Fault diagnosis method and system based on multi-source data association rule and graph neural network

Abstract

The invention discloses a fault diagnosis method and system based on a multi-source data association rule and a graph neural network. The method comprises the steps of extracting high-frequency operation data and low-frequency time sequence state data based on historical data, and establishing an equipment operation feature set; an Apriori algorithm is utilized to screen correlation characteristics to calculate a correlation relation, and a fault symptom set is constructed; and taking the association relationship of the features as an adjacent matrix embedded graph neural network, and training the constructed fuzzy graph neural network based on historical data to obtain a fault diagnosis model. The system comprises a data acquisition module, a preprocessing module, a feature extraction module, a feature screening module, a feature association relationship analysis module, a fault diagnosis model training module, a fault diagnosis module and a database storage module, and can perform multi-source data fusion analysis and training and updating of a fault diagnosis model. According to the method, the fault diagnosis model is constructed by combining multi-source data fusion, feature extraction, association relationship mining and the fuzzy graph neural network, so that more accurate and efficient fault diagnosis is realized.

Description

Technical Field

[0001] The present invention belongs to the field of fault diagnosis, and specifically relates to a fault diagnosis method and system based on multi-source data association rules and graph neural networks. Background Art

[0002] As industrial equipment continues to grow in complexity and intelligence, fault diagnosis faces increasing challenges. Traditional fault diagnosis methods often only consider a single type of operational data, making it difficult to fully reflect the equipment's operating status and potential faults. Therefore, how to effectively utilize multi-source data and extract fault signatures has become a research priority in the field of fault diagnosis.

[0003] Currently, multi-source data fusion and feature extraction technologies have been widely used in fault diagnosis. However, existing methods typically only consider simple associations between data when processing multi-source data, ignoring the complex dependencies and patterns between data. Therefore, how to more comprehensively mine the associations between multi-source data and build a generalizable fault diagnosis model is an urgent problem that needs to be solved. Summary of the Invention

[0004] To address the shortcomings of the prior art, this paper proposes a fault diagnosis method and system based on multi-source data association rules and graph neural networks. By combining multi-source data fusion, feature extraction, association mining, and fuzzy graph neural network fault diagnosis model construction, this method achieves more accurate and efficient fault diagnosis.

[0005] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0006] A fault diagnosis method based on multi-source data association rules and graph neural networks performs the following steps to obtain a fault diagnosis model, which is used to input real-time industrial field data into the model for real-time fault diagnosis and output diagnosis results. The method includes the following steps:

[0007] Step 1: Collect multi-source data on equipment operation that characterizes equipment operation status and potential faults, and perform data preprocessing; the multi-source data includes high-frequency operation data and low-frequency operation data;

[0008] Step 2: Extract time-domain and frequency-domain features from the high-frequency operation data, and combine them with the low-frequency operation data to form a device operation feature set to characterize the device's operating status and potential faults.

[0009] Step 3: Construct a set of fault symptoms, filter out features related to the operating status from historical fault data based on the Apriori algorithm, and calculate the correlation between the features;

[0010] Step 4: Construct a fault diagnosis model, embed the correlation between features into the graph neural network as the adjacency matrix of the graph neural network, and train the constructed fuzzy graph neural network based on historical data to obtain the fault diagnosis model.

[0011] The high-frequency operation data includes vibration data, and the low-frequency time series status data includes voltage, current, equipment inlet and outlet flow, temperature and pressure data.

[0012] The pretreatment includes

[0013] Data cleaning and preprocessing: Clean and preprocess the collected raw data, including removing outliers, filling missing values, and normalizing data to eliminate noise and anomalies in the data;

[0014] Data segmentation and labeling: The preprocessed data is segmented according to time series, and the corresponding equipment status of each data segment is labeled. The equipment status includes normal, inner race fault, outer race fault, and rolling element fault. These labeled data will be used for subsequent model training and verification.

[0015] The time domain feature extraction is to extract dimensional and dimensionless time domain features reflecting the vibration state and abnormal conditions of the equipment from the vibration data to form a sample set, and calculate the sample feature values ​​to form a time domain feature set; the sample features include absolute mean, variance, standard deviation, root amplitude, root mean square value, peak value, maximum value, minimum value, waveform index, peak index, pulse index, margin index, skewness and kurtosis statistics.

[0016] The frequency domain feature extraction is as follows: the vibration data is converted from the time domain to the frequency domain through the Fourier transform method, the spectrum features are extracted, and the amplitude and phase information of each frequency component are obtained to characterize the vibration mode and potential failure mode of the equipment at different frequencies.

[0017] Specifically, the step 3 includes:

[0018] Step 3.1: Calculation of association relationships:

[0019] Based on historical normal data, set thresholds for equipment operating characteristics, and if exceeded, it is considered a sign of failure;

[0020] Based on historical fault data, the Apriori algorithm is used to count the frequent features related to each fault type in the equipment operation feature set, and the correlation coefficient between the fault features and each fault type is calculated. The features with unsatisfactory correlation coefficients are screened out, and the remaining features are used to form the equipment fault symptom set.

[0021] The correlation between each fault feature in the equipment fault symptom set is calculated based on the Apriori algorithm to construct a fuzzy graph neural network model;

[0022] The apriori algorithm evaluates the strength and reliability of the association relationship by calculating support and confidence indicators;

[0023] Step 3.2: Dynamically update and optimize the association relationship matrix.

[0024] The step 4 specifically includes:

[0025] Step 4.1: Build a graph neural network model: Based on the calculated fault feature correlation matrix, build a fuzzy graph neural network model;

[0026] The model represents the operating status and potential faults of the equipment as a fuzzy graph structure, where nodes represent the fault characteristics of the equipment and edges represent the association between the characteristics.

[0027] Graph neural network formula: H = σ(AXW), where A is the adjacency matrix, X is the feature matrix, W is the weight matrix, and σ is the activation function;

[0028] Step 4.2: Train the network model: Use historical operation data to train the constructed fuzzy graph neural network, learn the complex dependencies and patterns between features, and continuously optimize the network parameters and structure.

[0029] A fault diagnosis system based on multi-source data association rules and graph neural networks includes industrial site data acquisition equipment and a host computer. The data acquisition equipment is a sensor or instrument installed at the industrial equipment end, which is used to collect real-time data from the industrial equipment and upload it to the host computer. The host computer includes the following modules, which cooperate with each other to perform real-time fault diagnosis on the real-time collected industrial site data;

[0030] The host computer includes the following modules:

[0031] Data acquisition module: collects equipment operation data and stores it;

[0032] Data preprocessing module: cleans and preprocesses the collected raw data;

[0033] Data feature extraction module: This module extracts features from pre-processed time series data, including time-frequency domain feature extraction for high-frequency vibration data, which together with the equipment's low-frequency operation data form an equipment operation feature set.

[0034] Data feature screening module: Based on the preset minimum support and minimum confidence, the Apriori algorithm is used to screen out fault features that are strongly correlated with the fault type and construct a feature set of equipment fault symptoms;

[0035] Data feature correlation analysis module: Utilizes historical operation data and calculates the correlation between fault symptoms based on the Apriori algorithm to construct a correlation matrix;

[0036] Fault diagnosis model training module: uses the constructed fault data set to train and update the fault diagnosis model;

[0037] Fault diagnosis module: Encapsulates the trained fault diagnosis model into a fault diagnosis module, performs fault diagnosis analysis on the real-time operation data of the equipment, and provides the diagnosis results;

[0038] Database storage module: stores the real-time operation data of the equipment collected by the data acquisition module, the minimum support and confidence threshold of the feature screening module, the fault diagnosis model parameters, and the fault diagnosis results of the fault diagnosis module.

[0039] The fault diagnosis model training module specifically includes:

[0040] The objective loss function of model training is multi-class cross entropy loss: L = -∑y i log(p i );

[0041] After model training is completed, it is necessary to set the global anomaly threshold and the local sensor anomaly threshold according to the actual operation of the system;

[0042] The specific process of the fault diagnosis module is to input real-time operation data into the fault diagnosis module to calculate the probability of each operation state, and output the fault diagnosis result corresponding to the maximum probability.

[0043] A computer-readable storage medium is used to store a computer program and data required for program execution. When the program is executed, the fault diagnosis system based on multi-source data association rules and neural networks is implemented.

[0044] The advantages and beneficial effects of the present invention are:

[0045] The present invention is a fault diagnosis method and system based on multi-source data association rules and graph neural networks. By combining technologies such as multi-source data fusion, feature extraction, association relationship mining, and fuzzy graph neural network model to build a fault diagnosis model, more accurate and efficient fault diagnosis is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a fault diagnosis method process based on multi-source data association rules and graph neural networks.

[0047] Figure 2 It is a fault diagnosis system architecture based on multi-source data association rules and graph neural networks. DETAILED DESCRIPTION

[0048] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, the specific implementation methods of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth many specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the invention. Therefore, the present invention is not limited to the specific implementation methods disclosed below.

[0049] like Figure 1 The flowchart of the method provided by the present invention includes the following steps:

[0050] Step 1: Obtain equipment operation data, including high-frequency operation data such as vibration, and low-frequency time series status data such as voltage, current, equipment inlet and outlet flow, temperature, and pressure from historical operation data, and perform data preprocessing;

[0051] Step 1.1: Data acquisition: Acquire multi-source data including vibration, temperature, pressure, flow, current, and voltage from historical operating data. This data can fully reflect the operating status and potential failures of the equipment.

[0052] Step 1.2: Data cleaning and preprocessing: Clean and preprocess the collected raw data, including removing outliers, filling missing values, and normalizing data to eliminate noise and anomalies in the data and improve data accuracy and consistency.

[0053] Step 1.3: Data Segmentation and Labeling: Segment the preprocessed data into time series and label each data segment with the corresponding device status. Device status includes normal, inner race fault, outer race fault, and rolling element fault as fault categories. This labeled data will be used for subsequent model training and validation.

[0054] Step 2: Construct the equipment operation feature set, extract time domain features and frequency domain features from the high-frequency vibration data, and combine them with the low-frequency operation status data to form the equipment operation feature set;

[0055] Step 2.1: Time Domain Feature Extraction: Extract dimensional and dimensionless time domain features from the vibration data, including but not limited to absolute mean, variance, standard deviation, root mean square amplitude, root mean square value, peak value, maximum value, minimum value, waveform index, peak index, pulse index, margin index, skewness, and kurtosis. These features can reflect the vibration status and abnormal conditions of the equipment.

[0056] 1024 data are selected sequentially from the vibration data x(n) to form a sample set, and some statistics are calculated according to the following formula to form a time domain feature set. Mean index: Mean absolute value: variance: Standard Deviation: Square root amplitude: RMS value: Peak value: x p =max|x(n)|, maximum value: x q =maxx(n), minimum value: x min =minx(n), waveform indicator: Peak index: .Pulse indicator: Margin index: Skewness: Kurtosis:

[0057] Step 2.2: Frequency Domain Feature Extraction: Vibration data is converted from the time domain to the frequency domain using signal processing methods such as Fourier transform. The calculated samples are consistent with the time domain samples, and spectral features, such as the amplitude and phase information of each frequency component, are extracted. These features can reveal the vibration patterns and potential failure modes of the equipment at different frequencies.

[0058] Step 2.3: Construct an operating status feature set. Combine low-frequency time series states such as voltage, current, inlet and outlet flow, temperature, and pressure with time-frequency domain features to form an equipment operating feature set that can comprehensively reflect the equipment's operating status and potential faults.

[0059] Step 3: Construct a set of fault symptoms, filter out features related to the operating status from historical fault data based on the Apriori algorithm, and calculate the correlation between the features;

[0060] Step 3.1: Correlation Calculation: Based on historical normal data, thresholds for equipment operating characteristics are set; exceeding these thresholds is considered a fault symptom. Based on historical fault data, the Apriori algorithm is used to count frequent features related to each fault type within the equipment operating characteristic set. The correlation coefficient between the fault characteristics and each fault type is calculated, and features with a correlation coefficient less than 0.2 are filtered out. The remaining features constitute the equipment fault feature set. The Apriori algorithm is used to calculate the correlation between each fault feature. The strength and reliability of the correlation are evaluated by calculating indicators such as support and confidence. The correlation between fault characteristics and fault types will be used to construct the fault symptom set, and the correlation between fault characteristics will be used to construct the fuzzy graph neural network model.

[0061] Support:

[0062] Confidence:

[0063] When calculating the correlation between features and various fault types, X and Y are the features and the fault type, respectively. When calculating the correlation between various features, X and Y are feature a and feature b, respectively.

[0064] Step 3.2: Dynamic Update and Optimization: Based on actual operational failure cases and experience, the association rule base is continuously updated and optimized. This improves the model's real-time performance and accuracy by dynamically adjusting thresholds and updating the association matrix.

[0065] Step 4: Construct a fault diagnosis model, embed the correlation between features into the graph neural network as the adjacency matrix of the graph neural network, and train the constructed fuzzy graph neural network based on historical data to obtain the fault diagnosis model.

[0066] Step 4.1: Build a graph neural network model: Based on the calculated fault feature correlation matrix, a fuzzy graph neural network model is constructed. This model represents the device's operating status and potential faults as a fuzzy graph structure, where nodes represent individual fault features and edges represent the relationships between features. This fuzzy graph neural network model can better handle complex data structures and patterns, improving the accuracy and efficiency of fault diagnosis.

[0067] Graph neural network formula: H = σ(AXW), where A is the adjacency relationship matrix, X is the feature matrix, i.e., the set of fault symptoms, W is the weight matrix, σ is the activation function, and H represents the result of feature extraction. Softmax is used to calculate the probability that the extracted features belong to each fault type, in preparation for model training.

[0068] Step 4.2: Train the network model: Use historical operational data to train the constructed fuzzy graph neural network to learn the complex dependencies and patterns between features. By continuously optimizing the network parameters and structure, the network's predictive performance and stability are improved.

[0069] like Figure 2 The following is a flow chart of the system provided by the present invention, where solid arrows represent the flow of real-time fault diagnosis data, and dashed arrows represent the flow of fault model training data. Therefore, the system can be divided into two main business processes: using historical data for model training and updating, and using real-time data for equipment fault diagnosis. The system mainly includes the following modules:

[0070] Data acquisition module: collects equipment operation data and stores it;

[0071] Data preprocessing module: cleans and preprocesses the collected raw data;

[0072] Data feature extraction module: This module extracts features from pre-processed time series data, including time-frequency domain feature extraction for high-frequency vibration data, which together with the equipment's low-frequency operation data form an equipment operation feature set.

[0073] Data feature screening module: Based on the preset minimum support and minimum confidence, the Apriori algorithm is used to screen out fault features that are strongly correlated with the fault type and construct a feature set of equipment fault symptoms;

[0074] Data feature correlation analysis module: Utilizes historical operation data and calculates the correlation between fault symptoms based on the Apriori algorithm to construct a correlation matrix;

[0075] Fault diagnosis model training module: Use the constructed fault data set to train and update the fault diagnosis model. The target loss function of model training is multi-classification cross entropy loss: L = -∑y i log(p i ) Among them, p i is the predicted probability of the i-th category, y i is the true label of the i-th category. After the model training is completed, the global anomaly threshold and the local sensor anomaly threshold need to be manually set according to the actual operation of the system.

[0076] Fault diagnosis module: The trained fault diagnosis model is encapsulated into a fault diagnosis module, which performs fault diagnosis analysis on the real-time operation data of the equipment and gives the diagnosis results. The specific process is to input the real-time operation data into the fault diagnosis module to calculate the probability of each operating state and output the fault diagnosis result corresponding to the maximum probability.

[0077] Database storage module: stores the real-time operation data of the equipment collected by the data acquisition module, the minimum support and confidence threshold of the feature screening module, the fault diagnosis model parameters, and the fault diagnosis results of the fault diagnosis module.

[0078] In practical applications, sensors and instruments installed on industrial equipment are used to collect high-frequency and low-frequency data on site, such as high-frequency vibration data and low-frequency operating condition data of various operating states of pumps, and input them into Figure 2 The system of the middle and upper computer automatically outputs the fault type label results.

[0079] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent structural change made to the above embodiment based on the technical essence of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A fault diagnosis method based on multi-source data association rules and graph neural network, characterized in that: The following steps are performed to obtain a fault diagnosis model, which is used to input the real-time collected industrial field data into the model for real-time fault diagnosis and output the diagnosis results. The method includes the following steps: Step 1: Collect multi-source data on equipment operation that characterizes equipment operation status and potential faults, and perform data preprocessing; the multi-source data includes high-frequency operation data and low-frequency operation data; Step 2: Extract time-domain and frequency-domain features from the high-frequency operation data, and combine them with the low-frequency operation data to form a device operation feature set to characterize the device's operating status and potential faults. Step 3: Construct a set of fault symptoms, filter out features related to the operating status from historical fault data based on the Apriori algorithm, and calculate the correlation between the features; Step 4: Construct a fault diagnosis model, embed the correlation between features into the graph neural network as the adjacency matrix of the graph neural network, and train the constructed fuzzy graph neural network based on historical data to obtain the fault diagnosis model.

2. A fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1, characterized in that: The high-frequency operation data includes vibration data, and the low-frequency time series status data includes voltage, current, equipment inlet and outlet flow, temperature and pressure data.

3. A fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1, characterized in that: The pretreatment includes Data cleaning and preprocessing: Clean and preprocess the collected raw data, including removing outliers, filling missing values, and normalizing data to eliminate noise and anomalies in the data; Data segmentation and labeling: The preprocessed data is segmented according to time series, and the corresponding equipment status of each data segment is labeled. The equipment status includes normal, inner race fault, outer race fault, and rolling element fault. These labeled data will be used for subsequent model training and verification.

4. The fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1 is characterized in that: The time domain feature extraction is to extract dimensional and dimensionless time domain features reflecting the vibration state and abnormal conditions of the equipment from the vibration data to form a sample set, and calculate the sample feature values ​​to form a time domain feature set; the sample features include absolute mean, variance, standard deviation, root amplitude, root mean square value, peak value, maximum value, minimum value, waveform index, peak index, pulse index, margin index, skewness and kurtosis statistics.

5. The fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1 is characterized in that: The frequency domain feature extraction is as follows: the vibration data is converted from the time domain to the frequency domain through the Fourier transform method, the spectrum features are extracted, and the amplitude and phase information of each frequency component are obtained to characterize the vibration mode and potential failure mode of the equipment at different frequencies.

6. The fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1 is characterized in that: Specifically, the step 3 includes: Step 3.1: Calculation of association relationships: Based on historical normal data, set thresholds for equipment operating characteristics, and if exceeded, it is considered a sign of failure; Based on historical fault data, the Apriori algorithm is used to count the frequent features related to each fault type in the equipment operation feature set, and the correlation coefficient between the fault features and each fault type is calculated. The features with unsatisfactory correlation coefficients are screened out, and the remaining features are used to form the equipment fault symptom set. The correlation between each fault feature in the equipment fault symptom set is calculated based on the Apriori algorithm to construct a fuzzy graph neural network model; The apriori algorithm evaluates the strength and reliability of the association relationship by calculating support and confidence indicators; Step 3.2: Dynamically update and optimize the association relationship matrix.

7. The fault diagnosis method based on multi-source data association rules and graph neural network according to claim 1 is characterized in that: The step 4 specifically includes: Step 4.1: Build a graph neural network model: Based on the calculated fault feature correlation matrix, build a fuzzy graph neural network model; The model represents the operating status and potential faults of the equipment as a fuzzy graph structure, where nodes represent the fault characteristics of the equipment and edges represent the association between the characteristics. Graph neural network formula: H = σ(AXW), where A is the adjacency matrix, X is the feature matrix, W is the weight matrix, and σ is the activation function; Step 4.2: Train the network model: Use historical operation data to train the constructed fuzzy graph neural network, learn the complex dependencies and patterns between features, and continuously optimize the network parameters and structure.

8. A fault diagnosis system based on multi-source data association rules and graph neural network, characterized in that: It includes data acquisition equipment and a host computer at the industrial site. The data acquisition equipment is a sensor and instrument installed at the industrial equipment end, which is used to collect real-time data of the industrial equipment and upload it to the host computer. The host computer includes the following modules, which cooperate with each other to perform real-time fault diagnosis on the real-time collected industrial field data; The host computer includes the following modules: Data acquisition module: collects equipment operation data and stores it; Data preprocessing module: cleans and preprocesses the collected raw data; Data feature extraction module: This module extracts features from pre-processed time series data, including time-frequency domain feature extraction for high-frequency vibration data, which together with the equipment's low-frequency operation data form an equipment operation feature set. Data feature screening module: Based on the preset minimum support and minimum confidence, the Apriori algorithm is used to screen out fault features that are strongly correlated with the fault type and construct a feature set of equipment fault symptoms; Data feature correlation analysis module: Utilizes historical operation data and calculates the correlation between fault symptoms based on the Apriori algorithm to construct a correlation matrix; Fault diagnosis model training module: uses the constructed fault data set to train and update the fault diagnosis model; Fault diagnosis module: Encapsulates the trained fault diagnosis model into a fault diagnosis module, performs fault diagnosis analysis on the real-time operation data of the equipment, and provides the diagnosis results; Database storage module: stores the real-time operation data of the equipment collected by the data acquisition module, the minimum support and confidence threshold of the feature screening module, the fault diagnosis model parameters, and the fault diagnosis results of the fault diagnosis module.

9. A fault diagnosis system based on multi-source data association rules and graph neural network according to claim 8, characterized in that: The fault diagnosis model training module specifically includes: The objective loss function of model training is multi-class cross entropy loss: L = -∑y i log(p i ); After model training is completed, it is necessary to set the global anomaly threshold and the local sensor anomaly threshold according to the actual operation of the system; The specific process of the fault diagnosis module is to input real-time operation data into the fault diagnosis module to calculate the probability of each operation state, and output the fault diagnosis result corresponding to the maximum probability.

10. A computer-readable storage medium for storing a computer program and data required for program execution, characterized in that When the program is executed, the fault diagnosis system based on multi-source data association rules and neural networks described in claim 8 is implemented.

Images