Fault evaluation management method and system based on AI Agent

Abstract

The invention relates to an AI Agent-based fault evaluation management method and system, and the method comprises the steps: obtaining the multi-source heterogeneous data of industrial equipment, such as vibration, temperature, pressure and lubricating oil quality parameters; constructing a first feature association matrix and a second feature association matrix, and combining the first feature association matrix and the second feature association matrix to generate a comprehensive feature association map; the AI Agent is used for analyzing the running state of the equipment in real time based on the atlas and outputting a fault evaluation result, so that whether a fault exists or not can be accurately judged, the fault type and the severity degree can be determined, and a processing scheme can also be given; according to the scheme, the accuracy and comprehensiveness of fault prediction detection are greatly improved, potential faults of the equipment can be responded in time, the downtime of the equipment is effectively shortened, then the fault evaluation management accuracy and the equipment maintenance efficiency of the industrial equipment are improved, and many risks such as economic losses caused by the faults are reduced.

Description

Technical Field

[0001] This application relates to the fields of intelligent manufacturing and industrial safety, and information technology, and in particular to a fault evaluation and management method and system based on AI Agent. Background Technology

[0002] Fault assessment and management is a crucial aspect of industrial equipment maintenance and operation. Its main objective is to identify potential faults and provide solutions by monitoring and analyzing the equipment's operating status. Traditional fault assessment and management methods typically rely on data collected by sensors, such as vibration, temperature, and pressure sensors. This data reflects changes in key equipment operating parameters. For example, monitoring the vibration frequency of a motor can determine if there is mechanical loosening; detecting the temperature of the lubricating oil can assess whether the equipment's friction condition is abnormal. When the monitored data exceeds the normal range, the system issues a warning signal, indicating a potential fault risk.

[0003] However, a single sensor can only provide information from a specific dimension, making it difficult to comprehensively capture the characteristics of complex fault modes. Furthermore, the integrated analysis of multi-sensor data typically requires the experience and knowledge of domain experts to achieve accurate fault diagnosis and prediction. This human-assisted approach has certain limitations in efficiency, especially when facing dynamically changing operating conditions or large-scale equipment clusters. Insufficient information integration or human factors can easily lead to decreased fault prediction accuracy or delayed response, resulting in inaccurate fault assessment and management of industrial equipment and low equipment maintenance efficiency. Summary of the Invention

[0004] The main purpose of this application is to provide a fault evaluation and management method and system based on AI Agent, which can use AI technology to predict the fault status of industrial equipment, improve the accuracy of fault prediction and diagnosis of industrial equipment, and thus improve the accuracy of fault evaluation and management and equipment maintenance efficiency.

[0005] To achieve the above objectives, embodiments of the present invention provide a fault evaluation and management method based on an AI Agent, the method comprising: acquiring multi-source heterogeneous data generated during the operation of industrial equipment, wherein the multi-source heterogeneous data includes vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters; Based on the vibration signal and the temperature signal, a first feature correlation matrix is ​​constructed, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the operating state of the equipment. Based on the pressure signal and the lubricating oil quality parameters, a second feature correlation matrix is ​​constructed, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters; A comprehensive feature association map is generated based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association map is used to integrate the global association characteristics between multi-source heterogeneous data; The AI ​​Agent uses the comprehensive feature association map to perform real-time analysis of the equipment's operating status and outputs fault evaluation results.

[0006] Accordingly, embodiments of this application also provide a fault assessment and management system based on an AI Agent, including: The acquisition module is used to acquire multi-source heterogeneous data generated during the operation of industrial equipment. The multi-source heterogeneous data includes vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters. The first feature processing module is used to construct a first feature correlation matrix based on the vibration signal and the temperature signal, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the device operating state; The second feature processing module is used to construct a second feature correlation matrix based on the pressure signal and the lubricating oil quality parameters, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters; The graph generation module is used to generate a comprehensive feature association graph based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association graph is used to integrate the global association characteristics between multi-source heterogeneous data; The analysis and processing module is used to perform real-time analysis of the equipment's operating status based on the comprehensive feature association map using the AI ​​Agent, and output fault evaluation results.

[0007] In summary, the technical solution of this application, by acquiring multi-source heterogeneous data from industrial equipment, such as vibration, temperature, pressure, and lubricating oil quality parameters, provides a comprehensive understanding of the equipment's operating status. The constructed first feature correlation matrix quantifies the interaction between vibration and temperature signals, while the second feature correlation matrix clarifies the coupling degree between pressure and lubricating oil quality parameters. The comprehensive feature correlation map generated by combining these two matrices integrates global correlation characteristics, providing detailed data support for fault analysis. Furthermore, an AI agent is used to analyze the equipment's operating status in real time based on this map and output fault evaluation results. This accurately determines the existence of faults, identifies the fault type and severity, and provides solutions. This significantly improves the accuracy and comprehensiveness of fault prediction and detection, enables timely response to potential equipment faults, effectively reduces equipment downtime, and ultimately improves the accuracy of fault evaluation management and equipment maintenance efficiency, reducing economic losses and other risks caused by faults. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 This is a schematic diagram of a scenario for the fault evaluation and management method based on AI Agent in the embodiments of this application; Figure 2 A flowchart is provided for an embodiment of the fault evaluation and management method based on an AI Agent. Figure 3 This is a schematic diagram illustrating the process of generating the first feature association matrix provided in an embodiment of this application; Figure 4 Another schematic diagram illustrating the process of generating the first feature association matrix provided in the embodiments of this application; Figure 5 This is a schematic diagram illustrating the process of generating the second feature association matrix provided in an embodiment of this application; Figure 6 Another schematic diagram illustrating the process of generating the second feature association matrix provided in the embodiments of this application; Figure 7 A schematic diagram illustrating the process of generating a comprehensive feature association map provided in this application embodiment; Figure 8 A flowchart illustrating the generation of fault evaluation results is provided for embodiments of this application. Figure 9 A schematic diagram of the structure of the fault evaluation management system based on AI Agent provided in the embodiments of this application; Figure 10 A schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] This application provides a fault evaluation and management method and system based on AI Agent, which will be described in detail below.

[0012] In this embodiment, the AI ​​Agent, or artificial intelligence agent, is a highly intelligent and autonomous software entity or program module designed to simulate human intelligent behavior to achieve specific goals. It possesses the ability to perceive its environment, receiving and understanding various types of information from the external environment, such as acquiring real-time data on equipment operation in fault assessment and management scenarios. Simultaneously, it has a knowledge reserve and learning mechanism, continuously updating and improving its knowledge system either based on pre-set rules and models or by constantly analyzing and processing new data and experience. The AI ​​Agent also possesses powerful decision-making capabilities, able to comprehensively analyze and reason based on perceived information and its accumulated knowledge, using specific algorithms and logic to make decisions that meet the target requirements. Furthermore, it has execution capabilities, taking corresponding actions after making a decision, such as automatically triggering repair operations or adjusting system parameters in fault handling scenarios. This allows it to operate flexibly in complex and ever-changing environments, effectively completing various specific tasks such as fault assessment and management, and greatly improving the automation and intelligence level of related work.

[0013] The fault evaluation and management method based on AI Agent provided in this application is an intelligent means of managing faults in the operating status of industrial equipment. First, it acquires multi-source heterogeneous data generated during the operation of the industrial equipment, including vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters. Next, a first feature correlation matrix is ​​constructed based on the vibration and temperature signals to quantify their interaction in the equipment's operating state, while a second feature correlation matrix is ​​constructed based on the pressure and lubricating oil quality parameters to quantify their coupling degree. Then, a comprehensive feature correlation map is generated based on these two feature correlation matrices to integrate the global correlation characteristics between the multi-source heterogeneous data. Finally, using an AI Agent based on this comprehensive feature correlation map, a multi-layer decision network constructed based on a deep reinforcement learning framework is used. The comprehensive feature correlation map is input and sequentially passed through each layer of the decision network for inference, outputting fault evaluation results including fault type, fault severity, and fault handling solutions. The results can also be verified based on a historical fault database, thereby generating a fault evaluation report and presenting it to the user through a visual interface. This achieves comprehensive evaluation and effective management of industrial equipment faults, solving problems such as information isolation and low manual efficiency in traditional methods, and improving the automation and intelligence level of fault management.

[0014] As shown in Figure 1, a fault evaluation and management method scenario based on AI Agent is provided. In the fault evaluation and management scenario based on AI Agent, in a large industrial manufacturing plant, it mainly includes factory equipment, data acquisition and processing center equipment, and AI Agent server. The equipment interacts with each other through the network.

[0015] The factory equipment includes large CNC machine tools, automated production line transmission equipment, various motors, and complex hydraulic systems. Each piece of equipment is equipped with multiple sensors to collect various data during operation. The data collected by these sensors constitutes multi-source heterogeneous data, which is then sent to the data acquisition and processing center.

[0016] Taking CNC machine tools as an example, key components are equipped with vibration sensors, temperature sensors, pressure sensors, and sensors for detecting coolant quality. Vibration sensors accurately sense the machine tool's vibration during operation, capturing vibration signals of different frequencies and amplitudes. For instance, when the cutting tool wears or the spindle bearing becomes slightly loose, the frequency components and amplitude of the vibration signal will change. Temperature sensors are distributed in areas prone to heat generation, such as the motor and spindle box, monitoring the equipment's temperature in real time. The collected temperature signals reflect the internal friction conditions and the operating temperature of electrical components. Pressure sensors are installed in the hydraulic system to monitor changes in hydraulic oil pressure; any blockage or leakage in the hydraulic circuit will cause abnormal pressure signals. Coolant quality sensors detect lubricating oil quality parameters such as the coolant's pH and impurity content. A decline in coolant quality affects the cooling effect on the cutting tool, thus impacting machining accuracy and even causing equipment malfunctions.

[0017] After receiving multi-source heterogeneous data from various devices, the data acquisition and processing center begins constructing a feature correlation matrix. First, for the vibration and temperature signals of the CNC machine tool, a first feature correlation matrix is ​​constructed based on these two signals. Then, the vibration signal undergoes time-frequency domain decomposition, and advanced signal processing algorithms, such as wavelet transform, are used to extract the frequency component distribution characteristics of the vibration signal. For example, the vibration signal is decomposed into components of different frequency bands to determine the energy distribution of each frequency band. Simultaneously, time-series modeling is performed on the temperature signal, using methods such as ARIMA (Autoregressive Integral Moving Average) models to extract the trend characteristics of the temperature signal, such as whether the temperature is gradually rising, falling, or fluctuating.

[0018] Then, the data acquisition and processing center cross-maps the frequency component distribution characteristics of the extracted vibration signal with the changing trend characteristics of the temperature signal to form a preliminary correlation matrix. This preliminary correlation matrix initially reflects the interaction between the vibration signal and the temperature signal during equipment operation, but it is not accurate enough. Next, nonlinear optimization algorithms, such as genetic algorithms and particle swarm optimization algorithms, are used to adjust the preliminary correlation matrix, optimizing parameters such as weight coefficients, so that the final first feature correlation matrix can more accurately quantify the interaction between the vibration signal and the temperature signal during equipment operation.

[0019] Similarly, the data acquisition and processing center constructs a second feature correlation matrix for the pressure signal and coolant quality parameters. The pressure signal is piecewise fitted, and the local extreme point distribution characteristics are extracted through the fitted curves, such as identifying the locations of peak and trough values ​​in the hydraulic system and their corresponding pressure values. Composition analysis is performed on the coolant quality parameters to determine the concentration distribution characteristics of key components, such as the specific values ​​of acidity / alkalinity and the distribution of impurities in the coolant. The local extreme point distribution characteristics of the pressure signal and the concentration distribution characteristics of the key components of the coolant quality parameters are jointly modeled to form an initial correlation matrix. Then, an adaptive weight allocation mechanism is used to optimize the initial correlation matrix, reasonably allocating weights according to the actual operating conditions of the equipment and the characteristics of the data, to obtain the final second feature correlation matrix. This matrix accurately quantifies the degree of coupling between the pressure signal and the coolant quality parameters.

[0020] After obtaining the first and second feature correlation matrices, the data acquisition and processing center merges them to form an initial correlation graph. Next, based on graph neural network technology, the initial correlation graph undergoes topological optimization to enhance its global feature representation capability. Graph neural networks can automatically learn the relationships between nodes in the graph, adjusting connection weights and other methods to better reflect the global correlation characteristics between multi-source heterogeneous data. Furthermore, by introducing a dynamic update mechanism, parameters such as node connection weights in the correlation graph are continuously updated according to real-time changes in device operating status, ultimately resulting in a comprehensive feature correlation graph. This comprehensive feature correlation graph integrates the global correlation characteristics between multi-source heterogeneous data from various sensors, providing a comprehensive and accurate data foundation for subsequent fault assessment by the AI ​​Agent.

[0021] The AI ​​Agent server is deployed with a deep reinforcement learning framework and a large-scale knowledge base. The deep reinforcement learning framework supports the AI ​​Agent's dynamic reasoning capabilities, while the large-scale knowledge base stores historical fault data and expert experience. The AI ​​Agent server takes a comprehensive feature association graph as input. Based on the deep reinforcement learning framework, the AI ​​Agent constructs a multi-layer decision network, with each layer corresponding to a specific fault evaluation task. The first layer is responsible for identifying whether the equipment has potential faults. When the comprehensive feature association graph is input to this layer, the AI ​​Agent, based on the various data association characteristics reflected in the graph and combined with the normal operation data patterns of similar equipment stored in the large-scale knowledge base, determines whether the CNC machine tool shows signs of fault. For example, if the frequency component distribution characteristics of the vibration signal have changed significantly compared to previous normal operation, and the temperature signal's trend also shows abnormalities, the AI ​​Agent initially judges that the equipment has a potential fault risk.

[0022] The second-layer decision network is used to determine the type of fault. Once the first-layer decision network identifies a potential fault, the second-layer decision network further analyzes the data details in the comprehensive feature association graph. If a significant increase in the high-frequency components of the vibration signal is detected, and the temperature signal continues to rise, combined with the feature records of different fault types in the knowledge base, AIAgent determines that the fault type is related to mechanical components, such as tool wear or spindle bearing overheating.

[0023] The third-layer decision network assesses the severity of the fault. Assuming the fault type is determined to be tool wear, the third-layer decision network evaluates the severity of the tool wear fault based on data features related to tool wear in the comprehensive feature association graph, such as the specific frequency variation amplitude of the vibration signal and the degree of temperature rise, as well as the judgment criteria for different degrees of tool wear in the knowledge base. If the frequency variation amplitude of the vibration signal is small and the temperature rise amplitude is also small, it is judged as mild tool wear; conversely, if the frequency variation amplitude is large and the temperature rise amplitude is significant, it is judged as severe tool wear.

[0024] The fourth-layer decision network generates specific processing solutions. For cases of minor tool wear, the AI ​​Agent suggests adjusting cutting parameters appropriately, such as reducing cutting speed and depth of cut, to extend tool life and maintain normal equipment operation. For cases of severe tool wear, the AI ​​Agent suggests immediately replacing the tool, inspecting and maintaining spindle bearings and other related components, and adjusting equipment operating parameters to accommodate the installation of the new tool.

[0025] During the reasoning process at each layer of the decision network, the parameters of each layer are continuously adjusted through an online learning mechanism to adapt to changes in the device's operating status. For example, as the device continues to operate, new operational data will continuously enrich the comprehensive feature association graph. The AI ​​Agent will adjust the parameters of the decision network in a timely manner based on this new data, making the fault evaluation results more accurate.

[0026] Based on the inference results of the AI ​​Agent, fault description information is generated, including the location of the fault (such as the spindle or tool of a CNC machine tool), the cause analysis (such as tool wear due to excessive cutting force or insufficient coolant), and the scope of impact (such as reduced machining accuracy or decreased production efficiency). Then, the fault description information is validated against a historical fault database to ensure its accuracy. For example, the generated fault cause analysis is compared with the causes of similar faults in the past; if the consistency is high, the fault description information is considered relatively accurate.

[0027] The AI ​​Agent combines the verified fault description information with the handling plan to generate a complete fault evaluation report. Finally, the fault evaluation report is presented to operators or maintenance personnel through a visual interface set up within the factory. Operators can view the detailed fault evaluation report by touching the screen to understand the equipment fault condition, cause, and handling plan. Maintenance personnel can then quickly take appropriate maintenance measures based on the information in the report to ensure that the equipment can be restored to normal operation as soon as possible.

[0028] For example, during an actual production process, a CNC machine tool suddenly experienced a decrease in machining accuracy. Based on data collected by sensors and after the aforementioned series of processing steps, the AI ​​Agent determined the fault type to be moderate tool wear, located at the cutting edge of the tool, caused by prolonged continuous cutting and insufficient coolant supply. The impact was a further decrease in machining accuracy and reduced production efficiency. The AI ​​Agent's proposed solution was to immediately replace the tool, check the coolant supply system for blockages or leaks, and adjust the equipment's operating parameters to accommodate the new tool. After viewing the fault assessment report through the visual interface, the operator immediately notified maintenance personnel to carry out repairs according to the proposed solution. The maintenance personnel quickly replaced the tool and checked the coolant supply system. After adjustments, the CNC machine tool returned to normal operation, ensuring the smooth operation of the production process.

[0029] Throughout the industrial manufacturing plant, the fault assessment and management method based on AI Agent continuously monitors and handles faults in various types of equipment. This effectively solves the problems of isolated information from single sensors and low efficiency of manual intervention in traditional fault assessment and management methods, significantly improving the automation and intelligence level of equipment maintenance and providing strong support for the efficient production of the factory.

[0030] refer to Figure 2 , Figure 2 This is a flowchart illustrating a fault assessment and management method based on an AI Agent provided in this application. The executing entity of this method can be a computer device, which can be a single computer device or a cluster of multiple computer devices. The computer device can be a terminal device or a server, etc. The fault assessment and management method based on an AI Agent provided in this application specifically includes: S10: Acquire multi-source heterogeneous data generated during the operation of industrial equipment, including vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters.

[0031] Industrial equipment refers to various mechanical and electrical equipment used in industrial production processes to complete various production tasks, manufacturing processes, material handling, and other operations. For example, common CNC machine tools, used for precise machining of metal workpieces, contain numerous mechanical structures and electrical control components and are considered industrial equipment. Large conveyor belts in factories, used to transport raw materials or finished products, are also a type of industrial equipment. Different industrial equipment has different operating mechanisms and work requirements. It can also include production equipment used in automobile engine manufacturing lines.

[0032] Multi-source heterogeneous data refers to a collection of data from different data sources (such as different types of sensors) with different data formats and properties. In the embodiments of this application, multi-source heterogeneous data includes vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters, etc.

[0033] Vibration signals are data collected by vibration sensors that reflect the vibration status of equipment. For example, they include information such as the vibration frequency, amplitude, and phase of a component during equipment operation. This information reflects the motion state and connection tightness of internal components. For instance, when the bearing of a rotating component wears, the frequency of the vibration signal changes, and the amplitude increases.

[0034] Temperature signals are acquired by temperature sensors and reflect the temperature status of various parts of the equipment. They can indicate the heat generation and dissipation within the equipment, as well as the operating temperature of components. For example, after a motor has been running for a long time, its temperature will rise, and the temperature sensor can monitor this temperature change.

[0035] Pressure signals are obtained using pressure sensors and are used to reflect pressure changes in equipment such as hydraulic and pneumatic systems, for example, the pressure value of hydraulic oil when a hydraulic press is working; lubricating oil quality parameters are data obtained by testing the lubricating oil in the equipment's lubrication system, including the viscosity, pH, and impurity content of the lubricating oil. Changes in these parameters affect the lubrication effect of the equipment and the wear of its components.

[0036] In this application, to achieve effective evaluation and management of industrial equipment failures, it is first necessary to comprehensively acquire various key data generated during equipment operation, i.e., multi-source heterogeneous data. These data sources are wide-ranging and diverse, covering vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters that reflect the equipment's operating status from different perspectives. By rationally deploying corresponding sensors at key locations within the industrial equipment, this data can be collected in real time, providing a rich and comprehensive data foundation for subsequent failure analysis. Different types of industrial equipment require targeted selection and deployment of sensors based on their own structure and operating characteristics to ensure accurate and useful data acquisition.

[0037] One implementation method is to use a distributed sensor network to acquire multi-source heterogeneous data. Different types of sensors are installed at various key parts of the industrial equipment. For example, vibration sensors are installed near bearings to accurately capture vibration signals; temperature sensors are installed around motors and heat-generating components to acquire temperature signals; pressure sensors are installed at pipeline connections in hydraulic or pneumatic systems to monitor pressure signals; and dedicated lubricating oil quality sensors are installed along the oil circulation path of the lubrication system to acquire lubricating oil quality parameters. Each sensor has independent data acquisition and preprocessing functions, capable of performing preliminary processing on the acquired raw data, such as filtering and amplification, and then transmitting the processed effective data to a data acquisition center for centralized storage and further processing via wired or wireless networks. This distributed sensor network layout can comprehensively and in real-time acquire multi-source heterogeneous data generated during the operation of industrial equipment, providing sufficient data support for subsequent fault evaluation and management. This method, through comprehensive and accurate data acquisition, avoids fault analysis biases caused by missing or inaccurate data, improving the reliability of fault evaluation and management.

[0038] S20: Based on the vibration signal and the temperature signal, construct a first feature correlation matrix, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the operating state of the equipment.

[0039] The first characteristic correlation matrix is ​​a matrix representation used to quantify the interaction between vibration and temperature signals during equipment operation. Through specific algorithms and data processing methods, it correlates and quantifies the relevant features of vibration and temperature signals, clearly showing their mutual influence and coordinated changes in the form of matrix element values. For example, if a certain frequency component of the vibration signal has a specific correlation with a certain temperature range of the temperature signal—such as when the vibration frequency is within a certain range, the temperature always changes within another specific range—this correlation will be reflected in the first characteristic correlation matrix as a corresponding numerical relationship.

[0040] In this embodiment, to gain a deeper understanding of the operating status and potential faults of industrial equipment, analyzing vibration and temperature signals alone is insufficient; a first feature correlation matrix needs to be constructed to quantify their interaction. By exploring the intrinsic relationship between vibration and temperature signals, a more comprehensive assessment can be made regarding the presence and cause of equipment faults. This is because, in actual equipment operation, vibration and temperature changes often have a certain mutual influence. For example, excessive vibration of components can lead to increased friction, resulting in higher temperatures; conversely, excessively high temperatures can affect the physical properties of components, altering their vibration characteristics. Constructing the first feature correlation matrix aims to accurately capture this interaction, enabling more precise fault analysis in the future.

[0041] In one embodiment, reference Figure 3 Step S20 may specifically include: S201: Perform time-frequency domain decomposition on the vibration signal to extract its frequency component distribution characteristics.

[0042] Time-frequency domain decomposition (TFD) is a technique that transforms a signal from a simple time-domain representation to a time-frequency domain representation that simultaneously incorporates both time and frequency information. It can reveal the frequency components of a signal at different times and their variations in greater detail. For vibration signals, TFD clearly shows the frequency distribution of vibration at various instants during equipment operation, thus uncovering more hidden features within the vibration signal. For example, in the operation of complex industrial equipment, the vibration signal is composed of multiple vibration components of different frequencies superimposed. TFD can separate these different frequency components and clarify their energy distribution at different time points, which is crucial for a deeper understanding of the equipment's vibration characteristics and potential fault correlations.

[0043] Frequency component distribution characteristics refer to the characteristics of a vibration signal after time-frequency domain decomposition, including the distribution of different frequency components throughout the entire signal time period and their respective energy proportions. For example, when a device is operating normally, time-frequency domain decomposition shows that the vibration signal has relatively stable and concentrated vibration components in the low-frequency range, while there are only a small number of weak vibration components in the high-frequency range. However, when the device malfunctions, such as when the imbalance of a rotating component worsens, the vibration components in the high-frequency range will increase, and their energy proportion will also rise accordingly. This change in the high-frequency vibration components is a manifestation of frequency component distribution characteristics.

[0044] In this embodiment, time-frequency domain decomposition of the vibration signal is a crucial preliminary step in constructing the first feature correlation matrix. This operation overcomes the limitations of simple time-domain analysis, allowing for a more comprehensive understanding of the intrinsic characteristics of the vibration signal. Observing a vibration signal solely in the time domain often only reveals the amplitude variation over time, failing to provide a deeper understanding of its frequency composition. Time-frequency domain decomposition breaks down the vibration signal according to different frequencies, clearly revealing the presence of each frequency component at different times and their relative strengths—that is, the frequency component distribution characteristics. This facilitates subsequent correlation analysis with temperature signals and accurate assessment of equipment operating status and potential faults.

[0045] From another perspective, different industrial equipment exhibits varying vibration signal frequency distribution characteristics during normal operation due to differences in their structure, operating principles, and workload. For instance, during normal stamping operations, large stamping equipment displays regular energy peaks in its vibration signal within a specific low-frequency range due to its periodic stamping motion. Conversely, high-precision machine tools require relatively low energy levels in the high-frequency band during normal operation to ensure machining accuracy. Therefore, by performing time-frequency domain decomposition on vibration signals and extracting their frequency component distribution characteristics, a more accurate reference standard for the operating status of specific equipment can be established, thereby enabling a more acute detection of abnormal changes during equipment operation.

[0046] Furthermore, the extracted frequency component distribution features can provide basic data for subsequent joint analysis with other signal features. For example, when constructing the first feature correlation matrix by correlating with the temperature signal, these frequency component distribution features can be matched and correlated with the temperature signal's trend characteristics according to specific rules, thereby more accurately quantifying the interaction between vibration signals and temperature signals in the equipment's operating state.

[0047] In one embodiment, wavelet transform can be used for time-frequency domain decomposition of vibration signals. Wavelet transform features multi-resolution analysis, enabling signal decomposition at different scales to effectively extract the frequency component distribution characteristics of vibration signals. The specific operation process is as follows: First, a suitable wavelet basis function is selected, such as a wavelet function from the commonly used Daubechies wavelet family, chosen based on the characteristics of the vibration signal and analysis requirements. Then, the vibration signal is used as input, and decomposed using a wavelet transform algorithm to obtain a series of wavelet coefficients at different scales. These wavelet coefficients correspond to the energy information of different frequency components at different times. By further processing and analyzing these wavelet coefficients, such as calculating the energy distribution of the wavelet coefficients at each scale, the frequency component distribution characteristics of the vibration signal can be extracted. In this way, the vibration signal can be accurately converted from the time domain to the time-frequency domain, clearly showing its frequency component distribution characteristics, providing strong data support for subsequent analysis. Its technical advantage lies in its ability to finely decompose vibration signals and accurately obtain their frequency component distribution characteristics, providing richer information for in-depth analysis of equipment operating status and fault detection.

[0048] S202: Perform time series modeling on the temperature signal to extract its trend characteristics.

[0049] Time series modeling is a method for mathematically modeling a series of data points (in this case, temperature signal data) based on their chronological order. It aims to describe the changes in data over time by establishing a suitable mathematical model, thereby uncovering hidden trends within the data. For temperature signals, time series modeling can capture how the temperature of an equipment rises, falls, or exhibits other regular patterns over time during operation, facilitating an understanding of the equipment's thermal characteristics and its correlation with other operating parameters.

[0050] Trend characteristics refer to the features extracted from temperature signals after time series modeling, including the overall trend, rate of change, and presence of periodicity in temperature changes over a period of time. For example, after a motor starts up, its temperature may first rise rapidly and then gradually stabilize; or in some periodically operating equipment, the temperature signal may show regular alternations of high and low temperatures according to a fixed cycle. These are all concrete manifestations of trend characteristics.

[0051] In this application, time-series modeling of temperature signals is performed to gain a deeper understanding of the equipment's operating status from another dimension. Temperature, as a crucial indicator of equipment operating conditions, often exhibits trends closely linked to various internal physical processes and potential faults. Extracting these trend characteristics through time-series modeling allows us to move beyond focusing solely on instantaneous temperature values ​​and grasp the patterns of temperature change over a more macroscopic time scale.

[0052] Different types of industrial equipment exhibit varying temperature signal trends due to differences in their operating modes and heat dissipation mechanisms. For example, large, continuously operating furnaces maintain a relatively stable high temperature over extended periods, with only brief fluctuations occurring when adding raw materials or performing certain maintenance operations. In contrast, automated packaging equipment that operates intermittently shows a rapid temperature rise at the start of each work cycle, followed by a rapid decline after completion, exhibiting a clear periodic trend. Therefore, by performing time-series modeling of temperature signals and extracting their trend characteristics, it is possible to establish temperature change reference models tailored to the specific characteristics of different equipment, thereby enabling more effective monitoring of equipment operation and the detection of potential faults.

[0053] Furthermore, the extracted trend features can be analyzed in conjunction with other data features such as the frequency component distribution features of vibration signals. For example, when constructing the first feature correlation matrix, associating the trend features of temperature signals with the frequency component distribution features of vibration signals according to specific rules can more comprehensively quantify the interaction between vibration signals and temperature signals in the operating state of equipment, providing a stronger basis for accurately determining whether equipment malfunctions.

[0054] In one embodiment, an Autoregressive Integral Moving Average (ARIMA) model can be used to model the temperature signal over time. First, the collected temperature signal data needs to be preprocessed, including data cleaning and handling of missing values, to ensure data quality and integrity. Then, based on the characteristics of the temperature signal's time-series data, the parameters of the ARIMA model are determined, such as the autoregressive order (p), the differencing order (d), and the moving average order (q). Through analysis and trial and error with historical temperature data, the most suitable parameter combination to describe the temperature signal's changing trend is found. Next, the preprocessed temperature signal data is substituted into the ARIMA model with the determined parameters for modeling, resulting in a mathematical model that describes the temperature signal's changing trend. Finally, by analyzing this mathematical model, such as calculating the deviation between the model's predicted and actual values ​​and observing changes in the model's slope, the changing trend characteristics of the temperature signal can be extracted. In this way, the changing patterns of the temperature signal over time can be effectively captured, and its changing trend characteristics can be extracted, providing important data support for subsequent correlation analysis and fault diagnosis. Its technical advantage lies in its ability to accurately describe the changing trend of temperature signals, providing more targeted information for in-depth analysis of equipment operating status and fault detection.

[0055] S203: Generate and construct a first feature correlation matrix based on the frequency component distribution characteristics and the change trend characteristics.

[0056] In this embodiment, generating and constructing the first feature correlation matrix based on the frequency component distribution characteristics and variation trend characteristics is a step of integrating and quantifying the key features of the vibration and temperature signals extracted earlier. By correlating these two different dimensions of features, the intrinsic relationship between vibration and temperature signals under equipment operating conditions can be examined from a more comprehensive perspective.

[0057] Specifically, the frequency component distribution characteristics of the vibration signal reveal the distribution of vibration across different frequency bands and the proportion of energy. Meanwhile, the temperature signal's trend reflects the temperature's variation over time. Corresponding these two characteristics to construct a matrix reveals that the occurrence of vibrations at certain frequencies corresponds to specific temperature trends, and vice versa. This correlation is crucial for accurately determining whether equipment malfunctions and their potential causes. For example, if an increase in high-frequency vibration components is observed during equipment operation, coupled with a continuously rising temperature trend, the elements in the first feature correlation matrix corresponding to the high-frequency vibration components and the temperature increase trend will reflect the strength of this correlation, suggesting the presence of vibration- and temperature-related potential faults, such as excessive friction of components leading to increased temperature and consequently, intensified vibration.

[0058] Moreover, constructing the first feature correlation matrix is ​​not simply a matter of listing features; it requires scientifically sound mapping rules and data processing methods. These rules and methods must accurately match and quantify the frequency component distribution characteristics with the changing trend characteristics, ensuring that each element in the matrix truly reflects the interaction between the vibration signal and the temperature signal, thus facilitating subsequent fault analysis.

[0059] In one embodiment, firstly, the frequency component distribution characteristics of the vibration signal and the trend characteristics of the temperature signal are normalized to ensure they fall within the same numerical range or data format, facilitating subsequent correlation operations. For example, the energy values ​​of the frequency components can be normalized, as can the temperature change rate in the temperature trend characteristics. Then, according to a pre-defined mapping rule, the normalized frequency component distribution characteristics and trend characteristics are cross-mapped. For instance, it can be set that when the vibration energy of a certain frequency component is within a specific range, it is correlated with a specific temperature change trend (such as the temperature rise rate being within a certain range), forming a preliminary correlation matrix. Finally, a nonlinear optimization algorithm, such as particle swarm optimization, is used to optimize and adjust the preliminary correlation matrix. By iteratively adjusting the element values ​​in the matrix, the matrix can more accurately quantify the interaction between the vibration signal and the temperature signal during equipment operation. The technical effect is that it can accurately construct a first feature correlation matrix, truly reflecting the interaction between the vibration signal and the temperature signal, providing more accurate data for subsequent fault analysis.

[0060] In one embodiment, reference Figure 4 Step S203 can be implemented by the following process: S2031: Cross-mapping the frequency component distribution characteristics with the change trend characteristics to form a preliminary correlation matrix.

[0061] Cross-mapping is a method of associating and corresponding different types of features according to specific rules. In this case, it involves associating the frequency component distribution features extracted from vibration signals with the trend features extracted from temperature signals based on a pre-defined logical relationship. This cross-mapping establishes a preliminary connection between the two features, enabling the subsequent formation of a matrix reflecting their relationship. For example, for the energy distribution of a specific frequency component of a vibration signal, finding a related trend in the temperature signal during equipment operation, such as a specific range of temperature rise rates, and then mapping them together—this is the basic idea behind cross-mapping.

[0062] The preliminary correlation matrix is ​​a matrix structure formed after cross-mapping the frequency component distribution characteristics with the variation trend characteristics. Each element of this matrix represents a preliminary correlation between a vibration signal of a specific frequency component and a temperature signal with a specific variation trend. However, the correlation at this stage is not precise enough; it only reflects a relatively broad correspondence based on the initial settings of the cross-mapping. For example, a certain element value in the matrix indicates that when vibration of a certain frequency component occurs, there is a certain probability that it will be accompanied by a temperature signal with a specific variation trend, but the specific correlation strength and accuracy still need further optimization.

[0063] In this embodiment, cross-mapping the frequency component distribution characteristics with the trend characteristics to form a preliminary correlation matrix is ​​an important intermediate step in constructing the first feature correlation matrix that accurately quantifies the interaction between vibration and temperature signals. Through this cross-mapping operation, the key features previously extracted independently from the vibration and temperature signals are integrated, allowing them to initially display the correlation between the two in a matrix form.

[0064] From the perspective of actual equipment operation, different frequency component distribution characteristics often have potential connections with different temperature change trends. For example, in some industrial equipment with periodically moving parts, when a certain low-frequency component in the vibration signal exhibits a relatively stable and energy-concentrated distribution, the corresponding temperature signal will show a slow upward trend with small fluctuations. Through cross-mapping, this correlation existing in actual operation can be reflected in a preliminary correlation matrix, providing a preliminary feature-based perspective for further in-depth analysis of equipment operating status and potential faults.

[0065] Moreover, the process of forming the preliminary correlation matrix is ​​not arbitrary. It requires a deep understanding of the equipment's operating principles, past fault data, and the inherent properties of the two signal characteristics to establish reasonable cross-mapping rules. Only in this way can the preliminary correlation matrix accurately reflect the initial correlation between vibration and temperature signals in the equipment's operating state, laying the foundation for subsequent optimization and adjustment steps.

[0066] In one embodiment, firstly, the frequency component distribution characteristics are classified and numbered in detail. For example, the frequency components of the vibration signal are divided into several frequency bands according to the frequency range, with each frequency band as a category and each category assigned a unique number. Similarly, the temperature signal's trend characteristics are processed in a similar way, classified and numbered according to factors such as the temperature change pattern (e.g., continuous rise, rise followed by fall, fluctuation, etc.) and the rate of change. Then, according to a pre-set cross-mapping rule, the corresponding numbers are associated to form a preliminary association matrix. For example, the rule is set such that when a certain frequency band number of the vibration signal and a certain trend number of the temperature signal appear simultaneously with a high frequency in the equipment's historical operating data, the positions corresponding to these two numbers are set to an initial association value in the preliminary association matrix (e.g., 1 indicates a certain association; 0 indicates no obvious association). In this way, the frequency component distribution characteristics and trend characteristics can be cross-mapped relatively systematically to form a preliminary association matrix, providing basic data for subsequent optimization. The technical effect is that it can initially establish the association relationship between the two features according to certain logic and rules, presenting it intuitively in matrix form, which facilitates further analysis and optimization.

[0067] S2032: The preliminary correlation matrix is ​​adjusted using a nonlinear optimization algorithm to obtain the first characteristic correlation matrix.

[0068] Nonlinear optimization algorithms are a class of algorithms used to solve optimization problems with nonlinear objective functions and constraints. In this scenario, such algorithms are used to adjust and optimize the element values ​​in the initial correlation matrix, enabling it to more accurately quantify the interaction between vibration and temperature signals during equipment operation. Common nonlinear optimization algorithms include genetic algorithms and particle swarm optimization (PSO) algorithms. They adjust the matrix element values ​​by iteratively searching for the optimal solution to achieve better optimization results. For example, PSO simulates the foraging behavior of a flock of birds, finding the optimal combination of element values ​​by having a group of particles continuously move and update their positions (i.e., matrix element values) in the solution space, allowing the matrix to more accurately reflect the relationships between features.

[0069] In this application, adjusting the preliminary correlation matrix using a nonlinear optimization algorithm is a crucial step. The aim is to obtain a first feature correlation matrix that can more accurately quantify the interaction between vibration and temperature signals. Although the preliminary correlation matrix has established a basic relationship between the two features, this relationship is not precise enough to fully and accurately reflect the actual interaction between vibration and temperature signals during equipment operation.

[0070] From a practical application perspective, equipment operating conditions are complex and constantly changing. Different operating conditions and the wear and tear of equipment components can all affect vibration and temperature signals, as well as the relationship between them. For example, as equipment operating time increases, wear on a component can alter the frequency distribution characteristics of the vibration signal, affecting the temperature signal's trend and thus changing their correlation. By adjusting the initial correlation matrix using a nonlinear optimization algorithm, the matrix element values ​​can be continuously optimized based on new equipment operating data and actual conditions. This allows the final first characteristic correlation matrix to adapt to these changes and more accurately reflect the interaction between vibration and temperature signals under the current equipment operating conditions.

[0071] Furthermore, different nonlinear optimization algorithms possess different characteristics and advantages, requiring comprehensive consideration based on the specific application scenario and data characteristics when selecting one. For instance, genetic algorithms have strong global search capabilities, making them suitable for finding optimal solutions in large solution spaces; while particle swarm optimization algorithms have relatively fast convergence speeds, making them more suitable for scenarios with high time requirements. Appropriately selecting and applying nonlinear optimization algorithms to adjust the initial correlation matrix can ensure the acquisition of a high-quality first-feature correlation matrix, providing more reliable data support for subsequent fault analysis and other tasks.

[0072] In one embodiment, taking the particle swarm optimization algorithm as an example, its implementation process is as follows: First, the element values ​​in the preliminary correlation matrix are used as the particle positions in the particle swarm optimization algorithm. Each particle represents a combination of matrix element values, and its initial position is the current element value of the preliminary correlation matrix. Then, based on the equipment operating data and a pre-set objective function (such as minimizing the prediction error, i.e., the error between the correlation between vibration and temperature signals predicted by the matrix and the actual observed correlation), the fitness value of each particle is calculated. The higher the fitness value, the more accurately the combination of matrix element values ​​represented by the particle reflects the interaction between vibration and temperature signals. Next, the particle positions (i.e., matrix element values) are updated iteratively according to the particle fitness value and the update rules of the particle swarm optimization algorithm (such as velocity update formula and position update formula). During each iteration, the particle continuously adjusts its position, that is, continuously adjusts the element values ​​in the preliminary correlation matrix to find a better combination of matrix element values. Finally, when the preset number of iterations is reached or the stopping condition is met (such as the fitness value reaching a certain threshold), the combination of matrix element values ​​corresponding to the particle position at this time is the element value of the optimized first feature correlation matrix. By using particle swarm optimization to adjust the initial correlation matrix in this way, a more accurate first feature correlation matrix can be obtained, quantifying the interaction between vibration and temperature signals during equipment operation. This provides more accurate data for subsequent fault analysis. The technical advantage lies in the ability to continuously optimize and adjust matrix element values, making the final first feature correlation matrix more accurately reflect the interaction between vibration and temperature signals during equipment operation, thus improving the accuracy of fault analysis.

[0073] S30: Based on the pressure signal and the lubricating oil quality parameters, construct a second feature correlation matrix, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters.

[0074] Among these, lubricating oil quality parameters are data obtained from testing the lubricating oil in the equipment's lubrication system, including the oil's viscosity, pH, and impurity content. Changes in these parameters affect the equipment's lubrication performance, which in turn affects component wear and the overall operating performance of the equipment. For example, a decrease in lubricating oil viscosity leads to insufficient lubrication, increasing the risk of component wear.

[0075] The second characteristic correlation matrix is ​​a matrix representation used to quantify the degree of coupling between pressure signals and lubricating oil quality parameters. Through specific algorithms and data processing methods, it correlates and quantifies the relevant features of pressure signals and lubricating oil quality parameters, clearly showing the mutual influence and synergistic changes between the two in the form of element values ​​in the matrix. For example, if there is a correlation between a certain pressure range of the pressure signal and a specific viscosity value of the lubricating oil quality parameter—for instance, if the viscosity of the lubricating oil always changes within another specific range when the pressure is within a certain range—this correlation will be reflected in the corresponding numerical relationship in the second characteristic correlation matrix.

[0076] In this application, for industrial equipment, pressure signals and lubricating oil quality parameters are also important factors reflecting its operating status, and there is a certain coupling relationship between them. Constructing a second characteristic correlation matrix aims to accurately quantify this coupling degree, in order to gain a deeper understanding of the equipment's operating status and potential faults from another perspective. In actual equipment operation, pressure changes affect the flow characteristics of lubricating oil, thus affecting its quality parameters; conversely, changes in lubricating oil quality parameters also affect the pressure distribution within the equipment, such as changes in lubricating oil viscosity leading to alterations in the pressure transmission efficiency of the hydraulic system. By constructing a second characteristic correlation matrix, this mutually influential relationship can be grasped more comprehensively, providing richer information for subsequent fault analysis.

[0077] In one embodiment, reference Figure 5 Step S30 may specifically include: S301: Perform segmented fitting on the pressure signal and extract its local extreme point distribution characteristics.

[0078] Piecewise fitting is a technique for handling continuously changing pressure signal data. It involves dividing the data into several continuous segments according to certain rules or algorithms, and then fitting each segment with a suitable mathematical function (such as a polynomial function). This method allows for a better capture of the changing patterns and characteristics of the pressure signal at different stages. For example, a pressure signal that fluctuates continuously with equipment operation can be divided into several parts based on factors such as time intervals or the magnitude of pressure changes. Each part can then be fitted separately to analyze its variations in greater detail.

[0079] Local extreme point distribution characteristics refer to the distribution of local maximum and minimum points within each segment of the fitted curve after segmenting the pressure signal, along with their corresponding pressure values. These extreme points reflect the turning points of the pressure signal's trend within a local region. For example, if the pressure first rises to a peak and then falls within a certain time period, this peak is a local extreme point. The distribution of numerous such extreme points throughout the entire pressure signal's time period constitutes the local extreme point distribution characteristics. This is crucial for understanding the fluctuation characteristics of the pressure signal during equipment operation and its correlation with other parameters.

[0080] In this embodiment, segmented fitting of the pressure signal to extract its local extreme point distribution characteristics is a crucial foundational step in constructing the second feature correlation matrix. As a key data point reflecting the operating status of equipment, the pressure signal often exhibits complex and fluctuating changes. Segmented fitting allows this complex process to be broken down into several relatively simple segments for analysis, thus providing a clearer understanding of the specific changing patterns of the pressure signal at different stages.

[0081] From the perspective of actual equipment operation, the pressure signal variation patterns of different industrial equipment vary during operation due to factors such as equipment structure, working principle, and operating conditions. For example, in hydraulically driven equipment, the pressure signal will exhibit different variation curves when performing different actions (such as clamping, lifting, etc.). During a certain action phase, the pressure rapidly rises to a high value, remains stable for a period of time, and then slowly decreases. By segmented fitting, the pressure change characteristics of these different phases can be accurately captured, and the distribution characteristics of local extreme points can be extracted, providing strong data support for subsequent correlation analysis with lubricating oil quality parameters.

[0082] Moreover, the extracted local extreme point distribution characteristics can help to better understand the potential relationship between pressure signals and other operating parameters. For example, when it is found that the time and location of a certain local extreme point in the pressure signal corresponds to certain changes in lubricating oil quality parameters, this suggests that there is a specific operating state or potential fault inside the equipment at that moment, providing important clues for further in-depth analysis of the equipment's operating status and potential faults.

[0083] In one embodiment, a polynomial fitting algorithm can be used to piecewise fit the pressure signal and extract the distribution features of local extrema. First, based on the time-series data of the pressure signal, the signal is divided into several continuous segments according to certain time intervals (e.g., at regular sampling points) or according to the magnitude of pressure value changes (e.g., when the pressure change exceeds a certain threshold). Then, for each segment, a polynomial function of appropriate degree (e.g., quadratic, cubic, etc.) is selected for fitting. The coefficients of the polynomial function are determined using fitting methods such as least squares, ensuring that the fitted curve closely approximates the pressure signal data within that segment. After fitting, the derivative of the fitted curve is calculated, and the points where the derivative is zero are identified as local extrema. The locations of these local extrema (corresponding time points or sampling points) and their pressure values ​​are recorded, thus obtaining the distribution features of local extrema. This method allows for relatively accurate piecewise fitting of the pressure signal and extraction of its local extrema distribution features, providing an accurate data foundation for the subsequent construction of the second feature correlation matrix. Its technical advantage lies in its ability to precisely analyze the changing patterns of pressure signals and accurately extract the distribution characteristics of local extreme points, providing richer information for a deeper understanding of equipment operating status and fault analysis.

[0084] S302: Perform component analysis on the quality parameters of the lubricating oil and extract the concentration distribution characteristics of its key components.

[0085] Composition analysis refers to the process of detecting and quantitatively analyzing various chemical components in lubricating oil using chemical analysis methods or related instruments and equipment. Lubricating oil is composed of a variety of different chemical components, and the types and contents of these components directly affect the performance of the lubricating oil and its lubrication effect on equipment. Through composition analysis, the specific content of each key component in the lubricating oil and their distribution can be determined, thereby providing a comprehensive understanding of the quality of the lubricating oil.

[0086] The key component concentration distribution characteristics refer to the characteristic information obtained after completing the component analysis of lubricating oil quality parameters, regarding the distribution of the concentrations of key components (such as viscosity modifiers, antioxidants, and detergents) that have a significant impact on equipment lubrication and operation within the overall lubricating oil, as well as the relative proportions between them. For example, in a lubricating oil sample, the concentration of antioxidants will vary in different areas, with higher concentrations in some places and lower concentrations in others. This distribution of antioxidant concentration and its relative proportions with other key component concentrations are manifestations of the key component concentration distribution characteristics. This is of great significance for evaluating the lubricating effect of lubricating oil on equipment, judging the operating status of equipment, and understanding its correlation with other parameters such as pressure signals.

[0087] In this application, performing component analysis on lubricating oil quality parameters to extract the concentration distribution characteristics of its key components is another important step in constructing the second feature correlation matrix. Lubricating oil plays a crucial role in equipment operation, and its quality directly affects the lubrication effect and service life of the equipment. Through component analysis, we can gain a deeper understanding of the chemical composition and concentration distribution of the lubricating oil, thereby providing a more accurate basis for subsequent correlation analysis with pressure signals.

[0088] Different types of industrial equipment have varying performance requirements for lubricating oils due to differences in their working environment and operational requirements. For example, equipment operating in high-temperature environments needs lubricating oils with high oxidation resistance, making the concentration distribution characteristics of antioxidants in the lubricating oil particularly important. Conversely, for some high-precision machining equipment, the concentration distribution characteristics of viscosity modifiers in the lubricating oil significantly impact the machining accuracy. Therefore, by analyzing the composition of lubricating oil quality parameters and extracting the concentration distribution characteristics of its key components, it is possible to establish lubricating oil quality reference standards tailored to the specific characteristics of different equipment, thereby more effectively monitoring the operating status of the equipment and identifying potential faults.

[0089] Furthermore, the extracted key component concentration distribution characteristics can be analyzed in conjunction with other data features such as the local extremum distribution characteristics of the pressure signal. For example, when constructing the second feature correlation matrix, associating the key component concentration distribution characteristics of the lubricating oil quality parameters with the local extremum distribution characteristics of the pressure signal according to specific rules can more comprehensively quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters, providing a stronger basis for accurately determining whether the equipment is faulty.

[0090] In one embodiment, spectral analysis technology can be used to analyze the composition of lubricating oil quality parameters and extract the concentration distribution characteristics of key components. First, a lubricating oil sample is collected and placed in a spectrometer. The spectrometer emits light of different wavelengths onto the lubricating oil sample; different chemical components in the lubricating oil will produce different optical effects such as absorption and reflection of light at different wavelengths. By detecting these optical effects and based on the known correspondence between chemical components and optical effects, the types and contents of various chemical components in the lubricating oil can be determined. Then, key components that have a significant impact on equipment lubrication and operation, such as viscosity modifiers, antioxidants, and detergents, are screened from these chemical components. The concentration values ​​of these key components and their distribution in the lubricating oil sample are recorded, thereby obtaining the concentration distribution characteristics of key components. In this way, the composition of lubricating oil quality parameters can be analyzed relatively accurately, and the concentration distribution characteristics of its key components can be extracted, providing important data support for subsequent correlation analysis and fault diagnosis. Its technical advantage lies in its ability to accurately analyze the chemical composition of lubricating oil and accurately extract the concentration distribution characteristics of key components, providing more targeted information for in-depth analysis of equipment operating status and fault detection.

[0091] S303: Construct a second feature correlation matrix based on the distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components.

[0092] The second characteristic correlation matrix can be a matrix representation constructed based on the distribution characteristics of local extreme points of the pressure signal and the concentration distribution characteristics of key components of lubricating oil quality parameters through specific mapping rules and data processing methods. This matrix aims to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters. Its element values ​​reflect the degree of correlation or mutual influence between the pressure signal at different local extreme points and the lubricating oil quality parameters of key components with different concentration distributions. For example, if the occurrence of a pressure signal at a certain local extreme point is often accompanied by the lubricating oil quality parameters of a key component with a specific concentration distribution (such as the concentration of a certain antioxidant being within a certain range), then the element values ​​corresponding to these two in the second characteristic correlation matrix will reflect this strong correlation.

[0093] In this embodiment, constructing a second feature correlation matrix based on the distribution characteristics of local extreme points and the concentration distribution characteristics of key components is a crucial step in integrating and quantifying the key features of the pressure signal and lubricating oil quality parameters extracted previously. By correlating these two different dimensions of features, the intrinsic relationship between the pressure signal and lubricating oil quality parameters under equipment operating conditions can be examined from a more comprehensive perspective.

[0094] Specifically, the distribution characteristics of local extreme points in the pressure signal reveal the turning points of pressure trends in different local areas and the corresponding pressure values. Meanwhile, the distribution characteristics of key component concentrations in lubricating oil quality parameters reflect the overall distribution of key components in the lubricating oil and their relative proportions. Corresponding these two characteristics to construct a matrix reveals that the appearance of certain local extreme points in the pressure signal corresponds to specific key component concentration distributions in lubricating oil quality parameters, and vice versa. This correlation is crucial for accurately determining whether equipment malfunctions and their potential causes. For example, if during equipment operation, a high pressure value is found at a local extreme point in the pressure signal, while the concentration of a key component in the lubricating oil quality parameters (such as antioxidants) is low, then in the second feature correlation matrix, the element values ​​corresponding to this local extreme point and the concentration of this key component will reflect the strength of this correlation, indicating potential pressure- and lubricating oil-related malfunctions, such as insufficient anti-oxidation properties of the lubricating oil leading to accelerated wear of equipment components.

[0095] Moreover, constructing the second feature correlation matrix is ​​not simply a matter of listing features; it requires scientifically sound mapping rules and data processing methods. These rules and methods must accurately match and quantify the distribution characteristics of local extreme points with the distribution characteristics of key component concentrations, ensuring that each element in the matrix truly reflects the degree of coupling between the pressure signal and the lubricating oil quality parameters. This provides a reliable data foundation for subsequent fault analysis and assessment.

[0096] In one embodiment, reference Figure 6 Step S303 can be implemented in the following way: S3031: Jointly model the distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components to form an initial correlation matrix.

[0097] Joint modeling refers to the organic combination of two types of feature information obtained from different perspectives: the distribution characteristics of local extrema points in pressure signals and the concentration distribution characteristics of key components in lubricating oil quality parameters. This is achieved through specific mathematical models and algorithms, constructing a model structure that can initially reflect the correlation between these two features. This process is not a simple feature superposition, but rather an integration based on their inherent relationships under equipment operating conditions, achieved through reasonable rules and calculation methods.

[0098] The initial correlation matrix is ​​a matrix form obtained after jointly modeling the distribution characteristics of local extrema and the concentration characteristics of key components. This matrix initially reflects the correlation between local extrema of the pressure signal and the concentration of key components in the lubricating oil quality parameters. However, this correlation is not precise enough; it only represents a relatively broad correspondence based on the initial settings of the joint modeling. For example, a certain element value in the matrix indicates that when a pressure signal occurs at a certain local extrema, there is a certain probability that it is accompanied by the presence of a specific concentration of a key component in the lubricating oil. However, the specific degree of correlation needs further optimization.

[0099] In this embodiment, jointly modeling the distribution features of local extreme points and the distribution features of key component concentrations to form an initial correlation matrix is ​​an important intermediate step in constructing the second feature correlation matrix that accurately quantifies the coupling degree between pressure signals and lubricating oil quality parameters. Through joint modeling, the key features previously extracted independently from pressure signals and lubricating oil quality parameters are integrated, allowing them to initially demonstrate the correlation between the two in a matrix form.

[0100] From the perspective of actual equipment operation, different local extreme point distribution characteristics often have a potential correlation with different key component concentration distribution characteristics. For example, in some hydraulic equipment, when the pressure signal reaches a high extreme point in a certain local area, the concentration of certain key components (such as anti-wear additives) in the corresponding lubricating oil quality parameters will show a specific trend, with a relative decrease in concentration. Through joint modeling, this correlation existing in actual operation can be reflected in the initial correlation matrix, providing a preliminary feature-based perspective for subsequent in-depth analysis of equipment operating status and potential faults.

[0101] Moreover, the process of forming the initial correlation matrix is ​​not arbitrary. It requires setting reasonable joint modeling rules based on a deep understanding of the equipment's operating principles, past fault data, and the properties of the two characteristics themselves. Only in this way can we ensure that the initial correlation matrix can accurately reflect the preliminary correlation between pressure signals and lubricating oil quality parameters in the equipment's operating state, laying the foundation for subsequent optimization and adjustment steps.

[0102] In one embodiment, a joint modeling method based on a multiple linear regression model can be used. First, the pressure values ​​of each local extreme point in the local extreme point distribution characteristics are used as independent variables, and the concentration values ​​of each key component in the key component concentration distribution characteristics are used as dependent variables. Then, based on a large amount of collected pressure signals and lubricating oil quality parameter data during equipment operation, a multiple linear regression algorithm is used to fit a linear model that describes the relationship between the two. In this model, the coefficient between each independent variable (local extreme point pressure value) and the dependent variable (key component concentration value) constitutes an element of the initial correlation matrix. For example, if the coefficient between a local extreme point pressure value and a key component concentration value is large, it indicates a strong statistical correlation between them, and a larger value is assigned to the corresponding position in the initial correlation matrix; conversely, a smaller value is assigned. In this way, the local extreme point distribution characteristics and key component concentration distribution characteristics can be jointly modeled in a relatively systematic way to form an initial correlation matrix, providing basic data for subsequent optimization. Its technical advantage lies in the ability to initially establish the correlation between the two characteristics according to certain logic and rules, presenting it intuitively in matrix form, facilitating further analysis and optimization.

[0103] S3032: The initial correlation matrix is ​​optimized through an adaptive weight allocation mechanism to obtain the final second feature correlation matrix.

[0104] An adaptive weight allocation mechanism is a mechanism that automatically adjusts the weights of each element in a matrix based on the actual operating conditions of the equipment, data characteristics, and the set optimization objectives. By continuously analyzing new data and feedback information, it dynamically assigns appropriate weights to the elements in the initial correlation matrix, enabling the matrix to more accurately quantify the coupling degree between pressure signals and lubricating oil quality parameters. For example, during equipment operation, if the actual correlation between certain local extreme points and the concentration of key components deviates from what the initial correlation matrix reflects, the adaptive weight allocation mechanism will adjust the weights of the corresponding elements based on this new information to improve the accuracy of the matrix.

[0105] The final second-feature correlation matrix is ​​obtained by optimizing the initial correlation matrix through an adaptive weighting mechanism. This matrix can more accurately quantify the coupling degree between pressure signals and lubricating oil quality parameters. Its element values ​​accurately reflect the correlation and mutual influence between pressure signals at different local extreme points and lubricating oil quality parameters of key components with different concentration distributions. Compared with the initial correlation matrix, it is more accurate and reliable in reflecting the intrinsic relationship between pressure signals and lubricating oil quality parameters under equipment operating conditions.

[0106] In this application, optimizing the initial correlation matrix through an adaptive weight allocation mechanism is a crucial step. The aim is to obtain a final second-feature correlation matrix that can more accurately quantify the coupling degree between the pressure signal and the lubricating oil quality parameters. Although the initial correlation matrix has preliminarily established the correlation between the two features, this relationship is not precise enough to fully and accurately reflect the true coupling between the pressure signal and the lubricating oil quality parameters during actual equipment operation.

[0107] From a practical application perspective, equipment operating conditions are complex and constantly changing. Different operating conditions and the wear and tear of equipment components can all affect pressure signals, lubricating oil quality parameters, and the relationships between them. For example, as equipment operating time increases, wear on a certain component can cause changes in the distribution characteristics of local extreme points in the pressure signal, and also affect the concentration distribution characteristics of key components in the lubricating oil quality parameters, thus altering their correlation. By optimizing the initial correlation matrix through an adaptive weight allocation mechanism, the matrix element values ​​can be continuously optimized based on new equipment operating data and actual conditions. This allows the final second feature correlation matrix to adapt to these changes and more accurately reflect the degree of coupling between the pressure signal and lubricating oil quality parameters under the current equipment operating conditions.

[0108] Moreover, the advantage of the adaptive weight allocation mechanism lies in its ability to flexibly adjust according to different equipment and data characteristics. Different types of industrial equipment exhibit varying patterns of pressure signal and lubricating oil quality parameter changes during operation. The adaptive weight allocation mechanism can address these differences by rationally allocating weights based on specific equipment operating data and optimization objectives, ensuring a high-quality final second-feature correlation matrix. This provides more reliable data support for subsequent fault analysis and other tasks.

[0109] In one embodiment: First, an optimization objective function is set, such as minimizing the prediction error, which is the error between the coupling relationship between the pressure signal and lubricating oil quality parameters predicted by the matrix and the actual observed coupling relationship. Then, based on the equipment operating data and the initial correlation matrix, an initial weight value for each element is calculated. These initial weight values ​​can be set based on some simple statistical methods or experience. Next, using an adaptive weight allocation algorithm (such as an adaptive weight allocation algorithm based on gradient descent), the weight value of each element is continuously updated according to the optimization objective function and new equipment operating data. During the update process, the algorithm dynamically adjusts the weight values ​​according to the prediction error under the current weight value and the changes in the data, so that the prediction error continues to decrease. Finally, when a preset stopping condition is reached (such as the prediction error being less than a certain threshold), the matrix at this point is the final optimized second characteristic correlation matrix. In this way, by using an adaptive weight allocation mechanism to optimize the initial correlation matrix, a final second characteristic correlation matrix that more accurately quantifies the coupling degree between the pressure signal and lubricating oil quality parameters can be obtained, providing more accurate data for subsequent fault analysis, etc. Its technical advantage lies in its ability to continuously optimize and adjust the matrix element values, so that the final second feature correlation matrix can more accurately reflect the coupling degree between the pressure signal and the lubricating oil quality parameters under the equipment operating conditions, thereby improving the accuracy of fault analysis.

[0110] S40: Generate a comprehensive feature association map based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association map is used to integrate the global association characteristics between multi-source heterogeneous data.

[0111] The comprehensive feature correlation graph is a graph-based representation that integrates information from related matrices such as the first and second feature correlation matrices. It is then optimized using specific graph structures and algorithms to consolidate the global correlation characteristics between multi-source heterogeneous data. In this graph, nodes can represent different data features (such as a frequency component of a vibration signal, a temperature range of a temperature signal, a pressure value of a pressure signal, or a key component of a lubricating oil quality parameter), while edges represent the relationships between nodes (such as mutual influence and synergistic changes between different data features). This graph format more intuitively displays the complex relationship network between multi-source heterogeneous data, providing a comprehensive perspective based on global correlation characteristics for subsequent fault analysis.

[0112] In this embodiment, although the first and second feature correlation matrices quantify the relationships between some data from different perspectives, in order to gain a more comprehensive and in-depth understanding of the operating status and potential faults of industrial equipment, they need to be merged to generate a comprehensive feature correlation map. By integrating the global correlation characteristics between multi-source heterogeneous data, the operating status of the equipment can be analyzed from a more macroscopic perspective, avoiding focusing only on local data relationships while ignoring the overall situation. Because in actual equipment operation, factors such as vibration, temperature, pressure, and lubricating oil quality are interconnected and mutually influential, generating a comprehensive feature correlation map aims to present these complex relationships in an intuitive and comprehensive way to better manage fault evaluation.

[0113] In one embodiment, reference Figure 7 Step S40 may specifically include: S401: The first feature correlation matrix and the second feature correlation matrix are fused to form an initial correlation map.

[0114] In this embodiment, fusion refers to combining the first and second feature association matrices, constructed from different data relationship perspectives, through specific mathematical operations or data processing methods. This process integrates the feature association information related to the device's operating status across different dimensions, forming a more comprehensive structure—the initial association map. This process is not a simple matrix addition or splicing, but rather a rational combination of information based on the inherent logic of device operation and the potential connections between various features, in order to more comprehensively reflect the various associations related to the device's operating status.

[0115] The initial correlation graph is a graph-theory-based structural representation obtained after fusing the first and second feature correlation matrices. In this graph, nodes can represent various data feature elements extracted from the two feature correlation matrices, such as specific frequency components of vibration signals, specific trends in temperature signals, local extrema of pressure signals, and key component concentrations of lubricating oil quality parameters. Edges represent the preliminary correlations between these nodes. These relationships are determined based on the integration of information from the two matrices during the fusion process. However, the correlations at this stage are relatively broad, only initially demonstrating the connections between different data features, and further optimization is needed.

[0116] In this embodiment, fusing the first feature correlation matrix and the second feature correlation matrix to form an initial correlation map is a crucial starting step in generating a comprehensive feature correlation map. By fusing these two matrices, the key feature correlation information previously obtained from the perspectives of the relationship between vibration and temperature signals and the relationship between pressure and lubricating oil quality parameters can be consolidated, laying the foundation for examining the equipment's operating status from a more macroscopic perspective.

[0117] From the perspective of actual equipment operation, the operating state of equipment is determined by the interaction of multiple factors. Vibration, temperature, pressure, and lubricating oil quality are not isolated phenomena but are interconnected. For example, in some industrial equipment, when a specific frequency component of the vibration signal exhibits abnormal changes, it is not only related to the trend of temperature signal changes but also indirectly affects the local extreme points of the pressure signal and the concentration of key components in the lubricating oil quality parameters through the internal mechanical structure and working principle of the equipment. By fusing the first feature correlation matrix and the second feature correlation matrix to form an initial correlation map, these potential relationships at different levels can be initially presented in a graphical form, providing a more comprehensive perspective for subsequent in-depth analysis of equipment operating status and discovery of potential faults. Moreover, the fusion process for forming the initial correlation map needs to be rationally designed based on a deep understanding of the equipment operating mechanism, the nature of each data feature, and past fault data. Only in this way can the initial correlation map accurately integrate the information from the two matrices, initially reflecting the correlation between different data features under the equipment operating state, thus preparing for subsequent steps such as topology optimization.

[0118] In one embodiment, a matrix fusion algorithm can be used to fuse the first feature association matrix and the second feature association matrix. First, the two matrices are aligned in dimensions. If their dimensions are inconsistent, zero elements are added to make them the same to facilitate subsequent fusion operations. Then, according to pre-defined fusion rules, such as weighted summation, the first and second feature association matrices are assigned weights (the weights are determined based on the importance of the data features represented by the two matrices in the device operation status analysis). The elements at corresponding positions are then summed according to their weights to obtain the fused matrix element values. These fused element values ​​are organized according to the node and edge representation methods of graph theory, with each element value serving as an attribute value for a node. The connection of edges is determined based on the relationships between element values, thus forming an initial association graph. In this way, the first and second feature association matrices can be fused relatively systematically to form an initial association graph, providing basic data for subsequent optimization. Its technical advantage lies in its ability to effectively integrate the information from two matrices according to certain rules and methods, and to initially display the correlation between different data features in the form of a graph, which facilitates further analysis and optimization.

[0119] S402: Perform topological optimization on the initial association graph to obtain an optimized association graph.

[0120] Topology optimization refers to the process of adjusting and improving the topological characteristics of an initial network graph, such as node connection methods, edge weight allocation, and overall network structure, using specific algorithms and techniques. Its purpose is to enhance the graph's global feature representation capability, enabling it to more accurately reflect the true relationships between different data features under device operating conditions. By optimizing the connections between nodes and the weights of edges, the graph can better capture the deeper, hidden relationships behind the data.

[0121] The optimized association graph is obtained by optimizing the topology of the initial association graph. Compared with the initial association graph, it is more accurate and in-depth in reflecting the relationships between different data features under the device's operating state. Its node connection method and edge weight allocation are more reasonable, and it can better show the mutual influence and coordinated changes among various factors in the device's operating state, providing a more reliable foundation for subsequent real-time updates and the final generation of a comprehensive feature association graph.

[0122] In this application, optimizing the topology of the initial correlation map is a crucial step in generating the comprehensive feature correlation map. Although the initial correlation map has initially demonstrated the correlation between different data features, this relationship is not precise enough and cannot fully and accurately reflect the true mutual influence and synergistic changes among various factors under the operating state of the equipment.

[0123] In actual equipment operation, the operating state is complex and constantly changing. Different operating conditions and the wear and tear of equipment components can all affect the correlation between various data features within the equipment. For example, as the equipment operates for longer periods, wear on a component can alter the correlation between data features such as vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters. Previously strong correlations may weaken, or previously insignificant correlations may become more prominent. By optimizing the topology of the initial correlation map, the map's topology can be adjusted based on new equipment operating data and actual conditions, allowing the map to adapt to these changes and more accurately reflect the true correlation between various data features under the equipment's operating state.

[0124] Furthermore, topology optimization requires the use of appropriate algorithms and techniques. Different algorithms and techniques have different characteristics and advantages, and their selection should be based on a comprehensive consideration of the specific characteristics of the equipment's operating data and the optimization objectives. For example, Graph Neural Networks (GNNs) are a commonly used technique for topology optimization. They can automatically learn the relationships between nodes in a graph and enhance the global feature representation capability of the graph by adjusting the connection weights between nodes, enabling the graph to better reflect the true correlations between different data features under the equipment's operating conditions.

[0125] In one embodiment, a graph neural network (GNN) can be used as an example to implement topology optimization. First, an initial association graph is provided as input to the GNN. Nodes in the GNN correspond to data feature nodes in the initial association graph, and edges correspond to the relationships between nodes. Then, by setting an appropriate GNN architecture, such as selecting a suitable number of layers and node update functions, the GNN begins to learn the relationships between nodes in the graph. During the learning process, the GNN continuously adjusts the connection weights between nodes and the attribute values ​​of the nodes themselves based on the input initial association graph data and the preset training objective (such as minimizing the prediction error, i.e., the error between the data feature relationships predicted by the optimized graph and the actual observed relationships). Through multiple iterations of training, the GNN gradually optimizes the topology of the graph, enabling the graph to better reflect the true relationships between different data features under device operating conditions. Finally, when the preset number of training iterations is reached or the stopping condition is met (such as the prediction error being less than a certain threshold), the graph at this point is the optimized association graph. By optimizing the topology of the initial correlation graph using a graph neural network, an optimized correlation graph can be obtained that more accurately reflects the true correlations between different data features under equipment operating conditions. This provides a more reliable foundation for subsequent real-time updates and the generation of the final comprehensive feature correlation graph. The technical advantage lies in improving the accuracy of the graph's reflection of the correlations between different data features under equipment operating conditions by optimizing the graph's topology, thus providing a more accurate basis for fault analysis and other applications.

[0126] S403: Update the optimized association map in real time to obtain the final comprehensive feature association map.

[0127] Real-time updates refer to the timely adjustment and updating of the optimized correlation map during equipment operation, based on continuously acquired new equipment operation data and real-time changes in the equipment's operating status. The purpose is to ensure that the map remains highly consistent with the current operating status of the equipment, accurately reflecting the latest correlations between different data features under the equipment's operating conditions. This provides the most accurate and real-time information support for subsequent map-based fault analysis and other operations.

[0128] The final comprehensive feature association map is obtained by updating the optimized association map in real time. This map integrates the latest correlations between all relevant data features obtained during equipment operation, possessing high real-time performance and accuracy. It provides a comprehensive, accurate, and real-time updated information foundation for AI Agents to perform equipment operation status analysis based on the map. It is the final result of generating the comprehensive feature association map and is crucial for accurately determining whether equipment has a fault and identifying fault-related information.

[0129] In this embodiment, updating the optimized correlation map in real time to obtain the final comprehensive feature correlation map is the final step in the entire process of generating the comprehensive feature correlation map, and it is also a key step to ensure that the map can adapt to the dynamic changes in the equipment's operating status. During operation, the equipment's operating status will continuously change, and new data features such as vibration, temperature, pressure, and lubricating oil quality will constantly be generated. These changes need to be reflected in the correlation map in a timely manner.

[0130] From a practical application perspective, various factors such as changes in operating conditions, the wear rate of equipment components, and the influence of the external environment can lead to real-time changes in the operating status of equipment. For example, in industrial production, sudden increases or decreases in load and fluctuations in ambient temperature can cause significant changes in data characteristics such as vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters, thereby altering the correlations between them. By updating the optimized correlation graph in real time, the node connection weights and node attribute values ​​of the graph can be adjusted promptly based on these new equipment operating data and actual conditions, ensuring that the graph reflects the latest correlations between different data characteristics under the equipment's operating state. Moreover, the real-time update process requires an effective data monitoring and feedback mechanism. Only by acquiring new equipment operating data in a timely manner and accurately determining which adjustments to the graph are needed can the final comprehensive feature correlation graph maintain its effectiveness and accuracy. For example, a sensor network can be set up to continuously monitor equipment operating data and transmit new data to the graph update module in a timely manner. The graph update module then updates the optimized graph according to preset update rules and algorithms.

[0131] In one embodiment, a dynamic update algorithm combined with a sensor network can be used to achieve real-time updates of the optimized correlation map. First, a comprehensive sensor network, such as vibration sensors, temperature sensors, pressure sensors, and lubricating oil quality detection sensors, is set up at key parts of the equipment. These sensors continuously collect new data during equipment operation. Then, the collected new data is transmitted to a data processing center. At the data processing center, the optimized correlation map is updated according to pre-set update rules and algorithms. For example, by comparing the new data with existing data features in the map, it can be determined which node connection weights need adjustment and which node attribute values ​​need updating. Specifically, if newly collected vibration signal data shows a significant increase in the vibration intensity of a certain frequency component, and this frequency component is correlated with other data features (such as a trend in temperature signals) in the optimized map, then the connection weights between these two nodes and the attribute values ​​of the relevant nodes are adjusted according to certain rules. In this way, the optimized map is continuously updated based on new equipment operating data, ultimately resulting in the final comprehensive feature correlation map. Its technical advantage lies in its ability to update the map in real time based on changes in equipment operating status, ensuring that the map always maintains its effectiveness and accuracy. This provides a comprehensive, accurate, and real-time updated information foundation for AI Agent's equipment operating status analysis based on the map, thereby improving the accuracy of fault analysis.

[0132] S50: Utilize the AI ​​Agent to perform real-time analysis of the equipment's operating status based on the comprehensive feature association map, and output fault evaluation results. An AI Agent is an artificial intelligence entity with autonomous perception, decision-making, and execution capabilities. In this scenario, it can analyze and judge the operating status of industrial equipment based on received comprehensive feature correlation maps and other data, using its own algorithms and models, and take corresponding actions (such as outputting fault evaluation results). It has learning capabilities, continuously improving its analytical and judgment abilities by receiving new data and feedback information. For example, when faced with different types of industrial equipment and complex operating conditions, the AI ​​Agent can make reasonable fault judgments and evaluations based on past experience data (stored in its internal or external knowledge base) and the currently received comprehensive feature correlation maps.

[0133] The fault evaluation result refers to the conclusion reached by the AI ​​Agent after analyzing the equipment's operating status, regarding whether a fault exists, the type of fault, the severity of the fault, and corresponding handling solutions. For example, if the AI ​​Agent determines that a fault exists in the equipment, it will clearly indicate which part of the equipment the fault occurs in, what type of fault it is (such as mechanical fault, electrical fault, etc.), the severity of the fault (such as minor, moderate, severe), and the specific handling solution provided for the fault (such as repair, replacement of parts, adjustment of parameters, etc.).

[0134] In this application, after generating the comprehensive feature correlation graph, the key step is to use an AI Agent to perform real-time analysis of the equipment's operating status. Leveraging its powerful analytical capabilities and the comprehensive data provided by the comprehensive feature correlation graph, the AI ​​Agent can deeply uncover potential problems in the equipment's operating status, accurately determine whether the equipment has a fault, and identify relevant fault information. Because the comprehensive feature correlation graph integrates the global correlation characteristics between multi-source heterogeneous data, the AI ​​Agent can take this macro-level and comprehensive perspective, combining its own algorithms and models, to perform detailed analysis of the equipment's operating status, thereby outputting accurate fault evaluation results and providing strong guidance for subsequent equipment maintenance and fault handling.

[0135] In one embodiment, reference Figure 8 Step S50 may specifically include: S501: A multi-layer decision network is constructed based on a deep reinforcement learning framework, where each layer of the decision network corresponds to a specific fault evaluation task.

[0136] Deep reinforcement learning is a technical framework that combines the powerful feature representation capabilities of deep learning with the decision optimization capabilities of reinforcement learning. It enables the AI ​​agent to continuously interact with its environment and learn optimal behavioral strategies based on feedback, thereby achieving effective handling of complex tasks. In this scenario, using a deep reinforcement learning framework to build a multi-layered decision network allows the AI ​​agent to better extract key features relevant to equipment fault evaluation from the vast amount of information contained in the comprehensive feature association graph, and make accurate decisions based on these features.

[0137] Multilayer decision networks are network models with a multi-layered structure built on a deep reinforcement learning framework. Each layer undertakes a specific function, corresponding to a specific fault evaluation task. For example, the first layer is responsible for the initial screening of the equipment's operating status to determine whether there are potential fault signs; the second layer, based on the initial judgment of potential faults, further determines the approximate type of fault; subsequent layers successively assess the severity of the fault and provide specific handling solutions. This hierarchical structure helps to decompose complex fault evaluation tasks into multiple relatively simple sub-tasks, improving the accuracy and efficiency of decision-making.

[0138] Specific fault assessment tasks refer to the specific task objectives set for different aspects and stages in the equipment fault assessment process. These tasks cover a series of operations, from determining whether a fault exists in the equipment to identifying the fault type, assessing the severity of the fault, and proposing a solution. Different tasks require different analytical methods and judgment criteria. By allocating them to decision networks at different levels, information from the comprehensive feature association graph can be used more effectively for processing.

[0139] In this embodiment, constructing a multi-layer decision network based on a deep reinforcement learning framework is a crucial foundational step in utilizing an AI Agent to perform real-time analysis of device operating status and output fault evaluation results. By constructing such a multi-layer network structure, the complex fault evaluation process can be rationally decomposed, allowing each level of the decision network to focus on completing a specific fault evaluation task, thereby improving the overall analysis efficiency and accuracy.

[0140] From the perspective of practical needs in equipment failure analysis, the operating status of equipment is affected by various factors, and the manifestations and severity of failures vary. For example, for a large industrial piece of equipment, the types of failures include wear of mechanical parts, electrical system failures, and abnormalities in the lubrication system. Each type of failure is reflected differently in the equipment operating data. For instance, wear of mechanical parts leads to changes in specific frequency components of vibration signals and temperature increases; electrical system failures manifest as abnormal fluctuations in current and voltage. By constructing a multi-layer decision network, the first layer can quickly determine whether there are potential signs of failure in the equipment based on the overall data characteristics in the comprehensive feature association graph, filtering out data from normal operating conditions. Then, the second-layer decision network can, for cases where a potential failure is initially identified, further determine the approximate type of failure based on more detailed information in the graph, such as changes in data characteristics related to different failure types, laying the foundation for more accurate subsequent analysis.

[0141] Moreover, deep reinforcement learning frameworks provide strong support for the construction and training of multi-layer decision networks. They allow the AI ​​Agent to adjust network parameters based on feedback information (such as comparisons between actual fault conditions and predicted results) as it continuously interacts with device operation data (provided through a comprehensive feature association graph). This enables each layer of the decision network to continuously optimize its ability to handle corresponding fault evaluation tasks, better adapting to the fault evaluation needs of different devices and various operating conditions.

[0142] In one embodiment, a Deep Q-Network (DQN) algorithm can be used to construct a multi-layer decision network based on a deep reinforcement learning framework. First, the number of layers and nodes in each layer of the multi-layer decision network is determined. This can be set according to the complexity of the specific fault assessment task and the characteristics of the equipment operation data. For example, for more complex industrial equipment fault assessment, a four-layer decision network can be set, with each layer having a different number of nodes to adapt to the analysis needs at different stages. Then, the data in the comprehensive feature association graph is preprocessed to meet the input requirements of the DQN algorithm, such as data normalization. Next, following the DQN algorithm flow, the network parameters, including weights, are initialized. During training, the AI ​​Agent selects decision actions in each layer of the decision network based on the currently input comprehensive feature association graph data (e.g., in the first layer, when determining whether a potential fault exists, selecting yes or no as the action), and updates the network parameters based on subsequent feedback (such as the results after verification of the actual fault situation). By continuously repeating this process, the multi-layer decision network gradually learns the optimal decision strategy for different fault assessment tasks. Its technical advantage lies in its ability to construct an effective multi-layer decision network using the deep Q-network algorithm, enabling it to accurately complete different fault evaluation tasks based on the comprehensive feature association map, thereby improving the efficiency and accuracy of fault evaluation.

[0143] S502: The comprehensive feature association map is used as input, and reasoning is performed through each layer of decision network in sequence to output the fault evaluation result. The fault evaluation result includes the fault type, fault severity and fault handling plan.

[0144] Reasoning refers to the process by which a multi-layered decision network, based on a comprehensive feature association graph of the input, uses its internally learned decision-making strategies and algorithmic rules to progressively analyze and judge the operating status of equipment. By performing reasoning operations at each layer of the decision network, key information related to fault evaluation can be extracted from the large amount of data features contained in the comprehensive feature association graph, and conclusions can be drawn about the fault type, severity, and handling solutions based on this information.

[0145] Fault evaluation results refer to the judgments made by the AI ​​Agent after a comprehensive analysis of the equipment's operating status, including whether a fault exists, if so, what type of fault it is, its severity, and the appropriate handling measures. For example, fault types include specific categories such as mechanical faults, electrical faults, and lubrication faults; fault severity can be divided into different levels such as minor, moderate, and severe; and handling measures involve specific operational plans such as repair, component replacement, and equipment parameter adjustment.

[0146] In this embodiment, the core step of using an AI Agent to implement equipment fault evaluation management involves taking the comprehensive feature association graph as input and passing it through each layer of the decision network for inference to output fault evaluation results. Through this process, the AI ​​Agent can fully utilize the comprehensive and detailed equipment operating status information provided by the comprehensive feature association graph, combined with the decision strategies learned by the multi-layer decision network, to conduct in-depth analysis and accurate judgment on whether the equipment has a fault and the specific circumstances of the fault.

[0147] From the perspective of actual equipment operation and fault handling, different types of faults manifest differently in equipment operation data, and their impact on the equipment also varies. For example, in mechanical faults, component wear is characterized by a continuous increase in specific frequency components of the vibration signal and a corresponding upward trend in the comprehensive feature correlation map. When the comprehensive feature correlation map is input into a multi-layer decision network, the first layer can identify potential fault signs based on the overall data characteristics. Then, the second layer, based on more detailed data features related to mechanical faults in the map, such as the range of vibration frequency changes and the degree of temperature increase corresponding to the worn component, further determines the fault type as component wear. Next, the third layer, based on data features related to the degree of wear in the map, such as the specific value of the increase in vibration frequency and the specific degree of temperature increase, assesses the severity of the fault, for example, classifying it as moderate wear. Finally, the fourth layer, based on the previously determined fault type and severity, combined with stored knowledge of handling schemes for different fault situations, generates a specific handling plan, such as replacing the worn component and inspecting and maintaining related components.

[0148] Moreover, during the inference process, the decision networks at each layer are interconnected and work collaboratively. The output of the previous layer serves as the input for the next layer, enabling the entire inference process to progressively and deeply analyze the equipment's operating status and fault conditions. Simultaneously, by continuously updating and optimizing the parameters of each layer of the decision network (such as during training based on a deep reinforcement learning framework), the accuracy and reliability of the inference results can be ensured, allowing it to better adapt to changes in equipment operating status and the fault evaluation needs of different devices.

[0149] In one embodiment, a combination of Long Short-Term Memory (LSTM) networks and Multilayer Perceptrons (MLPs) can be used to implement the reasoning functions of each decision network layer. First, the comprehensive feature association map is sequentially fed into the first-layer decision network as input data. This first-layer decision network uses an LSTM network structure, which effectively processes the temporal data features in the comprehensive feature association map (such as vibration signals and temperature signals changing over time), extracts preliminary information related to potential faults, and outputs an intermediate result. Then, the output of the first-layer decision network is used as part of the input to the second-layer decision network. This second-layer decision network uses an MLP structure, which further analyzes and processes the intermediate result output from the first layer, combining its learned feature knowledge about different fault types to determine the approximate type of fault and output another intermediate result. Next, the output of the second-layer decision network is used as part of the input to the third-layer decision network. This third-layer decision network also uses an MLP structure, which assesses the severity of the fault based on the input intermediate result and the data features in the map related to fault severity, and outputs the corresponding result. Finally, the output of the third-layer decision network is used as part of the input to the fourth-layer decision network. The fourth-layer decision network employs an MLP structure, which generates specific processing solutions based on the previously determined fault type and severity, combined with stored knowledge of processing schemes for different fault conditions. In this way, by combining LSTM and MLP, the inference functions of each layer of the decision network can be effectively realized, outputting accurate fault evaluation results and improving the accuracy and practicality of fault evaluation. Its technical advantage lies in its ability to accurately extract key information related to fault evaluation from the comprehensive feature association graph through the collaborative work and effective inference of each layer of the decision network, deriving fault evaluation results that conform to the actual situation, and providing strong guidance for equipment fault handling.

[0150] In one embodiment, the step "outputting fault evaluation results" may specifically include: The fault evaluation results are verified based on a historical fault database. A fault evaluation report is generated based on the verified fault evaluation results, and the fault evaluation report is presented to the user through a visual interface.

[0151] The historical fault database stores information on various faults that have occurred during the past operation of the equipment. This information covers various aspects, including the time of the fault, the equipment location, the type of fault, the severity of the fault, the corrective measures taken, and the effects of those corrective measures. It is a comprehensive record of the equipment's fault history, providing crucial reference for verifying current fault assessment results. For example, for a specific model of industrial equipment, the historical fault database records multiple fault events caused by motor overheating, including detailed information such as changes in motor temperature, vibration levels, the final repair measures taken, and the equipment's return to normal operation.

[0152] In this embodiment, verification refers to the process of comparing and verifying the fault evaluation results obtained by the AI ​​Agent based on comprehensive feature association graph analysis with similar fault cases stored in the historical fault database. This verification method aims to check whether the current fault evaluation results conform to the patterns and characteristics of similar past faults, thereby improving the accuracy and reliability of the fault evaluation results.

[0153] In this embodiment, verifying the fault evaluation results based on a historical fault database is a crucial step in ensuring the accuracy of the fault evaluation. When the AI ​​Agent infers from the comprehensive feature association graph through a multi-layer decision network and outputs fault evaluation results, these results, although based on current equipment operating data, contain certain uncertainties or biases. For example, due to new operating conditions or special circumstances, the equipment's fault behavior may differ from the past, yet similarities may remain. By comparing and verifying with the historical fault database, these potential problems can be identified.

[0154] From a practical application perspective, different equipment may experience various malfunctions during operation. Even the same type of malfunction can manifest differently in different individual pieces of equipment or under different operating environments. For example, for malfunctions involving wear and tear of mechanical parts, the vibration signal changes significantly on a large piece of equipment due to its heavier load, while on a smaller piece of equipment, the vibration signal changes are relatively smaller. By comparing the current malfunction evaluation results with similar cases in a historical malfunction database, the current results can be adjusted and improved based on past experience to ensure they better reflect actual conditions.

[0155] Furthermore, the richer and more detailed the information in the historical fault database, the more accurate and effective the verification of fault evaluation results will be. It can not only help verify whether the judgment on fault type, severity, etc., is correct, but also provide a reference for the subsequent formation of more accurate fault evaluation reports. For example, in terms of handling solutions, effective measures taken for similar faults in the past can be referenced to make the current handling solutions more reasonable and feasible.

[0156] In one embodiment, data matching and similarity calculation methods can be used to achieve verification based on a historical fault database. First, the fault evaluation results output by the AI ​​Agent are subjected to feature extraction, such as extracting key feature information like fault type, fault location, and fault severity, and then converted into a format that can be compared with data in the historical fault database. Then, a similar feature extraction operation is performed on each record in the historical fault database to ensure a unified comparison format. Next, for each key feature of the current fault evaluation result, the most similar record is searched in the historical fault database. This can be determined by calculating the similarity between features, such as using algorithms like Euclidean distance or cosine similarity to measure the degree of similarity between key features of different records. Once records with high similarity are found, the current fault evaluation result is compared with the fault handling measures and final handling effects in these similar records. If significant differences are found, further analysis of the reasons is needed, such as whether new situations in the current equipment have not been considered, thereby adjusting and improving the current fault evaluation result. In this way, the historical fault database can be used to effectively verify the fault evaluation results, improving their accuracy and reliability.

[0157] In this embodiment, the fault evaluation report is a comprehensive and systematic description of the equipment fault. It is generated based on verified fault evaluation results and includes detailed information such as the specific location of the fault, the fault type, the severity of the fault, the scope of its impact, and proposed solutions. This report is a crucial basis for equipment maintenance personnel or operators to understand the equipment fault status and take appropriate measures. For example, for a faulty CNC machine tool, the fault evaluation report would clearly indicate that the fault occurs in the spindle, the fault type is tool wear, the severity is moderate, it will affect machining accuracy and production efficiency, and the solution is to replace the tool and inspect related components.

[0158] A visual interface is a human-computer interaction interface that can display information in an intuitive form, such as graphics, charts, and text. In this scenario, it is used to present fault evaluation reports to users, enabling them to clearly and easily view and understand the report's content. For example, a visual interface can mark the location of the fault on a 3D model of the equipment, presenting it to the user in a more intuitive way; information such as fault type and severity can also be distinguished and displayed using different colors and icons, allowing users to quickly grasp key information.

[0159] In this application, the final step in the entire fault evaluation management process is to generate a fault evaluation report based on the verified fault evaluation results and present it to the user through a visual interface. This is also a crucial step in communicating the analysis results to relevant personnel so that action can be taken. Once the fault evaluation results have been verified and confirmed to be accurate, this information needs to be compiled into a standardized and clear fault evaluation report.

[0160] From a practical standpoint, different users have different needs and understandings of fault information. Equipment maintenance personnel are more concerned with the specific location, severity, and handling plan of the fault in order to quickly carry out repairs; operators are more concerned with the impact of the fault on the production process, such as whether it will cause production line shutdowns or affect product quality. By generating detailed fault evaluation reports and displaying them on a visual interface, the needs of different users can be met, allowing them to have a comprehensive understanding of the equipment fault situation.

[0161] Furthermore, the design of the visualization interface in this application can improve users' efficiency in obtaining information. A good visualization interface should have a simple and clear layout, easy-to-understand icons, and color coding. For example, using red to mark serious faults and green to mark normal status allows users to see the general condition of the equipment at a glance. At the same time, by setting interactive functions on the visualization interface, such as clicking on the location of the fault to view more detailed information, users can further understand the fault situation according to their needs, thereby making better decisions.

[0162] In one embodiment, based on the verified fault evaluation results, relevant information is organized according to a predetermined format using a programming language (such as Python) to form a structured fault evaluation report data file. For example, information such as fault location, type, severity, impact range, and handling plan are stored in different fields. Then, a suitable data visualization library, such as D3.js or Plotly, is selected, and its functions are used to create a visualization interface on a web page. In the design of the visualization interface, a 3D model of the device can be embedded (if available), and annotations can be made on the model based on the fault location information in the fault evaluation report. Information such as fault type and severity can be displayed using icons or text boxes of different colors, such as a red exclamation mark for a severe fault and a yellow triangle for a moderate fault. Interactive functions can be set, such as displaying detailed information when the mouse hovers over an icon, and clicking to view more related content. Finally, the generated fault evaluation report data file is associated with the visualization interface, so that when the user accesses the visualization interface, they can obtain and view the latest fault evaluation report content in real time. In this way, verified fault evaluation reports can be effectively presented to users, improving their efficiency in obtaining information and their understanding of equipment fault conditions.

[0163] The fault assessment and management method and system based on AI Agent proposed in this application have several significant technical advantages. First, by acquiring multi-source heterogeneous data during the operation of industrial equipment, including vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters, the operating status of the equipment can be comprehensively and accurately perceived from multiple dimensions, avoiding the problem of inaccurate judgment of equipment condition due to single or partial data. Next, the first feature correlation matrix, second feature correlation matrix, and comprehensive feature correlation map constructed based on this data effectively integrate the complex relationships between different data, quantify the interaction and coupling degree of various factors in equipment operation, and clearly present the inherent logic and potential correlations of the equipment's operating status, providing a solid and comprehensive data foundation for subsequent fault analysis.

[0164] This application also utilizes an AI Agent, built on a deep reinforcement learning framework, to perform real-time analysis of comprehensive feature association maps, enabling intelligent and automated diagnosis and assessment of equipment faults. It can accurately determine whether equipment is faulty, precisely identify the fault type, reasonably assess the severity of the fault, and provide targeted and feasible fault handling solutions. Based on the dynamic reasoning capabilities of the AI ​​Agent, it not only significantly improves the efficiency of fault evaluation and reduces the uncertainty and errors caused by human intervention, but also quickly responds to changes in equipment operating status, promptly captures potential fault hazards, effectively reduces the risk of equipment downtime due to faults, and ensures the continuity of industrial production.

[0165] Furthermore, during the output of fault evaluation results, the results are verified based on a historical fault database, further ensuring the accuracy and reliability of the fault evaluation. The final fault evaluation report is presented to users through a visual interface, displaying fault-related information in an intuitive and clear manner. This meets the needs of different users (such as equipment maintenance personnel and operators) for understanding the fault situation, facilitating their rapid implementation of appropriate measures. Ultimately, this improves the overall intelligence level, operational efficiency, and production benefits of industrial equipment fault management.

[0166] Accordingly, to better implement the above methods, embodiments of this application also provide a fault evaluation and management system based on an AI Agent. For example... Figure 9 As shown, the fault assessment and management system 80 based on AI Agent includes an acquisition module 801, a first feature processing module 802, a second feature processing module 803, a map generation module 804, and an analysis and processing module 805, as detailed below: The acquisition module 801 is used to acquire multi-source heterogeneous data generated during the operation of industrial equipment, including vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters. The first feature processing module 802 is used to construct a first feature correlation matrix based on the vibration signal and the temperature signal, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the device operating state; The second feature processing module 803 is used to construct a second feature correlation matrix based on the pressure signal and the lubricating oil quality parameters, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters; The map generation module 804 is used to generate a comprehensive feature association map based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association map is used to integrate the global association characteristics between multi-source heterogeneous data; The analysis and processing module 805 is used to perform real-time analysis of the equipment's operating status based on the comprehensive feature association map using the AI ​​Agent, and output fault evaluation results.

[0167] In one embodiment, the first feature processing module 802 is used for: The vibration signal is decomposed in the time and frequency domain to extract its frequency component distribution characteristics; The temperature signal is modeled over time to extract its trend characteristics; A first feature correlation matrix is ​​generated and constructed based on the frequency component distribution characteristics and the change trend characteristics.

[0168] In one embodiment, the first feature processing module 802 is used for: The frequency component distribution characteristics and the changing trend characteristics are cross-mapped to form a preliminary correlation matrix; The initial correlation matrix is ​​adjusted using a nonlinear optimization algorithm to obtain the first characteristic correlation matrix.

[0169] In one embodiment, the second feature processing module 803 is used for: The pressure signal is segmented and fitted to extract its local extreme point distribution characteristics; The quality parameters of the lubricating oil were analyzed to extract the concentration distribution characteristics of its key components. A second feature correlation matrix is ​​constructed based on the distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components.

[0170] In one embodiment, the second feature processing module 803 is used for: The distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components are jointly modeled to form an initial correlation matrix; The initial correlation matrix is ​​optimized using an adaptive weight allocation mechanism to obtain the final second feature correlation matrix.

[0171] In one embodiment, the map generation module 804 is used for: The first feature correlation matrix and the second feature correlation matrix are fused to form an initial correlation map; The initial association graph is topologically optimized to obtain an optimized association graph; The optimized association map is updated in real time to obtain the final comprehensive feature association map.

[0172] In one embodiment, the analysis and processing module 805 is used for: A multi-layer decision network is constructed based on a deep reinforcement learning framework, where each layer of the decision network corresponds to a specific fault evaluation task. The comprehensive feature association map is used as input, and reasoning is performed sequentially through each layer of decision network to output fault evaluation results, which include fault type, fault severity and fault handling plan.

[0173] In one embodiment, the analysis and processing module 805 is used for: The fault evaluation results are verified based on a historical fault database. A fault evaluation report is generated based on the verified fault evaluation results, and the fault evaluation report is presented to the user through a visual interface.

[0174] The implementation details of each module are provided in the preceding method embodiments and will not be repeated here. The technical effects achieved by each module and device are described in the foregoing method embodiments.

[0175] It should be noted that, in practical implementation, the above modules can be arbitrarily combined and integrated into one or more modules, or implemented as independent entities. Furthermore, the above modules can be implemented in hardware or as software functional modules. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. The aforementioned storage medium can be a read-only memory, a hard disk, or an optical disk, etc.

[0176] like Figure 10 As shown, this application embodiment also provides a computer device 90, characterized in that it includes a processor 901 and a memory 902, wherein the memory 902 stores a computer program, and when the computer program is executed by the processor 901, the processor 901 performs the steps of any of the methods described above.

[0177] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0178] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. A fault assessment and management method based on AI Agent, characterized in that, The method includes: Acquire multi-source heterogeneous data generated during the operation of industrial equipment, including vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters; Based on the vibration signal and the temperature signal, a first feature correlation matrix is ​​constructed, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the operating state of the equipment. Based on the pressure signal and the lubricating oil quality parameters, a second feature correlation matrix is ​​constructed, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters; A comprehensive feature association map is generated based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association map is used to integrate the global association characteristics between multi-source heterogeneous data; The AI ​​Agent uses the comprehensive feature association map to perform real-time analysis of the equipment's operating status and outputs fault evaluation results.

2. The method according to claim 1, characterized in that, Based on the vibration signal and the temperature signal, a first feature correlation matrix is ​​constructed, including: The vibration signal is decomposed in the time and frequency domain to extract its frequency component distribution characteristics; The temperature signal is modeled over time to extract its trend characteristics; A first feature correlation matrix is ​​generated and constructed based on the frequency component distribution characteristics and the change trend characteristics.

3. The method according to claim 2, characterized in that, A first feature correlation matrix is ​​generated and constructed based on the frequency component distribution characteristics and the trend characteristics, including: The frequency component distribution characteristics and the changing trend characteristics are cross-mapped to form a preliminary correlation matrix; The initial correlation matrix is ​​adjusted using a nonlinear optimization algorithm to obtain the first characteristic correlation matrix.

4. The method according to claim 2, characterized in that, Based on the pressure signal and the lubricating oil quality parameters, a second feature correlation matrix is ​​constructed, including: The pressure signal is segmented and fitted to extract its local extreme point distribution characteristics; The quality parameters of the lubricating oil were analyzed to extract the concentration distribution characteristics of its key components. A second feature correlation matrix is ​​constructed based on the distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components.

5. The method according to claim 4, characterized in that, Based on the distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components, a second feature correlation matrix is ​​constructed, including: The distribution characteristics of the local extreme points and the concentration distribution characteristics of the key components are jointly modeled to form an initial correlation matrix; The initial correlation matrix is ​​optimized using an adaptive weight allocation mechanism to obtain the final second feature correlation matrix.

6. The method according to claim 1, characterized in that, Based on the first feature correlation matrix and the second feature correlation matrix, a comprehensive feature correlation map is generated, including: The first feature correlation matrix and the second feature correlation matrix are fused to form an initial correlation map; The initial association graph is topologically optimized to obtain an optimized association graph; The optimized association map is updated in real time to obtain the final comprehensive feature association map.

7. The method according to claim 1, characterized in that, Using an AI Agent based on the comprehensive feature association map, the operating status of the equipment is analyzed in real time, and fault evaluation results are output, including: A multi-layer decision network is constructed based on a deep reinforcement learning framework, where each layer of the decision network corresponds to a specific fault evaluation task. The comprehensive feature association map is used as input, and reasoning is performed sequentially through each layer of decision network to output fault evaluation results, which include fault type, fault severity and fault handling plan.

8. The method according to claim 7, characterized in that, The multi-layer decision network includes four layers. The first layer of the decision network adopts a long short-term memory network, and the second, third, and fourth layers of the decision network adopt a multi-layer perceptron structure.

9. The method according to claim 7, characterized in that, Output fault evaluation results, including: The fault evaluation results are verified based on a historical fault database. A fault evaluation report is generated based on the verified fault evaluation results, and the fault evaluation report is presented to the user through a visual interface.

10. A fault assessment and management system based on AI Agent, characterized in that, The system includes: The acquisition module is used to acquire multi-source heterogeneous data generated during the operation of industrial equipment. The multi-source heterogeneous data includes vibration signals, temperature signals, pressure signals, and lubricating oil quality parameters. The first feature processing module is used to construct a first feature correlation matrix based on the vibration signal and the temperature signal, wherein the first feature correlation matrix is ​​used to quantify the interaction relationship between the vibration signal and the temperature signal in the device operating state; The second feature processing module is used to construct a second feature correlation matrix based on the pressure signal and the lubricating oil quality parameters, wherein the second feature correlation matrix is ​​used to quantify the degree of coupling between the pressure signal and the lubricating oil quality parameters; The graph generation module is used to generate a comprehensive feature association graph based on the first feature association matrix and the second feature association matrix, wherein the comprehensive feature association graph is used to integrate the global association characteristics between multi-source heterogeneous data; The analysis and processing module is used to perform real-time analysis of the equipment's operating status based on the comprehensive feature association map using the AI ​​Agent, and output fault evaluation results.

Images