129] The following components are connected to the I / O interface 305: an input section 306 including an audio input device, a button switch, etc.; an output section 307 including a liquid crystal display (LCD) and an audio output device, an indicator light, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed so that a computer program read therefrom is installed into the storage section 308 as needed.

[0130] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 309, and / or installed from the removable medium 311. When the computer program is executed by the CPU 301, various functions defined in the present invention are executed.

[0131] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings.

[0132] Specifically, the monitoring device of this embodiment includes a processor and a memory. A computer program is stored in the memory. When the computer program is executed by the processor, the deep learning-based vortex flowmeter monitoring method provided in the above embodiment is implemented.

[0133] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the monitoring device described in the above embodiment; or may exist independently without being assembled into the monitoring device. The above storage medium carries one or more computer programs, and when the above one or more computer programs are executed by a processor of the monitoring device, the monitoring device implements the deep learning-based vortex flowmeter monitoring method provided in the above embodiment.

[0134] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0135] As used in the above embodiments, the term "when..." may be interpreted to mean "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrases "upon determining..." or "if (the stated condition or event) is detected" may be interpreted to mean "if determining..." or "in response to determining..." or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)", depending on the context.

Claims

1. A vortex flowmeter monitoring method based on deep learning, characterized in that: Applied to a monitoring device, the method comprises: Acquire historical time series data of the vortex flowmeter, and preprocess the historical time series data to obtain standardized data; Based on a preset random number seed, the standardized data is divided into a training set and a validation set of preset proportions using Bernoulli distribution; According to the flow characteristics of the vortex flowmeter, the training set is segmented according to the preset flow threshold intervals, and a deep learning model is constructed based on the data in each flow threshold interval to obtain a flow model set; Using the verification set to verify multiple deep learning models in the traffic model set, and optimizing the traffic model set based on the verification error results; Collecting real-time data of the vortex flowmeter, and calling the corresponding deep learning model in the flow model set to perform calculation according to the flow threshold interval to which the real-time data belongs, to obtain a flow correction value; When the flow meter performance calculated based on the flow correction value is lower than a preset performance threshold, abnormal warning information is generated.

2. The method according to claim 1, characterized in that The step of obtaining the historical time series data of the vortex flowmeter and preprocessing the historical time series data to obtain standardized data specifically includes: Collecting raw data of the vortex flowmeter within a preset time period, wherein the raw data includes flow value, temperature value, pressure value and signal frequency; Performing outlier processing on the original data, calculating a local mean, and filling the identified outliers based on the local mean to obtain initialization data; Segmenting the initialization data according to a preset time window length to obtain a plurality of data segments; the preset time window length is determined based on the natural frequency characteristics of the vortex flowmeter; Each of the data segments is subjected to standardization processing to obtain the standardized data; the standardization processing includes maximum and minimum value normalization and Z-score standardization.

3. The method according to claim 1, characterized in that The step of segmenting the training set according to the preset flow threshold intervals according to the flow characteristics of the vortex flowmeter, and constructing a deep learning model based on the data in each flow threshold interval to obtain a flow model set specifically includes: Based on the theory of fluid mechanics, the working range of the vortex flowmeter is divided into a laminar flow range, a transitional flow range and a turbulent flow range; In each interval, multiple flow threshold points are determined according to the corresponding relationship between Reynolds number and Strouhal number; Segmenting the data of the training set according to the determined flow threshold points to obtain multiple sub-data sets; A deep neural network model consisting of convolutional layers, recurrent layers and fully connected layers is constructed for each sub-dataset to obtain a flow model set; the neural network structure of each interval is set with different convolution kernel sizes and numbers of recurrent units according to the different fluid characteristics.

4. The method according to claim 3, characterized in that Before the step of constructing a deep neural network model including a convolutional layer, a recurrent layer and a fully connected layer for each sub-data set to obtain a traffic model set, the method further includes: Extract the characteristic matrix of vortex signal within each flow threshold interval; Performing singular value decomposition on the characteristic matrix to obtain main eigenvectors; The network structure parameters of each interval are determined according to the distribution characteristics of the main eigenvectors.

5. The method according to claim 1, characterized in that: Before the step of using the validation set to validate the multiple deep learning models in the traffic model set and optimizing the traffic model set based on the validation error result, the method further includes: A loss function is constructed based on the physical characteristics of the vortex flowmeter; the loss function includes a flow prediction error term, a signal frequency consistency constraint term and a physical law constraint term; the flow prediction error term represents the deviation between the model prediction value and the actual value, the signal frequency consistency constraint term represents the correlation between the prediction result and the vortex signal frequency, and the physical law constraint term is constructed based on the Bernoulli equation and the continuity equation in fluid mechanics; A cross-validation method is used to evaluate the generalization performance of the traffic model set under different data segmentation schemes to obtain a generalization evaluation result; Based on the generalization evaluation results, the network structure and hyperparameters of each deep learning model in the traffic model set are adjusted and optimized.

6. The method according to claim 1, characterized in that After the step of collecting the real-time data of the vortex flowmeter and, according to the flow threshold interval to which the real-time data belongs, calling the corresponding deep learning model in the flow model set to perform calculations to obtain the flow correction value, the method further includes: Construct a time series-based condition assessment model, using flow correction values, temperature values, and pressure values ​​in multiple consecutive time windows as input to determine the equipment condition score; Probabilistically predicting the potential fault types of the equipment status scores based on the historical fault case library to obtain a fault prediction result; When it is determined according to the equipment status score that the vortex flowmeter has a performance degradation trend, a maintenance time and a maintenance suggestion are generated according to the fault prediction result.

7. The method according to claim 1, characterized in that Before the step of collecting the real-time data of the vortex flowmeter and, according to the flow threshold interval to which the real-time data belongs, calling the corresponding deep learning model in the flow model set to perform calculation to obtain the flow correction value, the method further includes: Obtaining the vortex signal frequency sequence and fluid parameter sequence within a preset time window; Calculating the power spectrum of the vortex signal frequency sequence by Fourier transform, and extracting the main frequency component; Calculating the Strouhal number under the current working condition based on the corresponding relationship between the main frequency component and the fluid parameter sequence; The flow threshold interval is dynamically adjusted according to the change trend of the Strouhal number.

8. A monitoring device, characterized in that: The monitoring device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the monitoring device to execute the method described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on the monitoring device, the monitoring device is caused to execute the method as claimed in any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is executed on a monitoring device, the monitoring device is caused to execute the method according to any one of claims 1 to 7.

Images