System and method for integrating intelligent fault diagnosis refueling machine

Abstract

The invention relates to the field of intelligent fault diagnosis, in particular to a system and method for integrating intelligent fault diagnosis of an oiling machine, and the method comprises the following steps: S1, integrating operation data obtained by sensors of an oil way, a circuit and associated equipment of the oiling machine in real time, and constructing a distributed framework based on a Spark core component and a Hadoop distributed file system; data storage is carried out through an HDFS; s2, continuously collected operation data of the oiling machine are input into the fault diagnosis model, and real-time anomaly detection is carried out; s3, analyzing the operation data through the fault prediction model, and predicting the fault probability of the oiling machine in the long-term change trend or the short-term change trend of the operation data; and S4, performing graded early warning according to the detected or predicted fault condition. According to the invention, the fault detection accuracy can be improved, real-time monitoring and early warning are realized, diagnosis and maintenance are facilitated, and the operation management level of the gas station is improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent fault diagnosis, and in particular to a system and method for integrated intelligent fault diagnosis of a fuel dispenser. Background Art

[0002] In modern society, as the core equipment of gas stations, the stable operation of gas dispensers is crucial to ensuring energy supply and traffic safety. Traditional gas dispenser fault detection mainly relies on manual inspection and experience judgment, which has many disadvantages. On the one hand, manual inspection is inefficient and it is difficult to monitor the operating status of the gas dispenser in real time. It can often be handled only after the fault occurs, resulting in interruption of gas station operations and affecting customer experience and economic benefits. On the other hand, the accuracy of experience judgment is limited, and complex fault conditions may not be accurately identified, which can easily lead to misdiagnosis or missed diagnosis. As the functions of gas dispensers continue to increase, their structure has become more complex, involving oil circuits, circuits, and multiple related equipment. In terms of oil circuits, abnormal parameters such as pipeline pressure, oil and gas concentration, oil pump flow and flow rate may indicate the occurrence of faults. In the circuit part, the working status of motors, solenoid valves, and circuit boards plays a key role in the normal operation of gas dispensers. In addition, the failure of equipment such as encoders, printers, IC card readers, and keyboard boards will also affect the overall performance of gas dispensers.

[0003] For the gas pumps currently used on the market, when a gas pump has a malfunction, the station manager first calls the manufacturer for consultation and describes the approximate malfunction of the gas pump, and then sends maintenance personnel to the gas station for on-site inspection. Sometimes the maintenance personnel do not carry enough spare parts, soft faults, or cannot find the problem points, making it difficult to repair in a short period of time. The manufacturer's maintenance personnel also have varying levels of judgment on the faults, affecting the normal operation of the gas station. Summary of the invention

[0004] To solve the above problems, the present invention provides a system and method for integrated fault intelligent diagnosis of fuel dispensers, which can improve the accuracy of fault detection, realize real-time monitoring and early warning, facilitate diagnosis and maintenance, and improve the operation and management level of gas stations.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A method for integrated fault intelligent diagnosis of a fuel dispenser comprises the following steps:

[0007] S1: Integrate the real-time operating data obtained by sensors of the fuel circuit, circuit and related equipment of the fuel dispenser, and use the distributed framework built based on Spark core components and Hadoop distributed file system to store data through HDFS;

[0008] S2: Input the continuously collected operation data of the fuel dispenser into the fault diagnosis model to perform real-time anomaly detection;

[0009] S3: Analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data;

[0010] S4: Provide graded warnings based on detected or predicted fault conditions.

[0011] Furthermore, the operating data of the oil circuit includes: pipeline pressure in the hydraulic area of ​​the fuel dispenser, oil and gas concentration in the space, flow rate of the oil pump and flow meter, pipeline and filter;

[0012] The circuit operation data includes: current values ​​and voltage values ​​of the motor, solenoid valve and circuit board in the fuel dispenser computer box;

[0013] The circuit board includes: an APP board, a main control board, a key display board, a power board, an IC card board and a driver board;

[0014] The associated equipment operation data includes: the communication status of the fuel dispenser application panel, the microprocessor code in the main control board, and the working status of the encoder, printer, IC card reader and keyboard board;

[0015] The integration includes: performing data fusion through Kalman filtering after data standardization and time synchronization.

[0016] Furthermore, the method of continuously collecting the operating data of the fuel dispenser and inputting it into the fault diagnosis model for real-time abnormality detection includes the following steps:

[0017] S21: According to the operation data characteristics and fault diagnosis requirements of the fuel dispenser, and based on the fuzzy rule base established according to expert experience and historical fault data, the defuzzified output value is obtained through the fuzzy logic algorithm;

[0018] S22: using the output of the fuzzy logic algorithm as input data and the preprocessed raw data as input features for further analysis and diagnosis through the random forest algorithm;

[0019] S23: Verify the accuracy and reliability through test set data or actual operation data, evaluate the fault diagnosis model regularly, and optimize the model based on the evaluation results.

[0020] Furthermore, the method of obtaining the defuzzified output value by using a fuzzy logic algorithm based on the operation data characteristics and fault diagnosis requirements of the fuel dispenser and the fuzzy rule base established based on expert experience and historical fault data includes the following steps:

[0021] S211: According to the operating characteristics and fault characteristics of the fuel dispenser, and based on expert experience or historical data, a fuzzy logic rule is defined, wherein the fuzzy set of input variables is the operating parameters of the fuel dispenser, and the fuzzy set of output variables is the fuzzy description of the fault type or fault degree;

[0022] S212: Calculate the degree to which the input data belongs to each fuzzy set through the membership function;

[0023] S213: According to the defined fuzzy logic rules, the fuzzified input data is inferred to obtain a fuzzy output result, and during the inference process, the membership degree of the input data and the weight of the rule are calculated to obtain a fuzzy set of output variables;

[0024] S214: Defuzzifying the fuzzy output result by the centroid method to obtain the probability distribution of the fault type or the output numerical representation of the fault degree.

[0025] Furthermore, the output result of the fuzzy logic algorithm is used as input data, and the pre-processed raw data is used as input features, and further analysis and diagnosis are performed through the random forest algorithm; the following steps are included:

[0026] S221: using the output result of the fuzzy logic algorithm as input data of the random forest algorithm, and using the preprocessed raw data as input features of the random forest algorithm;

[0027] S222: randomly selecting data as a training set, and training the random forest model with the training set data, constructing each decision tree by randomly sampling and feature selecting the training data, and performing splitting by a recursive method during the construction process until a stop condition is met;

[0028] S223: Input the input data into the random forest model, determine the fault type by majority voting according to the prediction results of each decision tree, and determine the fault degree by averaging.

[0029] Furthermore, the method of analyzing the operating data by using a fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operating data comprises the following steps:

[0030] S31: Based on the time series analysis results, a long-term trend prediction model is constructed through the time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data;

[0031] S32: Build a short-term trend prediction model through machine learning algorithms to handle large-scale data and nonlinear relationships;

[0032] S33: The prediction results of the two algorithms are fused by weighted average method to construct a comprehensive fault prediction model.

[0033] Furthermore, the method of constructing a long-term trend prediction model based on the time series analysis results by using a time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data includes the following steps:

[0034] S311: determining the range of autoregressive moving average model parameters by observing the autocorrelation function and the partial autocorrelation function, and determining the best parameter combination by grid search;

[0035] S312: The autoregressive moving average model with determined parameters is trained using training data. The autoregressive moving average model formula is:

[0036] Where: y t is the time series data, c is the constant term, φ i is the autoregressive coefficient, θ j is the moving average coefficient, ∈ t is a white noise sequence, p is the autoregressive order, and q is the moving average order;

[0037] S313: Input the operating data into the autoregressive moving average model, and predict the future long-term trend based on the autoregressive, differential and moving average relationship of the historical data.

[0038] Furthermore, the short-term change trend prediction model is constructed by machine learning algorithm to process large-scale data and nonlinear relationships; including the following steps:

[0039] S321: extracting statistical features, time domain features, and frequency domain features from the preprocessed raw data, performing dimensionality reduction through principal component analysis, calculating mutual information between the features and the fault labels, and retaining the features of the mutual information;

[0040] S322: transforming the fuel dispenser fault prediction into a binary classification problem, i.e., normal state and fault state, through a support vector machine algorithm;

[0041] S323: Determine the radial basis function as the kernel function of the support vector machine, and determine the penalty coefficient and kernel parameter by a cross-validation method;

[0042] S324: The preprocessed raw data is subjected to feature engineering to extract features and then input into the support vector machine model. The model learns the short-term change pattern in the data based on the input features, and the output result is the short-term prediction value of the fuel dispenser operation data in the future.

[0043] Furthermore, the step of providing a graded warning according to the detected or predicted fault conditions includes the following steps:

[0044] S41: determining key indicators for fault classification based on the operation data and common fault types of the fuel dispenser, and classifying the fault levels into primary fault, secondary fault and tertiary fault according to the determined fault indicators;

[0045] S42: Different push configurations are performed according to the fault level, and different push channels are established for the gas station manager and the manufacturer's maintenance personnel: when a level 1 fault is detected, the warning information is pushed to the gas station manager; when a level 2 or 3 fault is detected, it is pushed to the gas station manager and the manufacturer's maintenance personnel at the same time.

[0046] A system for integrated fault intelligent diagnosis of a fuel dispenser, comprising a data acquisition module, a fault diagnosis module, a fault prediction module and an early warning module connected in communication, wherein:

[0047] The data acquisition module is used to obtain real-time operating data from the fuel circuit, circuit and associated equipment sensors of the fuel dispenser, and integrate and store the data;

[0048] The fault diagnosis module is used to continuously collect the operation data of the fuel dispenser and input it into the fault diagnosis model for real-time anomaly detection;

[0049] The fault prediction module is used to analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data;

[0050] The early warning module is used to issue graded early warnings based on detected or predicted fault conditions, and maintenance personnel can remotely view and make repair decisions based on the early warning conditions.

[0051] The beneficial effects of the present invention are:

[0052] 1. Through a variety of sensors, the data of the fuel circuit, circuit and related equipment of the fuel dispenser can be fully collected, which can reflect the operation of the fuel dispenser from multiple angles, avoid misjudgment that may be caused by a single data source, and greatly improve the accuracy of fault detection. Continuous data collection makes it possible to monitor the operation status of the fuel dispenser in real time, understand the working status of each component of the fuel dispenser at any time, detect abnormal conditions in time and take corresponding measures.

[0053] 2. The fault diagnosis model built by using a distributed architecture combined with two complementary algorithms can quickly and accurately diagnose the fault type and location of the fuel dispenser. Accurate fault diagnosis allows maintenance personnel to understand the fault situation in advance and prepare the corresponding maintenance tools and spare parts, thereby greatly improving maintenance efficiency and reducing downtime.

[0054] 3. By analyzing the long-term and short-term trend of operating data, the failure probability of the fuel dispenser at different time points in the future can be predicted. Preventive measures can be taken before the failure occurs to avoid the impact of the failure on the gas station operation. The failure prediction model provides a basis for the gas station to formulate a reasonable maintenance plan. According to the predicted failure probability and time, maintenance time and resources can be reasonably arranged to improve the reliability and availability of equipment.

[0055] 4. Provide graded warnings according to the severity of the fault, which are divided into three levels. Minor faults are pushed to the gas station manager, who can arrange for the station staff to handle them in a timely manner; faults that require manufacturer repairs are pushed to maintenance personnel so that they can respond quickly. Different levels of warnings allow relevant personnel to take different actions according to the urgency of the fault and improve response speed. Maintenance personnel can remotely view the fault situation based on the warning situation, make maintenance judgments, and confirm the necessary tools for maintenance. This greatly improves the accuracy and efficiency of maintenance and reduces unnecessary on-site investigation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The present invention is a flow chart of a method for integrated fault intelligent diagnosis of a fuel dispenser.

[0057] Figure 2 It is a schematic diagram of an integrated fault intelligent diagnosis fuel dispenser system. DETAILED DESCRIPTION

[0058] See also Figure 1 and Figure 2 As shown, the present invention is about

[0059] A method for integrated fault intelligent diagnosis of a fuel dispenser comprises the following steps:

[0060] S1: Integrate the real-time operating data obtained by sensors of the fuel circuit, circuit and related equipment of the fuel dispenser, and use the distributed framework built based on Spark core components and Hadoop distributed file system to store data through HDFS;

[0061] S2: Input the continuously collected operation data of the fuel dispenser into the fault diagnosis model to perform real-time anomaly detection;

[0062] S3: Analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data;

[0063] S4: Provide graded warnings based on detected or predicted fault conditions.

[0064] Specifically, the distributed framework built based on the Spark core components and the Hadoop distributed file system stores data through HDFS, including: configuring the master node and slave nodes of Spark, storing the collected data in HDFS, continuously writing real-time data to HDFS, and submitting the written Spark application to the Spark cluster for operation. Monitor the running status of the Spark job, including task progress, resource usage and performance indicators. Optimize according to the monitoring results, including adjusting the parallelism, memory allocation and data partitioning of the job to improve the efficiency and performance of the job.

[0065] Furthermore, the operating data of the oil circuit includes: pipeline pressure in the hydraulic area of ​​the fuel dispenser, oil and gas concentration in the space, flow rate of the oil pump and flow meter, pipeline and filter;

[0066] The circuit operation data includes: current values ​​and voltage values ​​of the motor, solenoid valve and circuit board in the fuel dispenser computer box;

[0067] The circuit board includes: an APP board, a main control board, a key display board, a power board, an IC card board and a driver board;

[0068] The associated equipment operation data includes: the communication status of the fuel dispenser application panel, the microprocessor code in the main control board, and the working status of the encoder, printer, IC card reader and keyboard board;

[0069] The integration includes: performing data fusion through Kalman filtering after data standardization and time synchronization.

[0070] Specifically, the pipeline pressure collection includes installing the pressure sensor at the key position of the hydraulic area pipeline to ensure that it is firmly installed and does not affect the normal operation of the pipeline; connecting the pressure sensor through the data acquisition module, setting the sampling frequency, and collecting pipeline pressure data in real time. The spatial oil and gas concentration collection includes considering the ventilation conditions and possible oil and gas leakage points when installing the oil and gas concentration sensor to ensure the accuracy of the detection; similarly, the oil and gas concentration sensor is connected through the data acquisition module to regularly collect oil and gas concentration data. The oil pump flow collection includes selecting a turbine flowmeter or an electromagnetic flowmeter according to the working characteristics and flow range of the oil pump, and the data acquisition module is connected to the flow sensor to record the flow data of the oil pump in real time. The flow rate of the flowmeter, pipeline and filter includes installing the flow rate sensor at the appropriate positions of the flowmeter, pipeline and filter respectively to ensure that the flow rate of the fluid can be accurately measured and the flow rate data is collected at certain time intervals.

[0071] The current and voltage values ​​of the motor, solenoid valve and circuit board in the tanker computer box include detecting the working current and voltage of the motor, solenoid valve and all circuit boards through current transformers and voltage sensors, installing the current transformers and voltage sensors on the corresponding circuit loops to ensure that they are installed correctly and do not affect the normal operation of the circuit; the data acquisition module connects the current transformer and voltage sensor to collect current and voltage data in real time, and performs data processing and analysis. The communication status of the tanker application panel includes establishing a stable connection with the tanker application board through serial communication or Ethernet communication; reading the error code of the monitoring microprocessor in the mainboard, encoder, printer, IC card reader, keyboard board and other working status information through the communication interface.

[0072] The data standardization includes standardizing various data from different sensors. Since the data formats, units and ranges of different sensors may be different, they need to be uniformly converted into a standardized format, such as converting the unit of the pressure sensor from Pascal (Pa) to Megapascal (MPa), converting the unit of the flow sensor from liter / minute (L / min) to cubic meter / hour (m 3 / h), etc. Determine the standardization range and accuracy requirements of the data, and set them according to the working characteristics of the fuel dispenser and the needs of fault diagnosis. For example, standardize the pressure data to the range of 0-10MPa, and retain the accuracy to two decimal places.

[0073] The time synchronization includes: ensuring that the data from different sensors are synchronized in time. Because the various components of the fuel dispenser may generate data at different time points, it will be difficult to accurately analyze the relationship between them without time synchronization. Mark the data by means of timestamps to ensure that all data have the same time reference. Use a high-precision clock source to synchronize the sensors and regularly calibrate the time of the sensors, such as synchronizing the data acquisition system through a global positioning system (GPS) clock source or a network time protocol (NTP) server.

[0074] The data fusion is to fuse different types of data after standardization and time synchronization. The Kalman filter method in the data fusion algorithm is used to comprehensively process the data from multiple sensors to obtain more accurate and reliable fused data. For example, for the data fusion of pressure sensors and flow sensors, different weights are set according to their importance in fault diagnosis, and then the fused pressure-flow data is obtained by weighted average method. For the data fusion of multiple similar sensors, the Kalman filter method is used to predict and update the sensor measurement values ​​to obtain a more accurate state estimation.

[0075] Furthermore, the method of continuously collecting the operating data of the fuel dispenser and inputting it into the fault diagnosis model for real-time abnormality detection includes the following steps:

[0076] S21: According to the operation data characteristics and fault diagnosis requirements of the fuel dispenser, and based on the fuzzy rule base established according to expert experience and historical fault data, the defuzzified output value is obtained through the fuzzy logic algorithm;

[0077] S22: using the output of the fuzzy logic algorithm as input data and the preprocessed raw data as input features for further analysis and diagnosis through the random forest algorithm;

[0078] S23: Verify the accuracy and reliability through test set data or actual operation data, evaluate the fault diagnosis model regularly, and optimize the model based on the evaluation results.

[0079] Specifically, in step S23, the fault diagnosis model is evaluated regularly, and the accuracy and reliability of the model are verified using test set data or actual operation data. The model is evaluated by calculating performance indicators of the model, such as accuracy, recall rate, F1 value, etc. For example, the actual fault data over a period of time is used as a test set to evaluate the model. According to the evaluation results, the model is optimized, the fuzzy logic rules, the parameters of the random forest are adjusted, or new features are added. If it is found that the model has a low diagnostic accuracy for a certain type of fault, the fuzzy logic rules can be adjusted or features related to the fault type can be added. As the tanker runs and data accumulates, the fault diagnosis model is continuously updated. The model is retrained regularly, or the model is updated in real time using online learning. For example, the random forest model is retrained once a month to adapt to changes in the operating status of the tanker.

[0080] Furthermore, the method of obtaining the defuzzified output value by using a fuzzy logic algorithm based on the operation data characteristics and fault diagnosis requirements of the fuel dispenser and the fuzzy rule base established based on expert experience and historical fault data includes the following steps:

[0081] S211: According to the operating characteristics and fault characteristics of the fuel dispenser, and based on expert experience or historical data, a fuzzy logic rule is defined, wherein the fuzzy set of input variables is the operating parameters of the fuel dispenser, and the fuzzy set of output variables is the fuzzy description of the fault type or fault degree;

[0082] S212: Calculate the degree to which the input data belongs to each fuzzy set through the membership function;

[0083] S213: According to the defined fuzzy logic rules, the fuzzified input data is inferred to obtain a fuzzy output result, and during the inference process, the membership degree of the input data and the weight of the rule are calculated to obtain a fuzzy set of output variables;

[0084] S214: Defuzzifying the fuzzy output result by the centroid method to obtain the probability distribution of the fault type or the output numerical representation of the fault degree.

[0085] Specifically, in step S211, the fuzzy roadbed rules, such as the definition rule "If the pressure is too high and the flow is too low, there may be a fuel pump failure". Determine the fuzzy sets of input variables and output variables. The input variables can be operating parameters of the fuel dispenser, such as pressure, flow, current, etc.; the output variables can be fuzzy descriptions of the fault type or fault degree. For example, the input variable pressure can be divided into three fuzzy sets of "low", "medium" or "high", and the output variable fault degree can be divided into three fuzzy sets of "primary", "secondary" or "tertiary".

[0086] In step S212, the pre-processed input data is fuzzified, that is, mapped into the corresponding fuzzy set. The degree to which the input data belongs to each fuzzy set is calculated by the membership function. For example, for pressure data, the triangular membership function can be used to determine the degree to which it belongs to the three fuzzy sets of "low", "medium" and "high".

[0087] In step S213, the fuzzy output result is realized by a fuzzy reasoning engine. For example, fuzzy reasoning is performed using the Mamdani reasoning method or the Sugeno reasoning method. The fuzzy set of the output variable includes that if the pressure of the input data belongs to the "high" fuzzy set to a degree of 0.8, the flow belongs to the "low" fuzzy set to a degree of 0.6, and the weight of the rule "if the pressure is too high and the flow is too low, there may be an oil pump failure" is 0.7, then it can be calculated that the degree of the output variable failure degree belongs to the "third level" fuzzy set to a degree of 0.8×0.6×0.7=0.336.

[0088] In step S214, the fuzzy output result is defuzzified to obtain a clear output value. The defuzzification method of the centroid method or the maximum membership method can be used. For example, the centroid of the output variable fuzzy set can be calculated by the centroid method as the defuzzified output value. The defuzzified output result is the probability distribution of the fault type or the numerical representation of the fault degree. For example, the defuzzified output result may be "the probability of oil pump failure is 0.7, and the probability of other failures is 0.3".

[0089] Furthermore, the output result of the fuzzy logic algorithm is used as input data, and the pre-processed raw data is used as input features, and further analysis and diagnosis are performed through the random forest algorithm; the following steps are included:

[0090] S221: using the output result of the fuzzy logic algorithm as input data of the random forest algorithm, and using the preprocessed raw data as input features of the random forest algorithm;

[0091] S222: randomly selecting data as a training set, and training the random forest model with the training set data, constructing each decision tree by randomly sampling and feature selecting the training data, and performing splitting by a recursive method during the construction process until a stop condition is met;

[0092] S223: Input the input data into the random forest model, determine the fault type by majority voting according to the prediction results of each decision tree, and determine the fault degree by averaging.

[0093] Specifically, in step S221, for example, the fault type probability distribution output by the fuzzy logic algorithm and the parameters such as pressure, flow, and current in the original data are used as inputs of the random forest algorithm.

[0094] In step S222, a portion of the data is randomly selected as a training set for training the random forest model. Determine the parameters of the random forest, such as the number of trees, the depth of the tree, the method for selecting the split feature, etc. For example, set the random forest to consist of 100 decision trees, the depth of the tree is 10, and use information gain as the method for selecting the split feature. The random forest model is trained with the training set data, and each decision tree is constructed by random sampling and feature selection of the training data. In the process of constructing the decision tree, the splitting is performed recursively until the stopping condition is met. For example, the splitting is stopped when the number of samples in the node is less than a certain threshold or the depth of the tree reaches a set value.

[0095] In step S223, new data is predicted according to the trained random forest model. The input data is input into the random forest model, and the prediction results of each decision tree are voted or averaged to obtain the final prediction result. For example, the fault type is determined by majority voting, that is, the fault type with the most votes is selected as the prediction result; for the prediction of the degree of fault, the average method is used, that is, the average value of the degree of fault predicted by all decision trees is calculated as the final prediction result.

[0096] Furthermore, the method of analyzing the operating data by using a fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operating data comprises the following steps:

[0097] S31: Based on the time series analysis results, a long-term trend prediction model is constructed through the time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data;

[0098] S32: Build a short-term trend prediction model through machine learning algorithms to handle large-scale data and nonlinear relationships;

[0099] S33: The prediction results of the two algorithms are fused by weighted average method to construct a comprehensive fault prediction model.

[0100] Specifically, in step S33, according to the fusion strategy of the time series analysis algorithm and the machine learning algorithm. The weighted average method is used to determine the weights according to the performance of the two algorithms on historical data. The prediction results of the two algorithms are fused to obtain a comprehensive prediction value. Then, according to the preset fault threshold or rule, the failure probability of the tanker at different time points in the future is calculated. For example, if the predicted pipeline pressure exceeds a certain proportion of the normal range, or the rate of change of the flow exceeds a certain threshold, it is considered that the tanker has a higher probability of failure. The fault prediction model is evaluated through test data. Common evaluation indicators such as root mean square error (RMSE), mean absolute error (MAE), accuracy, recall rate, F1 value, etc. are calculated to measure the prediction accuracy and reliability of the model. According to the evaluation results, the model is optimized. For example, adjust the parameters of the time series analysis algorithm, the structure and parameters of the machine learning algorithm, the weight of the fusion strategy, etc. Or improve the performance of the model by increasing the amount of data, improving feature engineering, trying different algorithm combinations, etc.

[0101] Furthermore, the method of constructing a long-term trend prediction model based on the time series analysis results by using a time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data includes the following steps:

[0102] S311: determining the range of autoregressive moving average model parameters by observing the autocorrelation function and the partial autocorrelation function, and determining the best parameter combination by grid search;

[0103] S312: The autoregressive moving average model with determined parameters is trained using training data. The autoregressive moving average model formula is:

[0104] Where: y t is the time series data, c is the constant term, φ i is the autoregressive coefficient, θ j is the moving average coefficient, ∈ t is a white noise sequence, p is the autoregressive order, and q is the moving average order;

[0105] S313: Input the operating data into the autoregressive moving average model, and predict the future long-term trend based on the autoregressive, differential and moving average relationship of the historical data.

[0106] Specifically, in step S311, the autoregressive moving average model (ARIMA) consists of an autoregressive term (AR), a difference term (I) and a moving average term (MA), generally expressed as ARIMA (p, d, q), where p is the autoregressive order, d is the number of differences, and q is the moving average order. The autocorrelation function (ACF) and the partial autocorrelation function (PACF) are observed to preliminarily determine the parameter range. If the ACF decays exponentially and the PACF is truncated after the pth order, an AR (p) process may exist; if the ACF is truncated after the qth order and the PACF decays exponentially, an MA (q) process may exist. Use grid search to determine the best parameter combination. By traversing different parameter combinations, the Akaike Information Criterion (AIC) or Bayesian Information Criterion (BIC) under each combination is calculated, and the parameter combination with the smallest AIC or BIC value is selected as the best parameter.

[0107] In step S313, the output result is a long-term trend prediction value of the fuel dispenser operation data in the future period of time, such as the predicted values ​​of pipeline pressure, flow rate, etc. in the next week.

[0108] Furthermore, the short-term change trend prediction model is constructed by machine learning algorithm to process large-scale data and nonlinear relationships; including the following steps:

[0109] S321: extracting statistical features, time domain features, and frequency domain features from the preprocessed raw data, performing dimensionality reduction through principal component analysis, calculating mutual information between the features and the fault labels, and retaining the features of the mutual information;

[0110] S322: transforming the fuel dispenser fault prediction into a binary classification problem, i.e., normal state and fault state, through a support vector machine algorithm;

[0111] S323: Determine the radial basis function as the kernel function of the support vector machine, and determine the penalty coefficient and kernel parameter by a cross-validation method;

[0112] S324: The preprocessed raw data is subjected to feature engineering to extract features and then input into the support vector machine model. The model learns the short-term change pattern in the data based on the input features, and the output result is the short-term prediction value of the fuel dispenser operation data in the future.

[0113] Specifically, in step S321, the statistical features are used to calculate the mean, variance, maximum value, minimum value, etc. For example, the pressure data P has a mean of The variance is

[0114] The time domain characteristics, such as the peak value, peak-to-peak value, root mean square value, etc. of the signal. For pressure data P, the peak value is P peak =max(P i), the peak-to-peak value is P pp =P peak -P min , the root mean square value is

[0115] The frequency domain features are spectrum features obtained by Fourier transform. The time series data is subjected to fast Fourier transform (FFT) to obtain spectrum distribution.

[0116] Use principal component analysis (PCA) to reduce the dimension. Assume that the original feature matrix is ​​X, and calculate the covariance matrix Solve the eigenvalues ​​and eigenvectors, select the eigenvectors with larger eigenvalues ​​to form a new feature space, project the original data into the new feature space to achieve dimensionality reduction. Calculate the mutual information between the features and the fault labels, and select the features with larger mutual information to retain.

[0117] In step S322, the binary classification algorithm separates data of different categories by finding a hyperplane. For the prediction of tanker failure, it is converted into a binary classification problem, namely, normal state and failure state.

[0118] In step S323, in constructing the support vector machine model, the radial basis function kernel function formula is K(x i , x j ) = exp(-γ||x i -x j || 2 ), where γ is the kernel parameter, and the model parameters are determined, including the penalty coefficient C and the kernel parameter γ. The optimal parameter combination can be selected by cross-validation method.

[0119] In step S324, the output result is a short-term trend forecast value of the fuel dispenser operation data in the short future time, such as the forecast value of various parameters in the next hour, and the predicted fault state (normal or faulty).

[0120] Furthermore, the step of providing a graded warning according to the detected or predicted fault conditions includes the following steps:

[0121] S41: determining key indicators for fault classification based on the operation data and common fault types of the fuel dispenser, and classifying the fault levels into primary fault, secondary fault and tertiary fault according to the determined fault indicators;

[0122] S42: Different push configurations are performed according to the fault level, and different push channels are established for the gas station manager and the manufacturer's maintenance personnel: when a level 1 fault is detected, the warning information is pushed to the gas station manager; when a level 2 or 3 fault is detected, it is pushed to the gas station manager and the manufacturer's maintenance personnel at the same time.

[0123] Specifically, in step S41, the key indicators are indicators such as the degree of pressure abnormality, flow deviation, and the number of abnormal equipment states. For pressure abnormalities, the percentage of pressure deviation exceeding the normal range is set as the basis for classification. For example, a pressure deviation within 10% is a level one fault, 10% to 30% is a level two fault, and more than 30% is a level three fault. The three levels, for example, if only one device state has a slight abnormality, such as a printer out of paper, it can be determined as a level one fault; if the pressure deviation is within a medium range and there are multiple device states abnormalities, such as encoder failure and some keys on the keyboard board malfunctioning, it can be determined as a level two fault; if the pressure is completely out of control or a major circuit fault occurs, such as a motor burnout, it can be determined as a level three fault.

[0124] In step S42, different push channels are established for gas station managers and maintenance personnel. For example, early warning push is performed by using SMS, mobile application push, email, etc. For gas station managers, mobile application push is used to ensure that they can receive early warning information of level 1 faults in a timely manner; for maintenance personnel, a combination of SMS and email is used to ensure that they can receive early warning information under any circumstances.

[0125] Different push configurations are made according to the fault level. When a level 1 fault is detected, the warning information is pushed to the gas station manager. When a level 2 or level 3 fault is detected, it is pushed to the gas station manager and maintenance personnel at the same time. In the push information, the fault level, fault description and preliminary handling suggestions are clearly stated. For example, for a level 1 fault, push "Level 1 fault: The printer is out of paper, please replenish paper in time." For a level 3 fault, you can push "Level 3 fault: Pressure out of control, it may be a fuel pump failure, please ask maintenance personnel to handle it as soon as possible."

[0126] When the manufacturer's maintenance personnel receive a push notification of a secondary or tertiary fault, they can view the fault status of the fuel dispenser through the remote monitoring platform and display the fault information in an intuitive manner on the remote monitoring platform. The remote monitoring platform is in the form of a web page or a dedicated mobile phone application. The platform should be able to display the fuel dispenser's operating data, fault code, equipment status and other information in real time. For example, maintenance personnel can view the real-time data of the pressure sensor, the current and voltage values ​​of the motor, the working status of the printer, etc. on the platform. Ensure that the fuel dispenser's operating data can be transmitted to the remote monitoring platform in real time.

[0127] Use data synchronization technology to ensure that the data on the remote monitoring platform is consistent with the actual operating data of the fuel dispenser. For example, use timed synchronization or event-triggered synchronization to ensure that maintenance personnel can see the latest fault conditions. Display fault information in an intuitive way on the remote monitoring platform. Use charts, graphical interfaces, or fault code lists to display fault conditions. For example, for abnormal pressure faults, draw a curve of pressure changes over time, and mark the abnormal points and abnormal ranges. For equipment status faults, use icons or colors to mark the faulty equipment to facilitate maintenance personnel to quickly locate the fault.

[0128] The manufacturer's maintenance personnel make maintenance judgments based on the fault information on the remote monitoring platform. If the fault is minor, they will remotely guide the gas station manager to perform simple processing. If the fault requires on-site repair, determine the maintenance plan and the required tools. Establish a maintenance knowledge base on the remote monitoring platform, which includes common fault handling methods, maintenance steps and required tools. The maintenance knowledge base can be classified according to the model and fault type of the gas station, making it convenient for maintenance personnel to quickly find relevant information. For example, for a certain model of gas station, when a pressure sensor fails, the maintenance knowledge base can provide fault cause analysis, maintenance steps and a list of required tools, such as wrenches, screwdrivers, new pressure sensors, etc.

[0129] Maintenance personnel make maintenance judgments based on the fault information and maintenance knowledge base on the remote monitoring platform. First, determine the possible cause and scope of the fault. Then evaluate the severity of the fault and the urgency of the repair. If the fault is minor, remotely guide the gas station staff to perform simple processing, such as restarting the equipment, replacing consumables, etc. If the fault requires on-site repair, determine the maintenance plan and required tools. For example, for a level 2 fault, maintenance personnel can determine that a circuit board may be faulty and need to carry tools such as a multimeter and a soldering iron for on-site repair. For a level 3 fault, large components such as oil pumps may need to be replaced, and maintenance personnel need to prepare the corresponding tools and spare parts in advance.

[0130] A system for integrated fault intelligent diagnosis of a fuel dispenser, comprising a data acquisition module, a fault diagnosis module, a fault prediction module and an early warning module connected in communication, wherein:

[0131] The data acquisition module is used to obtain real-time operating data from the fuel circuit, circuit and associated equipment sensors of the fuel dispenser, and integrate and store the data;

[0132] The fault diagnosis module is used to continuously collect the operation data of the fuel dispenser and input it into the fault diagnosis model for real-time anomaly detection;

[0133] The fault prediction module is used to analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data;

[0134] The early warning module is used to issue graded early warnings based on detected or predicted fault conditions, and maintenance personnel can remotely view and make repair decisions based on the early warning conditions.

[0135] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for integrated fault intelligent diagnosis of fuel dispensers, characterized in that: The following steps are involved: S1: Integrate the real-time operating data obtained by sensors of the fuel circuit, circuit and related equipment of the fuel dispenser, and use the distributed framework built based on Spark core components and Hadoop distributed file system to store data through HDFS; S2: Input the continuously collected operation data of the fuel dispenser into the fault diagnosis model to perform real-time anomaly detection; S3: Analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data; S4: Provide graded warnings based on detected or predicted fault conditions.

2. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 1, characterized in that: The operating data of the oil circuit includes: pipeline pressure in the hydraulic area of ​​the fuel dispenser, oil and gas concentration in the space, oil pump flow rate and flow rate of the flow meter, pipeline and filter; The circuit operation data includes: current values ​​and voltage values ​​of the motor, solenoid valve and circuit board in the fuel dispenser computer box; The circuit board includes: an APP board, a main control board, a key display board, a power board, an IC card board and a driver board; The associated equipment operation data includes: the communication status of the fuel dispenser application panel, the microprocessor code in the main control board, and the working status of the encoder, printer, IC card reader and keyboard board; The integration includes: performing data fusion through Kalman filtering after data standardization and time synchronization.

3. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 1, characterized in that: The method comprises the following steps: continuously collecting the operation data of the fuel dispenser and inputting it into the fault diagnosis model for real-time abnormality detection: S21: According to the operation data characteristics and fault diagnosis requirements of the fuel dispenser, and based on the fuzzy rule base established according to expert experience and historical fault data, the defuzzified output value is obtained through the fuzzy logic algorithm; S22: using the output of the fuzzy logic algorithm as input data and the preprocessed raw data as input features for further analysis and diagnosis through the random forest algorithm; S23: Verify the accuracy and reliability through test set data or actual operation data, evaluate the fault diagnosis model regularly, and optimize the model based on the evaluation results.

4. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 3, characterized in that: The method includes the following steps: obtaining a defuzzified output value by using a fuzzy logic algorithm based on the operation data characteristics and fault diagnosis requirements of the fuel dispenser and a fuzzy rule base established based on expert experience and historical fault data: S211: According to the operating characteristics and fault characteristics of the fuel dispenser, and based on expert experience or historical data, a fuzzy logic rule is defined, wherein the fuzzy set of input variables is the operating parameters of the fuel dispenser, and the fuzzy set of output variables is the fuzzy description of the fault type or fault degree; S212: Calculate the degree to which the input data belongs to each fuzzy set through the membership function; S213: According to the defined fuzzy logic rules, the fuzzified input data is inferred to obtain a fuzzy output result, and during the inference process, the membership degree of the input data and the weight of the rule are calculated to obtain a fuzzy set of output variables; S214: Defuzzifying the fuzzy output result by the centroid method to obtain the probability distribution of the fault type or the output numerical representation of the fault degree.

5. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 3, characterized in that: The method uses the output result of the fuzzy logic algorithm as input data, and uses the pre-processed raw data as input features, and further analyzes and diagnoses through the random forest algorithm; including the following steps: S221: using the output result of the fuzzy logic algorithm as input data of the random forest algorithm, and using the preprocessed raw data as input features of the random forest algorithm; S222: randomly selecting data as a training set, and training the random forest model with the training set data, constructing each decision tree by randomly sampling and feature selecting the training data, and performing splitting by a recursive method during the construction process until a stop condition is met; S223: Input data into the random forest model, determine the fault type by majority voting according to the prediction results of each decision tree, and determine the fault degree by averaging.

6. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 1, characterized in that: The method of analyzing the operation data by using a fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data comprises the following steps: S31: Based on the time series analysis results, a long-term trend prediction model is constructed through the time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data; S32: Build a short-term trend prediction model through machine learning algorithms to handle large-scale data and nonlinear relationships; S33: The prediction results of the two algorithms are fused by weighted average method to construct a comprehensive fault prediction model.

7. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 6, characterized in that: The method of constructing a long-term trend prediction model based on the time series analysis results by using a time series analysis algorithm to capture the long-term trend and seasonal characteristics of the data includes the following steps: S311: determining the range of autoregressive moving average model parameters by observing the autocorrelation function and the partial autocorrelation function, and determining the best parameter combination by grid search; S312: The autoregressive moving average model with determined parameters is trained using training data. The autoregressive moving average model formula is: Where: y t is the time series data, c is the constant term, φ i is the autoregressive coefficient, θ j is the moving average coefficient, ∈ t is a white noise sequence, p is the autoregressive order, and q is the moving average order; S313: Input the operating data into the autoregressive moving average model, and predict the future long-term trend based on the autoregressive, differential and moving average relationship of the historical data.

8. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 6, characterized in that: The method constructs a short-term trend prediction model by using a machine learning algorithm to process large-scale data and nonlinear relationships; the method includes the following steps: S321: extracting statistical features, time domain features, and frequency domain features from the preprocessed raw data, performing dimensionality reduction through principal component analysis, calculating mutual information between the features and the fault labels, and retaining the features of the mutual information; S322: transforming the fuel dispenser fault prediction into a binary classification problem, i.e., normal state and fault state, through a support vector machine algorithm; S323: Determine the radial basis function as the kernel function of the support vector machine, and determine the penalty coefficient and kernel parameter by a cross-validation method; S324: The preprocessed raw data is subjected to feature engineering to extract features and then input into the support vector machine model. The model learns the short-term change pattern in the data based on the input features, and the output result is the short-term prediction value of the fuel dispenser operation data in the future.

9. The method for integrated fault intelligent diagnosis of a fuel dispenser according to claim 1, characterized in that: The method of providing a graded warning according to the detected or predicted fault conditions comprises the following steps: S41: determining key indicators for fault classification based on the operation data and common fault types of the fuel dispenser, and classifying the fault levels into primary fault, secondary fault and tertiary fault according to the determined fault indicators; S42: Different push configurations are performed according to the fault level, and different push channels are established for the gas station manager and the manufacturer's maintenance personnel: when a level 1 fault is detected, the warning information is pushed to the gas station manager; when a level 2 or 3 fault is detected, it is pushed to the gas station manager and the manufacturer's maintenance personnel at the same time.

10. A system for integrated fault intelligent diagnosis of a fuel dispenser, used to execute a method for integrated fault intelligent diagnosis of a fuel dispenser as claimed in any one of claims 1 to 9, characterized in that: It includes a data acquisition module, a fault diagnosis module, a fault prediction module and an early warning module connected by communication, wherein: The data acquisition module is used to obtain real-time operating data from the fuel circuit, circuit and associated equipment sensors of the fuel dispenser, and integrate and store the data; The fault diagnosis module is used to continuously collect the operation data of the fuel dispenser and input it into the fault diagnosis model for real-time anomaly detection; The fault prediction module is used to analyze the operation data through the fault prediction model to predict the failure probability of the fuel dispenser in the long-term change trend or short-term change trend of the operation data; The early warning module is used to issue graded early warnings based on detected or predicted fault conditions, and maintenance personnel can remotely view and make repair decisions based on the early warning conditions.

Images