Equipment fault intelligent diagnosis system based on knowledge graph and deep learning

Abstract

The invention relates to the technical field of equipment fault diagnosis, in particular to an intelligent equipment fault diagnosis system based on a knowledge graph and deep learning, which comprises a data acquisition module, a knowledge graph construction module, a deep learning reasoning module, a diagnosis output module and a feedback optimization module. According to the system, real-time reflection of the equipment operation state is realized through multi-source heterogeneous data acquisition and knowledge graph dynamic modeling, and the fault recognition capability is improved in combination with a double-branch deep learning network and an attention mechanism. A diagnosis result is displayed in a graphical interface, and closed-loop optimization is supported. The device state change can be comprehensively captured, the complex working condition adaptability is enhanced, and the diagnosis accuracy and the intelligent level are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent diagnosis and fault detection, and specifically is an intelligent equipment fault diagnosis system based on knowledge graph and deep learning. Background Art

[0002] The present invention relates to the technical field of intelligent diagnosis of equipment failures, and specifically to an intelligent diagnosis system for equipment failures based on knowledge graphs and deep learning.

[0003] Existing intelligent equipment fault diagnosis systems have played a significant role in improving equipment reliability, reducing downtime, and lowering maintenance costs. However, they still have limitations in knowledge integration capabilities, data correlation analysis, and adaptability to complex operating conditions, which impact diagnostic accuracy and efficiency. For example, Publication No. CN115235612B discloses an intelligent servo motor fault diagnosis system and method. This system uses Fourier transforms and context encoders to extract frequency-domain statistical features from vibration signals, and combines them with a convolutional neural network model to perform deep implicit feature mining on the vibration signals, enabling accurate diagnosis of servo motor faults. This technical solution primarily relies on a single data source (the vibration signal) and lacks the ability to integrate and analyze heterogeneous data from multiple sources. Consequently, it may not fully capture changes in equipment operating status under complex operating conditions. Furthermore, this solution fails to fully utilize knowledge graph technology to systematically model equipment operating knowledge and fault modes, making it difficult to deeply reason about and explain the causes of faults.

[0004] In addition, a method for intelligent fault diagnosis and early warning of distribution switch cabinets with publication number CN118839205B realizes early identification and accurate early warning of thermal faults in distribution switch cabinets by collecting historical thermal fault cases and constructing a digital model for early warning evaluation. However, this technical solution is mainly based on statistical analysis of historical data and lacks the ability to respond in real time to dynamic changes during equipment operation. At the same time, the extraction of fault features by this solution relies on manually set parameters and thresholds, and may show low adaptability and generalization ability when faced with new faults or unknown working conditions. In addition, this solution does not introduce technical means that combine deep learning with knowledge graphs, making it difficult to achieve efficient identification and intelligent reasoning of complex fault patterns.

[0005] The above issues demonstrate that existing intelligent equipment fault diagnosis systems still have shortcomings in terms of multi-source data fusion, dynamic real-time response, and intelligent reasoning. Therefore, the present invention provides an intelligent equipment fault diagnosis system based on knowledge graphs and deep learning. This system aims to improve the comprehensiveness, accuracy, and adaptability of equipment fault diagnosis by constructing an equipment operation knowledge graph, integrating multi-source heterogeneous data, and utilizing deep learning technology to achieve accurate fault identification and intelligent reasoning under complex operating conditions. This system meets the needs of modern industry for intelligent fault diagnosis systems. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent equipment fault diagnosis system based on knowledge graph and deep learning to address the deficiencies of the existing technology mentioned in the above background technology in terms of multi-source data fusion, dynamic real-time response and intelligent reasoning capabilities.

[0007] The technical solution of the present invention comprises a data acquisition module, a knowledge graph construction module, a deep learning inference module, and a diagnostic output module. The data acquisition module collects multi-source heterogeneous data generated during equipment operation, including vibration signals, temperature signals, current signals, voltage signals, and environmental parameter signals. These signals are acquired in real time via a sensor network and filtered, normalized, and timestamp-aligned by a preprocessing unit. The knowledge graph construction module automatically constructs a knowledge graph of equipment operation based on the collected data. Nodes in the knowledge graph represent key equipment components and their states, while edges represent logical or causal relationships between components. Node and edge attributes store relevant operating parameters and historical failure modes. The deep learning inference module utilizes a hybrid model combining an improved convolutional neural network and a recurrent neural network to perform feature extraction and pattern recognition on the input multi-source data. It also performs inference analysis based on the logical relationships in the knowledge graph, thereby enabling accurate fault identification under complex operating conditions. The diagnostic output module presents the inference results in a visual format and generates a diagnostic report that includes the fault type, possible cause, and repair recommendations.

[0008] Furthermore, the data acquisition module design includes multiple submodules, each corresponding to a specific type of signal. For example, the vibration signal acquisition submodule uses a high-precision accelerometer, installed near the equipment's key moving components to ensure it can capture high-frequency vibration information. The temperature signal acquisition submodule uses a thermocouple sensor, installed near the equipment's heat-generating components. The cold end of the thermocouple is connected to the data acquisition unit via a compensation circuit to eliminate the effects of ambient temperature fluctuations. The current and voltage signal acquisition submodules are implemented using Hall-effect sensors, which are bolted to the equipment's power supply circuits. The signals are transmitted to the data acquisition unit via shielded cables. All signal acquisition submodules are connected to the central processing unit via industrial Ethernet, ensuring real-time and reliable data transmission.

[0009] Furthermore, the knowledge graph construction module realizes real-time modeling of the equipment operating status through a dynamic update mechanism. The initial knowledge graph is constructed based on the equipment's design documents and historical maintenance records, where the initial attribute values of the nodes are derived from the equipment's standard operating parameters. As equipment operating data is continuously collected, the node attributes and edge weights in the knowledge graph will be dynamically adjusted according to the actual operating conditions. For example, when the temperature signal of a certain component frequently exceeds the normal range, the "abnormal probability" attribute value of the node will gradually increase, and the edge weights related to it will also increase accordingly to reflect the increased impact of the component on the overall system stability. In addition, the knowledge graph also introduces a rule-based reasoning engine to verify whether the newly added data conforms to known physical laws and logical constraints.

[0010] Furthermore, the core of the deep learning inference module is a two-branch network structure, with one branch responsible for processing time series data and the other for processing image or spectrum data. The time series branch uses a bidirectional long short-term memory network (BiLSTM) to extract implicit features by capturing the temporal dependencies of device operating states. The image or spectrum branch uses a modified residual convolutional network (ResNet) to extract spatial features by analyzing signal spectrograms or device operating state graphs. The outputs of the two branches are weighted and fused using an attention mechanism to ultimately generate a comprehensive feature vector. The comprehensive feature vector is then input into a fully connected layer and matched with logical relationships in the knowledge graph to determine whether the current device state deviates from the normal range. If an anomaly is detected, the fault diagnosis process is triggered.

[0011] Furthermore, a series of experimental tests were designed to evaluate the performance of the deep learning inference module. The testing environment included both simulation experiments and actual equipment operation scenarios. The simulation experiments were implemented by building a virtual equipment operation platform. The platform used software simulation tools to generate equipment operation data under different operating conditions, including normal operation data and data on various failure modes. The actual equipment operation scenario selected key equipment on an industrial production line as test subjects, and real-world operation data was obtained through long-term continuous monitoring. The tests focused on verifying the inference module's adaptability to new faults or unknown operating conditions, as well as its ability to integrate and analyze heterogeneous data from multiple sources.

[0012] Furthermore, the design of the diagnostic output module prioritizes user-friendly human-computer interaction. Diagnostic results are displayed in a graphical interface divided into three sections: the first section provides an overview of the device's operating status, displaying the real-time operating parameters and health index of key components; the second section provides fault details, listing the detected fault type and its corresponding confidence score; and the third section provides maintenance recommendations, generating specific repair steps and precautions based on historical cases and logical reasoning within the knowledge graph. Diagnostic reports can be exported to PDF format for easy archiving and reference by technicians.

[0013] Furthermore, the system includes a feedback optimization module for continuous improvement of diagnostic performance. This module iteratively optimizes the knowledge graph and deep learning model by collecting user evaluation data on diagnostic results. For example, when a user indicates a false positive in a diagnostic result, the feedback optimization module reanalyzes the relevant data and adjusts the attribute values of the corresponding node in the knowledge graph or the parameter settings of the deep learning model. This closed-loop optimization mechanism ensures that the system's diagnostic accuracy continues to improve over time.

[0014] The present invention provides an improved intelligent equipment fault diagnosis system based on knowledge graph and deep learning. Compared with the existing technology, it has the following characteristics and advantages:

[0015] By employing a data acquisition module and a knowledge graph construction module, we achieve comprehensive collection and systematic modeling of multi-source heterogeneous data during equipment operation, resolving the problem of a single data source failing to fully capture changes in equipment operating status. The knowledge graph's dynamic update mechanism enables the system to reflect the actual operating status of equipment in real time, enhancing its adaptability to complex operating conditions.

[0016] By employing a deep learning inference module, combined with time series analysis and image feature extraction techniques, we achieve multi-level feature mining of equipment operating status. The introduction of a dual-branch network structure and attention mechanism significantly enhances the system's ability to integrate and analyze multi-source data, enabling it to demonstrate greater adaptability and generalization capabilities when faced with new faults or unknown operating conditions.

[0017] By utilizing the diagnostic output module and the feedback optimization module, the system provides an intuitive display of diagnostic results and a continuous optimization mechanism. The diagnostic output module's graphical interface allows technicians to quickly understand equipment operating status and fault information, while the feedback optimization module continuously improves the system's diagnostic performance through a closed-loop mechanism, meeting modern industry's demand for intelligent fault diagnosis systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is the overall structural block diagram of the system of the present invention.

[0019] Figure 2 This is a schematic diagram of the composition of the data acquisition module.

[0020] Figure 3 A dynamically updated flowchart for building modules for the knowledge graph.

[0021] Figure 4 This is a dual-branch network structure diagram of the deep learning reasoning module.

[0022] Figure 5 This is a schematic diagram of the interface of the diagnostic output module. DETAILED DESCRIPTION

[0023] The present invention provides an intelligent equipment fault diagnosis system based on knowledge graph and deep learning, the overall structure of which is as follows: Figure 1 As shown in Figure 1, the system includes a data acquisition module, a knowledge graph construction module, a deep learning inference module, a diagnostic output module, and a feedback optimization module. These modules are connected to the central processing unit via industrial Ethernet, forming a complete closed-loop system. The specific implementation of the system will be described in detail below with reference to the accompanying drawings.

[0024] The data acquisition module is the foundation of the entire system and is used to obtain multi-source heterogeneous data generated during the operation of the equipment in real time. Figure 2 As shown in the schematic diagram, the data acquisition module consists of multiple submodules, including a vibration signal acquisition submodule, a temperature signal acquisition submodule, a current and voltage signal acquisition submodule, and an environmental parameter acquisition submodule. The vibration signal acquisition submodule uses a high-precision accelerometer and is mounted near key moving parts of the equipment, such as the motor shaft or gearbox housing, to ensure that it can capture high-frequency vibration information. The accelerometer is bolted to the surface of the equipment and transmits its signal to the central processing unit via a shielded cable. The temperature signal acquisition submodule uses a thermocouple sensor and is mounted near heat-generating components of the equipment, such as the motor windings or bearing housing. The cold end of the thermocouple is connected to the data acquisition unit via a compensation circuit to eliminate the influence of ambient temperature fluctuations on the measurement results. The current and voltage signal acquisition submodule uses Hall-effect sensors, which are bolted to the power supply circuit of the equipment. The signals are also transmitted to the central processing unit via a shielded cable. The environmental parameter acquisition submodule is installed in the surrounding environment of the equipment to monitor external conditions such as humidity and air pressure. All submodules are connected to the central processing unit via Industrial Ethernet to ensure real-time and reliable data transmission. The data acquisition module also includes a pre-processing unit for filtering, normalizing and time-stamp aligning the received original signal so that subsequent modules can further analyze it.

[0025] The knowledge graph construction module automatically constructs the equipment operation knowledge graph based on the data collected by the data acquisition module. The construction process of the initial knowledge graph is as follows: Figure 3As shown, key information is first extracted from the device's design documents and historical maintenance records to initialize the nodes and edges in the knowledge graph. Nodes represent key components of the device and their status, such as motors, bearings, and gearboxes. Node attribute values are derived from the device's standard operating parameters. Edges represent logical or causal relationships between components, such as the drive relationship between a motor and a bearing. As device operating data is continuously collected, the node attributes and edge weights in the knowledge graph are dynamically adjusted based on actual operating conditions. For example, if a component's temperature signal frequently exceeds the normal range, the node's "abnormal probability" attribute value will gradually increase, and the weights of its associated edges will also increase accordingly, reflecting the component's increasing impact on overall system stability. Furthermore, the knowledge graph incorporates a rule-based reasoning engine to verify that newly added data conforms to known physical laws and logical constraints. For example, if an abnormal temperature rise is detected for a component, the reasoning engine will check for other potential contributing factors, such as current overload or poor heat dissipation, thereby ensuring the accuracy and consistency of the knowledge graph.

[0026] The deep learning reasoning module is the core part of the system, and its structure is as follows Figure 4 As shown in Figure 2, this module utilizes a two-branch network structure, with one branch processing time series data and the other processing image or spectrum data. The time series branch uses a bidirectional long short-term memory (BiLSTM) network to extract implicit features by capturing the temporal dependencies of the device's operating status. For example, the BiLSTM can analyze vibration signal trends over time and identify potential periodic fault patterns. The image or spectrum branch uses a modified residual convolutional network (ResNet) to extract spatial features by analyzing the signal spectrogram or device operating status graph. For example, a ResNet can extract features from the vibration signal's spectrogram and identify abnormal signals within a specific frequency range. The outputs of the two branches are weighted and fused using an attention mechanism to generate a comprehensive feature vector. This comprehensive feature vector is then input into a fully connected layer, where it is matched against logical relationships in the knowledge graph to determine whether the current device status deviates from the normal range. If an anomaly is detected, the fault diagnosis process is triggered. For example, if the comprehensive feature vector is highly similar to the feature vector of a fault pattern in the knowledge graph, the system determines that the device has a corresponding fault and initiates the subsequent fault analysis and diagnosis process.

[0027] The design of the diagnostic output module focuses on the friendliness of human-computer interaction. Its interface layout is as follows: Figure 5As shown. The diagnostic results are displayed in the form of a graphical interface, which is divided into three parts. The first part is an overview of the equipment's operating status, showing the real-time operating parameters of each key component and its health index. For example, the interface can display the motor's speed, temperature, and vibration amplitude, as well as the health index corresponding to these parameters. The second part is the fault details, which lists the detected fault types and their corresponding confidence scores. For example, the interface can display the fault type of early wear of the bearing, and the confidence score of the fault is 85%. The third part is the maintenance recommendation, which generates specific maintenance steps and precautions based on historical cases and logical reasoning in the knowledge graph. For example, for early wear of the bearing, the system will recommend replacing the lubricating oil and checking whether the bearing is loose. The diagnostic report can be exported to PDF format for archiving and reference by technicians.

[0028] The feedback optimization module iteratively optimizes the knowledge graph and deep learning model by collecting user evaluation data on diagnostic results. For example, when a user points out that a diagnostic result is a false positive, the feedback optimization module reanalyzes the relevant data and adjusts the attribute values of the corresponding node in the knowledge graph or the parameter settings of the deep learning model. This closed-loop optimization mechanism ensures that the system can continuously improve its diagnostic accuracy as time goes by. For example, in a certain diagnosis, the system mistakenly judged that the motor had an overheating fault, but user feedback indicated that the actual cause was excessively high ambient temperature. Based on this information, the feedback optimization module adjusts the "abnormal probability" attribute value of the motor node in the knowledge graph and optimizes the parameters of the deep learning model to avoid similar misjudgments.

[0029] In practical applications, this system can be deployed on key equipment on industrial production lines, such as CNC machine tools or wind turbines. The data acquisition module obtains equipment operating data in real time through a sensor network and transmits the data to the central processing unit. The knowledge graph construction module dynamically updates the knowledge graph based on the collected data to ensure that it can reflect the actual operating status of the equipment in real time. The deep learning reasoning module extracts features and recognizes patterns from multi-source data, and performs reasoning analysis based on the logical relationships in the knowledge graph, thereby achieving accurate identification of faults under complex working conditions. The diagnostic output module presents the reasoning results in a visual manner and generates a diagnostic report containing the fault type, possible causes, and maintenance recommendations. The feedback optimization module continuously improves the system's diagnostic performance through a closed-loop mechanism to meet the needs of modern industry for intelligent fault diagnosis systems.

[0030] In order to better enable relevant personnel in this technical field to fully understand and implement the present invention, the specific implementation principle of the present invention is supplemented below with reference to a specific application scenario.

[0031] Before deploying the system, the sensor network must be installed and configured on the target device. For example, in an industrial CNC machine tool, the high-precision accelerometer in the vibration signal acquisition submodule is mounted on the machine tool's spindle housing and bolted securely to ensure stable signal transmission. The thermocouple sensor in the temperature signal acquisition submodule is installed near the motor windings, with its cold end connected to the central processing unit (CPU) via a compensation circuit to eliminate the effects of ambient temperature fluctuations. The Hall-effect sensor in the current and voltage signal acquisition submodule is attached to the power supply line, monitoring real-time changes in the motor's electrical parameters during operation. The environmental parameter acquisition submodule is deployed around the machine tool to collect external environmental data such as humidity and air pressure. All sensors are connected to the CPU via shielded cables, and real-time data transmission is achieved via industrial Ethernet. The data acquisition module's preprocessing unit then filters, normalizes, and timestamp-aligns the received raw signals, providing standardized data input for subsequent analysis.

[0032] Next, the knowledge graph construction module initializes the knowledge graph based on the equipment's design documents and historical maintenance records. For example, for a CNC machine tool, the nodes in the knowledge graph include key components such as the spindle, motor, and bearings, while edges represent the logical or causal relationships between these components, such as how the spindle drives the bearings. Initial node attribute values are derived from the equipment's standard operating parameters, such as the spindle's rated speed and the motor's operating temperature range. As equipment operating data is continuously collected, the node attributes and edge weights in the knowledge graph are dynamically adjusted. For example, if the spindle's vibration signal frequently exceeds the normal range, the "abnormal probability" attribute value of the spindle node gradually increases, and the weights of the associated edges also increase accordingly. Furthermore, a rule-based reasoning engine verifies whether newly added data conforms to physical laws. For example, if an abnormally high spindle temperature is detected, the reasoning engine checks for other related factors, such as insufficient coolant flow or excessive load, to ensure the accuracy and consistency of the knowledge graph.

[0033] Subsequently, the deep learning inference module performs feature extraction and pattern recognition on multi-source heterogeneous data. Taking the spindle vibration signal as an example, the time series branch uses a bidirectional long short-term memory network (BiLSTM) to analyze the vibration signal's changing trends over time and identify potential periodic fault patterns. Simultaneously, the image or spectrum branch uses a modified residual convolutional network (ResNet) to extract features from the vibration signal's spectrogram and identify abnormal signals within a specific frequency range. The outputs of the two branches are weighted and fused using an attention mechanism to generate a comprehensive feature vector. This feature vector is then input into a fully connected layer to be matched against logical relationships in the knowledge graph. For example, if the comprehensive feature vector is highly similar to the feature vector of a spindle wear fault pattern in the knowledge graph, the system determines that the spindle has a corresponding fault and initiates the subsequent fault analysis process.

[0034] The diagnostic output module presents the inference results to the user in a graphical interface. This interface is divided into three sections: the first section provides an overview of the equipment's operating status, such as spindle speed, temperature, and vibration amplitude, along with the corresponding health index for these parameters; the second section lists the detected fault type and its confidence score, such as early wear of the spindle bearing, with a confidence score of 85%; and the third section provides repair recommendations, such as replacing the lubricant and checking for loose bearings. The diagnostic report can be exported to PDF format for easy archiving and reference by technicians.

[0035] Finally, the feedback optimization module continuously improves the system's diagnostic performance through a closed-loop mechanism. For example, when a user points out a false positive in a diagnostic result, the feedback optimization module reanalyzes the relevant data and adjusts the attribute values of the corresponding node in the knowledge graph or the parameter settings of the deep learning model. For example, in a particular diagnosis, the system mistakenly identified a motor overheating fault, but user feedback indicated that the actual cause was excessively high ambient temperature. Based on this information, the feedback optimization module adjusts the "abnormal probability" attribute value of the motor node in the knowledge graph and optimizes the parameters of the deep learning model to avoid similar misjudgments.

[0036] Through the above steps, this system can achieve accurate identification and intelligent reasoning of equipment failures under complex working conditions, meeting the needs of modern industry for intelligent fault diagnosis systems.

[0037] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0038] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent equipment fault diagnosis system based on knowledge graph and deep learning, characterized by: It includes a data acquisition module, a knowledge graph construction module, a deep learning reasoning module and a diagnostic output module. The data acquisition module is used to collect multi-source heterogeneous data generated during the operation of the equipment. The knowledge graph construction module automatically constructs the equipment operation knowledge graph based on the collected data. The deep learning reasoning module uses a hybrid model combining an improved convolutional neural network and a recurrent neural network to perform feature extraction and pattern recognition on the input multi-source data, and performs reasoning analysis based on the logical relationship in the knowledge graph. The diagnostic output module presents the reasoning results in a visual manner and generates a diagnostic report.

2. The intelligent equipment fault diagnosis system according to claim 1, characterized in that: The data acquisition module includes a vibration signal acquisition submodule, a temperature signal acquisition submodule, a current and voltage signal acquisition submodule, and an environmental parameter acquisition submodule. The vibration signal acquisition submodule uses a high-precision accelerometer, the temperature signal acquisition submodule uses a thermocouple sensor, and the current and voltage signal acquisition submodule is implemented through a Hall effect sensor. All submodules are connected to the central processing unit via industrial Ethernet.

3. The intelligent equipment fault diagnosis system according to claim 2, characterized in that: The data acquisition module further includes a pre-processing unit for performing filtering, normalization and time stamp alignment operations on the received original signal.

4. The intelligent equipment fault diagnosis system according to claim 1, characterized in that: The knowledge graph construction module realizes real-time modeling of the equipment operation status through a dynamic update mechanism. The initial knowledge graph is constructed based on the equipment's design documents and historical maintenance records, and the node attributes and edge weights are dynamically adjusted according to changes in the equipment operation data.

5. The intelligent equipment fault diagnosis system according to claim 4, characterized in that: The knowledge graph construction module introduces a rule-based reasoning engine to verify whether the newly added data conforms to known physical laws and logical constraints.

6. The intelligent equipment fault diagnosis system according to claim 1, characterized in that: The core of the deep learning inference module is a two-branch network structure, one of which is a time series branch that uses a bidirectional long short-term memory network (BiLSTM) to process time series data, and the other is an image or spectrum branch that uses an improved residual convolutional network (ResNet) to process image or spectrum data. The outputs of the two branches are weighted and fused through an attention mechanism.

7. The intelligent equipment fault diagnosis system according to claim 6, characterized in that: The comprehensive feature vector of the deep learning reasoning module is input into the fully connected layer and matched with the logical relationship in the knowledge graph to determine whether the device status deviates from the normal range.

8. The intelligent equipment fault diagnosis system according to claim 1, characterized in that: The interface of the diagnostic output module is divided into three parts: equipment operating status overview, fault details and maintenance suggestions. The diagnostic report can be exported in PDF format.

9. The intelligent equipment fault diagnosis system according to claim 1, characterized in that: It also includes a feedback optimization module for collecting user evaluation data on diagnostic results and iteratively optimizing the knowledge graph and deep learning model.

10. The intelligent equipment fault diagnosis system according to claim 9, characterized in that: The feedback optimization module reanalyzes relevant data and adjusts the attribute values of corresponding nodes in the knowledge graph or the parameter settings of the deep learning model when a false alarm is detected.

Images