Reciprocating type natural gas compressor fault diagnosis system

Abstract

The invention relates to the technical field of fault data processing, and provides a reciprocating natural gas compressor fault diagnosis system which comprises a data acquisition module, a data preprocessing module, a feature extraction module, a knowledge graph construction module, a fault analysis module, a maintenance decision module, an early warning module and a man-machine interaction module. The data acquisition module is used for acquiring operation parameters and historical data of the compressor in real time and transmitting the operation parameters and the historical data to the data preprocessing module; the preprocessing module is used for filtering, denoising and normalizing the data; the feature extraction module extracts fault features; the knowledge graph construction module establishes a relation graph of compressor faults, symptoms, reasons and maintenance measures according to historical data. The fault analysis module performs fault diagnosis, prediction and severity evaluation through a deep learning model; the maintenance decision module matches maintenance measures according to the analysis result and generates a plan; the early warning module generates early warning information when the condition is triggered; and the man-machine interaction module supports graph editing and maintenance.

Description

Technical Field

[0001] The present invention relates to the field of fault data processing, and more particularly to a reciprocating natural gas compressor fault diagnosis system. Background Art

[0002] Reciprocating compressor is an important gas delivery equipment, widely used in oil, natural gas, chemical and other industries. It compresses low-pressure gas to high pressure through reciprocating motion, and plays a vital role in industrial production. However, due to its complex mechanical structure and harsh working environment, reciprocating compressors are prone to problems such as valve failure, piston wear and bearing damage during operation. Timely diagnosis of faults and effective maintenance are of great significance to avoid equipment damage, reduce downtime and reduce maintenance costs. Therefore, research and development of efficient fault diagnosis technology is essential to improve the safety and reliability of reciprocating compressors.

[0003] Existing fault diagnosis technologies mainly rely on the analysis of physical quantities such as vibration signals and pressure signals. These methods usually use steps such as data acquisition, feature extraction, and pattern recognition to monitor the health status of equipment and identify the type of fault by establishing a discrimination model between normal and fault states. In recent years, with the development of artificial intelligence technology, machine learning and deep learning methods have been widely used in fault diagnosis. Through big data analysis and intelligent algorithms, these methods can automatically extract fault features from complex signal data, classify and predict them, and improve the accuracy and efficiency of diagnosis.

[0004] In reference 1 (Fault Diagnosis of a Reciprocating Compressor Air Valve Based on Deep Learning, 2020), researchers proposed a deep learning-based fault diagnosis method for reciprocating compressor air valves. This method first collects the operating data of the reciprocating compressor through a single-channel vibration signal, and then uses a one-dimensional convolutional neural network (1-DCNN) to learn the fault features and compare them with traditional feature extraction methods. The results show that deep learning methods have high diagnostic accuracy when processing nonlinear and non-stationary signals of complex systems. In further research, the authors applied multi-channel vibration signals and pressure signals to a two-dimensional convolutional neural network (2-DCNN) and found that removing useless channel signals can improve the accuracy and efficiency of diagnosis. This study shows that deep learning technology has great potential for application in compressor fault diagnosis and provides an effective solution for the intelligent diagnosis of complex equipment.

[0005] In reference 2 (A new method for reciprocating compressor fault diagnosis based on indicator diagram feature extraction, 2023), researchers proposed a fault diagnosis method based on indicator diagram feature extraction for key components of reciprocating compressors, such as valves, piston rings, and packings. This method extracts characteristic values ​​with clear physical meanings, such as average suction pressure, average exhaust pressure, indicator diagram area, and center of gravity coordinates, by collecting indicator diagram data reflecting the thermal process in the cylinder, and establishes a feature threshold database under different states. During the diagnosis process, the health status of the compressor is judged by comparing the deviation between the extracted features and the threshold. Experimental results show that this method can judge the fault status of the compressor with an accuracy of 98.3%, and achieve an accuracy of 86.86% when identifying single and multiple faulty components. However, due to the similarity of the impact of different faults on the features, the diagnostic accuracy is low in some cases.

[0006] Although Reference 2 has made some progress in feature extraction and fault classification, there are still some limitations. For example, this method only analyzes the features of the indicator diagram and fails to comprehensively consider other multi-source data, such as vibration signals, temperature signals, and equipment operation history records. Therefore, when the compressor is operating under complex working conditions, incomplete feature extraction or inaccurate diagnosis may occur. In addition, the method in Reference 2 has limited recognition accuracy under multiple fault conditions, making it difficult to accurately distinguish and evaluate the severity of multiple faults. Summary of the invention

[0007] In order to overcome the above-mentioned defects of the prior art, the present invention provides a reciprocating natural gas compressor fault diagnosis system, which constructs a multi-level knowledge graph through data collection, preprocessing and feature extraction, and uses a deep learning model to perform fault diagnosis, prediction and severity assessment, combined with maintenance decision and early warning modules, to achieve comprehensive monitoring and intelligent maintenance management of compressor faults. It effectively solves the problem of inaccurate diagnosis caused by relying only on a single data source and feature extraction in the prior art, and improves the intelligent level of multi-fault identification and maintenance management.

[0008] To achieve the above object, the present invention provides the following technical solutions: A reciprocating natural gas compressor fault diagnosis system includes a data acquisition module, a data preprocessing module, a feature extraction module, a knowledge graph construction module, a fault analysis module, a maintenance decision module, an early warning module and a human-computer interaction module, wherein the data acquisition module is connected to the data preprocessing module and the human-computer interaction module respectively, the data preprocessing module is connected to the feature extraction module, the feature extraction module is connected to the knowledge graph construction module and the fault analysis module respectively, the knowledge graph construction module is bidirectionally connected to the fault analysis module, the knowledge graph construction module is connected to the maintenance decision module and the human-computer interaction module respectively, the fault analysis module is connected to the early warning module and the human-computer interaction module respectively, and the early warning module is connected to the human-computer interaction module; the knowledge graph construction module is used to construct a multi-level knowledge graph of compressor fault-symptom-cause-maintenance measures based on historical fault data, maintenance records and equipment operation parameters, and dynamically adjust node weights and relationship paths according to diagnosis results and prediction results; The fault analysis module is used to comprehensively analyze the fault information of the compressor, and includes a fault diagnosis unit, a fault prediction unit and a fault severity assessment unit; wherein the fault diagnosis unit is used to input the time domain features, frequency domain features and time-frequency domain features provided by the feature extraction module into the deep belief network model for feature learning and fault classification, and output the fault diagnosis result; the fault prediction unit is used to predict future faults based on the long short-term memory network algorithm using the time domain features and time-frequency domain features as inputs, and evaluate the prediction accuracy, and the formula is: ; In the formula, is the prediction accuracy evaluation value, is the actual fault occurrence time, is the predicted failure time, is the maximum allowable time error, is the time weight function; The fault severity evaluation unit is used to extract factors related to the fault based on the node information in the knowledge graph, and generate the fault severity value by calculating these factors to evaluate the comprehensive impact of the current fault on the operation of the equipment. The calculation formula of the fault severity is: ; In the formula, is the fault severity, is the fault influencing factor, is the failure probability factor, is the fault expansion factor, is the fault tolerance time.

[0009] As a further solution of the present invention, a data acquisition module is used to collect operating parameters of a reciprocating natural gas compressor in real time, wherein the operating parameters include compressor cylinder pressure, cylinder temperature, crankshaft speed, vibration signal, bearing temperature, lubricating oil pressure, intake pressure, and exhaust pressure; the data acquisition module adopts a distributed data acquisition architecture, including a field data acquisition unit and a central data processing unit, the field data acquisition unit is installed at various key positions of the compressor, and communicates with the central data processing unit through an industrial bus; each field data acquisition unit includes a sensor for collecting operating parameters.

[0010] As a further solution of the present invention, the data acquisition module is also used to download the operating parameters, historical fault data and corresponding maintenance records of the equipment in the past period of time in the big data analysis platform; the data preprocessing module is used to filter, denoise and normalize the operating parameters collected by the data acquisition module.

[0011] As a further scheme of the present invention, a feature extraction module is used to extract fault features from the preprocessed operating parameters, and the fault features include time domain features, frequency domain features and time-frequency domain features; the time domain features extracted by the feature extraction module include mean, variance, peak factor, pulse factor, margin factor, waveform factor, signal energy, peak, root mean square value, and signal-to-noise ratio; the frequency domain features include main frequency, sideband, harmonic component, power spectral density, frequency center, frequency variance, frequency band energy ratio, and harmonic distortion; the time-frequency domain features include wavelet packet decomposition coefficients, inherent mode functions of empirical mode decomposition, instantaneous frequency, instantaneous amplitude, time-frequency spectrum, time-frequency energy distribution, and time-frequency representation with variable time-frequency resolution.

[0012] As a further solution of the present invention, the knowledge graph construction module includes the following steps: Step A1, defining a fault node, a symptom node, a cause node and a maintenance measure node, wherein the fault node represents a specific compressor fault type, the symptom node represents an abnormal state of an operating parameter, the cause node represents a specific cause of the fault, and the maintenance measure node represents a method for handling the fault, and each node is set with a specific label according to a preset attribute; Step A2, according to the operating characteristics and fault evolution law of each component of the compressor, establish a relationship path between each node, connect the fault node with the corresponding symptom node, connect the symptom node with the corresponding cause node, and connect the cause node with the corresponding maintenance measure node; Step A3, based on historical fault data and maintenance records, the initial weights of each node and its relationship path are set according to the performance of the node and path in the historical data. The weights are used to represent the importance of each node and path in fault diagnosis and prediction.

[0013] As a further solution of the present invention, the fault diagnosis unit uses a multi-source data fusion algorithm to perform feature fusion processing based on the time domain features, frequency domain features and time-frequency domain features obtained from the feature extraction module, and uses the fused feature data to construct a deep belief network model to classify the fault types, wherein the fault diagnosis accuracy of the deep belief network is evaluated, and the calculation formula is: ; In the formula, is the fault diagnosis accuracy, The number of samples for which the deep belief network model identified a specific fault type that actually existed in the compressor, The number of samples where the deep belief network model accurately identifies the type of fault that does not actually exist in the compressor, is the number of samples that the deep belief network model incorrectly identifies as having a certain fault type, is the number of samples where the deep belief network model fails to identify the actual fault type, and is the weight coefficient.

[0014] As a further solution of the present invention, the fault prediction unit is used to predict possible future faults based on historical fault data and current operating parameters. The fault prediction module uses a long short-term memory network model to perform fault prediction, processes the time domain features and time-frequency domain features obtained from the feature extraction module, and uses these feature data as input to the long short-term memory network model to perform time series learning and pattern recognition of historical data to predict the type of faults that may occur in the compressor in the future and the time when the faults will occur.

[0015] As a further solution of the present invention, according to the analysis results of the fault diagnosis unit and the fault prediction unit, the node weights and relationship paths in the knowledge graph are dynamically adjusted, including the following steps: Step S1, respectively obtain the current diagnosis result and prediction result from the fault diagnosis unit and the fault prediction unit, compare these results with the fault nodes and symptom nodes in the knowledge graph one by one, confirm whether the diagnosed fault type, fault severity and predicted symptom characteristics are consistent with the node information in the graph, and compare the reliability of the diagnosis result and the prediction result in combination with the fault diagnosis accuracy and the fault prediction accuracy, and record the matching results and deviation information; Step S2: if the diagnosis result and prediction result are consistent with the graph node information and the fault diagnosis accuracy and fault prediction accuracy both exceed the set threshold, the weight value of the corresponding node is increased according to the preset weight increment factor; if they are consistent but the accuracy does not reach the set threshold, the adjustment range of the weight increment factor is reduced; if the diagnosis result or prediction result is inconsistent with the graph node information and the accuracy reaches the set threshold, the weight value of the corresponding node is reduced according to the preset weight reduction factor; if they are inconsistent and the accuracy is lower than the set threshold, the adjustment range of the reduction factor is reduced, and the adjusted weight information and deviation information are stored in the graph database; Step S3, when the diagnosis result and the prediction result are inconsistent and there is a difference between the fault diagnosis accuracy and the fault prediction accuracy, the result with higher accuracy is preferentially used as the reference path according to the set accuracy threshold, and a new relationship path is established between the symptom nodes actually diagnosed or predicted and these cause nodes. At the same time, the original low-accuracy path is set as the suboptimal path, its priority is adjusted, and the optimization process and results are recorded in the graph database.

[0016] As a further solution of the present invention, the fault impact factor represents the degree of deviation of the current fault from the compressor operating parameters; the fault probability factor represents the probability of a specific fault type occurring under the current operating state; the fault expansion factor represents the potential impact of the current fault on other components of the compressor; the fault tolerance time represents the time interval from the current fault state to the time when the equipment must be shut down for maintenance.

[0017] As a further solution of the present invention, the maintenance decision module is used to determine the corresponding maintenance measures by matching the association between the current cause node and the maintenance measure node in the current knowledge graph. When the fault severity exceeds the set priority threshold, the maintenance measures in the knowledge graph are triggered, and a maintenance plan is generated according to the current status of the equipment and historical maintenance records, and the fault data and maintenance records are uploaded to the big data analysis platform; the early warning module is used to generate early warning information according to the diagnosis and prediction results generated by the fault analysis module output; the human-computer interaction module is used to display the compressor operating status, fault diagnosis results and early warning information, and visualize the knowledge graph through the human-computer interaction module, allowing users to view detailed information on each node and its relationship, and manually edit and maintain nodes and relationship paths.

[0018] Compared with the prior art, the reciprocating natural gas compressor fault diagnosis system of the present invention has the following beneficial effects: 1. The present invention improves the accuracy of fault diagnosis and the ability to identify multiple fault states. Traditional methods usually rely on a single signal source, such as vibration or pressure signals, while the present invention improves the comprehensiveness of fault feature extraction by integrating multiple signal sources including vibration signals, temperature signals, and historical maintenance records. On this basis, the deep belief network and long short-term memory network deep learning models are used to effectively deal with the nonlinear and non-stationary characteristics of complex signals, and realize the identification of compressor fault types and the prediction of future faults.

[0019] 2. The present invention constructs a multi-level knowledge graph based on compressor faults, symptoms, causes and maintenance measures, and dynamically adjusts the weights and relationship paths of nodes in the graph to achieve a comprehensive assessment of the impact of faults; the maintenance decision module combines the knowledge graph and diagnostic results to generate a personalized maintenance plan, and automatically triggers early warnings and maintenance recommendations when the fault severity exceeds the set threshold; compared with traditional maintenance decisions that rely on manual experience, the present invention significantly improves the intelligence level and maintenance efficiency of compressor fault handling, and reduces the unplanned downtime of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a structural schematic diagram of the reciprocating natural gas compressor fault diagnosis system of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Reference Figure 1 ,A reciprocating natural gas compressor fault diagnosis system includes a data acquisition module, a data preprocessing module, a feature extraction module, a knowledge graph construction module, a fault analysis module, a maintenance decision module, an early warning module and a human-computer interaction module.

[0023] The data acquisition module in the embodiment of the present invention is respectively connected to the data preprocessing module and the human-computer interaction module, the data preprocessing module is connected to the feature extraction module, the feature extraction module is respectively connected to the knowledge graph construction module and the fault analysis module, the knowledge graph construction module is bidirectionally connected to the fault analysis module, the knowledge graph construction module is respectively connected to the maintenance decision module and the human-computer interaction module, the fault analysis module is respectively connected to the early warning module and the human-computer interaction module, and the early warning module is connected to the human-computer interaction module.

[0024] The data acquisition module is used to collect the operating parameters of the reciprocating natural gas compressor in real time, and the operating parameters include compressor cylinder pressure, cylinder temperature, crankshaft speed, vibration signal, bearing temperature, lubricating oil pressure, intake pressure, and exhaust pressure; the data acquisition module adopts a distributed data acquisition architecture, including a field data acquisition unit and a central data processing unit. The field data acquisition unit is installed at various key positions of the compressor and communicates with the central data processing unit through an industrial bus; each field data acquisition unit includes a sensor for collecting operating parameters.

[0025] The data acquisition module is also used to download the equipment's operating parameters, historical fault data and corresponding maintenance records over the past period of time in the big data analysis platform.

[0026] The data preprocessing module is used to filter, denoise and normalize the operating parameters collected by the data acquisition module; the data preprocessing module adopts an adaptive filtering algorithm; it can automatically adjust the filtering parameters according to the characteristics of different operating parameters to improve the filtering effect; the denoising process adopts the wavelet threshold denoising method, and realizes signal denoising by selecting the optimal wavelet basis and threshold function; the normalization process adopts the Z-score standardization method to make the parameters of different dimensions comparable.

[0027] The feature extraction module is used to extract fault features from the preprocessed operating parameters, and the fault features include time domain features, frequency domain features and time-frequency domain features; the feature extraction module adopts a multi-scale feature extraction method, which can extract fault features from different scales and angles, thereby improving the comprehensiveness and representativeness of the features.

[0028] The time domain features extracted by the feature extraction module in the embodiment of the present invention include mean, variance, peak factor, pulse factor, margin factor, waveform factor, signal energy, peak, root mean square value, and signal-to-noise ratio; the frequency domain features include main frequency, sideband, harmonic component, power spectral density, frequency center, frequency variance, frequency band energy ratio, and harmonic distortion; the time-frequency domain features include wavelet packet decomposition coefficients, intrinsic mode functions of empirical mode decomposition, instantaneous frequency, instantaneous amplitude, time-frequency spectrum, time-frequency energy distribution, and time-frequency representation with variable time-frequency resolution.

[0029] The knowledge graph construction module is used to construct a multi-level knowledge graph of compressor failure-symptom-cause-maintenance measures based on historical fault data, maintenance records and equipment operating parameters, and dynamically adjust node weights and relationship paths according to diagnosis results and prediction results.

[0030] The knowledge graph construction module in the embodiment of the present invention includes the following steps: Step A1, defining a fault node, a symptom node, a cause node and a maintenance measure node, wherein the fault node represents a specific compressor fault type, the symptom node represents an abnormal state of an operating parameter, the cause node represents a specific cause of the fault, and the maintenance measure node represents a method for handling the fault, and each node is set with a specific label according to a preset attribute; Step A2, according to the operating characteristics and fault evolution law of each component of the compressor, establish a relationship path between each node, connect the fault node with the corresponding symptom node, connect the symptom node with the corresponding cause node, and connect the cause node with the corresponding maintenance measure node; Step A3, based on historical fault data and maintenance records, the initial weights of each node and its relationship path are set according to the performance of the node and path in the historical data. The weights are used to represent the importance of each node and path in fault diagnosis and prediction.

[0031] The fault analysis module is used to comprehensively analyze the fault information of the compressor, and includes a fault diagnosis unit, a fault prediction unit and a fault severity assessment unit.

[0032] The fault diagnosis unit in the embodiment of the present invention is used to input the time domain features, frequency domain features and time-frequency domain features provided by the feature extraction module into the deep belief network model for feature learning and fault classification, including the following steps: Step B1, constructing a deep belief network model, which is composed of multiple stacked restricted Boltzmann machines; the number of nodes of each layer of restricted Boltzmann machines is adaptively determined according to the input feature dimension and the number of fault types; Step B2, pre-training the deep belief network using unlabeled samples to initialize network parameters; pre-training uses contrastive divergence algorithm to optimize network parameters by minimizing reconstruction error; Step B3, fine-tune the deep belief network using labeled samples to obtain the final classifier; the fine-tuning training uses a back propagation algorithm to optimize the network parameters by minimizing the cross entropy loss function; Step B4, input the extracted fault features into the trained classifier, and output the fault diagnosis results, including the fault type and fault location.

[0033] The fault diagnosis unit in the embodiment of the present invention uses a multi-source data fusion algorithm to perform feature fusion processing based on the time domain features, frequency domain features and time-frequency domain features obtained from the feature extraction module, and uses the fused feature data to construct a deep belief network model to classify the fault types, wherein the calculation formula for the fault diagnosis accuracy of the deep belief network is:

[0034] In the formula, The number of samples for which the deep belief network model identified a specific fault type that actually existed in the compressor, The number of samples where the deep belief network model accurately identifies the type of fault that does not actually exist in the compressor, is the number of samples that the deep belief network model incorrectly identifies as having a certain fault type, is the number of samples that the deep belief network model fails to identify the actual fault type, and α and β are weight coefficients.

[0035] The calculation formula for the fault diagnosis accuracy in the embodiment of the present invention takes into account the importance of different fault types and can more reasonably evaluate the diagnostic performance, wherein α is the weight coefficient of a specific fault type, which is used to adjust the importance of the model diagnosis results according to the degree of influence of the faults of different components of the compressor on the operation of the equipment; β is the weight coefficient of the non-critical fault type, which is used to balance the importance of the diagnosis results of relatively less important fault types.

[0036] The fault prediction unit is used to predict possible future faults based on historical fault data and current operating parameters. The fault prediction module uses a long short-term memory network model to perform fault prediction, processes the time domain features and time-frequency domain features obtained from the feature extraction module, and uses these feature data as input to the long short-term memory network model to perform time series learning and pattern recognition of historical data, and predict the type of faults that may occur in the compressor in the future and the time when the faults will occur.

[0037] The fault prediction unit in the embodiment of the present invention comprises the following steps: Step C1, constructing a long short-term memory network model, including an input layer, multiple LSTM layers, a fully connected layer, and an output layer; the number of LSTM layers and the number of neurons in each layer are adaptively determined according to the length of the input sequence and the complexity of the prediction task; Step C2, using historical fault data to train the long short-term memory network; the training data uses a sliding window method to generate time series samples, and the network parameters are optimized by minimizing the mean square error loss function; Step C3, input the current operating parameters into the trained network model to predict possible future failures.

[0038] The fault severity evaluation unit is used to extract factors related to the fault based on the node information in the knowledge graph, and generate the fault severity value by calculating these factors to evaluate the comprehensive impact of the current fault on the operation of the equipment. The calculation formula of the fault severity is:

[0039] In the formula, is the fault influencing factor, is the failure probability factor, is the fault expansion factor, is the fault tolerance time.

[0040] The fault impact factor in the embodiment of the present invention indicates the degree of deviation of the current fault from the compressor operating parameters; the fault probability factor indicates the probability of a specific fault type occurring under the current operating state; the fault expansion factor indicates the potential impact of the current fault on other components of the compressor; the fault tolerance time indicates the time interval from the current fault state to the time when the equipment must be shut down for maintenance.

[0041] The fault impact factor in the embodiment of the present invention indicates the direct impact of a specific fault type on the key performance parameters of the compressor. It is calculated by the deviation degree of each feature data extracted in the feature extraction module, and reflects the impact of the fault on the overall performance of the equipment. The formula is:

[0042] In the formula, For the The actual value of a performance parameter under fault conditions; It is the benchmark value of the performance parameter under normal operating conditions; is the total number of performance parameters involved in the calculation.

[0043] The fault probability factor in the embodiment of the present invention represents the probability value of a specific fault occurring in the device under the current operating state. The factor reflects the possibility of the occurrence of the fault type under the current state by combining the prediction accuracy of the fault prediction unit with the historical fault data in the knowledge graph. The formula is:

[0044] In the formula, The prediction accuracy calculated for the fault prediction unit; It represents the conditional probability of a fault occurring under the current equipment state. By matching the time domain features and time-frequency domain feature data extracted from the fault prediction unit with the feature patterns of the corresponding fault types in the historical fault database, the similarity between the current data and the occurrence patterns of each fault type is calculated, and then the conditional probability value of a specific fault under the current data conditions is obtained; The frequency of the current equipment status appearing in past operation records.

[0045] The fault expansion factor in the embodiment of the present invention represents the secondary impact that the current fault type may have on other components or systems of the compressor. The factor is calculated by analyzing the association strength and relationship path between the fault node and other symptom nodes in the knowledge graph, and is used to measure the possibility of fault expansion. The calculation formula is:

[0046] In the formula, The fault node and The strength of the association between symptom nodes ranges from 0 to 1; is the trigger weight of the jth symptom node, indicating the probability of the symptom being triggered under the current fault type; is the number of symptom nodes involved in the calculation.

[0047] The fault tolerance time in the embodiment of the present invention indicates the time interval from the current fault state to the time when the equipment must be shut down for maintenance. The fault tolerance time is based on the historical fault evolution data of the equipment and the current operating state. It is calculated by analyzing the development trend of similar faults in historical data and current real-time data, and reflects the upper limit of the time that the equipment continues to operate safely in the current fault state. The formula is:

[0048] In the formula, The fault tolerance time limit is set, that is, the time point when the equipment performance index reaches irreversible damage; The current time point.

[0049] According to the analysis results of the fault diagnosis unit and the fault prediction unit in the embodiment of the present invention, the node weights and relationship paths in the knowledge graph are dynamically adjusted, including the following steps: Step S1, respectively obtain the current diagnosis result and prediction result from the fault diagnosis unit and the fault prediction unit, compare these results with the fault nodes and symptom nodes in the knowledge graph one by one, confirm whether the diagnosed fault type, fault severity and predicted symptom characteristics are consistent with the node information in the graph, and compare the reliability of the diagnosis result and the prediction result in combination with the fault diagnosis accuracy and the fault prediction accuracy, and record the matching results and deviation information; Step S2: if the diagnosis result and prediction result are consistent with the graph node information and the fault diagnosis accuracy and fault prediction accuracy both exceed the set threshold, the weight value of the corresponding node is increased according to the preset weight increment factor; if they are consistent but the accuracy does not reach the set threshold, the adjustment range of the weight increment factor is reduced; if the diagnosis result or prediction result is inconsistent with the graph node information and the accuracy reaches the set threshold, the weight value of the corresponding node is reduced according to the preset weight reduction factor; if they are inconsistent and the accuracy is lower than the set threshold, the adjustment range of the reduction factor is reduced, and the adjusted weight information and deviation information are stored in the graph database; Step S3, when the diagnosis result and the prediction result are inconsistent and there is a difference between the fault diagnosis accuracy and the fault prediction accuracy, the result with higher accuracy is preferentially used as the reference path according to the set accuracy threshold, and a new relationship path is established between the symptom nodes actually diagnosed or predicted and these cause nodes. At the same time, the original low-accuracy path is set as the suboptimal path, its priority is adjusted, and the optimization process and results are recorded in the graph database.

[0050] The maintenance decision module is used to determine the corresponding maintenance measures by matching the association between the current cause node and the maintenance measure node in the current knowledge graph. When the fault severity exceeds the set priority threshold, the maintenance measures in the knowledge graph are triggered, and a maintenance plan is generated based on the current status of the equipment and historical maintenance records. The fault data and maintenance records are then uploaded to the big data analysis platform.

[0051] The early warning module is used to generate early warning information based on the diagnosis and prediction results generated by the fault analysis module.

[0052] The human-computer interaction module is used to display the compressor operating status, fault diagnosis results and warning information, and to visualize the knowledge graph through the human-computer interaction module, allowing users to view detailed information on each node and its relationships, and to manually edit and maintain the nodes and relationship paths.

[0053] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0054] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A reciprocating natural gas compressor fault diagnosis system, comprising a data acquisition module, a data preprocessing module, a feature extraction module, a knowledge graph construction module, a fault analysis module, a maintenance decision module, an early warning module and a human-computer interaction module, characterized in that: The data acquisition module is connected to the data preprocessing module and the human-computer interaction module respectively, the data preprocessing module is connected to the feature extraction module, the feature extraction module is connected to the knowledge graph construction module and the fault analysis module respectively, the knowledge graph construction module is bidirectionally connected to the fault analysis module, the knowledge graph construction module is connected to the maintenance decision module and the human-computer interaction module respectively, the fault analysis module is connected to the early warning module and the human-computer interaction module respectively, and the early warning module is connected to the human-computer interaction module; the knowledge graph construction module is used to construct a multi-level knowledge graph of compressor fault-symptom-cause-maintenance measures based on historical fault data, maintenance records and equipment operation parameters, and dynamically adjust node weights and relationship paths according to diagnosis results and prediction results; The fault analysis module is used to comprehensively analyze the fault information of the compressor, and includes a fault diagnosis unit, a fault prediction unit and a fault severity assessment unit; wherein the fault diagnosis unit is used to input the time domain features, frequency domain features and time-frequency domain features provided by the feature extraction module into the deep belief network model for feature learning and fault classification, and output the fault diagnosis result; the fault prediction unit is used to predict future faults based on the long short-term memory network algorithm using the time domain features and time-frequency domain features as inputs, and evaluate the prediction accuracy, and the formula is: ; In the formula, is the prediction accuracy evaluation value, is the actual fault occurrence time, is the predicted failure time, is the maximum allowable time error, is the time weight function; The fault severity evaluation unit is used to extract factors related to the fault based on the node information in the knowledge graph, and generate the fault severity value by calculating these factors to evaluate the comprehensive impact of the current fault on the operation of the equipment. The calculation formula of the fault severity is: ; In the formula, is the fault severity, is the fault influencing factor, is the failure probability factor, is the fault expansion factor, is the fault tolerance time.

2. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The data acquisition module is used to collect the operating parameters of the reciprocating natural gas compressor in real time, and the operating parameters include compressor cylinder pressure, cylinder temperature, crankshaft speed, vibration signal, bearing temperature, lubricating oil pressure, intake pressure, and exhaust pressure; the data acquisition module adopts a distributed data acquisition architecture, including a field data acquisition unit and a central data processing unit. The field data acquisition unit is installed in various parts of the compressor and communicates with the central data processing unit through an industrial bus; each field data acquisition unit includes a sensor for collecting operating parameters.

3. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The data acquisition module is also used to download the operating parameters, historical fault data and corresponding maintenance records of the equipment in the past period of time in the big data analysis platform; the data preprocessing module is used to filter, denoise and normalize the operating parameters collected by the data acquisition module.

4. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The feature extraction module is used to extract fault features from the preprocessed operating parameters, wherein the fault features include time domain features, frequency domain features and time-frequency domain features; The time domain features extracted by the feature extraction module include mean, variance, peak factor, pulse factor, margin factor, waveform factor, signal energy, peak, root mean square value, and signal-to-noise ratio; the frequency domain features include main frequency, sideband, harmonic component, power spectrum density, frequency center, frequency variance, frequency band energy ratio, and harmonic distortion; the time-frequency domain features include wavelet packet decomposition coefficients, intrinsic mode functions of empirical mode decomposition, instantaneous frequency, instantaneous amplitude, time-frequency spectrum, time-frequency energy distribution, and time-frequency representation with variable time-frequency resolution.

5. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The knowledge graph construction module includes the following steps: Step A1, defining a fault node, a symptom node, a cause node and a maintenance measure node, wherein the fault node represents a specific compressor fault type, the symptom node represents an abnormal state of an operating parameter, the cause node represents a specific cause of the fault, and the maintenance measure node represents a method for handling the fault, and each node is set with a specific label according to a preset attribute; Step A2, according to the operating characteristics and fault evolution law of each component of the compressor, establish a relationship path between each node, connect the fault node with the corresponding symptom node, connect the symptom node with the corresponding cause node, and connect the cause node with the corresponding maintenance measure node; Step A3, based on historical fault data and maintenance records, the initial weights of each node and its relationship path are set according to the performance of the node and path in the historical data. The weights are used to represent the importance of each node and path in fault diagnosis and prediction.

6. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The fault diagnosis unit uses a multi-source data fusion algorithm to perform feature fusion processing based on the time domain features, frequency domain features and time-frequency domain features obtained from the feature extraction module, and uses the fused feature data to construct a deep belief network model to classify the fault types, wherein the fault diagnosis accuracy of the deep belief network is evaluated, and the calculation formula is: ; In the formula, is the fault diagnosis accuracy, The number of samples for which the deep belief network model identified a specific fault type that actually existed in the compressor, The number of samples where the deep belief network model accurately identifies the type of fault that does not actually exist in the compressor, is the number of samples that the deep belief network model incorrectly identifies as having a certain fault type, is the number of samples where the deep belief network model fails to identify the actual fault type, and is the weight coefficient.

7. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The fault prediction unit is used to predict future faults based on historical fault data and current operating parameters. The fault prediction module uses a long short-term memory network model to perform fault prediction, processes the time domain features and time-frequency domain features obtained from the feature extraction module, and uses these feature data as input to the long short-term memory network model to perform time series learning and pattern recognition of historical data, and predict the type of future compressor faults and the time when the faults will occur.

8. The reciprocating natural gas compressor fault diagnosis system according to any one of claims 1 to 7, characterized in that: According to the analysis results of the fault diagnosis unit and the fault prediction unit, the node weights and relationship paths in the knowledge graph are dynamically adjusted, including the following steps: Step S1, respectively obtain the current diagnosis result and prediction result from the fault diagnosis unit and the fault prediction unit, compare these results with the fault nodes and symptom nodes in the knowledge graph one by one, confirm whether the diagnosed fault type, fault severity and predicted symptom characteristics are consistent with the node information in the graph, and compare the reliability of the diagnosis result and the prediction result in combination with the fault diagnosis accuracy and the fault prediction accuracy, and record the matching results and deviation information; Step S2: if the diagnosis result and prediction result are consistent with the graph node information and the fault diagnosis accuracy and fault prediction accuracy both exceed the set threshold, the weight value of the corresponding node is increased according to the preset weight increment factor; if they are consistent but the accuracy does not reach the set threshold, the adjustment range of the weight increment factor is reduced; if the diagnosis result or prediction result is inconsistent with the graph node information and the accuracy reaches the set threshold, the weight value of the corresponding node is reduced according to the preset weight reduction factor; if they are inconsistent and the accuracy is lower than the set threshold, the adjustment range of the reduction factor is reduced, and the adjusted weight information and deviation information are stored in the graph database; Step S3, when the diagnosis result and the prediction result are inconsistent and there is a difference between the fault diagnosis accuracy and the fault prediction accuracy, the result with higher accuracy is preferentially used as the reference path according to the set accuracy threshold, and a new relationship path is established between the symptom nodes actually diagnosed or predicted and these cause nodes. At the same time, the original low-accuracy path is set as the suboptimal path, its priority is adjusted, and the optimization process and results are recorded in the graph database.

9. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The fault impact factor indicates the degree of deviation of the current fault from the compressor operating parameters; the fault probability factor indicates the probability of a specific fault type occurring under the current operating state; The fault expansion factor represents the potential impact of the current fault on other components of the compressor; The fault tolerance time represents the time interval from the current fault state to the time when the equipment must be shut down for maintenance.

10. The reciprocating natural gas compressor fault diagnosis system according to claim 1, characterized in that: The maintenance decision module is used to determine the corresponding maintenance measures by matching the association between the current cause node and the maintenance measure node in the current knowledge graph. When the fault severity exceeds the set priority threshold, the maintenance measures in the knowledge graph are triggered, and a maintenance plan is generated according to the current status of the equipment and historical maintenance records, and the fault data and maintenance records are uploaded to the big data analysis platform; the early warning module is used to generate early warning information according to the generated diagnosis and prediction results output by the fault analysis module; The human-computer interaction module is used to display the compressor operating status, fault diagnosis results and warning information, and visualize the knowledge graph through the human-computer interaction module, allowing users to view detailed information on each node and its relationship, and manually edit and maintain the nodes and relationship paths.

Images