Fault Identification Method Based on Intelligent Model

Abstract

The present invention relates to the technical field of fault identification, and particularly to a fault identification method based on an intelligent model, which includes collecting comprehensive state information of a device and its environment by using multi-modal high-precision sensing elements, and formulating a fusion strategy to form multi-dimensional tensor data to provide a basis for subsequent analysis; by constructing a semantic space, using generative adversarial network technology to mine fault embryo patterns and conduct traceability analysis; based on the fault embryo patterns and traceability results, constructing a hidden order model reflecting the dynamic changes of the device operation architecture; by using an association rule recognition algorithm to find the hidden associations between fault embryos and features, constructing an inference prediction engine for fault prediction to improve the accuracy and timeliness of fault prediction; combining the inference prediction results, establishing an adaptive defense mechanism to identify and process the fault states of the device operation architecture, and ensuring the safety and stability of the device operation.

Description

Technical Field

[0001] The present invention relates to the technical field of fault identification, and particularly to a fault identification method based on an intelligent model. Background Art

[0002] There are still some challenges in current fault identification, including the following aspects: incomplete monitoring, traditional fault identification methods may not be able to comprehensively monitor the comprehensive state information of equipment and its environment, resulting in limited accuracy and timeliness of fault identification; difficult to mine fault association rules, the association rules between faults and features are often deeply hidden, and traditional fault identification methods are difficult to effectively mine these rules, affecting the accuracy of fault prediction; the fault identification method is not intelligent enough, traditional fault identification methods often rely on manual experience and rules, lacking intelligent processing means, resulting in limited efficiency and accuracy of fault identification. Therefore, the present invention proposes a fault identification method based on an intelligent model. Summary of the Invention

[0003] The purpose of the present invention is to solve the problems in the background art, and propose a fault identification method based on an intelligent model.

[0004] To achieve the above purpose, the present invention adopts the following technical solutions:

[0005] A fault identification method based on an intelligent model, including:

[0006] S1. Use multi-modal high-precision sensing elements to construct a multi-dimensional field perception structure for collecting the comprehensive state information of equipment and its environment; for the obtained field data, formulate corresponding fusion coding strategies to obtain the fused multi-dimensional tensor data;

[0007] S2. Use natural language processing and knowledge graph technology to construct a semantic space and generate a semantic fusion feature data set; in the constructed semantic space, use generative adversarial network technology to mine fault embryo patterns; if a fault embryo pattern is identified, use the entity relationship network in the semantic space for traceability analysis;

[0008] S3. Based on the identified fault embryo patterns and their traceability analysis results, construct a hidden order model;

[0009] S4. Based on the hidden order model, use the association rule identification algorithm to further find the association rules between the already identified fault embryo patterns and the semantic fusion feature data set; according to the obtained association rules and the state of the current equipment operation architecture, construct an inference prediction engine based on hidden order association and perform inference prediction;

[0010] S5. Regard the normal state of the equipment operation architecture as itself and the fault state as non-self; combine the inference prediction results to establish an adaptive defense mechanism.

[0011] Furthermore, the multi-dimensional field specifically includes the microscopic electromagnetic field, the weak gravitational field, as well as the macroscopic geographical environment and spatial location; the multi-modal high-precision sensing components specifically include the near-field optical sensor, the gravitational wave sensor, and the satellite positioning and geographic information system.

[0012] Furthermore, the process of S1 includes:

[0013] Using the near-field optical sensor to capture the change in the distribution of the microscopic electromagnetic field on the surface of the device; using the gravitational wave sensor to monitor the weak gravitational field fluctuations caused by the change in the mass distribution of the large device cluster; combining the satellite positioning and geographic information system to obtain the macroscopic geographical environment and spatial location information where the device is located; obtaining the microscopic electromagnetic field data on the surface of the device, the weak gravitational field fluctuation data of the large device cluster, and the macroscopic geographical environment and spatial location data where the device is located;

[0014] Collecting various text description data during the operation of the device;

[0015] Mapping the microscopic electromagnetic field data into the complex space, and using the real part and the imaginary part of the complex number to represent the coupling relationship between different physical quantities; for the weak gravitational field fluctuation data, using the wavelet transform algorithm for encoding processing to form a data vector with high-dimensional features; for the macroscopic geographical space data, converting the geographical information into a topological feature vector through topological mapping;

[0016] For the text description data, using natural language processing technology to convert it into a word vector representation;

[0017] Using the tensor product operation to fuse the feature vectors obtained by these different encoding methods into a unified multi-dimensional tensor data.

[0018] Furthermore, S2 includes:

[0019] Extracting the entities related to the device and the relationships between entities from the fused multi-dimensional tensor data to construct a fault knowledge graph;

[0020] Performing a semantic association operation on various features in the fused multi-dimensional tensor data with the entities already constructed in the knowledge graph, assigning semantic labels to each data feature, thereby forming a semantic space that fuses multi-source data and domain knowledge, and generating a semantic fusion feature data set; among them, various features in the fused multi-dimensional tensor data include microscopic electromagnetic field features, weak gravitational field features, macroscopic geographical space features, and text description data features;

[0021] The generator is responsible for generating samples of possible faulty embryo patterns, and the discriminator distinguishes whether the samples of faulty embryo patterns are real faulty embryos; among them, when generating samples of faulty embryo patterns, the generator comprehensively uses the operating states of the equipment reflected by the information in each field of the generated semantic fusion feature dataset;

[0022] Through continuous adversarial training, the generator learns the potential faulty embryo feature patterns in the semantic fusion feature dataset;

[0023] Along the relationship paths between entities in the knowledge graph, trace the origin and development context of the faulty embryo; among them, the origin of the faulty embryo traces back to multiple aspects, including the origin of the equipment itself, the origin of the external environment, and the origin of human operations; the development context of the faulty embryo includes the initial evolution of the fault, the expansion of the influence range, and the changes at the equipment operation level;

[0024] Through calculation and analysis, obtain the relevant quantitative values for the traceability analysis.

[0025] Furthermore, the hidden order model regards the entire equipment operation architecture as a dynamic composite system, in which there are complex and hidden interaction relationships among the various elements; among them, the equipment operation architecture refers to the comprehensive equipment operation system composed of equipment components, software management programs, operators, and the operating environment; the elements of the equipment operation architecture cover equipment components, software management programs, operators, and the operating environment.

[0026] Furthermore, S3 includes:

[0027] Quantify each element in the equipment operation architecture, and use nonlinear dynamic equations to analyze the dynamic change relationships among the elements;

[0028] During the construction process, add the information of each element in the equipment operation architecture to the parameter settings of the hidden order model, and combine the relevant quantitative values given by the faulty embryo pattern and its traceability analysis results, and associate them with the hidden order model parameters;

[0029] Determine the normal state or potential fault state of the equipment operation architecture through the hidden order model: define the threshold for the normal state; compare and analyze the output value of the hidden order model with the defined threshold. If the output value of the hidden order model is within the normal state threshold range, that is, the output value is greater than or equal to the normal state lower threshold and less than or equal to the normal state upper threshold, it is determined that the current state of the equipment operation architecture meets the standard of normal operation and is regarded as the normal state; if the output value of the hidden order model exceeds the normal state threshold range, that is, the output value is less than the normal state lower threshold or greater than the normal state upper threshold, it indicates that there is a problem with the equipment operation architecture and is regarded as the potential fault state;

[0030] Meanwhile, capture the order structure hidden inside the operation architecture of the capture device that causes the transition from the normal state to the fault state.

[0031] Furthermore, S4 includes:

[0032] Obtain the recognized fault embryo patterns and the semantic fusion feature dataset: The fault embryo patterns are stored in vector form, and each element in the fault embryo pattern vector represents the corresponding fault feature identifier and quantization value. The semantic fusion feature dataset is stored in the form of a data table, with each row representing a data sample and each column representing a fusion feature;

[0033] Mine frequent item sets in the integrated fault embryo pattern vector and semantic fusion feature dataset through the Apriori algorithm, and set the minimum support and confidence thresholds to identify the feature combinations that frequently and simultaneously appear before the fault occurs. These feature combinations further constitute and screen out the association rules for the occurrence of the fault;

[0034] Combine the obtained association rules with the state of the current device operation architecture to construct an inference prediction engine based on hidden order association; use the inference prediction engine to infer and predict future fault occurrences according to the state of the current device operation architecture and the association rules: If it is monitored that the key monitoring parameters of the device operation architecture satisfy the association rules, the inference prediction engine predicts the probability, time, and type of the fault occurrence; among them, the key monitoring parameters are important indicators for monitoring and evaluating the device state in the device operation architecture and are obtained in real time through the sensor network;

[0035] Meanwhile, use the Bayesian inference method to correct the prediction results.

[0036] Furthermore, mine frequent item sets in the integrated fault embryo pattern vector and semantic fusion feature dataset through the Apriori algorithm, and set the minimum support and confidence thresholds to identify the feature combinations that frequently and simultaneously appear before the fault occurs, including:

[0037] Integrate the faulty embryo pattern vector with the semantic fusion feature dataset to ensure that each element in the faulty embryo pattern vector matches and corresponds to the corresponding fusion feature in the semantic fusion feature dataset; convert the integrated dataset into the form of a transactional database processed by the Apriori algorithm, where each transaction represents a set containing multiple features, and these features come from the faulty embryo pattern vector and the semantic fusion feature dataset; set the minimum support threshold for filtering frequent item sets, where the support represents the frequency of an item set appearing in all transactions; set the minimum confidence threshold for filtering association rules; utilize the iterative process of the Apriori algorithm, starting from a single item, gradually generating frequent item sets containing more items, where the generated frequent item sets represent the feature combinations that often appear simultaneously before a fault occurs; in each iteration, generate candidate item sets based on the frequent item sets generated in the previous round, and calculate the support of each candidate item set; retain the candidate item sets with support not lower than the minimum support threshold as the new frequent item sets for generating more candidate item sets in the next round; repeat the iterative process until no new frequent item sets can be generated; for each frequent item set, generate all possible association rules, and calculate the support and confidence of each association rule; retain the association rules with support and confidence not lower than the set thresholds as the final fault occurrence association rules; utilize the association relationship between the feature combinations represented by the generated association rules and the occurrence of faults.

[0038] Further, S5 includes:

[0039] Obtain the inference and prediction results of the hidden order model, and set its own tolerance threshold set; when the detected probability value of the fault occurrence deviates from the range of the set own tolerance threshold set, then activate the adaptive defense mechanism to identify and process the fault state of the device operation architecture; generate an adaptive regulation strategy according to the type and severity of the fault.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows: By constructing a multi-modal high-precision sensing element to build a multi-dimensional field sensing structure, the comprehensive monitoring of the device and its environment is realized, and the accuracy and timeliness of fault identification are improved; by using natural language processing and knowledge graph technology to construct a semantic space and mine the faulty embryo pattern, the early detection and traceability analysis of faults are realized, which helps to timely discover and handle potential faults; by constructing a hidden order model and an inference prediction engine, the association rules between faults and features can be further mined, improving the accuracy of fault prediction. By introducing an adaptive defense mechanism, a regulation strategy can be generated according to the type and severity of the fault, effectively ensuring the safety and stability of the device operation. Description of the Drawings

[0041] Figure 1Flowchart of the fault identification method based on an intelligent model proposed by the present invention. Detailed implementation manners

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0043] Refer to Figure 1 , the fault identification method based on an intelligent model includes:

[0044] S1. Use multi-modal high-precision sensing elements to construct a multi-dimensional field perception structure for collecting comprehensive state information of the device and its environment; for the obtained field data, formulate corresponding fusion coding strategies to obtain the fused multi-dimensional tensor data;

[0045] Among them, the multi-dimensional field includes a microscopic electromagnetic field, a weak gravitational field, and a macroscopic geographical environment and spatial position. The multi-modal high-precision sensing elements include a near-field optical sensor (for capturing changes in the distribution of the microscopic electromagnetic field), a gravitational wave sensor (for monitoring weak gravitational field fluctuations), and a satellite positioning and geographic information system (for obtaining macroscopic geographical environment and spatial position information);

[0046] S2. Use natural language processing and knowledge graph technology to construct a semantic space and generate a semantic fusion feature dataset; in the constructed semantic space, use generative adversarial network technology to mine fault embryo patterns; if a fault embryo pattern is identified, use the entity relationship network in the semantic space for traceability analysis to understand the origin and development context (possible causes and paths) of the fault;

[0047] S3. Based on the identified fault embryo patterns and their traceability analysis results, construct a hidden order model; this model regards the entire device operation architecture as a dynamic composite system, in which there are complex and hidden interaction relationships among various elements;

[0048] Among them, the device operation architecture refers to a comprehensive device operation system composed of device components, software management programs, operators, and the operating environment. For example, for an automated production line device, the device operation architecture includes all elements related to device operation such as the mechanical devices of the production line, the program software that controls the operation of the production line, the operating workers, and the plant environment where the production line is located; the elements of the device operation architecture cover device components (such as motors, sensors, valves, etc.), software management programs, operators, and the operating environment (temperature, humidity, electromagnetic interference, etc.). These elements interact with and influence each other, jointly determining the operating state of the device; in addition, the hidden order model is used to capture and describe these dynamic transformation behaviors hidden inside the device operation architecture.

[0049] S4. Based on the hidden order model, using the association rule recognition algorithm, further search for the association rules between the already identified fault embryo patterns and the semantic fusion feature data set; according to the analyzed association rules and the current state of the device operation architecture, construct an inference prediction engine based on hidden order association, and conduct inference prediction to achieve accurate prediction of future fault occurrences.

[0050] S5. Regard the normal state of the device operation architecture as self and the fault state as non-self; combined with the inference prediction results, establish an adaptive defense mechanism.

[0051] Among them, referring to the self / non-self recognition in the biological immune system (that is, an organism can distinguish its own cells from foreign invading pathogens), an adaptive defense mechanism is established. This defense mechanism can automatically adjust the device operation parameters or take other preventive measures according to the inference prediction results to avoid or reduce the impact of faults.

[0052] It should be further noted that in the specific implementation process, a multi-modal high-precision sensing element is used to construct a multi-dimensional field perception structure to collect the comprehensive state information of the device and its environment; for the obtained field data, the process of formulating a corresponding fusion coding strategy to obtain the fused multi-dimensional tensor data is as follows:

[0053] Capture the change in the microscopic electromagnetic field distribution on the device surface using a near-field optical sensor: Select a near-field optical sensor based on the principle of a scanning near-field optical microscope, arrange the sensors on the device surface in a uniformly distributed manner, with a sensor distribution density of 1 sensor per square centimeter, and calibrate the sensors using a standard calibration block with a calibration accuracy of 0.1 microvolts per meter to ensure accurate capture of the change in the microscopic electromagnetic field distribution; Monitor the weak gravitational field fluctuations caused by changes in the mass distribution of a large device cluster using a gravitational wave sensor: Adjust the sensitivity range of the gravitational wave sensor according to the mass and distribution characteristics of the large device cluster; Exclude the influence of other interference factors in the environment (such as the tiny fluctuations of the Earth's own gravitational field, the interference of the gravitational force of other nearby objects, etc.) on the monitoring of weak gravitational field fluctuations by setting up a shielding device and using a differential measurement method; Combine satellite positioning and geographic information system to obtain the macroscopic geographical environment and spatial location information of the device: Use a satellite positioning system that integrates GPS and Beidou multi-systems, and the data in the geographic information system is updated once an hour, and the data source is the integration of official surveying and mapping data and real-time crowdsourcing data; Obtain the microscopic electromagnetic field data on the device surface, the weak gravitational field fluctuation data of the large device cluster, and the macroscopic geographical environment and spatial location data of the device; It is understandable that these data come from different fields, contain detailed indications of the device's operating state, and comprehensively reflect the comprehensive state of the device and its surrounding environment;

[0054] Collect various text description data during the operation of the device, including log files, operation procedures, maintenance records, fault reports, etc., to reflect the background information of the device's operating state;

[0055] Map the microscopic electromagnetic field data to the complex space, and use the real and imaginary parts of the complex number to represent the coupling relationship between different physical quantities: Use a non-linear mapping function (where a, b, and c are constants determined according to the device characteristics) to map the microscopic electromagnetic field data to the complex space; For the weak gravitational field fluctuation data, use a wavelet transform algorithm for encoding processing to form a data vector with high-dimensional features: For example, use the Daubechies wavelet basis function for transformation; For the macroscopic geographical space data, convert the geographical information into a topological feature vector through topological mapping: Map based on the network topological structure in graph theory, use the geographical coordinates in the geographical information as nodes, the connection relationship between geographical regions as edges, the node attribute as the coordinate value, and the edge attribute as the connection weight, and determine the weight size according to the distance;

[0056] For the text description data, use natural language processing technology to convert it into a word vector representation: Use the Word2Vec pre-trained model to process the text description data, and standardize the specific abbreviations related to the device during the processing;

[0057] Using the tensor product operation, the eigenvectors obtained by these different coding methods are fused into a unified multi-dimensional tensor data, comprehensively retaining the key information of each field data: The weight setting principle in the tensor product operation is set according to the importance degree of each eigenvector to the device operation state. For example, the weight of the micro-electromagnetic field data is 0.3, the weight of the weak gravitational field fluctuation data is 0.2, the weight of the macro-geospatial data is 0.2, and the weight of the text description data is 0.3. The minimum-maximum normalization process is performed on the fused multi-dimensional tensor data to normalize the data to the [0,1] interval.

[0058] It should be further noted that in the specific implementation process, natural language processing and knowledge graph technology are used to construct a semantic space and generate a semantic fusion feature data set; in the constructed semantic space, generative adversarial network technology is used to mine the fault embryo pattern; if a fault embryo pattern is identified, the process of using the entity relationship network in the semantic space for traceability analysis is as follows:

[0059] Extract the entities related to the device (device components, fault types, repair measures) and the relationships between entities (causal relationships or association relationships) from the fused multi-dimensional tensor data to construct a fault knowledge graph; it can be understood that the sources of these entities and relationships are relatively extensive, including both the information presented in the text description of the device operation data and the potential connections implied in the other field information in the fused multi-dimensional tensor data;

[0060] Specifically, A1. Adopt a method combining rules and machine learning to extract device-related entities from the fused multi-dimensional tensor data: Create a predefined dictionary for device components, fault types, and repair measures. For device components, the dictionary contains the names of common parts of the device and their synonyms; Use an entity recognition model based on a convolutional neural network to perform a preliminary scan of the fused multi-dimensional tensor data. Divide the fused multi-dimensional tensor data into windows of a fixed length. For example, each window contains 100 data points. The kernel size of the entity recognition model based on the convolutional neural network is set to 3, and the stride is 1. Extract features through multiple convolutional layers and pooling layers; For the data part with matching items in the predefined dictionary, directly mark it as the corresponding entity. For the part that is not matched but is recognized as a potential entity by the entity recognition model, conduct manual review and confirmation; A2. Adopt a method based on logical rules and data statistical analysis to determine the relationships between entities (causal relationships or association relationships): For causal relationships, set logical rules. For example, if the temperature of a certain component of the device increases (reflected by sensor data) and then a device failure occurs (reflected by fault report data), establish a causal relationship from the temperature increase to the occurrence of the fault; For association relationships, calculate the co-occurrence frequency between entities. Taking device components and fault types as an example, count the number of times a specific fault type appears when a certain component appears within a certain time window (such as 1 day). When the co-occurrence frequency exceeds the set threshold (such as 0.3), determine that there is an association relationship. At the same time, based on the timestamp order in the fused multi-dimensional tensor data, preliminarily judge whether there may be a tendency of causal relationship;

[0061] Perform semantic association operations on various features in the fused multi-dimensional tensor data with the entities already constructed in the knowledge graph, assign semantic labels to each data feature, thereby forming a semantic space that fuses multi-source data and domain knowledge, and generating a semantic fusion feature dataset; Among them, various features in the fused multi-dimensional tensor data include micro-electromagnetic field features, weak gravitational field features, macro-geospatial features, and text description data features; It can be understood that in this semantic space, the data features not only retain the original numerical or attribute information, but also establish connections with the knowledge graph entities through semantic labels. Finally, a semantic fusion feature dataset is generated, and the semantic fusion feature dataset contains the feature information encoded from the original multi-source data and clear semantic meanings;

[0062] Specifically, B1. Use a method based on semantic similarity calculation to semantically associate various features in the fused multi-dimensional tensor data with the entities already constructed in the knowledge graph: Use a pre-trained word vector model specifically trained for the device domain (such as the Word2Vec model pre-trained with a device domain corpus); for each data feature, calculate its cosine similarity with the entity word vectors in the knowledge graph. For example, for a feature vector representing device temperature, calculate its cosine similarity with the entity word vector of "device overheating failure" in the knowledge graph; assign semantic labels according to the value of the semantic similarity. If the cosine similarity is greater than 0.8, assign a strong association semantic label. If it is between 0.5 and 0.8, assign a medium association semantic label. If it is less than 0.5, assign a weak association semantic label. B2. Integrate the data features and the knowledge graph entities according to the semantic labels to generate a semantic fusion feature dataset: For the data features with strong association semantic labels, directly merge them with the corresponding entities as an element in the semantic fusion feature dataset; for the data features with medium and weak associations, perform weighted merging according to their relationship weights with the entities (calculated through the co-occurrence frequency of the association relationships) to form a semantic fusion feature dataset.

[0063] The generator is responsible for generating possible fault embryo pattern samples, and the discriminator distinguishes whether the fault embryo pattern samples are real fault embryos; among them, when generating fault embryo pattern samples, the generator comprehensively uses the device operating states reflected by the information in each field in the generated semantic fusion feature dataset.

[0064] Through continuous adversarial training, the generator learns the potential fault embryo feature patterns in the semantic fusion feature dataset; it can be understood that these potential fault embryo feature patterns may exist before the faults are significantly manifested and are early signs of fault development.

[0065] Specifically, C1. The generator adopts a multi-layer perceptron structure: the number of nodes in the input layer is determined according to the dimension of the semantic fusion feature dataset. For example, if the dataset has 100 features, then there are 100 nodes in the input layer; the hidden layer is set to 3 layers. The number of nodes in the first layer is 200, and the ReLU activation function is used. The number of nodes in the second layer is 150, and the ReLU activation function is used. The number of nodes in the third layer is 100, and the tanh activation function is used; the number of nodes in the output layer is determined according to the coding dimension of the faulty embryo pattern sample. If the faulty embryo pattern sample is represented by a 50-dimensional vector, then there are 50 nodes in the output layer; C2. The discriminator adopts a probability-based discrimination method: for the input faulty embryo pattern sample, calculate the probability that it belongs to a real faulty embryo; input the faulty embryo pattern sample into the discriminator composed of multi-layer perceptrons. The number of nodes in the input layer of the discriminator is the same as the number of nodes in the output layer of the generator (such as 50); the hidden layer is set to 2 layers, and the number of nodes in each layer is 80, and the ReLU activation function is used; the number of nodes in the output layer is 1, and the sigmoid activation function is used. The output value represents the probability that the sample is a real faulty embryo. When the probability is greater than 0.5, it is judged as a real faulty embryo; at the same time, set the discrimination dimension of the severity of the faulty embryo pattern: judge the severity according to the numerical size of the relevant features (such as features related to device components) in the faulty embryo pattern sample. For example, if the feature value representing the temperature of the device component exceeds 50% of the normal range, it is considered that the severity of the faulty embryo pattern is relatively high; C3. The learning rate of the adversarial training is set to 0.001, and the learning rate is adjusted according to the scale of the semantic fusion feature dataset. If the scale of the semantic fusion feature dataset is small (such as less than 1000 samples), then the learning rate is increased to 0.005. If the scale of the semantic fusion feature dataset is large (such as more than 5000 samples), then the learning rate can be reduced to 0.0005; the batch size of the training is set to 32, and the batch size can be adjusted according to the hardware resources (such as GPU video memory). If the video memory is large, then the batch size is increased to 64. If the video memory is small, then it can be reduced to 16; the number of iterations is set to 1000 times. During the training process, the model is evaluated every 100 iterations. According to the loss function values of the generator and the discriminator, it is judged whether the model converges. If the change in the loss function value is less than 0.01 in three consecutive evaluations, it is considered that the model converges and the training is stopped in advance;

[0066] Trace the origin and development context of the faulty embryo along the relationship paths (causal relationships and association relationships) between entities in the knowledge graph; among them, the origin of the faulty embryo can be traced back to multiple aspects, including the origin of the equipment itself, the origin of the external environment, and the origin of human operation; for example, for the origin of the equipment itself, if through in-depth semantic association analysis, it is found that there are defects in the manufacturing process of the engine pistons produced in the current batch, it indicates that there are defects in the initial manufacturing of the equipment components, thus constituting a possible origin of the faulty embryo. On the other hand, if the data in the semantic space shows that the tooth surface of the gear set of the equipment is severely worn after a long period of high-load operation, it indicates that the long-term operation of the equipment has caused wear and aging of the equipment components, constituting a possible origin of the faulty embryo; for the origin of the external environment, when the equipment is in a harsh macro-geographical environment with high humidity and high salt spray in coastal areas, it accelerates the corrosion process of the metal components of the equipment, thereby triggering the origin of the fault. In addition, through semantic association analysis, it is revealed that when the grid voltage in the area where the equipment is operating is unstable, it will cause an impact on the electrical system of the equipment, thus becoming another external environment origin of the faulty embryo; for the origin of human operation, if the operation records in the semantic space show that the operator fails to start the equipment in accordance with the specified sequence, or makes incorrect equipment parameter configurations during the operation of the equipment, it will directly lead to the generation of the faulty embryo; the development context of the faulty embryo includes the initial evolution of the fault, the expansion of the influence range, and the changes at the equipment operation level; for example, for the initial evolution of the fault, if there are slight abnormalities in the vibration parameters and temperature of the equipment-related components in the semantic space, it indicates that the wear of the equipment components is in the initial stage, showing slight surface wear, and gradually intensifying to form a clear development context; for the expansion of the influence range, if the association relationships between components in the semantic space show that the faulty embryo spreads from a single component to other related components, such as the failure of the engine piston affecting the force on the connecting rod, and then affecting the operation of the crankshaft, the influence range of the fault is gradually expanding, forming a coherent development context; for the changes at the equipment operation level, if the fluctuations in the overall operation indicators of the equipment in the semantic space reflect that the faulty embryo affects multiple components, resulting in a decline in the operation performance of the equipment, such as a decrease in output power and an increase in energy consumption, it demonstrates the development context of the fault from local to overall and gradually expanding;

[0067] During root cause analysis, it specifically includes the following aspects: D1. If there are multiple possible relationship paths, select the relationship path for priority tracing based on the credibility of the relationship path: Determine the credibility of the relationship path by calculating the product of the strengths of the relationships between entities on the relationship path. If there are three entity relationships on a relationship path, and the strength values of the three entity relationships are 0.8, 0.7, and 0.6 respectively, then the credibility of this relationship path is 0.8 x 0.7 x 0.6 = 0.336; Arrange the relationship paths in descending order of credibility and select the path for priority tracing; D2. Quantitatively represent the origin of the fault: For the origin within the device itself, quantify the degree of device aging based on the device usage time and the device service life. Let the device service life be T and the usage time be t, then the device aging degree is expressed as t / T; For the origin in the external environment, conduct quantitative analysis based on the deviation degree of the current environment from the normal working environment range of the device. If the normal working environment range of the device is [h1, h2] and the current environment is h0, then the environmental impact degree is expressed as |h0 - (h1 + h2) / 2| / (h2 - h1); For the origin of human operation, conduct quantitative analysis based on the operation frequency and operation correctness. When the operation frequency is higher than 50% of the normal operation frequency, the operation frequency impact factor is 0.5. When the operation error rate is p, the operation correctness impact factor is 1 - p. The comprehensive impact degree of the origin of human operation is expressed as: the product of the operation frequency impact factor and the operation correctness impact factor; D3. Quantitatively describe the development context of the fault: For the evolution stage in the initial stage of the fault, conduct quantitative evaluation based on the change speed of the characteristics of the fault embryo pattern. Measure it by calculating the change rate of the characteristic values of the fault embryo pattern per unit time. If the change rate is larger, it indicates that the initial evolution process of the fault is more rapid; For the expansion of the influence range, use the number of device components or the number of device function modules involved as a quantitative indicator. If the initial fault only affects one device component, and as the fault continues to develop, its influence range expands to three components, then calculate (3 - 1) / 1 to obtain the expansion degree of the influence range, that is, it expands by 2 times; For the changes at the device operation level, conduct quantitative analysis based on the changes in device operation parameters (i.e., efficiency, power). If the initial power of the device is g1 and its power drops to g2 as the fault evolves, then the change degree at the device operation level is expressed as (g1 - g2) / g1;

[0068] In summary, after identifying the potential fault embryo pattern, it is necessary to further analyze the interaction between the fault embryo pattern and the entire device operation architecture in order to more accurately predict the occurrence of faults.

[0069] It should be further noted that in the specific implementation process, based on the identified fault embryo pattern and its root cause analysis results, the process of constructing the hidden order model is as follows:

[0070] Quantify each element in the device operation architecture and analyze the dynamic change relationships between elements using non-linear dynamics equations; for different types of elements, select corresponding quantification indicators: for device components, quantify them according to their performance indicators, for software management programs, quantify them according to indicators such as code complexity and execution efficiency, for operators, quantify them according to operation proficiency and operation frequency, and for the operating environment, quantify them according to environmental parameters; after completing the element quantification, select relevant non-linear dynamics equation types according to the nature and interaction relationships of the elements. When there are different relationships such as competition and cooperation, energy transfer, and information interaction between elements, different types of non-linear dynamics equations are respectively corresponding; for example, for the competition and collaborative work of different device components for resources, consider the deformation of the Lotka-Volterra equation. If it is an energy transfer or information interaction relationship, refer to the Hamiltonian equation or the dynamics equation related to information entropy.

[0071] During the construction process, add the information of each element in the device operation architecture to the parameter settings of the hidden order model (that is, when the fault embryo mode shows that the fault probability of any element in the device operation architecture is relatively high, then convert the information of this element into the coefficient of the non-linear dynamics equation), and combine the relevant quantification values given by the fault embryo mode and its traceability analysis results, and associate them with the hidden order model parameters.

[0072] Determine the normal state or potential fault state of the device operation architecture through the hidden order model: define the threshold for the normal state; compare and analyze the output value of the hidden order model with the defined threshold. If the output value of the hidden order model is within the normal state threshold range, that is, the output value is greater than or equal to the lower limit threshold of the normal state and less than or equal to the upper limit threshold of the normal state, then it is determined that the current state of the device operation architecture meets the standard of normal operation and is regarded as the normal state. At this time, the device can probably operate stably and reliably, and each performance indicator is within the expected reasonable range; if the output value of the hidden order model exceeds the normal state threshold range, that is, the output value is less than the lower limit threshold of the normal state or greater than the upper limit threshold of the normal state, it indicates that there may be potential problems in the device operation architecture and is regarded as the potential fault state. Once this situation occurs, further in-depth analysis is required.

[0073] Meanwhile, capture the hidden order structure within the operation architecture of the capture device that causes the transition from the normal state to the fault state: abstract the hidden order structure from the element relationships of the device operation architecture, specifically including the energy flow order structure, software interaction order structure, environmental action order structure, and human-machine operation order structure: for the energy transmission between device components, establish a mathematical model based on the energy flow in the circuit and the energy transfer in mechanical transmission, and then find the hidden order structure therein; for the interaction relationship between the software management program and device components, analyze the hidden order structure from the aspects of instruction flow and data flow; for the impact of the operating environment on device components, reveal the hidden order structure between the operating environment and device components by monitoring environmental parameters and analyzing their effects on each device component; for the interaction between the operator and device components, reveal the hidden interaction order structure between the operator and device components by analyzing the process in which the operator receives device status information and makes control inputs accordingly;

[0074] Specifically, after establishing the hidden order model that describes the dynamic transition behavior of the device operation architecture, this model needs further verification and optimization to ensure that it can accurately predict the occurrence of faults.

[0075] It should be further noted that in the specific implementation process, based on the hidden order model, use the association rule recognition algorithm to further find the association rules between the already identified fault embryo patterns and the semantic fusion feature data set; according to the obtained association rules and the current state of the device operation architecture, construct an inference prediction engine based on hidden order association, and the process of inference prediction is as follows:

[0076] Obtain the already identified fault embryo patterns and the semantic fusion feature data set: the fault embryo patterns are stored in vector form, each element in the fault embryo pattern vector represents the corresponding fault feature identifier and quantization value, the semantic fusion feature data set is stored in the form of a data table, each row represents a data sample, and each column represents a fusion feature;

[0077] Mining frequent item sets in the integrated fault embryo pattern vector and semantic fusion feature dataset through the Apriori algorithm, and setting the minimum support and confidence thresholds to identify the feature combinations that frequently and simultaneously appear before the fault occurs. These feature combinations further constitute and screen out the association rules for fault occurrence: Integrate the fault embryo pattern vector and the semantic fusion feature dataset to ensure that each element in the fault embryo pattern vector matches and corresponds to the corresponding fusion feature in the semantic fusion feature dataset; Convert the integrated dataset into the form of a transactional database processed by the Apriori algorithm, where each transaction represents a set containing multiple features, and these features come from the fault embryo pattern vector and the semantic fusion feature dataset; Set the minimum support threshold for screening frequent item sets, where the support represents the frequency of an item set appearing in all transactions; Set the minimum confidence threshold for screening association rules; Utilize the iterative process of the Apriori algorithm, starting from single items, and gradually generate frequent item sets containing more items, where the generated frequent item sets represent the feature combinations that often appear simultaneously before the fault occurs; In each iteration, generate candidate item sets based on the frequent item sets generated in the previous round, and calculate the support of each candidate item set; Retain the candidate item sets with support not lower than the minimum support threshold as new frequent item sets for generating more candidate item sets in the next round; Repeat the iterative process until no new frequent item sets can be generated; For each frequent item set, generate all possible association rules, and calculate the support and confidence of each association rule; Retain the association rules with support and confidence not lower than the set thresholds as the final association rules for fault occurrence; Utilize the association relationship between the feature combinations represented by the generated association rules and the fault occurrence;

[0078] Combine the obtained association rules with the state of the current device operation architecture to construct an inference prediction engine based on hidden order association; Utilize the inference prediction engine to infer and predict future fault occurrences according to the state of the current device operation architecture and the association rules: If it is monitored that the key monitoring parameters of the device operation architecture satisfy the association rules, the inference prediction engine predicts the probability, time, and type of fault occurrence; Among them, the key monitoring parameters are important indicators used to monitor and evaluate the device state in the device operation architecture, and are obtained in real time through the sensor network; According to the feature combinations and historical fault data in the association rules, respectively use machine learning algorithms, time series prediction models, and classification algorithms to comprehensively predict the probability, time, and type of fault occurrence;

[0079] Meanwhile, the Bayesian inference method is used to correct the prediction results to improve the accuracy and reliability of the prediction: the prior probability is obtained based on the statistics of historical fault data and classified; the likelihood function is determined based on the relationship between the currently monitored key monitoring parameters and the occurrence of faults, and the posterior probability is calculated according to the Bayesian formula, and the posterior probability is converted into the correction values of the probability of fault occurrence, time, and fault type.

[0080] It should be further noted that in the specific implementation process, the normal state of the device operation architecture is regarded as itself, and the fault state is regarded as non-self; the process of establishing an adaptive defense mechanism in combination with the inference prediction results is as follows:

[0081] Obtain the inference prediction results of the hidden order model and set its own tolerance threshold set; when the probability value of the detected fault occurrence deviates from the range of the set own tolerance threshold set, the adaptive defense mechanism is started to identify and process the fault state of the device operation architecture; according to the type and severity of the fault, an adaptive regulation strategy is generated; among them, the regulation strategy includes measures such as maintenance and replacement of device components, dynamic adjustment of device operation parameters, optimization of work processes, and reallocation of resources; for example, when it is predicted that a certain key component is about to fail, the operation mode of the device is automatically adjusted to reduce the load of the component, and at the same time, maintenance personnel are arranged to prepare for replacing the component in advance to ensure that the device can smoothly transition when the fault occurs and reduce the downtime and losses.

[0082] It should be understood that in various embodiments of the present application, the magnitudes of the serial numbers of the above processes do not mean the order of execution, and the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0083] It should be understood that determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information.

[0084] The above is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0085] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. Fault identification method based on an intelligent model, characterized in that: S1. Use multi-modal high-precision sensing elements to construct a multi-dimensional field perception structure for collecting comprehensive state information of the device and its environment; for the acquired field data, formulate corresponding fusion coding strategies to obtain the fused multi-dimensional tensor data; S2. Use natural language processing and knowledge graph technology to construct a semantic space and generate a semantic fusion feature data set; In the constructed semantic space, use generative adversarial network technology to mine fault embryo patterns; if a fault embryo pattern is identified, use the entity relationship network in the semantic space for traceability analysis; S3. Based on the identified fault embryo patterns and their traceability analysis results, construct a hidden order model; S4. Based on the hidden order model, use the association rule recognition algorithm to further find the association rules between the already identified fault embryo patterns and the semantic fusion feature data set; based on the analysis results of the association rules and the state of the current device operation architecture, construct an inference prediction engine based on hidden order association and perform inference prediction; S5. Regard the normal state of the device operation architecture as itself and the fault state as non-self; Combine the inference prediction results to establish an adaptive defense mechanism; Among them, the process of S2 includes: Extract the entities related to the device and the relationships between entities from the fused multi-dimensional tensor data to construct a fault knowledge graph; Perform semantic association operations on various features in the fused multi-dimensional tensor data with the entities already constructed in the knowledge graph, assign semantic labels to each data feature, thereby forming a semantic space that fuses multi-source data and domain knowledge, and generating a semantic fusion feature data set; among them, various features in the fused multi-dimensional tensor data include microscopic electromagnetic field features, weak gravitational field features, macroscopic geographical space features, and text description data features; The generator is responsible for generating possible fault embryo pattern samples, and the discriminator distinguishes whether the fault embryo pattern samples are real fault embryos; among them, when the generator generates fault embryo pattern samples, it comprehensively uses the device operation states reflected by the information of each field in the generated semantic fusion feature data set; Through continuous adversarial training, use the generator to learn the potential fault embryo feature patterns in the semantic fusion feature data set; Follow the relationship paths between entities in the knowledge graph to trace the origin and development context of the fault embryo; among them, the origin of the fault embryo is traced back to multiple aspects, including the origin of the device itself, the origin of the external environment, and the origin of human operations; the development context of the fault embryo includes the initial evolution of the fault, the expansion of the influence range, and the changes at the device operation level; Through calculation and analysis, obtain the relevant quantitative values of the traceability analysis; Among them, the process of S3 includes: Quantify each element in the device operation architecture and use non-linear dynamics equations to analyze the dynamic change relationships between the elements; During the construction process, add the information of each element in the device operation architecture to the parameter settings of the hidden order model, and combine the relevant quantitative values given by the fault embryo pattern and its traceability analysis results, and associate them with the hidden order model parameters; Determine the normal state or potential fault state of the device operation architecture through the hidden order model: Define the threshold for the normal state; Compare and analyze the output value of the hidden order model with the defined threshold. If the output value of the hidden order model is within the normal state threshold range, that is, the output value is greater than or equal to the lower threshold of the normal state and less than or equal to the upper threshold of the normal state, it is determined that the current state of the device operation architecture meets the standard of normal operation and is regarded as the normal state; If the output value of the hidden order model exceeds the normal state threshold range, that is, the output value is less than the lower threshold of the normal state or greater than the upper threshold of the normal state, it indicates that there is a problem with the device operation architecture and is regarded as the potential fault state; At the same time, capture the hidden order structure inside the device operation architecture that causes the transition from the normal state to the fault state.

2. The fault identification method based on an intelligent model according to claim 1, wherein: In S1, the multi-dimensional field specifically includes the microscopic electromagnetic field, the weak gravitational field, and the macroscopic geographical environment and spatial position; The multi-modal high-precision sensing elements specifically include near-field optical sensors, gravitational wave sensors, and satellite positioning and geographic information systems.

3. The fault identification method based on an intelligent model according to claim 1, wherein: The process of S1 includes: Use near-field optical sensors to capture changes in the distribution of the microscopic electromagnetic field on the device surface; Use gravitational wave sensors to monitor the weak gravitational field fluctuations caused by changes in the mass distribution of large device clusters; Combine satellite positioning and geographic information systems to obtain the macroscopic geographical environment and spatial position information where the device is located; Obtain the microscopic electromagnetic field data on the device surface, the weak gravitational field fluctuation data of large device clusters, and the macroscopic geographical environment and spatial position data where the device is located; Collect various text description data during the operation of the device; Map the microscopic electromagnetic field data to the complex space, and use the real and imaginary parts of the complex number to represent the coupling relationship between different physical quantities; For the weak gravitational field fluctuation data, use the wavelet transform algorithm for encoding processing to form a data vector with high-dimensional features; For the macroscopic geographical space data, transform the geographical information into a topological feature vector through topological mapping; For the text description data, use natural language processing technology to transform it into a word vector representation; Use the tensor product operation to fuse the feature vectors obtained by these different encoding methods into a unified multi-dimensional tensor data.

4. The fault identification method based on an intelligent model according to claim 1, characterized in that: In S3, the hidden order model regards the entire device operation architecture as a dynamic composite system, in which there are complex and hidden interaction relationships among the elements; Among them, the device operation architecture refers to a comprehensive device operation system composed of device components, software management programs, operators, and the operating environment; The elements of the device operation architecture cover device components, software management programs, operators, and the operating environment.

5. The fault identification method based on an intelligent model according to claim 1, characterized in that: S4 includes: Obtain the identified fault embryo pattern and semantic fusion feature data set: The fault embryo pattern is stored in vector form, and each element in the fault embryo pattern vector represents the corresponding fault feature identifier and quantization value. The semantic fusion feature data set is stored in the form of a data table, with each row representing a data sample and each column representing a fusion feature; Mining frequent item sets in the integrated fault embryo pattern vectors and semantic fusion feature datasets through the Apriori algorithm, and setting minimum support and confidence thresholds to identify the feature combinations that frequently and simultaneously appear before a fault occurs. These feature combinations further constitute and screen out the association rules for fault occurrence; Combining the obtained association rules with the state of the current device operation architecture to construct an inference prediction engine based on hidden order association; using the inference prediction engine, according to the state of the current device operation architecture and the association rules, to infer and predict future fault occurrence: if it is monitored that the key monitoring parameters of the device operation architecture satisfy the association rules, the inference prediction engine predicts the probability, time, and type of fault occurrence; among them, the key monitoring parameters are important indicators for monitoring and evaluating the device state in the device operation architecture and are obtained in real time through the sensor network; At the same time, using the Bayesian inference method to correct the prediction results.

6. The fault identification method based on an intelligent model according to claim 5, wherein: The mining of frequent item sets in the integrated fault embryo pattern vectors and semantic fusion feature datasets through the Apriori algorithm, and setting minimum support and confidence thresholds to identify the feature combinations that frequently and simultaneously appear before a fault occurs, includes: Integrating the fault embryo pattern vectors and semantic fusion feature datasets to ensure that each element in the fault embryo pattern vectors matches and corresponds to the corresponding fusion features in the semantic fusion feature datasets; converting the integrated dataset into the form of a transactional database processed by the Apriori algorithm, where each transaction represents a set containing multiple features, and these features come from the fault embryo pattern vectors and semantic fusion feature datasets; setting a minimum support threshold for screening frequent item sets, where the support represents the frequency of an item set appearing in all transactions; setting a minimum confidence threshold for screening association rules; using the iterative process of the Apriori algorithm, starting from single items, gradually generating frequent item sets containing more items, where the generated frequent item sets represent the feature combinations that often appear simultaneously before a fault occurs; in each iteration, generating candidate item sets based on the frequent item sets generated in the previous round and calculating the support of each candidate item set; retaining the candidate item sets with support not lower than the minimum support threshold as new frequent item sets for generating more candidate item sets in the next round; repeating the iterative process until no new frequent item sets can be generated; for each frequent item set, generating all possible association rules and calculating the support and confidence of each association rule; retaining the association rules with support and confidence not lower than the set thresholds as the final association rules for fault occurrence; using the association relationship between the feature combinations represented by the generated association rules and fault occurrence.

7. The fault identification method based on an intelligent model according to claim 1, wherein: The S5 includes: Obtaining the inference prediction results of the hidden order model and setting its own tolerance threshold set; when the detected probability value of fault occurrence deviates from the range of the set own tolerance threshold set, then starting the adaptive defense mechanism to identify and process the fault state of the device operation architecture; generating an adaptive regulation strategy according to the type and severity of the fault.

Images