Fault diagnosis positioning technology

Abstract

The invention relates to the technical field of intelligent maintenance, and discloses a fault diagnosis positioning technology, which comprises a data acquisition and alarm module for acquiring equipment operation state data and generating a dynamic alarm signal; the integrated analysis processing module is used for analyzing, processing and identifying the abnormal state of the system; the multi-source fault positioning module is used for analyzing the identified abnormal state, extracting fault features through an optimization algorithm and positioning parts which have faults and possibly have faults in combination with a deep learning model; and the three-dimensional fault visualization module is used for marking the position and the reason of the fault component in the model. Through cooperative work of the data acquisition and alarm module, the integrated analysis processing module, the multi-source fault positioning module and the three-dimensional fault visualization module, automation of the whole process of fault diagnosis is realized, and the diagnosis efficiency is greatly improved. Compared with the existing problems of dependence on manual checking and simple alarm and difficulty in accurate positioning, the method effectively solves the defects of low positioning efficiency and slow response of a traditional method.

Description

Technical Field

[0001] The present invention relates to the field of intelligent maintenance technology, and in particular to a fault diagnosis and positioning technology. Background Art

[0002] At present, fault detection mainly relies on manual troubleshooting or simple automatic alarm mechanisms. However, these traditional methods are usually difficult to accurately locate the source of the fault, resulting in low troubleshooting efficiency, long equipment repair time, and high operating costs.

[0003] In traditional methods, when a device fails, dispersed sensors are used to collect operating data, which is then transmitted to a background monitoring system for preliminary screening and aggregation. However, such methods generally lack integrated data analysis capabilities and intelligent positioning functions, making it difficult to meet the fault diagnosis needs of complex systems.

[0004] Existing fault detection systems are usually unable to achieve efficient and accurate positioning when dealing with complex and changeable equipment failures. Especially for deep-seated software and hardware failures, the limitations of system detection require a lot of manual intervention, which consumes a lot of time and resources to solve the problem. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a fault diagnosis and positioning technology, which solves the problems that the existing fault detection system cannot accurately locate the source of the fault, lacks integrated data analysis capabilities and intelligent positioning functions, and requires a large amount of manual intervention when dealing with complex and changeable software and hardware failures, resulting in low troubleshooting efficiency.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A fault diagnosis and positioning technology, including: Data collection and alarm module, used to collect equipment operation status data and generate dynamic alarm signals; Integrated analysis and processing module, used to build nonlinear analysis model based on equipment operation status, and identify abnormal status of the system through analysis and processing; Multi-source fault location module, which is used to analyze the identified abnormal conditions, extract fault features through optimization algorithms and locate components that have or may fail in combination with deep learning models; The 3D fault visualization module is used to generate a 3D model of the equipment based on the fault location results and mark the location and cause of the faulty components in the model.

[0007] Preferably, in the data collection and alarm module, the equipment operation status data includes temperature, vibration, pressure, and current, and a dynamic threshold is used to determine whether to trigger an alarm, wherein the dynamic threshold is calculated based on the mean and standard deviation of historical operation data.

[0008] Preferably, the integrated analysis processing module describes the equipment operation state by constructing a nonlinear analysis model, the state change of the nonlinear analysis model is described by a state transfer function, and the model includes a fault disturbance term for representing the state deviation caused by the fault.

[0009] Preferably, the integrated analysis processing module analyzes the system state by constructing a positive definite function. When the derivative value of the function is negative, the system is in a normal state. When the derivative value is positive, the system is judged to be abnormal and enters the fault analysis stage.

[0010] Preferably, the multi-source fault location module selects fault features using an optimization algorithm, and the optimization algorithm improves fault location accuracy while minimizing the number of features through an objective function with regularization constraints.

[0011] Preferably, the multi-source fault location module constructs a graph model based on the association relationship between equipment components, and the graph model includes nodes and edges, wherein nodes represent equipment components and edges represent dynamic coupling relationships between equipment components, and the graph model is analyzed through a graph neural network.

[0012] Preferably, the graph neural network models the dynamic coupling relationship of equipment components through a multi-layer message passing mechanism, and the node features of each layer are obtained by aggregation and linear transformation of neighboring node features, and finally the failure possibility of each equipment component is output.

[0013] Preferably, the multi-source fault location module fuses the collected multi-source data, and uses a filtering algorithm to filter out random noise and outliers in the multi-source data. The filtering algorithm is dynamically adjusted based on the deviation between the observed value at the current moment and the predicted state to generate a high-precision state estimate.

[0014] Preferably, the three-dimensional fault visualization module generates a labeling of the faulty component in the three-dimensional model of the equipment according to the fault location result, and the labeling displays the position of the faulty component in a highlighted manner and provides a textual description of the cause of the fault.

[0015] Preferably, the three-dimensional fault visualization module generates a final model of the device by superimposing a labeling matrix with the original three-dimensional model, wherein the labeling matrix includes index information of the faulty component and labeling style parameters for dynamically adjusting the labeling style.

[0016] The present invention provides a fault diagnosis and location technology. It has the following beneficial effects: 1. The present invention uses a data acquisition and alarm module, an integrated analysis and processing module, a multi-source fault location module, and a three-dimensional fault visualization module. Through the collaborative work between modules, the whole process from data acquisition to fault location to visualization is automated, which significantly improves the efficiency of fault diagnosis. Compared with the problem that it is difficult to accurately locate the source of the fault in the prior art that relies on manual troubleshooting and simple alarm methods, the present invention effectively solves the shortcomings of low fault location efficiency and long response time in traditional methods.

[0017] 2. The present invention uses a combination of dynamic threshold judgment mechanism and deep learning algorithm to intelligently analyze and process the collected multi-source data. By dynamically adjusting the threshold to adapt to changes in the operating status of the equipment, the sensitivity and accuracy of fault detection are further improved. Compared with the shortcomings of the prior art that lack integrated data analysis capabilities and rely on fixed threshold judgments to cause frequent false alarms, the present invention can more efficiently adapt to the changing operating environment of complex equipment.

[0018] 3. The present invention dynamically evaluates the equipment status and identifies the abnormal state of the system by constructing a nonlinear analysis model and a positive definite function. The multi-level modeling method adopted can not only capture the complex relationship between the equipment state variables, but also accurately identify the potential failure risks. Compared with the existing technology that relies on a lot of manual intervention to complete the analysis of complex software and hardware failures, the present invention reduces manual participation and realizes accurate analysis and early warning of complex failures.

[0019] 4. The present invention uses a three-dimensional fault visualization module to intuitively present the fault location results through highlighting, dynamic adjustment and text description. By superimposing the annotation matrix to generate an intuitive three-dimensional model, it helps operation and maintenance personnel to quickly locate faulty components and shorten the troubleshooting time. Compared with the existing technology that only relies on data reports or static graphics display, the present invention solves the shortcomings of the existing methods that the information presentation is not intuitive and it is difficult to effectively guide maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the system architecture of the present invention; DETAILED DESCRIPTION

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

[0022] In order to better understand the present invention, the above contents are described in detail below in conjunction with specific embodiments.

[0023] Please refer to the attached Figure 1 , Embodiment 1: The embodiment of the present invention provides a fault diagnosis and positioning technology, including: a data acquisition and alarm module, which is used to collect equipment operation status data and generate dynamic alarm signals; an integrated analysis and processing module, which is used to build a nonlinear analysis model based on the equipment operation status, and identify the abnormal state of the system through analysis and processing; a multi-source fault positioning module, which is used to analyze the identified abnormal state, extract fault features through an optimization algorithm, and locate components that have or may fail in combination with a deep learning model; a three-dimensional fault visualization module, which is used to generate a three-dimensional model of the equipment according to the fault positioning results, and mark the location and cause of the faulty component in the model.

[0024] In this embodiment, the data collection and alarm module collects multi-dimensional status data of the equipment in real time by arranging multiple types of sensors (such as temperature, vibration, pressure, current, etc.) at key parts of the equipment. The module has a built-in dynamic threshold judgment mechanism, which dynamically calculates the alarm threshold by analyzing the mean and standard deviation of historical operation data. When the operating status of the equipment deviates from the normal range, the module will automatically trigger an alarm and record abnormal data, providing basic support for subsequent analysis and positioning; The integrated analysis and processing module builds a nonlinear analysis model based on the equipment's operating data to describe the dynamic coupling relationship between the equipment's components. The dynamic changes in equipment operation are modeled through the state transfer function, and fault disturbance terms are introduced to identify key state changes that may cause abnormalities. At the same time, the system stability is further analyzed by constructing a positive definite function to accurately determine whether the equipment's operating status is normal. When system instability or abnormal deviation is detected, the module passes the abnormal information to the subsequent positioning module; The multi-source fault location module uses an optimization algorithm to perform feature selection on the collected multi-source data, remove redundant information and extract feature information with a high degree of correlation with the fault. Then, the dynamic relationship between equipment components is constructed through a graph model, where nodes represent equipment components and edges represent dynamic coupling relationships. Combining deep learning technology and the multi-layer message passing mechanism of graph neural networks, the module can effectively analyze the interaction characteristics between components and ultimately locate specific components that may or have failed. The 3D fault visualization module highlights the faulty components in the 3D model of the equipment based on the positioning results. By superimposing the annotation matrix, the location, status and possible causes of the faulty components can be dynamically displayed. The highlighting method is intuitive and easy to understand, and the text description helps operation and maintenance personnel quickly understand equipment problems, speed up the troubleshooting and handling process, and improve the overall operation and maintenance efficiency of the system.

[0025] In the data collection and alarm module, the equipment operation status data includes temperature, vibration, pressure, and current, and the dynamic threshold is used to determine whether to trigger an alarm. The dynamic threshold is calculated based on the mean and standard deviation of historical operation data.

[0026] In this embodiment, the operation status data of the equipment is collected and combined with the dynamic threshold judgment mechanism to generate an alarm signal, which provides data support for subsequent integrated analysis and fault location, ensuring that the entire system can identify potential problems in a timely manner and take further treatment measures.

[0027] The data acquisition and alarm module relies on multiple distributed sensor networks, which are deployed in key parts of the equipment and are responsible for collecting a variety of operating status data. These data include but are not limited to key parameters such as temperature, vibration, pressure, and current. These parameters are acquired in real time through sensors and form a complete multidimensional data set to describe the current operating status of the equipment. In one possible implementation, these sensors are placed in key components of the equipment or areas where failures may occur, such as motor bearings, pipeline interfaces, or core components of control units.

[0028] Specifically, the collected data is processed through the built-in dynamic threshold judgment mechanism. The setting of dynamic threshold is based on the historical operation data of the device, combined with the normal operation conditions of the device and the differences in workload, and calculated using statistical methods. The calculation formula of dynamic threshold is as follows:

[0029] in, Indicates Dynamic thresholds for sensors, Indicates sensor The historical average of represents its historical standard deviation, is the sensitivity factor, which is adjusted according to the specific operating requirements and application scenarios of the equipment. For example, for equipment components operating in a high-load environment, The value of may be larger to increase the sensitivity of the system.

[0030] In another implementation, the dynamic threshold can also be adjusted according to the operating conditions of different devices. For example, for equipment components that are in a constant working state for a long time, the dynamic threshold may be set more strictly, while for components that are frequently started and stopped or have large load changes, the calculation of the dynamic threshold will be appropriately relaxed to avoid frequent triggering of false alarms. As an option, the adjustment of the dynamic threshold can also be combined with external environmental parameters, such as the temperature and humidity of the equipment operating environment, to further optimize the alarm mechanism.

[0031] In this embodiment, when the collected data value exceeds the dynamic threshold, the system will generate an alarm signal and record the alarm information in the fault data log as an important input data source for the subsequent integration analysis and processing module. Specifically, the alarm signal contains a variety of information content, such as the time when the alarm is triggered, the identification of the relevant sensor, the operating parameters that trigger the alarm, and the specific amplitude of the threshold.

[0032] In some embodiments, in order to improve the accuracy and real-time performance of data collection, the data collection and alarm module can also use multi-channel data fusion technology. In a specific implementation, multi-channel data is processed by timestamp synchronization and signal strength weighted fusion to eliminate abnormal fluctuations caused by signal delay or interference. As a possible implementation method, the fused data can effectively reflect the operating status of the device and provide a more accurate basis for the calculation of dynamic thresholds and alarm judgment.

[0033] In addition, in another possible implementation, the data acquisition and alarm module can also use different types of sensors for different equipment component types and operation scenarios. For example, for high-speed rotating equipment, high-precision vibration sensors can be used first; for large electrical equipment, the layout density of current and temperature sensors can be increased to enhance the comprehensiveness and reliability of data acquisition.

[0034] In general, the data acquisition and alarm module undertakes the tasks of basic data input and preliminary anomaly detection in the entire fault diagnosis and location technology. Its dynamic threshold judgment mechanism can quickly detect anomalies in the equipment's operating status and provide efficient and reliable data support for subsequent analysis, processing and fault location.

[0035] The integrated analysis and processing module describes the equipment operation status by constructing a nonlinear analysis model. The state change of the nonlinear analysis model is described by the state transfer function, and the model contains a fault disturbance term to represent the state offset caused by the fault. The integrated analysis and processing module analyzes the system state by constructing a positive definite function. When the derivative value of the function is negative, the system is in a normal state. When the derivative value is positive, the system is judged to be abnormal and enters the fault analysis stage.

[0036] In this embodiment, the integrated analysis and processing module receives multi-dimensional data from the data acquisition and alarm module, analyzes the dynamic characteristics of the equipment operation status, and provides processed abnormal data for the multi-source fault location module. The integrated analysis and processing module describes the equipment operation status by constructing a nonlinear analysis model, uses a positive definite function to evaluate the stability and abnormal status of the system, and outputs possible fault characteristics or offset parameters.

[0037] In this embodiment, the nonlinear analysis model is used to characterize the operating state of the device and its changing relationship. The model describes the dynamic behavior of the system through the state transfer function, which is in the following form: in, is the time derivative of the device state variable, indicating the rate of change of the state; is the current state vector of the device, which contains multiple state variables, such as temperature, vibration frequency, etc.; is an external input signal, which represents the control variable or external stimulus of the device; is a nonlinear state transfer function, which represents the dynamic changes of the system under normal conditions; is the fault disturbance term, which reflects the abnormal state deviation of the system caused by the fault.

[0038] In some embodiments, the state transfer function The specific form can be designed according to the operating principle and dynamic characteristics of the equipment. For example, for a vibration system, the state transfer function can adopt a second-order nonlinear vibration equation; for a multivariable control system, the function form may involve a nonlinear expression of the coupling relationship.

[0039] Specifically, the fault disturbance term It is an important part of the nonlinear analysis model and can be estimated through statistical analysis of historical equipment data or real-time monitoring. In one possible implementation, It can indicate the magnitude of the impact of certain specific failure modes, such as the performance degradation caused by aging of a component.

[0040] The integrated analysis and processing module further evaluates the stability of the system state through a positive definite function. In general, a positive definite function is constructed as an energy function of the device state variable to measure the stability of the system. The typical form of a positive definite function is:

[0041] in, is a positive definite function, used to describe the energy of the current state of the device; is a symmetric positive definite matrix whose elements are used to weight the contributions of different state variables. Represents the transposed matrix.

[0042] In one possible implementation, the positive definite function The derivative of is used to determine the changing trend of the system state, and its calculation formula is: in, For time The derivative of is used to describe the rate of change of the system state energy; is the state vector The gradient of Represents the state vector About Time The derivative of Specifically, when When , the system is in a stable state; when When the system is judged to be unstable and enter the fault analysis stage; As an option, the positive definite function can be adjusted according to the dynamic characteristics of different equipment. For example, for high dynamic load equipment, additional damping terms can be introduced to enhance the sensitivity to vibration-type faults. In another possible implementation, the construction of the positive definite function can also be combined with the calculation of the multivariate covariance matrix to improve the description of the coupled variables; In some embodiments, the integrated analysis and processing module can also generate a set of continuous system stability assessment data by calculating the value and derivative value of the positive definite function in real time. These data can be used to monitor the changing trend of the system's operating state and predict possible unstable areas or abnormal states in advance. For example, in a high-temperature operation scenario, when the system state variable (such as temperature) gradually approaches the boundary value of the positive definite function, the module can issue an early warning signal to provide a reference for the possibility of faults for the multi-source fault location module.

[0043] As an extended implementation method, the integrated analysis and processing module can modify the nonlinear analysis model by introducing external environmental parameters (such as ambient temperature, humidity, etc.), thereby improving the adaptability to the system operation state. For example, in an environment with high humidity, the module can adjust the weights of certain sensitive variables in the state transfer function to enhance the accuracy of fault analysis.

[0044] Therefore, the integrated analysis and processing module can effectively identify the system's operating status changes and abnormal areas through the construction of nonlinear analysis models and the use of positive definite functions, and provide reliable input data for the multi-source fault location module. Its design not only improves the accuracy of system operating status assessment, but also provides strong technical support for the early detection and location of equipment faults.

[0045] The multi-source fault location module uses an optimization algorithm to select fault features. The optimization algorithm improves the fault location accuracy while minimizing the number of features through an objective function with regularization constraints.

[0046] In this embodiment, the multi-source fault location module receives the abnormal state data output by the integrated analysis and processing module, and selects fault features through an optimization algorithm to minimize the number of features while improving the positioning accuracy, thereby more effectively locating components that may or have failed in the system. The optimization algorithm achieves a trade-off selection of fault features through an objective function, and combines regularization constraints to ensure the robustness and performance of the model.

[0047] Among them, the multi-source fault location module first preprocesses the input data, including data normalization, noise removal, and correlation analysis. Specifically, the normalization step is used to unify the numerical range of each fault feature to the same scale to avoid the unbalanced impact of some features with a larger numerical range on the objective function of the optimization algorithm. In one possible implementation, the normalization formula is as follows:

[0048] in, represents the normalized eigenvalue, represents the original eigenvalue, and represent the mean and standard deviation of the features respectively.

[0049] The optimization algorithm uses an objective function to select fault features. The objective function not only takes into account the fault location accuracy, but also penalizes redundant features through regularization terms to reduce model complexity and improve the robustness of fault location. The expression of the objective function is as follows: in, represents the objective function value of the optimization algorithm; Represents a set of features; Indicates the fault location accuracy based on the current feature set; A weight factor representing the strength of regularization, used to control model complexity.

[0050] In one possible implementation, the optimization algorithm may use a Bayesian optimization method to find the optimal subset of fault features under specific conditions by searching for combinations of different feature subsets. In some embodiments, the optimization process dynamically adjusts the regularization strength to adapt to different data scales and feature dimensions. For example, when there are a large number of features, the regularization strength may be appropriately increased. to significantly reduce the impact of redundant features.

[0051] Specifically, when the dimension of the fault feature is high, the multi-source fault location module removes redundant features through correlation analysis. As an option, the correlation analysis can be calculated based on the mutual information or Pearson correlation coefficient between the features.

[0052] For the mutual information method, the formula is as follows: in, Representation characteristics and goals The mutual information of express and Joint probability distribution; and Respectively and The marginal probability distribution of .

[0053] In another possible implementation, the multi-source fault location module can also be combined with deep learning technology to further enhance the ability to extract fault features. Specifically, a graph neural network (GNN) can be used to model the dynamic coupling relationship between equipment components. Nodes represent equipment components, edges represent the association relationship between components, and node features are updated and aggregated through a multi-layer message passing mechanism. For example, for a certain mechanical equipment, the node feature can be the vibration amplitude of the component, and the edge weight can be the coupling stiffness between the components. Through multiple iterations, the graph neural network can effectively extract the dynamic characteristics between components and output the fault probability distribution of each component.

[0054] In general, the combination of optimization algorithms and deep learning technology can further improve the overall performance of the multi-source fault location module. As an extended application, the fault feature extraction results of the module can also be input into other analysis tools, such as predictive maintenance models or real-time alarm systems, to provide richer data support for the intelligent management of equipment operation.

[0055] Therefore, the multi-source fault location module can not only effectively extract features that are highly correlated with the fault, but also reasonably constrain the number of features, providing accurate and efficient input data for the subsequent three-dimensional fault visualization module. It undertakes the important functions of feature selection and data optimization in the entire fault diagnosis process, and plays an important role in improving the accuracy and efficiency of fault diagnosis.

[0056] The multi-source fault location module builds a graph model based on the association relationship between equipment components. The graph model includes nodes and edges, where nodes represent equipment components and edges represent the dynamic coupling relationship between equipment components. The graph model is analyzed through a graph neural network.

[0057] The graph neural network models the dynamic coupling relationship between equipment components through a multi-layer message passing mechanism. The node features of each layer are obtained by aggregating and linearly transforming the features of neighboring nodes, and finally outputs the failure possibility of each equipment component.

[0058] In this embodiment, the integrated analysis and processing module provides dynamic characteristic data of the equipment state, and the multi-source fault location module analyzes the complex dynamic relationship between the equipment components by further modeling these data. And a graph model is constructed based on the association relationship between the equipment components to reveal the hidden fault propagation path and the coupling characteristics between the components in the equipment structure. Through the analysis of the graph model, the components that may fail in the system can be accurately identified, and the accuracy of fault location can be further improved. Among them: the graph model consists of nodes and edges, the nodes are used to represent the various components of the equipment, and the edges describe the dynamic coupling relationship between these components. The dynamic coupling relationship can be defined based on physical connection, signal transmission path or functional dependency. For example, in some mechanical equipment, the node can represent a rotating shaft, a bearing or a gear, and the edge represents the transmission connection relationship between the components. In a possible implementation, the weight of the edge can also reflect the coupling strength between the equipment components, such as the transmission efficiency or the proportion of vibration energy transmission.

[0059] Specifically, the graph model is analyzed through a graph neural network. A graph neural network is a deep learning model that can directly process graph structured data. It extracts complex relationships between device components through multi-layer message passing and aggregation of node features. In one possible implementation, node features can include device operating parameters such as vibration frequency, temperature change, and pressure value, while edge features can describe the signal transmission strength or coupling stiffness between components.

[0060] In general, graph neural networks use a multi-layer message passing mechanism to model the dynamic coupling relationship of device components. In each layer, node features are updated and calculated by aggregating the features of its neighboring nodes. The update formula for node features is as follows: in, For Node In the Feature representation of the layer; For Node In the Feature representation of the layer; For Node The set of neighbor nodes; is an aggregation function used to aggregate the features of neighbor nodes, such as sum or average; For the The weight matrix of the layer; For the The bias vector of the layer; is the activation function; In a possible implementation, the aggregation function may select weighted average or maximum pooling to adapt to different data distribution characteristics. For example, in a vibration signal propagation scenario, weighted average can better reflect the actual energy distribution between nodes.

[0061] In some embodiments, the final output of the graph neural network is the feature representation of each node, which can be further converted into the failure probability distribution of the node. Specifically, by applying a classifier to the output node features, the probability of failure of each component can be predicted. For example, for a node The failure probability can be expressed as: in, For Node The probability of failure; For Node Features in the last layer of the graph neural network; and are the weights and biases of the classifier.

[0062] As an option, the classifier can use a simple linear layer or introduce more nonlinear structures to improve the accuracy of fault classification. In some embodiments, the historical data of the node can also be combined as an additional feature input to improve the recognition ability of long-term fault modes.

[0063] In another possible implementation, the edge features of the graph neural network can also be updated dynamically. The update mechanism of edge features is similar to that of node features, capturing the relationship changes between nodes through feature aggregation of neighboring nodes. This mechanism is particularly suitable for fault propagation scenarios, such as the chain vibration effect caused by bearing damage in mechanical systems.

[0064] In general, the combination of graph models and graph neural networks can effectively improve the accuracy of multi-source fault location. As an extended technology, the fault location results can also be compared and verified with the data output by the integrated analysis and processing module, thereby improving the overall reliability of the system. In this way, the multi-source fault location module can not only effectively identify potential faults in complex equipment, but also provide high-quality input data for the subsequent three-dimensional fault visualization module.

[0065] The multi-source fault location module fuses the collected multi-source data and uses a filtering algorithm to filter out random noise and outliers in the multi-source data. The filtering algorithm dynamically adjusts based on the deviation between the observed value at the current moment and the predicted state to generate a state estimate.

[0066] In this embodiment, the multi-source fault location module uses the multi-source data provided by the data acquisition and alarm module, combined with the results of the integrated analysis and processing module, to further analyze and process the equipment status. In general, multi-source data contains observations from multiple sensors, which may cause data deviations and outliers due to environmental noise, sensor failures, or signal interference. In order to ensure the accuracy and reliability of the data, the multi-source fault location module introduces a filtering algorithm to effectively filter out random noise and outliers in the data through dynamic adjustment, and finally generates a high-precision state estimate to provide reliable data support for subsequent fault diagnosis.

[0067] The core of multi-source data fusion lies in the design and application of filtering algorithms. The filtering algorithm dynamically adjusts the data by combining the current observation value with the predicted state of the system. Specifically, the filtering algorithm adopts the implementation method based on Kalman filtering, and its state estimation is updated by the following formula: in, Indicates the current time state estimation; Represents the state estimate predicted at the previous moment; Represents the observation value at the current moment; Represents the observation matrix, which is used to map state variables to the observation space; Represents the filter gain, which is used to dynamically adjust the impact of observations on the estimate.

[0068] In general, the filter gain The calculation formula is: in, Represents the state prediction covariance matrix, which is the covariance matrix of the state estimation error obtained in the prediction step, describing the size of the prediction error. The dimension is consistent with the state vector, reflecting the uncertainty of the state prediction; Represents the observation matrix The transpose of is used to reflect the observed value into the state variable space, Represents the observation noise covariance matrix, which is used to describe the distribution characteristics of random noise in the observation process, usually determined by the measurement noise of the sensor. The dimension is the same as the observation value. Represents the inverse matrix of the weighted sum of the covariance matrix. This term contains the prediction covariance (i.e., the uncertainty of the state prediction) and the observation noise covariance (i.e., the uncertainty of the measurement error). The weighted ratio is obtained by the inversion operation, which is used to dynamically adjust the weight of the influence of the observation value on the state estimation.

[0069] In one possible implementation, the filtering algorithm adjusts the covariance matrix and To adapt to different data noise levels and equipment operating environments. For example, when the sensor signal noise is large, you can appropriately increase The value of can reduce the influence of observations on state estimation; in the case of less noise, it can reduce to increase the weight of the observations.

[0070] Specifically, the filtering algorithm can not only eliminate random noise, but also detect and remove outliers. For example, in a mechanical system running under high load, if the output value of a sensor suddenly exceeds the normal range, the filtering algorithm can identify the abnormal signal based on the deviation between the predicted value and the observed value of the current state, and dynamically adjust the state estimation to avoid the influence of the outlier on the system analysis.

[0071] In some embodiments, in order to improve the efficiency of the filtering algorithm, multi-source data fusion can be implemented in a distributed computing manner. Specifically, multi-source data can be divided into different groups according to data source and type, and each group is independently processed by the filtering algorithm, and finally the results of the group processing are summarized to generate an overall high-precision state estimate. In another possible implementation, the filtering algorithm can also be combined with an adaptive parameter adjustment strategy to dynamically optimize the filtering parameters according to the real-time collected data to further improve the robustness of data processing.

[0072] As an option, the filtering algorithm can also be combined with a deep learning model to pre-process the abnormal patterns of multi-source data. For example, by training on historical data, the deep learning model can identify potential abnormal signal features and use these features to guide the parameter adjustment of the filtering algorithm. This combination can further improve the accuracy and reliability of multi-source data fusion.

[0073] In another possible implementation, the filtering algorithm can also be applied to dynamic weight allocation to quantify the credibility of different data sources. For example, for some critical sensors, higher weights can be assigned to make them have a greater impact on the state estimation results; while for data sources with greater noise, their weights can be reduced to reduce interference with the overall estimation.

[0074] Through the above technology, the multi-source fault location module can efficiently fuse multi-source data, filter out noise and outliers, and thus provide strong support for the accurate estimation of the equipment operation status. The flexibility and dynamic adjustment capability of the filtering algorithm ensure that the module can adapt to the complex and changing operating environment and provide accurate data input for subsequent fault location and three-dimensional visualization.

[0075] The 3D fault visualization module generates annotations of faulty components in the 3D model of the equipment based on the fault location results. The annotations highlight the location of the faulty components and provide a text description of the cause of the fault.

[0076] The 3D fault visualization module generates the final model of the device by superimposing the annotation matrix with the original 3D model. The annotation matrix contains the index information of the faulty parts and the annotation style parameters, which are used to dynamically adjust the annotation style.

[0077] In this embodiment, the 3D fault visualization module uses the fault location results provided by the multi-source fault location module to process and annotate the 3D model of the equipment. Generally, the module displays the diagnosis results by generating annotations of the faulty components in the 3D model, and intuitively presents the fault information to the operation and maintenance personnel. The annotations include the location of the faulty components and the related fault causes, and are visually enhanced by highlighting, color changes, or animation effects to help the operation and maintenance personnel quickly identify and handle the problem components.

[0078] Specifically, it processes based on the 3D model of the original device. The 3D model is usually generated from the CAD data of the device, representing the various components of the device and their geometric position relationships. Specifically, this module first uses the component index information provided in the fault location result to locate the corresponding components in the 3D model and generate a labeling matrix. The structure of the labeling matrix includes the index information of the faulty component and the labeling style parameters, which are used to define the specific method of labeling, such as highlight color, transparency change or flashing frequency.

[0079] In one possible implementation, the construction of the labeling matrix can be performed using the following formula:

[0080] in, Represents the final generated three-dimensional model; a three-dimensional model representing the original equipment; represents the highlight intensity factor; Represents the annotation matrix, which contains the index information and style parameters of the faulty components; Highlight intensity factor The value of is adjusted dynamically according to the severity of the fault. For example, for critical components or high priority faults, Set to a larger value for a more visual effect, or use a smaller value for lower priority failures to reduce visual noise.

[0081] Specifically, the annotation style parameters include not only highlight colors, but also transparency and animation effects. For example, in some embodiments, when a serious fault is detected in a component of the device, the component will be highlighted in red in the 3D model with a flashing effect to alert the operator. In another possible implementation, the internal structure of the component can be displayed in combination with different transparency settings to help the operator more intuitively understand the fault location and its relationship with surrounding components.

[0082] As an option, the 3D fault visualization module can also provide text descriptions associated with the annotations. These descriptions usually include the name of the faulty component, the cause of the fault, and the recommended treatment measures. For example, in a 3D model of an electrical device, when an overload problem is detected on a circuit board, the text description can display "Component: Circuit Board A, Cause of Fault: Overload, Recommended Action: Check circuit connections and replace fuses."

[0083] In another possible implementation, the annotation matrix can also be dynamically associated with the operating status data of the equipment to achieve real-time visual updates. For example, when the status parameters of a component change, the annotation style in the 3D model will also be updated synchronously. Specifically, the color intensity or flashing frequency of the annotation can be adjusted by monitoring the real-time changes in parameter values. For example, when the temperature of a component gradually rises and approaches the warning value, its color can gradually transition from green to red, indicating that the risk of failure is increasing.

[0084] In some embodiments, the 3D fault visualization module can also support interactive functions, such as allowing maintenance personnel to click on a highlighted faulty component to view its detailed information or historical data. As an extended function, interactive operations can also trigger linkage with other systems, such as sending maintenance tasks to a work order management system or requesting a more detailed diagnostic report.

[0085] Through the above methods, the 3D fault visualization module can intuitively display fault information and provide accurate fault location and cause analysis for operation and maintenance personnel. The design of the annotation matrix and its dynamic adjustment function ensure the adaptability and practicality of the model in a complex operating environment, providing strong support for the rapid troubleshooting and repair of equipment faults. Example

[0086] In this embodiment, by strengthening data fusion and analysis capabilities, an artificial intelligence engine is combined with a big data platform to construct a real-time data stream intelligent analysis and pattern mining method for early identification of potential equipment risks and proactive intervention to prevent fault escalation. In this embodiment, the integrated analysis and processing module is deeply integrated with the enterprise-level big data warehouse, and a deep learning model is used to extract features and learn association rules for massive monitoring data, ultimately achieving prediction and proactive intervention of potential faults.

[0087] Generally, a large amount of monitoring data is generated during the operation of the equipment, including state parameters such as temperature, pressure, vibration, and power consumption. After these data are collected by various sensors, they are uploaded to the integration analysis and processing module via the data acquisition and alarm module. In this embodiment, the integration analysis and processing module captures these monitoring data in real time by connecting to the enterprise-level big data warehouse, and performs intelligent analysis with the help of deep learning algorithms.

[0088] Specifically, the integrated analysis and processing module includes the following key steps: Access and processing of real-time data streams; Feature extraction and pattern mining of deep neural networks; Prediction and early warning of potential failures.

[0089] Access and processing of real-time data streams In this embodiment, the integration analysis and processing module captures real-time data streams by connecting to the enterprise-level big data warehouse. Data processing is divided into the following steps: 1. Data stream access: Data is uploaded from multi-source sensors to the enterprise-level big data warehouse in the form of message queues. As a possible implementation method, message queues use high-throughput message middleware such as Kafka to transmit high-frequency monitoring data to ensure real-time performance. The data captured in real time includes device status parameters, operation history records, and external environmental conditions.

[0090] 2. Data preprocessing: In order to improve the efficiency of data analysis, the original monitoring data needs to be standardized and cleaned. Data cleaning includes removing missing values, eliminating abnormal data points and unifying data formats.

[0091] Feature extraction and pattern mining of deep neural networks The integrated analysis and processing module uses deep neural networks to extract features and mine patterns from the processed monitoring data. The structure of the deep neural network includes an input layer, multiple hidden layers, and an output layer, which are used to extract features and output prediction results. .

[0092] 1. Data input:

[0093] in, Indicates time Real-time monitoring data, including various status parameters.

[0094] 2. Feature extraction: Deep neural networks extract features from data through multiple hidden layers. As an option, the hidden layer can use a convolutional neural network (CNN) to extract local features in the data, or a long short-term memory network (LSTM) to capture the dynamic changes in time series. The feature update formula of the hidden layer is: in, For the Feature representation of the hidden layer; and are the weight matrix and bias vector respectively; is an activation function, such as the ReLU function.

[0095] 3. Pattern Mining: The high-dimensional features extracted by the deep neural network are input into the association rule mining module to identify the relationship between key features. For example, the trained model is used to identify the correlation between certain features (such as hard disk read and write delay and temperature rise) and specific failure events. The mathematical expression of the association rule is as follows: in, is the confidence of the association rule; Means that it happens at the same time and The proportion of data; Indicates an event The proportion of occurrence.

[0096] Prediction and early warning of potential failures In this embodiment, prediction of potential faults and active intervention are achieved through the analysis results of the deep learning model.

[0097] 1. Fault prediction: When the characteristic pattern matches the historical failure pattern to a high degree, the system generates a failure warning signal. For example, server hard disk read and write delays and temperature rise are identified as key patterns before downtime, and the system can issue an alarm several hours in advance. The output formula of the prediction result is: in, is the result vector of fault prediction; is the output of the last hidden layer; and are the weights and biases of the output layer.

[0098] 2. Active intervention: When the system detects a potential failure, it can trigger automated actions to prevent the failure from escalating. For example, when it detects that a hard drive is at risk of failure, the system can automatically migrate data to a spare drive and notify the administrator to replace the device.

[0099] In some embodiments, the integrated analysis processing module can also combine real-time external environmental data (such as temperature, humidity and power fluctuations) to correct the prediction results. For example, in a high temperature operating environment, the normal read and write delay range of the hard disk may need to be dynamically adjusted to avoid misjudgment.

[0100] In another possible implementation, the system can continuously optimize the deep learning model through an adaptive training mechanism. For example, the system uses newly collected monitoring data to update the model weights online to improve the accuracy of fault prediction. Through this embodiment, the integrated analysis and processing module can process massive amounts of data in real time, automatically identify potential equipment failure modes and take intervention measures in advance, avoiding system downtime and equipment damage. At the same time, by combining deep neural networks and association rule mining technology, comprehensive analysis and accurate prediction of equipment status are achieved, providing strong technical support for intelligent operation and maintenance.

[0101] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A fault diagnosis and location technology, characterized in that: include: Data collection and alarm module, used to collect equipment operation status data and generate dynamic alarm signals; Integrated analysis and processing module, used to build nonlinear analysis model based on equipment operation status, and identify abnormal status of the system through analysis and processing; Multi-source fault location module, which is used to analyze the identified abnormal conditions, extract fault features through optimization algorithms and locate components that have or may fail in combination with deep learning models; The 3D fault visualization module is used to generate a 3D model of the equipment based on the fault location results and mark the location and cause of the faulty components in the model.

2. A fault diagnosis and positioning technology according to claim 1, characterized in that: In the data collection and alarm module, the equipment operation status data includes temperature, vibration, pressure, and current, and a dynamic threshold is used to determine whether to trigger an alarm, wherein the dynamic threshold is calculated based on the mean and standard deviation of historical operation data.

3. A fault diagnosis and location technology according to claim 1, characterized in that: The integrated analysis processing module describes the equipment operation status by constructing a nonlinear analysis model. The state change of the nonlinear analysis model is described by a state transfer function, and the model includes a fault disturbance term for indicating the state deviation caused by the fault.

4. A fault diagnosis and location technology according to claim 3, characterized in that: The integrated analysis processing module analyzes the system state by constructing a positive definite function. When the derivative value of the function is negative, the system is in a normal state. When the derivative value is positive, the system is judged to be abnormal and enters the fault analysis stage.

5. A fault diagnosis and location technology according to claim 1, characterized in that: The multi-source fault location module selects fault features using an optimization algorithm, and the optimization algorithm improves the fault location accuracy while minimizing the number of features through an objective function with regularization constraints.

6. A fault diagnosis and location technology according to claim 1, characterized in that: The multi-source fault location module constructs a graph model based on the association relationship between equipment components. The graph model includes nodes and edges, wherein nodes represent equipment components and edges represent dynamic coupling relationships between equipment components. The graph model is analyzed through a graph neural network.

7. A fault diagnosis and location technology according to claim 6, characterized in that: The graph neural network models the dynamic coupling relationship of equipment components through a multi-layer message passing mechanism. The node features of each layer are obtained by aggregation and linear transformation of neighbor node features, and finally the failure possibility of each equipment component is output.

8. A fault diagnosis and location technology according to claim 1, characterized in that: The multi-source fault location module fuses the collected multi-source data and uses a filtering algorithm to filter out random noise and abnormal values ​​in the multi-source data. The filtering algorithm is dynamically adjusted based on the deviation between the observed value at the current moment and the predicted state to generate a high-precision state estimate.

9. A fault diagnosis and location technology according to claim 1, characterized in that: The three-dimensional fault visualization module generates a label of the faulty component in the three-dimensional model of the device according to the fault location result. The label displays the position of the faulty component in a highlighted manner and provides a textual description of the cause of the fault.

10. A fault diagnosis and location technology according to claim 9, characterized in that: The three-dimensional fault visualization module generates a final model of the device by superimposing a labeling matrix with the original three-dimensional model. The labeling matrix contains index information of the faulty components and labeling style parameters for dynamically adjusting the labeling style.

Images