Equipment maintenance fault intelligent analysis method and system based on Internet of Things

Abstract

The invention relates to the technical field of the Internet of Things and artificial intelligence, and discloses an equipment maintenance fault intelligent analysis method and system based on the Internet of Things, and the method comprises the steps: representing an entity and a relation as a time function, and constructing a time sequence knowledge graph; monitoring vector change in the knowledge embedding space, detecting a concept drift phenomenon and mining an equipment performance evolution mode; constructing an equipment state transition model by using a recurrent neural network, predicting the future state of the equipment and evaluating the fault risk; collecting feedback processing knowledge conflicts and updating the model; generating multi-time-scale fault risk early warning and maintenance suggestions; according to the method, the knowledge acquisition efficiency is improved, dynamic knowledge support of the full life cycle of the equipment is provided, early recognition of gradient faults is realized, the method has adaptive knowledge updating capability, multi-dimensional fault analysis is provided, maintenance resource configuration is optimized, and the equipment management efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical fields of the Internet of Things and artificial intelligence. More specifically, it relates to an intelligent analysis method and system for equipment maintenance faults based on the Internet of Things. Background Art

[0002] With the rapid development of industry and intelligent manufacturing, Internet of Things technology has been widely applied in the field of industrial equipment monitoring and maintenance. The stable operation of industrial equipment is a key factor in ensuring production continuity and enterprise benefits. Therefore, an efficient equipment maintenance and fault warning system is of great significance. Traditional equipment maintenance mainly relies on regular inspections and passive response-based maintenance strategies, that is, maintenance is carried out only after the equipment fails. This method not only causes production interruptions but also may result in significant economic losses.

[0003] In recent years, data-driven predictive maintenance technologies have gradually emerged. These technologies collect equipment operation data through sensors and use statistical analysis or machine learning methods to evaluate the health status of the equipment and predict possible fault risks. However, these methods mainly focus on data pattern recognition and lack the in-depth utilization of equipment maintenance knowledge and the ability to model long-term evolution trends. Especially for industrial equipment with a long service life and complex fault modes, it is difficult to accurately grasp the complete evolution law of equipment status over time simply by relying on short-term data analysis.

[0004] As an effective tool for knowledge representation and management, knowledge graphs have demonstrated powerful knowledge organization and reasoning capabilities in multiple fields. Traditional knowledge graph technologies mainly focus on the representation of static knowledge and construct domain knowledge networks through entities and relationships. However, in the field of equipment maintenance, knowledge such as the performance characteristics, fault modes, and maintenance methods of equipment evolves continuously with equipment aging, environmental changes, and technological updates. This dynamic change characteristic is difficult to effectively express using static knowledge graphs, resulting in the disconnection between the information in the knowledge base and actual applications.

[0005] Therefore, there is a need for an integrated solution that can effectively integrate Internet of Things data collection, knowledge graph technology, time series feature modeling, and intelligent fault analysis to improve equipment maintenance efficiency, reduce maintenance costs, and extend the service life of equipment. Summary of the Invention

[0006] The present invention provides an intelligent analysis method and system for equipment maintenance faults based on the Internet of Things, which solves the technical problem in related technologies of lacking the in-depth utilization of equipment maintenance knowledge and the ability to model long-term evolution trends.

[0007] The present invention provides an intelligent analysis method for equipment maintenance faults based on the Internet of Things, including the following steps: Automatically extract entities and relationships from equipment maintenance-related documents to construct an initial knowledge graph; Based on the constructed initial knowledge graph, represent entities and relationships as functions of time to construct a temporal knowledge graph; Utilize the constructed temporal knowledge graph to detect the concept drift phenomenon and mine the equipment performance evolution pattern by monitoring the changes of entity and relationship vectors in the knowledge embedding space over time; Based on the constructed temporal knowledge graph and current observed data, use a recurrent neural network to construct an equipment state transition model to predict the future state of the equipment and evaluate the fault risk; Collect the feedback of maintenance personnel on the system prediction results, handle knowledge conflicts, and update the temporal knowledge graph and prediction model; Combine the equipment real-time state data and the updated temporal knowledge graph to generate fault risk warnings and maintenance suggestions at multiple time scales.

[0008] In a preferred embodiment, the step of automatically extracting entities and relationships from equipment maintenance-related documents includes: Preprocess maintenance manuals, historical fault records, maintenance reports, and unstructured texts; Adopt a named entity recognition method combining conditional random fields and bidirectional long short-term memory networks to identify and extract three core entities: equipment components, fault phenomena, and maintenance methods; Use a relationship extraction algorithm with remote supervision and attention mechanism to identify the semantic relationships between entities; Integrate the extracted entities and relationships into a unified knowledge framework and perform consistency checks.

[0009] In a preferred embodiment, the step of representing entities and relationships as functions of time includes: Introduce time attributes to entities and relationships in the knowledge graph, and expand the traditional triple knowledge representation to a quadruple; Adopt a tensor decomposition method to combine entity vectors, relationship vectors, and time vectors to obtain a time-aware embedding representation; Define temporal relationship reasoning rules to support reasoning with time constraints on the temporal knowledge graph.

[0010] In a preferred embodiment, the step of detecting the concept drift phenomenon and mining the equipment performance evolution pattern includes: Calculate the semantic displacement of an entity at two time points. When the semantic displacement exceeds a preset threshold, it is determined that the calculated entity has concept drift; Use a recurrent variational autoencoder to automatically extract the periodic and trend patterns of knowledge evolution from the temporal knowledge graph; Construct a knowledge evolution visualization module to present the detected concept drift and evolution patterns in an intuitive way.

[0011] In a preferred embodiment, the steps of predicting the future state of the device and evaluating the failure risk include: Based on the temporal knowledge graph, reconstruct the knowledge state of the device at specific historical time points; Use a recurrent neural network to construct a device state transition model. The input of the model is the device state vector and relevant knowledge graph information at the current time point, and the output is the state prediction at future time points; Calculate the similarity between the predicted state and known failure modes, and generate the probability distribution of different types of failures.

[0012] In a preferred embodiment, the steps of handling knowledge conflicts and updating the temporal knowledge graph and prediction model include: Calculate the credibility of conflicting knowledge based on the number of supporting evidences and the reliability of sources; Demote or eliminate knowledge with low credibility; Add new knowledge with high credibility; For knowledge with timeliness differences, update the time attribute instead of directly replacing it; Adopt an incremental learning method to update the parameters of the state transition model and the failure risk assessment model.

[0013] In a preferred embodiment, the steps of generating failure risk warnings and maintenance suggestions at multiple time scales include: Based on the state transition model, generate device state predictions at different time scales; Determine the warning level according to the matching degree between the predicted state and the failure mode and the time urgency; Based on the failure risk warning results and the maintenance knowledge in the temporal knowledge graph, automatically generate targeted maintenance suggestions; Based on the device's historical maintenance records, estimate the time and resources required for maintenance to assist in formulating the maintenance plan.

[0014] In a preferred embodiment, the time attribute includes time points, time intervals, or periodic characteristics. A time point represents a specific moment when the knowledge holds, a time interval represents the start and end times when the knowledge is valid, and a periodic characteristic represents a seasonal failure mode.

[0015] In a preferred embodiment, the cyclic variational autoencoder includes an encoder and a decoder. The encoder encodes the sequence of temporal entity representations into the distribution parameters of latent variables, and the decoder reconstructs the sequence based on the sampled latent variables and historical information.

[0016] In a preferred embodiment, an IoT-based intelligent analysis system for device maintenance faults is used to execute the IoT-based intelligent analysis method for device maintenance faults, including: The multi-source text knowledge extraction and graph construction module is used to automatically extract entities and relationships from device maintenance-related documents and construct an initial knowledge graph; The temporal knowledge graph construction module, based on the constructed initial knowledge graph, represents the entities and relationships therein as time functions and constructs a temporal knowledge graph; The concept drift detection module is used to utilize the constructed temporal knowledge graph, detect the concept drift phenomenon and mine the device performance evolution pattern by monitoring the changes of entity and relationship vectors in the knowledge embedding space over time; The device state prediction module, based on the constructed temporal knowledge graph and current observation data, uses a recurrent neural network to construct a device state transition model, predict the future state of the device and evaluate the failure risk; The knowledge update module is used to collect the feedback of maintenance personnel on the system prediction results, handle knowledge conflicts and update the temporal knowledge graph and prediction model; The fault warning module is used to generate fault risk warnings and maintenance suggestions at multiple time scales by combining the real-time state data of the device and the updated temporal knowledge graph.

[0017] The beneficial effects of the present invention are as follows: The knowledge acquisition efficiency is significantly improved: Through weak supervision learning and remote supervision techniques, the system can automatically extract device fault-related entities and relationships from a large amount of unstructured text. Compared with traditional manual construction methods, the knowledge acquisition efficiency is increased by several times, the accuracy is significantly improved, and the labor cost of knowledge graph construction and maintenance is greatly reduced.

[0018] Dynamic knowledge support for the entire life cycle of the device: Through temporal knowledge graph technology, the system can characterize the evolution of knowledge over time, capture the state characteristics and fault modes of the device at different usage stages, provide targeted knowledge support for the maintenance of the entire life cycle of the device, and avoid knowledge timeliness problems.

[0019] Early identification of gradual faults: Combining the evolution pattern mining technology of the cyclic variational autoencoder, the system can identify the subtle change trends of device performance, discover the early signs of gradual faults in advance. Compared with traditional methods, the fault warning lead time is significantly increased, providing a sufficient time window for preventive maintenance.

[0020] Adaptive knowledge update: Through the maintenance feedback collection and knowledge conflict handling mechanism, the system realizes the self-evolution of knowledge, can continuously learn from actual maintenance cases and optimize the knowledge structure, so that the system continuously improves its accuracy and practicality during the usage process.

[0021] Multi-dimensional fault analysis ability: The system can not only analyze the current fault, but also trace the development process of the fault and predict the future evolution trend through temporal knowledge reasoning, providing a comprehensive reference for maintenance decision-making and improving the accuracy and comprehensiveness of fault diagnosis.

[0022] Optimized allocation of maintenance resources: Based on the fault warning and maintenance advice generation functions on multiple time scales, the system can support the advance planning and optimized allocation of maintenance resources. Compared with passive reactive maintenance, the average maintenance cost is significantly reduced, the equipment availability rate is significantly improved, and the equipment management efficiency of the enterprise is greatly enhanced. Brief Description of the Drawings

[0023] Figure 1 is a flowchart of the intelligent analysis method for equipment maintenance faults based on the Internet of Things according to the present invention; Figure 2 is a line chart of the improvement of knowledge acquisition efficiency according to the present invention; Figure 3 is a bar chart of the comparison of the accuracy of concept drift detection of the temporal knowledge graph according to the present invention; Figure 4 is a bar chart of the performance of the fault prediction accuracy on different equipment types according to the present invention; Figure 5 is a radar chart of the fault risk assessment performance according to the present invention; Figure 6 is a bubble chart of the optimization effect of maintenance resources according to the present invention. Detailed Embodiments

[0024] Now, the subject matter described herein will be discussed with reference to exemplary embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0025] In at least one embodiment of the present invention, an intelligent analysis method for equipment maintenance faults based on the Internet of Things is disclosed, as Figure 1 shown, including the following steps: Step 1: Automatically extract entities and relationships from equipment maintenance-related documents to construct an initial knowledge graph; Specifically, it includes the following sub-steps: Step 1.1: Text preprocessing; Preprocess unstructured texts such as maintenance manuals, historical fault records, and maintenance reports, including text tokenization, stop word removal, part-of-speech tagging, and syntactic analysis.

[0026] During the preprocessing process, a specific dictionary is constructed for professional terms in the field of equipment maintenance to improve the accuracy of term recognition.

[0027] Step 1.2, entity recognition and extraction; A named entity recognition method combining Conditional Random Field (CRF) and Bidirectional Long Short-Term Memory Network (BiLSTM) is used to identify and extract three types of core entities: equipment components, fault phenomena, and maintenance methods from the preprocessed text. The entity recognition model is initially trained with a small number of manually annotated seed entities, and then the model performance is further optimized by combining remote supervision technology and using a larger scale of unannotated text.

[0028] In the embodiments of the present application, the BiLSTM-CRF model calculates the conditional probability of the label sequence, where there is a transition score from one label to another, and each position label output by the network has an emission score.

[0029] This model can learn the context features of words and the transition constraints between labels, improving the accuracy of identifying equipment professional terms.

[0030] When processing complex equipment maintenance documents, domain pre-trained language models such as BERT or its variants can be introduced to further improve the accuracy of entity recognition. The pre-trained model can be fine-tuned for domain adaptation through equipment maintenance manuals and technical documents to better understand the professional terms and expressions in the field of equipment maintenance.

[0031] Step 1.3, relation extraction; Based on the extracted entities, a relation extraction algorithm using remote supervision and attention mechanism is used to identify the semantic relations between entities.

[0032] First, a small number of predefined relation templates are used to extract sentences containing known entity pairs from the text as training samples, and then a convolutional neural network with attention mechanism is used to automatically learn the relation representation between entities.

[0033] Finally, relation types such as "components have fault phenomena", "fault phenomena require maintenance methods", and "components contain sub-components" are extracted.

[0034] Step 1.4, knowledge graph integration and consistency check; Integrate the extracted entities and relations into a unified knowledge framework to form an initial knowledge graph.

[0035] Adopt rule-based and statistical consistency check methods to identify and process knowledge conflicts, including entity disambiguation, relation redundancy elimination, and logical consistency verification.

[0036] Finally, the integrated knowledge data is stored in the graph database to form a graph structure that can be queried and reasoned about.

[0037] Step 2: Based on the constructed initial knowledge graph, represent entities and relationships as time functions to construct a temporal knowledge graph. Specifically, it includes the following sub-steps: Step 2.1: Temporal attribute modeling. Introduce temporal attributes for entities and relationships in the knowledge graph, and expand the traditional triple knowledge representation <entity1, relationship, entity2> to a quadruple <entity1, relationship, entity2, temporal attribute>.

[0038] The temporal attribute can be a time point (representing a specific moment when the knowledge holds) or a time interval (representing the start and end times when the knowledge is valid), and can also include periodic characteristics (such as seasonal failure patterns).

[0039] In some embodiments, the temporal attribute can be further divided into creation time, update time, start time of validity period, and end time of validity period to support more refined temporal knowledge management.

[0040] For example, for some seasonal failure patterns, their historical occurrence times, expected validity periods, and periodic characteristic parameters can be recorded.

[0041] Step 2.2: Time-aware embedding representation. Represent entities and relationships in the knowledge graph as time functions and instead of static vectors.

[0042] Specifically, a tensor decomposition method is used to combine the entity vector and the relationship vector with the time vector to obtain a time-aware embedding representation: ; ; where, represents the time-aware embedding representation of the entity at time which reflects the semantic features of the entity changing over time; represents the time-aware embedding representation of the relationship at time which reflects the semantic features of the relationship changing over time; is the basic embedding vector of the entity, representing the semantic representation of the entity in the initial state; is the basic embedding vector of the relationship, representing the semantic representation of the relationship in the initial state; is a time parameter, which can be a timestamp or an encoded representation of a time interval; is the time mapping matrix of the entity, which is used to map the time parameter to the entity semantic space and control how time affects the semantic changes of the entity; is the time mapping matrix of the relationship, which is used to map the time parameter to the relationship semantic space and control how time affects the semantic changes of the relationship; is the entity time influence intensity parameter, which is used to adjust the degree of influence of time on the semantic changes of the entity; is the relationship time influence intensity parameter, which is used to adjust the degree of influence of time on the semantic changes of the relationship.

[0043] This representation method enables the same entity or relationship to have different semantic representations at different time points, thereby capturing the evolution of concepts over time.

[0044] The system can also adopt a non-linear mapping function to replace the above linear combination method, and construct a more complex time mapping relationship through a multi-layer perceptron network: ; Among them, represents the time-aware entity representation obtained by using a multi-layer perceptron network; is the basic embedding vector of the entity, representing the static semantic features of the entity in the initial state; is the time parameter, which can be a timestamp or an encoded representation of a specific time point; represents a multi-layer perceptron network, which consists of multiple layers of neural networks and can capture the non-linear interaction relationship between the entity vector and time; represents the entity vector and the time parameter as the output vector obtained after feeding into the multi-layer perceptron, representing the non-linear influence of time on the entity semantics. This non-linear mapping method can express more complex time evolution patterns than simple linear combinations, and is particularly suitable for modeling the non-linear changes of device component performance over time.

[0045] Step 2.3, definition of temporal relationship reasoning rules; Define a set of temporal relationship reasoning rules to support reasoning about time constraints on a temporal knowledge graph.

[0046] These rules include time point inclusion relationships, time interval overlap relationships, temporal constraints of causal relationships, etc.

[0047] For example, if relationship R1 holds at time t1, relationship R2 holds at time t2, and t1 precedes t2, then there may be a causal relationship where R1 causes R2.

[0048] Step 3: Using the constructed temporal knowledge graph, detect the concept drift phenomenon and mine the device performance evolution pattern by monitoring the changes of entity and relationship vectors in the knowledge embedding space over time; Specifically, it includes the following sub-steps: Step 3.1: Concept drift detection; By monitoring the changes of entity and relationship vectors in the knowledge embedding space over time, identify the drifts of concept definitions and relationship strengths.

[0049] For an entity , calculate its semantic displacement between two time points and : ; where represents the semantic displacement of entity between the two time points; is the basic embedding vector of the entity; represents the first time point; represents the second time point; represents the vector representation of entity at time point ; represents the vector representation of entity at time point ; represents the Euclidean distance between the two vectors (other distance metrics can also be used) to quantify the magnitude of the semantic change of the entity.

[0050] When the semantic displacement exceeds the preset threshold, it is determined that the entity has a concept drift. Similarly, the semantic displacement of the relationship can be calculated to detect the change of the relationship strength.

[0051] Through cluster analysis, classify the detected concept drifts into two modes: gradual (slow continuous change) and abrupt (rapid discrete change).

[0052] Step 3.2: Evolution pattern mining; Use the Recurrent Variational Autoencoder (RVAE) to automatically extract the periodic and trend patterns of knowledge evolution from the temporal knowledge graph. RVAE combines the generative ability of the variational autoencoder and the sequence modeling ability of the recurrent neural network, and can effectively capture the temporal dependence relationships in the sequence data.

[0053] The encoder part of the RVAE model encodes the time-series entity representation sequence into the distribution parameters of latent variables, and the decoder part reconstructs the sequence based on the sampled latent variables and historical information. The model is trained by minimizing a combined loss function of reconstruction error and KL divergence.

[0054] By analyzing the learned latent space structure, the system can identify the evolution laws of device performance parameters, such as patterns like linear degradation trends, periodic fluctuations, and stage changes, providing a basis for fault prediction.

[0055] In addition, this application also provides another implementation method. A time-series model with self-attention mechanism can be used to replace the RVAE to directly extract long-term and short-term dependencies from time-series data. This method calculates the correlations between different time points in the sequence through the self-attention layer, without explicit sequence modeling, and is suitable for capturing dependencies between distant time points.

[0056] Step 3.3, Knowledge evolution visualization; Construct a knowledge evolution visualization module to present the detected concept drift and evolution patterns in an intuitive way.

[0057] The visualization content includes entity semantic drift trajectory diagrams, relationship strength change curves, evolution pattern clustering views, etc., helping maintenance personnel understand the change laws of device knowledge over time.

[0058] Step 4, Based on the constructed time-series knowledge graph and current observation data, use a recurrent neural network to construct a device state transition model to predict the future state of the device and evaluate the fault risk; Specifically, it includes the following sub-steps: Step 4.1, Historical state reconstruction; Based on the time-series knowledge graph, realize the reconstruction of the knowledge state of the device at specific historical time points. Given time point t, the system retrieves the corresponding entity and relationship representations at this time point and , and reconstructs the knowledge graph snapshot at this time point to support the query and analysis of historical fault cases and maintenance records.

[0059] A knowledge graph snapshot refers to the state of the knowledge graph at a specific time point , which contains all entity information at this time point and the relationship information between them; The reconstruction process refers to the process in which the system extracts and restores the complete knowledge structure at this time point from the time-series knowledge graph according to the time parameter .

[0060] Step 4.2, State transition model construction; Construct a device state transition model using a Recurrent Neural Network (RNN) to capture the evolution law of the device state over time. The model input is the device state vector at the current time point t and the relevant knowledge graph information, and the output is the state prediction at the future time point.

[0061] According to an embodiment of the present application, the device state vector includes key parameters collected from Internet of Things sensors, such as temperature, vibration, pressure, etc., and high-level features derived from these parameters; the knowledge graph information includes the time-aware representation of entities and relationships related to the current device.

[0062] In some embodiments, a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU) can be used to replace the basic RNN to better handle long-term dependencies. The LSTM model can effectively learn long-term dependencies in long sequences by introducing memory cells and multiple gating mechanisms, and is particularly suitable for modeling the long-term operating state of devices.

[0063] In addition, for prediction tasks with different time scales, a multi-scale state transition model can be constructed. For example, for short-term prediction, fine-grained time steps and more sensor features are adopted; for long-term prediction, coarse-grained time steps and more historical trend features are adopted.

[0064] Step 4.3, Fault risk assessment; Based on the prediction results of the state transition model and the fault mode information in the temporal knowledge graph, evaluate the risk of the device failing in the future. Calculate the similarity between the predicted state and the known fault modes, and generate the probability distribution of different types of faults.

[0065] The similarity calculation method adopted in this application is based on a distance metric function, and various distance metric methods such as Euclidean distance, cosine similarity, or Mahalanobis distance can be selected.

[0066] The system calculates the distance between the predicted state and various fault modes and converts it into a probability distribution through the softmax function to realize the risk quantification of multiple possible fault types.

[0067] The system can also combine the historical fault occurrence frequency of the device as prior knowledge and use the Bayesian framework to calculate the posterior fault probability to improve the accuracy of risk assessment.

[0068] In addition, the system can also generate a confidence interval for the fault risk, provide a measure of the uncertainty of the risk assessment, and help maintenance personnel more comprehensively understand the reliability of the prediction results.

[0069] Step 5: Collect the feedback from maintenance personnel on the system prediction results, handle knowledge conflicts, and update the temporal knowledge graph and prediction model; Specifically, it includes the following sub-steps: Step 5.1: Maintenance feedback collection; Design a maintenance feedback collection interface to collect the evaluation of maintenance personnel on the system prediction results and suggestions, as well as new information discovered during the actual maintenance process.

[0070] The feedback content includes the accuracy score of fault diagnosis, the effectiveness evaluation of maintenance methods, newly discovered fault phenomena and causes, etc.

[0071] Step 5.2: Knowledge conflict detection and handling; Compare the maintenance feedback with the information in the existing knowledge graph to detect potential knowledge conflicts.

[0072] When the new feedback conflicts with the existing knowledge, the system processes it through the following strategies: Calculate the credibility of the conflicting knowledge based on the number of supporting evidences and the reliability of the sources; Demote or eliminate the knowledge with low credibility; Add the new knowledge with high credibility; For the knowledge with time - effectiveness differences, update its time attribute instead of directly replacing it.

[0073] The system can adopt fuzzy logic or Bayesian network methods to handle knowledge uncertainty.

[0074] By introducing confidence levels or probability values, the system can represent and reason about uncertain knowledge, rather than simply making binary true - false judgments. For example, the association relationship between a certain fault and a certain component can be represented as a relationship with a confidence level of 0.8, and this confidence level can be dynamically adjusted as more evidences accumulate.

[0075] Step 5.3: Model parameter update; Based on the accumulated maintenance feedback data, regularly update the parameters of the state transition model and the fault risk assessment model.

[0076] Adopt an incremental learning method to integrate new knowledge while retaining the original knowledge, avoiding the problem of catastrophic forgetting.

[0077] Step 6: Combine the real - time status data of the equipment and the updated temporal knowledge graph to generate fault risk warnings and maintenance suggestions at multiple time scales; Specifically, it includes the following sub - steps: Step 6.1: Scale - based status prediction; Based on the state transition model, generate equipment status predictions at different time scales (short - term, medium - term, long - term).

[0078] Short-term prediction focuses on the acute failure risk in the next few hours to days; Medium-term prediction focuses on the failure development trend in the next few weeks; Long-term prediction focuses on the performance degradation law of the equipment on a monthly to quarterly basis.

[0079] Step 6.2, determination of the warning level; Determine the warning level according to the matching degree between the prediction status and the failure mode and the time urgency.

[0080] The warning levels are divided into four levels from low to high: normal, attention, warning, and danger, and each level corresponds to different handling strategies and response time limits.

[0081] Step 6.3, generation of maintenance suggestions; Based on the failure risk warning result and the maintenance knowledge in the time series knowledge graph, automatically generate targeted maintenance suggestions.

[0082] The content of the maintenance suggestions includes possible failure causes, components to be checked, recommended maintenance methods, and required tools and materials.

[0083] The system will also estimate the time and resources required for maintenance based on the equipment's historical maintenance records to assist in formulating the maintenance plan.

[0084] The system can combine factors such as the importance of the equipment, spare part supply situation, and availability of maintenance personnel to generate multiple maintenance plans and sort them, providing multiple options for decision-makers.

[0085] In addition, the system can also recommend the best maintenance time window according to the equipment's usage plan to minimize the impact of maintenance activities on production.

[0086] For complex equipment systems, the system can also provide collaborative maintenance suggestions, combining the maintenance tasks of multiple related equipment for execution, reducing the number of maintenance downtimes, and improving the overall maintenance efficiency.

[0087] Application example of this embodiment: To better illustrate the actual application effect of the device maintenance fault intelligent analysis method based on the Internet of Things provided by this application, the following will be described in detail in combination with the actual application case of a cold rolling mill unit in a steel plant.

[0088] Application scenario: A steel plant has multiple cold rolling mill units. These devices have complex structures, including multiple subsystems such as hydraulic systems, transmission systems, and electrical control systems, and there are complex mutual influence relationships among the systems.

[0089] These devices operate in a harsh environment, work at high load all year round, and equipment failures occur frequently and the maintenance cost is high.

[0090] The traditional regular maintenance method cannot cope with sudden failures caused by equipment aging and performance degradation, while reactive maintenance often leads to long-term downtime of the production line, resulting in huge economic losses.

[0091] The factory decided to adopt the method provided in this application to build an intelligent analysis system for equipment maintenance faults based on the Internet of Things, realizing intelligent maintenance of the whole life cycle of the cold rolling mill unit.

[0092] The system collects equipment operation data in real time through Internet of Things sensors, and constructs a time series knowledge graph in combination with historical maintenance records to achieve early fault warning and predictive maintenance.

[0093] Implementation process example: Multi-source text knowledge extraction and graph construction: The system first collected about 8,500 text documents such as maintenance manuals, fault records, and maintenance reports of the cold rolling mill unit of this factory in the past five years.

[0094] In the text preprocessing stage, the system performs word segmentation, stop word removal, and syntactic analysis on these documents, and combines a special dictionary for cold rolling equipment to improve the accuracy of term recognition.

[0095] Through entity recognition by the BiLSTM-CRF model, the system identified 2,364 equipment component entities (such as "roll bearing", "hydraulic cylinder", "sensor", etc.), 976 fault phenomenon entities (such as "abnormal vibration", "pressure fluctuation", "too high temperature", etc.) and 843 maintenance method entities (such as "replace the sealing ring", "adjust the clearance", "clean the filter", etc.) from the preprocessed text.

[0096] Based on the entity recognition results, the system applied a relation extraction algorithm with remote supervision and attention mechanism to extract about 12,000 entity relations, including relation types such as "roll bearing has abnormal vibration", "abnormal vibration requires bearing replacement", etc. After consistency checking and redundancy elimination, an initial knowledge graph containing more than 4,000 entities and more than 10,000 relations was finally constructed. The knowledge acquisition efficiency improvement is as Figure 2 shown, demonstrating the performance change of the knowledge extraction method based on weak supervision learning with the increase in the number of labeled samples.

[0097] Temporal attribute extension and concept drift detection: The system adds time attributes to the entities and relations in the knowledge graph, converting them into a quadruple structure. For example, "roll bearing has abnormal vibration on March 15, 2021" indicates that this relation was recorded at a specific time point. Through the time-aware embedding representation method, the system represents entities and relations as time functions rather than static vectors.

[0098] By monitoring semantic displacements in the embedding space, the system detects multiple concept drift phenomena. For example, the definition of the fault phenomenon of "abnormal vibration of roll bearings" has changed over time. In the early stage (before 2018), it was mainly related to improper installation, while recently (after 2020), it is more related to insufficient lubrication. In addition, the system also finds that the effectiveness of certain maintenance methods decreases with the increase in the service life of the equipment. For example, the effect of "adjusting the hydraulic valve" significantly decreases for equipment used for more than 8 years.

[0099] Using the cyclic variational autoencoder, the system mines various evolution patterns from historical data, including the seasonal fluctuation pattern of roll bearing temperature, the linear degradation trend of hydraulic system pressure, and the stage change pattern of transmission system vibration. The comparison of the concept drift detection accuracy of the temporal knowledge graph is as Figure 3 shown, demonstrating the performance change of the knowledge extraction method based on weak supervision learning with the increase in the number of labeled samples.

[0100] State prediction and fault warning: Based on the constructed temporal knowledge graph and state transition model, the system can predict the future state of the equipment. The performance of the fault prediction accuracy on different equipment types is as Figure 4 shown, demonstrating the comparison of the fault prediction accuracy of the method of the present invention with traditional methods on different types of equipment, as well as the percentage increase in accuracy.

[0101] In an actual application, the system monitors that the pressure fluctuation frequency of the hydraulic system of Cold Rolling Mill No. 1 gradually increases and the temperature slowly rises, and compares it with the fault patterns in historical data. It is predicted that there is a 78% probability that the hydraulic system of this equipment will fail within the next 14 days, mainly due to the aging of the hydraulic pump seal ring.

[0102] The system generates a "warning" level warning and recommends checking the status of the hydraulic pump seal ring during the planned shutdown. The estimated repair time is 4 hours, and the required materials include specific model seal rings and hydraulic oil.

[0103] The maintenance personnel carried out the inspection during the planned shutdown according to the system's suggestion and indeed found signs of aging of the hydraulic pump seal ring. After timely replacement, possible sudden failures were avoided. The performance of the fault risk assessment is as Figure 5 shown, comparing the performance of different fault risk assessment methods on multiple key performance indicators.

[0104] Knowledge update and system optimization: After the repair was completed, the maintenance personnel submitted the actual findings and handling situations through the feedback interface, confirming the accuracy of the system diagnosis. At the same time, a new discovery was added: the phenomenon that the sealing ring of this type of hydraulic pump ages faster in a high-temperature environment. The system added this new knowledge to the temporal knowledge graph and updated the parameters of the relevant fault models. The effect of the repair resource optimization is as Figure 6 shown, demonstrating the effect of the repair resource optimization for equipment groups of different scales.

[0105] During the next three months of operation, the early warning accuracy rate of the system for similar hydraulic systems increased from the original 78% to 86%, and the early warning lead time also increased from an average of 14 days to 19 days, providing a more sufficient time window for the repair plan arrangement. Verification of technical effects:

[0106] After the cold rolling mill unit applied this system for one year, significant technical effects were achieved: Improvement in the fault prediction accuracy rate: The data of the fault prediction accuracy rate of the system for the main fault types are shown in Table 1: Table 1: The fault prediction accuracy rate of the system for the main fault types;

[0107] Improvement in equipment availability and repair cost: The comparison data of the equipment operation indicators before and after the application of the system are shown in Table 2: Table 2: Comparison of the equipment operation indicators before and after the application of the system;

[0108] From the above data, it can be seen that the intelligent analysis method for equipment repair faults based on the Internet of Things provided by this application has achieved significant technical effects in practical applications. It not only greatly improves the accuracy rate and lead time of fault prediction, but also significantly improves the equipment availability and reduces the repair cost, verifying the practicality and effectiveness of the method of this application.

[0109] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more forms of equivalent embodiments, all of which fall within the protection scope of this embodiment.

Claims

1. An intelligent analysis method for equipment maintenance faults based on the Internet of Things, characterized in that, It includes the following steps: Automatically extract entities and relationships from device maintenance-related documents to construct an initial knowledge graph; Based on the constructed initial knowledge graph, represent entities and relationships as time functions to construct a temporal knowledge graph; Utilize the constructed temporal knowledge graph to detect concept drift phenomena and mine device performance evolution patterns by monitoring the changes of entity and relationship vectors in the knowledge embedding space over time; Based on the constructed temporal knowledge graph and current observation data, use a recurrent neural network to construct a device state transition model, predict the future state of the device, and evaluate the failure risk; Collect feedback from maintenance personnel on the system prediction results, handle knowledge conflicts, and update the temporal knowledge graph and prediction model; Combine the device real-time state data and the updated temporal knowledge graph to generate failure risk warnings and maintenance suggestions at multiple time scales.

2. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, wherein, The step of automatically extracting entities and relationships from device maintenance-related documents includes: Preprocess maintenance manuals, historical failure records, maintenance reports, and unstructured texts; Adopt a named entity recognition method combining conditional random fields and bidirectional long short-term memory networks to identify and extract three types of core entities: device components, failure phenomena, and maintenance methods; Use a relationship extraction algorithm with remote supervision and attention mechanism to identify semantic relationships between entities; Integrate the extracted entities and relationships into a unified knowledge framework and perform consistency checks.

3. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, characterized in that, The step of representing entities and relationships as time functions includes: Introduce time attributes to entities and relationships in the knowledge graph, and expand the traditional triple knowledge representation to quadruples; Adopt a tensor decomposition method to combine entity vectors, relationship vectors, and time vectors to obtain a time-aware embedding representation; Define temporal relationship reasoning rules to support reasoning with time constraints on the temporal knowledge graph.

4. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, characterized in that, The step of detecting concept drift phenomena and mining device performance evolution patterns includes: Calculate the semantic displacement of an entity at two time points. When the semantic displacement exceeds a preset threshold, it is determined that the calculated entity has concept drift; Use a recurrent variational autoencoder to automatically extract the periodic and trend patterns of knowledge evolution from the temporal knowledge graph; Construct a knowledge evolution visualization module to present the detected concept drift and evolution patterns in an intuitive way.

5. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, wherein The step of predicting the future state of the device and evaluating the failure risk includes: Based on the temporal knowledge graph, realize the reconstruction of the knowledge state of the device at a specific historical time point; Use a recurrent neural network to construct a device state transition model. The input of the model is the device state vector and relevant knowledge graph information at the current time point, and the output is the state prediction at a future time point; Calculate the similarity between the predicted state and known failure modes to generate the probability distribution of different types of failures.

6. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, characterized in that The step of handling knowledge conflicts and updating the temporal knowledge graph and prediction model includes: Calculate the credibility of conflicting knowledge based on the number of supporting evidences and the reliability of sources; Demote or eliminate knowledge with low credibility; Add new knowledge with high credibility; For knowledge with timeliness differences, update the time attribute instead of directly replacing it; Adopt an incremental learning method to update the parameters of the state transition model and the failure risk assessment model.

7. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 1, characterized in that The steps of generating fault risk warnings and maintenance suggestions for multiple time scales include: Based on the state transition model, generating equipment state predictions for different time scales; Determining the warning level according to the matching degree between the predicted state and the fault mode and the time urgency; Automatically generating targeted maintenance suggestions based on the fault risk warning results and the maintenance knowledge in the temporal knowledge graph; Based on the historical maintenance records of the equipment, estimating the time and resources required for maintenance to assist in formulating the maintenance plan.

8. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 3, characterized in that The time attribute includes time points, time intervals or periodic characteristics. A time point represents a specific moment when the knowledge holds, a time interval represents the start and end times when the knowledge is valid, and a periodic characteristic represents a seasonal fault mode.

9. The intelligent analysis method for equipment maintenance faults based on the Internet of Things according to claim 4, characterized in that The recurrent variational autoencoder includes an encoder and a decoder. The encoder encodes the sequence of temporal entity representations into the distribution parameters of the latent variables, and the decoder reconstructs the sequence based on the sampled latent variables and historical information.

10. An intelligent analysis system for equipment repair faults based on the Internet of Things, which is used to execute the intelligent analysis method for equipment repair faults based on the Internet of Things according to any one of claims 1-9, characterized in that, It includes: A multi-source text knowledge extraction and graph construction module, which is used to automatically extract entities and relationships from equipment maintenance-related documents and construct an initial knowledge graph; A temporal knowledge graph construction module, which represents the entities and relationships in the constructed initial knowledge graph as time functions to construct a temporal knowledge graph; A concept drift detection module, which is used to utilize the constructed temporal knowledge graph to detect the concept drift phenomenon and mine the equipment performance evolution pattern by monitoring the changes of entity and relationship vectors in the knowledge embedding space over time; An equipment state prediction module, which constructs an equipment state transition model using a recurrent neural network based on the constructed temporal knowledge graph and the current observation data, predicts the future state of the equipment and evaluates the fault risk; A knowledge update module, which is used to collect the feedback of maintenance personnel on the system prediction results, handle knowledge conflicts and update the temporal knowledge graph and the prediction model; A fault warning module, which is used to generate fault risk warnings and maintenance suggestions for multiple time scales by combining the real-time state data of the equipment and the updated temporal knowledge graph.

Images