Industrial internet of things-based core component performance prediction method, system, device and medium

Abstract

This application discloses a method, system, device, and medium for predicting the performance of core components based on the Industrial Internet of Things (IIoT), relating to the technical field. This application is implemented through a five-platform architecture, including a user platform, service platform, management platform, sensor network platform, and production object platform. It acquires multi-dimensional feature vectors of core components in real time during their operating cycle and performs feature alignment processing. A clustering algorithm is used to cluster the performance degradation patterns of core components, generating prototype vectors reflecting different degradation stages. A knowledge graph of the performance degradation evolution path is established by combining equipment operating condition labels. Prediction results are calculated using Bayes' theorem reasoning method combined with the knowledge graph. Simultaneously, the relationship edge weights and confidence levels are updated based on the prediction and actual results. This enables accurate prediction of core component performance degradation, improving prediction accuracy and reliability, extending equipment lifespan, optimizing maintenance strategies, and reducing unplanned downtime and maintenance costs.

Description

Technical Field

[0001] This application relates to the field of performance prediction technology for industrial components, and in particular to methods, systems, devices and media for performance prediction of core components based on the Industrial Internet of Things. Background Technology

[0002] With the continuous development of Industrial Internet of Things (IIoT) technology, performance prediction of core components of industrial equipment is becoming increasingly important in the fields of smart manufacturing and predictive maintenance. Accurately predicting the performance degradation trend of core components not only helps improve the reliability of equipment operation but also effectively reduces economic losses caused by unplanned downtime. In recent years, data-driven equipment condition monitoring and performance prediction methods have gradually become a research hotspot, and related patents have emerged. However, existing technologies still have significant shortcomings in achieving high-precision and highly generalizable performance prediction of core components.

[0003] Existing technologies still have significant shortcomings in terms of refined modeling, dynamic adaptability, and the synergistic driving force of knowledge and data for performance prediction at the core component level. Especially in the context of the Industrial Internet of Things (IIoT), facing massive amounts of real-time data, complex coupled operating conditions, and the uncertainty of component degradation processes, existing methods struggle to achieve high-precision, adaptive, and interpretable performance prediction.

[0004] Therefore, there is an urgent need for a core component performance prediction method, system, device, and medium based on the Industrial Internet of Things, which can deeply integrate sensor data, physical mechanisms, and operational knowledge to achieve dynamic modeling and long-term trend prediction of core component performance degradation, so as to meet the needs of modern intelligent manufacturing for intelligent upgrading of equipment health management. Summary of the Invention

[0005] In order to at least overcome the above-mentioned shortcomings in the prior art, the purpose of this application is to provide a method, system, device and medium for predicting the performance of core components based on the Industrial Internet of Things to solve the above problems.

[0006] Firstly, this application provides a method for predicting the performance of core components based on the Industrial Internet of Things (IIoT). The method is applied to a five-platform architecture including a user platform, a service platform, a management platform, a sensor network platform, and a production object platform. The method is stored on and executed by the management platform, and includes: Obtain the original multidimensional feature vectors of core components of industrial equipment; Based on the original feature vector, feature alignment is performed to obtain a real-time multidimensional feature vector; Based on the aligned real-time multidimensional feature vector, a clustering algorithm is used to obtain prototype vectors of decay patterns representing different decay stages. Based on the prototype vector, combined with the equipment condition tags obtained synchronously with the component during operation, a knowledge graph of the performance degradation evolution path is obtained by modeling through a relationship network. Based on the real-time multidimensional feature vector, the performance prediction of the component is obtained through evidence reasoning and combined with the knowledge graph, and the confidence of the knowledge graph is updated according to the latest prediction.

[0007] In one possible implementation, performance predictions for components are obtained based on real-time multidimensional feature vectors, through evidence-based reasoning, and in conjunction with a knowledge graph. The confidence level of the knowledge graph is then updated based on the latest predictions. Specifically, this includes: Based on the latest real-time multidimensional feature vectors, through evidence reasoning and in conjunction with knowledge graphs, the posterior probability of a component evolving into each prototype vector is calculated within a specified time period. Based on the posterior probability, the prototype vector with the highest posterior probability is obtained as the final performance prediction result. Based on the discrepancy between the final performance prediction results and the actual observation results of the core components of industrial equipment, the confidence of the state transition probability weights of directed relation edges and the influence weights of causal relation edges in the knowledge graph are updated.

[0008] In one possible implementation, each prototype vector is defined as a state entity in the knowledge graph, and the equipment condition label is defined as a condition entity in the knowledge graph.

[0009] Establish directed edges that point from one state entity to another, with weights representing state transition probabilities and the time required for the transition to occur.

[0010] And establish a causal relationship edge from a conditional entity to a directed relation edge, which is used to quantify the influence of equipment operating conditions on state transition probability.

[0011] Based on conditional entities, directed relation edges, and causal relation edges, a weighted multidimensional relation network is obtained, which is then used as a knowledge graph.

[0012] In one possible implementation, the relational network is modeled, where the equipment condition label is a one-dimensional or multi-dimensional discrete or continuous variable, including at least one of equipment load rate, equipment temperature, ambient temperature, ambient humidity, and lubricating oil content.

[0013] The influence weights of causal relationship edges are modeled using a non-linear mapping function, which maps equipment condition labels to the weights of state transition probabilities using adjustment coefficients.

[0014] In one possible implementation, evidence-based reasoning specifically includes: using Bayes' theorem, combined with the state entities, condition entities, and relational edges defined in the knowledge graph, to calculate the posterior probability of the core component evolving into each target prototype vector within a specified future time period.

[0015] Bayes' theorem first transforms a knowledge graph into a Bayesian network, where state entities are nodes in the Bayesian network and directed edges are conditional probabilistic dependencies between nodes.

[0016] Real-time multidimensional feature vectors are input into the Bayesian network as observational evidence.

[0017] In knowledge graphs, the posterior probability of evolving into each prototype vector is calculated based on observational evidence using a variable elimination algorithm.

[0018] The prototype vector with the highest posterior probability is selected as the most likely decay mode at present.

[0019] Starting from the most likely decay pattern in the knowledge graph, find the corresponding evolution path, combine the probability weights on the path, predict the decay state entity in the next stage and its occurrence time, and generate performance prediction results.

[0020] In one possible implementation, the confidence update specifically includes: When the actual observation results or maintenance records of the core components are received, the actual observation results or maintenance records will be used as the true labels.

[0021] Compare the consistency between the performance prediction results and the actual labels.

[0022] If consistent, increase the confidence level of the causal relationship edge upon which the prediction depends.

[0023] If there is a discrepancy, the confidence level of the causal relationship edge upon which the prediction depends is reduced, and the relevant probability weights are adjusted according to the degree of difference.

[0024] In one possible implementation, the relevant probability weights are adjusted according to the degree of difference. Then, using gradient descent, the mean square error between the performance prediction and the actual observation is minimized through backpropagation to obtain the state transition probability weights and the influence weights of the causal relationship edges.

[0025] Secondly, this application provides a core component performance prediction system based on the Industrial Internet of Things (IIoT). The system adopts a five-platform system architecture, including a user platform, a service platform, a management platform, a sensor network platform, and a production object platform. The management platform includes a data acquisition module, a data preprocessing module, a decay pattern recognition module, a knowledge graph construction module, and a performance prediction and update module, all connected sequentially. The data acquisition module is configured as a data acquisition terminal deployed on the core components of industrial equipment to acquire the original multidimensional feature vectors of the core components of industrial equipment. The data preprocessing module is configured to perform feature alignment operation on the original multidimensional feature vector based on a unified timestamp reference to obtain a real-time multidimensional feature vector. The decay pattern recognition module is configured to obtain prototype vectors of decay patterns representing different decay stages by using a clustering algorithm based on the aligned real-time multidimensional feature vectors. The knowledge graph construction module is configured to model the performance degradation evolution path by combining the prototype vector with the equipment condition tags obtained synchronously with the core component in the running state, and by modeling through a relationship network. The performance prediction and update module is configured to obtain a performance prediction of a component based on the real-time multidimensional feature vector, through evidence reasoning and in conjunction with a knowledge graph, and update the confidence level of the knowledge graph based on the latest prediction; calculate the posterior probability of the core component evolving into each prototype vector within a specified time period based on the latest acquired real-time multidimensional feature vector, through evidence reasoning and in conjunction with the knowledge graph; and obtain the prototype vector with the highest posterior probability as the final performance prediction result based on the posterior probability.

[0026] Thirdly, this application provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of any one of the methods for predicting the performance of core components based on the Industrial Internet of Things.

[0027] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any one of the methods described in the Industrial Internet of Things-based core component performance prediction method.

[0028] In summary, the beneficial effects that this application can achieve are: This application provides a method and system for predicting the performance of core components based on the Industrial Internet of Things. It realizes data acquisition and processing through a five-platform architecture, identifies performance degradation patterns by combining clustering algorithms, constructs a dynamically evolving knowledge graph, and realizes adaptive updates of prediction results based on Bayesian inference. It solves the problems of low prediction accuracy and poor adaptability to working conditions of traditional methods. It has the advantages of modeling the performance degradation process of core components, adapting to complex working condition changes, integrating multi-source data, and improving prediction accuracy and generalization ability. Attached Figure Description

[0029] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the method steps in an embodiment of this application; Figure 2 This is a schematic diagram of the method flow of an embodiment of this application; Figure 3 This is a schematic diagram of the knowledge graph in an embodiment of this application; Figure 4 This is a schematic diagram of the evidence reasoning and updating process in an embodiment of this application; Figure 5 This is a schematic diagram of the five-platform system architecture of an embodiment of this application; Figure 6 This is a schematic diagram of the management platform structure according to an embodiment of this application. Detailed Implementation

[0030] To address the shortcomings of existing technologies, such as reliance on fixed threshold settings for fault risk prediction and a lack of online learning and model self-updating capabilities for core component performance prediction, this invention provides a core component performance prediction method based on the Industrial Internet of Things (IIoT). This method acquires multi-dimensional feature vectors of core components in real-time during their operating cycle and performs feature alignment processing. It then employs a clustering algorithm to cluster performance degradation patterns of core components, generating prototype vectors reflecting different degradation stages. A knowledge graph of performance degradation evolution paths is established by combining equipment operating condition labels. Prediction results are calculated using Bayesian reasoning methods combined with the knowledge graph. Simultaneously, the influence weights of state transition probabilities and causal relationships in the knowledge graph are dynamically updated based on actual feedback. This not only accurately captures the nonlinear characteristics of core component performance degradation but also achieves deep integration of multimodal data and domain knowledge in the prediction model. This improves prediction accuracy while enhancing the model's generalization ability and adaptability, solving the problem of insufficient prediction of dynamic evolution of core component performance under complex operating conditions in current technologies.

[0031] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0032] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0033] Example 1 Please refer to the following: Figure 1 , 2 The above is a schematic diagram of the steps of the core component performance prediction method based on the Industrial Internet of Things provided in the embodiment of the present invention. Further, the core component performance prediction method based on the Industrial Internet of Things may specifically include the contents described in steps S1 to S5.

[0034] Step S1: Obtain the original multidimensional feature vectors of the core components of industrial equipment.

[0035] Step S2: Perform feature alignment based on the original feature vector to obtain a real-time multidimensional feature vector.

[0036] Step S3: Based on the aligned real-time multidimensional feature vector, a clustering algorithm is used to perform a degradation pattern clustering operation on the performance degradation state of the core component throughout its entire life cycle, and prototype vectors representing degradation patterns at different degradation stages are obtained.

[0037] Step S4: Based on the prototype vector, combined with the equipment condition tags obtained synchronously with the core component in its operating state, a knowledge graph of the performance degradation evolution path is obtained by modeling through a relationship network.

[0038] Step S5: Based on the real-time multidimensional feature vector, obtain the performance prediction through evidence reasoning and in combination with the knowledge graph, and update the confidence of the knowledge graph based on the latest prediction; Based on the latest acquired real-time multidimensional feature vectors, through evidence reasoning and in conjunction with the knowledge graph, the posterior probability of the core component evolving into each prototype vector is calculated within a specified time period. Based on the posterior probability, the prototype vector with the highest posterior probability is obtained as the final performance prediction result.

[0039] In the implementation of this application embodiment, the original multidimensional feature vector of the core component during its operating cycle is obtained in real time from a data acquisition terminal on the core component of the industrial equipment. The data acquisition terminal is installed in key parts of the industrial equipment, such as motors and bearings, and periodically collects data on multiple dimensions such as vibration, temperature, and pressure to form a multidimensional feature vector. For example, the motor speed is 1000 r / min, the temperature is 80℃, the current is 10A, and the vibration amplitude is 0.1mm.

[0040] For the original multidimensional feature vector, a feature alignment operation is performed based on a unified timestamp reference to generate a real-time multidimensional feature vector synchronized with the current prediction time. The timestamp reference is set to be once per minute, and timestamp calibration is performed on all feature vectors to ensure that data at the same time point can be aligned. For example, when the timestamp is 10:00:00, the obtained real-time multidimensional feature vector is [1000, 80, 10, 0.1].

[0041] Based on the aligned real-time multidimensional feature vectors, a clustering algorithm is used to perform degradation pattern clustering on the performance degradation state of the core component throughout its entire lifecycle, identifying and outputting prototype vectors representing typical states at different degradation stages. The clustering algorithm employs the K-means algorithm, dividing the feature vectors in the historical data into five categories, representing normal operation, early wear, mid-term wear, late wear, and failure states, respectively. For example, the prototype vector for mid-term wear is [1480, 80, 11, 0.06].

[0042] Based on the prototype vector and combined with the equipment condition tags acquired synchronously with the core component during operation, a relationship network modeling operation is performed to construct a knowledge graph of the performance degradation evolution path. Equipment condition tags include load rate, ambient temperature, etc. The relationship between equipment condition tags and prototype vectors is analyzed using linear regression to establish the knowledge graph. For example, when the load rate increases by 10%, the probability of transitioning from a mid-stage wear state to a late-stage wear state increases by 5%.

[0043] Based on the real-time multidimensional feature vectors, an evidence-based reasoning method is employed, combined with the knowledge graph, to calculate and output the performance prediction results of the core components. Simultaneously, the relation edge weights and confidence levels of the knowledge graph are dynamically updated based on the latest prediction results. The evidence-based reasoning method uses a Bayesian network, taking the real-time multidimensional feature vectors as input to calculate the posterior probability of each prototype vector. For example, if the probability that the current motor is in a mid-stage wear state is 80%, it is predicted that it will enter a late-stage wear state within the next week.

[0044] This invention discloses a method, system, device, and medium for predicting the performance of core components of industrial equipment based on the Industrial Internet of Things (IIoT). It constructs a relational network knowledge graph that integrates the influence of equipment operating conditions, enabling dynamic and accurate modeling and prediction of component performance degradation processes. This invention achieves component performance prediction through data processing, pattern recognition, knowledge graph construction, and evidence-based reasoning.

[0045] Example 2 Based on Embodiment 1, the following is a flowchart illustrating the performance prediction method for core components based on the Industrial Internet of Things provided in this embodiment of the invention. Further, the performance prediction method for core components based on the Industrial Internet of Things may specifically include the following contents.

[0046] When implementing the embodiments of this application, please refer to Figure 2 The diagram illustrates a specific method flow of an embodiment of this application.

[0047] Step S1: Obtain the original multidimensional feature vectors of the core components of industrial equipment.

[0048] In the implementation of this application embodiment, different sensor arrays are deployed on the equipment of the production object platform to collect physical parameters and generate a spatiotemporally aligned feature matrix to obtain the original multidimensional feature vector of the industrial equipment; The system receives feature data from various industrial devices via the cloud of the Internet of Things (IoT) as raw multidimensional feature vectors. Each feature vector includes the device ID, the timestamp of the data collection, and data such as the vibration spectrum, temperature gradient, and current harmonic distortion rate of the industrial devices collected according to user needs, forming an N-dimensional raw feature vector. We encapsulate the data from a single device at a single time point into an N-dimensional original feature vector, denoted as . Where ID is the unique identifier of the device. This refers to data on the physical characteristics of the device collected based on the specific needs of the user.

[0049] Step S2: Perform feature alignment based on the original feature vector to obtain a real-time multidimensional feature vector.

[0050] In implementing the embodiments of this application, feature alignment is required first based on the original feature vectors, which is existing technology and will not be elaborated upon in this application. It should be understood that the original multidimensional feature vectors across devices are uniformly represented and converted into universal features that are aligned in space and time, thus solving the data distribution offset caused by differences in models and operating conditions of equipment in different factories.

[0051] At this point, we will denote the aligned data as... The ID and timestamp are removed, and only the standardized physical features are retained. The aligned data forms a real-time multidimensional feature vector, denoted as . .

[0052] Aligning timestamps is an existing technique and will not be elaborated here.

[0053] Step S3: Based on the aligned real-time multidimensional feature vector, a clustering algorithm is used to perform a degradation pattern clustering operation on the performance degradation state of the core component throughout its entire life cycle, and prototype vectors representing degradation patterns at different degradation stages are obtained.

[0054] In the implementation of this application embodiment, based on the aligned real-time feature vector, the typical decay patterns of core components are extracted from massive real-time features using a spectral clustering algorithm, and cluster centers are output as prototype vectors of the decay patterns, thus transforming the real-time multidimensional feature vector into K prototype vectors. The decay pattern clustering operation transforms scattered feature points into cluster center vectors, extracting physical failure knowledge from massive real-time device data. Simultaneously, it compresses storage dimensions, identifying common decay patterns. Clustering operations are existing technology and will not be elaborated upon in this embodiment. It should be understood that this embodiment presets K decay patterns according to user needs, with the goal of... The clustering is divided into K clusters. If the user presets the decay pattern as "normal → slight decay → moderate decay → severe decay", then K=4. After the spectral clustering algorithm, the center vector of each cluster is the prototype vector, that is, the prototype vectors of the K decay patterns are obtained. The specific steps are as follows: Quantize any two real-time multidimensional feature vectors using a similarity matrix. and The similarity between them is such that the closer the similarity is, the higher the similarity score, and the two real-time multidimensional feature vectors are in the same decay stage.

[0055] Similarity matrix Using the Gaussian kernel function, we can obtain, expressed as ,in ; yes and The squared Euclidean distance is mainly used to measure differences in features; It is bandwidth that controls the rate at which similarity decays with distance. The larger the similarity, the lower its sensitivity to distance, making it suitable for industrial data with large feature fluctuations.

[0056] if and Belonging to the same stage of decline, with little difference in characteristics, Small, Close to 1; conversely, if they belong to different stages, Close to 0.

[0057] Then based on the similarity matrix Calculate its Laplacian matrix. Define a diagonal matrix as a metric matrix. , The Laplace matrix L can be expressed as follows: The Laplace matrix is ​​used to transform a network of sample similarities into a "graph structure." Its core principle is to preserve the topological relationships of the data, i.e., the continuity of the decay phase, through matrix operations.

[0058] Eigenvalue decomposition is performed on the Laplacian matrix L to extract lower-dimensional features that characterize the clustering structure, and the characteristic equation is solved. , to obtain eigenvalues and the corresponding feature vectors Furthermore, since the eigenvalues ​​of the Laplacian matrix are related to the connectivity of the graph, the first K smallest eigenvalues ​​are close to 0, corresponding to the K connected components in the graph, which are the K decay stages. Their eigenvectors can effectively distinguish different clusters. Therefore, the eigenvectors corresponding to the first K smallest non-zero eigenvalues ​​are selected. , form a matrix from the eigenvectors .

[0059] For matrix The row vectors are clustered using K-means to obtain K cluster centers, which are the prototype vectors representing different stages of decay. , Corresponding to the initial health state, This corresponds to the critical state before failure.

[0060] Step S4: Based on the prototype vector, combined with the equipment condition tags obtained synchronously with the core component in its operating state, a knowledge graph of the performance degradation evolution path is obtained by modeling through a relationship network.

[0061] In one possible implementation, please refer to Figure 3 Each prototype vector is defined as a state entity in the knowledge graph, and the equipment condition label is defined as a condition entity in the knowledge graph.

[0062] Establish directed edges that point from one state entity to another, with weights representing state transition probabilities and the time required for the transition to occur.

[0063] And establish a causal relationship edge from a conditional entity to a directed relation edge, which is used to quantify the influence of equipment operating conditions on state transition probability.

[0064] Based on conditional entities, directed relation edges, and causal relation edges, a weighted multidimensional relation network is obtained, which is then used as a knowledge graph.

[0065] In the implementation of this application, a knowledge graph with probability weights is constructed using prototype vectors and operating condition labels as entities to quantify state transitions and the impact of operating conditions.

[0066] Each of the prototype vectors is defined as a state entity in the knowledge graph, and the equipment condition label is defined as a condition entity in the knowledge graph.

[0067] prototype vector Find the corresponding state entity and record it as Based on historical equipment operating condition tags and according to user requirements, define M tags such as load rate, ambient temperature, and lubricant content, and denote them as follows: , .

[0068] For example, the prototype vector for normal operation is [1000, 80, 10, 0.1], which is defined as state entity A; the equipment condition label with a load rate of 80% is defined as condition entity B.

[0069] Establish directed edges pointing from one state entity to another, with weights representing state transition probabilities and the time required for the transition to occur. The directed edges represent state transitions from one state entity to another. Pointing to state entity The side note is , The transition probability is denoted as From historical data arrive The average time consumed was obtained transfer time .

[0070] For example, from state entities Pointing to state entity The directed edges represent the transition from the normal working state to the initial wear state, with a transition probability of 10% and a required time of 3 months.

[0071] A causal relationship edge is established from a conditional entity to a directed relation edge, which is used to quantify the influence of the equipment operating condition on the state transition probability, thereby constructing the knowledge graph.

[0072] Causal relationship edges represent the influence of operating conditions on transition probabilities, from arrive The side note is Its influence weight is denoted as .

[0073] By integrating entities and relation edges, a weighted multidimensional relation network is formed, thus constructing a knowledge graph.

[0074] For example, conditional entity B on directed relation edges The impact weight, based on historical data, is 1.2, meaning that when the load rate is 80%, the probability of transitioning from normal working state to initial wear state increases by 20%.

[0075] In one possible implementation, a relational network is modeled, where... The equipment operating condition label is a one-dimensional or multi-dimensional discrete or continuous variable, including at least one of the following: equipment load rate, equipment temperature, ambient temperature, ambient humidity, and lubricating oil content.

[0076] The influence weights of causal relationship edges are modeled using a non-linear mapping function, which maps equipment condition labels to the weights of state transition probabilities using adjustment coefficients.

[0077] When implementing the embodiments of this application, The equipment operating condition label is a one-dimensional or multi-dimensional discrete or continuous variable, including but not limited to at least one of the following: equipment load rate, equipment temperature, ambient temperature, ambient humidity, and lubricating oil content. For example, the equipment load rate is 85%, the equipment temperature is 80℃, and the ambient humidity is 60%.

[0078] Furthermore, the influence weight of the causal relationship edge is modeled using a nonlinear mapping function, and the equipment operating condition label is mapped to the adjustment coefficient of the state transition probability weight through a neural network algorithm.

[0079] Step S5: Based on the real-time multidimensional feature vector, through evidence reasoning and in conjunction with the knowledge graph, a performance prediction is obtained, and the confidence level of the knowledge graph is updated based on the latest prediction. Please refer to [reference needed]. Figure 4 This illustrates the process of updating evidence reasoning.

[0080] Based on the latest acquired real-time multidimensional feature vectors, through evidence reasoning and in conjunction with the knowledge graph, the posterior probability of the core component evolving into each prototype vector is calculated within a specified time period. Based on the posterior probability, the prototype vector with the highest posterior probability is obtained as the final performance prediction result.

[0081] In one possible implementation, evidence reasoning specifically includes: using Bayes' theorem, combined with the state entities, condition entities and relation edges defined in the knowledge graph, to obtain the posterior probability of the core component evolving into each target prototype vector within a specified future time period; The Bayes theorem first transforms the knowledge graph into a Bayesian network, where the state entities are nodes in the Bayesian network and the directed edges are conditional probabilistic dependencies between nodes. The real-time multidimensional feature vector is input into the Bayesian network as observation evidence; In the knowledge graph, based on the observational evidence, the posterior probability of evolving into each prototype vector is calculated using a variable elimination algorithm; Select the prototype vector with the highest posterior probability as the most likely decay mode at present; Starting from the most likely decay mode in the knowledge graph, find the corresponding evolution path, combine the probability weights on the path, predict the decay state entity in the next stage and its occurrence time, and generate the performance prediction result.

[0082] In the implementation of this application embodiment, Bayesian inference is used to calculate the posterior probability of the core component evolving into each target prototype vector within a specified future time period, in conjunction with the state entities, condition entities, and relation edges defined in the knowledge graph. For example, the probability of a motor evolving into a mid-term wear state within the next month is calculated.

[0083] The knowledge graph is converted into a Bayesian network, where state entity A serves as a node in the Bayesian network, and directed edges represent the conditional probabilities between nodes. Dependencies. For example, state entities. As nodes, directed relation edges Represents a node To the node The conditional probability dependency.

[0084] With the real-time multidimensional feature vector The observational evidence is input into the Bayesian network. For example, a real-time multidimensional feature vector [1000, 80, 10, 0.1] is input as observational evidence.

[0085] Based on the observed evidence, the posterior probability of evolving into each prototype vector is calculated using a variable elimination algorithm. For example, the probability of the motor evolving into the intermediate wear state is calculated to be 80%.

[0086] The prototype vector with the highest posterior probability is selected as the most likely decay mode at present. For example, the intermediate wear state has the highest posterior probability, so the intermediate wear state is selected as the most likely decay mode at present.

[0087] The evolution path corresponding to the prototype vector with the highest posterior probability is determined, and the probability weights on the path are combined to predict the entity in the next stage of the decay state and its occurrence time, thereby generating the performance prediction result. For example, it is predicted that the motor will enter the late wear state within the next two weeks.

[0088] In one possible implementation, the confidence of the state transition probability weights of directed relation edges and the influence weights of causal relation edges in the knowledge graph is updated based on the deviation between the final performance prediction results and the actual observation results of the core components of industrial equipment.

[0089] In the implementation of this application embodiment, the confidence levels of the state transition probability weights of directed relation edges and the influence weights of causal relation edges in the knowledge graph are updated based on the deviation between the performance prediction results and the actual observation results of the core components of industrial equipment. For example, when the prediction results are inconsistent with the actual observation results, the confidence levels of the relevant directed relation edges are reduced.

[0090] In one possible implementation, the confidence update specifically includes: When the actual observation results or maintenance records of the core components are received, the actual observation results or maintenance records will be used as the true labels.

[0091] Compare the consistency between the performance prediction results and the actual labels.

[0092] If consistent, increase the confidence level of the causal relationship edge upon which the prediction depends.

[0093] If there is a discrepancy, the confidence level of the causal relationship edge upon which the prediction depends is reduced, and the relevant probability weights are adjusted according to the degree of difference.

[0094] In the implementation of this application embodiment, when the actual observation results or maintenance records of the core components are received, they are used as the true label. For example, the actual condition of the motor is mid-term wear.

[0095] Compare the performance prediction results with the actual labels. For example, if the prediction result is mid-term wear and the actual label also shows mid-term wear, they are consistent.

[0096] If the two are consistent, then increase the confidence level of the causal relationship edges associated with this prediction. For example, increase the confidence level of directed relationship edges. The confidence level.

[0097] If there is a discrepancy, the confidence level of the causal relationship edges associated with that prediction is reduced, and the corresponding probability weights are adjusted according to the degree of difference between the two. For example, the confidence level of directed relationship edges is reduced. The confidence level is determined, and the state transition probability is adjusted.

[0098] In one possible implementation, the relevant probability weights are adjusted according to the degree of difference. Then, using gradient descent, the mean square error between the performance prediction and the actual observation is minimized through backpropagation to obtain the state transition probability weights and the influence weights of the causal relationship edges.

[0099] In the implementation of this application embodiment, the gradient descent method is used to minimize the mean square error between the performance prediction result and the actual observation result. Backpropagation is used to calculate the state transition probability weights and the influence weights of the causal relationship edges. For example, the directed relationship edges are adjusted using the gradient descent method. The state transition probability weights are used to minimize the mean square error between the predicted results and the actual observation results.

[0100] Evidence-based reasoning is performed by fusing data, knowledge, and decision-making. Based on the real-time device characteristics and combined with the knowledge graph, the confidence level of the knowledge graph is updated.

[0101] Example 3 This is the third embodiment of the present invention. Based on embodiments 1 and 2, please refer to the following references. Figure 5 This embodiment provides a core component performance prediction system based on the Industrial Internet of Things, applied to a five-platform system architecture, which includes a user platform, a service platform, a management platform, a sensor network platform, and a production object platform.

[0102] Please refer to the following: Figure 6 The core component performance prediction system based on the Industrial Internet of Things is executed on the management platform and includes a data acquisition module, a data preprocessing module, a decay pattern recognition module, a knowledge graph construction module, and a performance prediction and update module connected in sequence.

[0103] The data acquisition module is configured as a data acquisition terminal deployed on the core components of industrial equipment to acquire the original multidimensional feature vectors of the core components of industrial equipment. The data preprocessing module is configured to perform feature alignment operation on the original multidimensional feature vector based on a unified timestamp reference to obtain a real-time multidimensional feature vector. The decay pattern recognition module is configured to obtain prototype vectors of decay patterns representing different decay stages by using a clustering algorithm based on the aligned real-time multidimensional feature vectors. The knowledge graph construction module is configured to model the performance degradation evolution path by combining the prototype vector with the equipment condition tags obtained synchronously with the core component in the running state, and by modeling through a relationship network. The performance prediction and update module is configured to obtain a performance prediction of a component based on the real-time multidimensional feature vector, through evidence reasoning and in conjunction with a knowledge graph, and update the confidence level of the knowledge graph based on the latest prediction; calculate the posterior probability of the core component evolving into each prototype vector within a specified time period based on the latest acquired real-time multidimensional feature vector, through evidence reasoning and in conjunction with the knowledge graph; and obtain the prototype vector with the highest posterior probability as the final performance prediction result based on the posterior probability.

[0104] In one possible implementation, the data acquisition module is configured to acquire the original multidimensional feature vector of the core component in real time during its operating cycle through a data acquisition terminal deployed on the core component of the industrial equipment. The data dimensions of the original multidimensional feature vector include the time-domain statistical characteristics and frequency-domain statistical characteristics of the vibration signal, the mean and fluctuation values ​​of the temperature signal, and the peak and valley values ​​of the pressure signal.

[0105] The data preprocessing module is configured to perform feature alignment operations on the original multidimensional feature vectors based on a unified timestamp benchmark to generate real-time multidimensional feature vectors synchronized with the current prediction time point. The decay pattern recognition module is configured to apply decay pattern clustering operations to the real-time multidimensional feature vectors to identify and output prototype vectors representing typical states of different decay stages.

[0106] The knowledge graph construction module is configured to perform relational network modeling operations based on prototype vectors and equipment condition labels to construct a knowledge graph of performance degradation evolution path. In the knowledge graph, each prototype vector is defined as a state entity, the equipment condition label is defined as a condition entity, and directed relation edges with probability weights representing state transitions and causal relation edges representing the influence of operating conditions are established.

[0107] The knowledge graph is stored in the graph database submodule, where state entities and condition entities are stored as nodes, and directed relation edges and causal relation edges are stored as edges. Each edge includes type, direction, weight value and confidence score as its attributes.

[0108] The performance prediction and update module is configured to calculate and output the performance prediction results of the core components based on the latest acquired real-time multidimensional feature vectors using evidence reasoning methods combined with the knowledge graph, and to update the confidence of the relation edge weights in the knowledge graph based on the comparison between the performance prediction results and the actual observation results.

[0109] In the implementation of this application embodiment, the timestamp reference alignment operation refers to the time synchronization processing of asynchronous data collected by different sensors to ensure the correlation of multidimensional data at the same time scale.

[0110] Decay pattern clustering refers to unsupervised clustering of multidimensional feature vectors to identify typical performance degradation stages, and different decay patterns are represented by cluster center vectors.

[0111] In a knowledge graph, state entities refer to performance degradation state nodes defined by cluster center vectors, while condition entities refer to operating condition nodes composed of equipment load rate, temperature, and environmental parameters. Specifically, a graph-structured database can be used for storing and retrieving entities and relationships. Directed relation edges connect two state entities and are labeled with state transition probabilities and time weights. Causal relation edges connect condition entities to directed relation edges and are labeled with operating condition influence coefficients. Specifically, adjacency matrices or adjacency lists can be used to store and update the weights of relation edges.

[0112] Compared to existing technologies that rely on fixed thresholds for fault warning, which struggle to adapt to the nonlinear degradation characteristics under complex operating conditions, this solution quantifies state transition probabilities and operating condition influence coefficients using a dynamically updated knowledge graph, enabling adaptive adjustment of prediction model parameters. Existing technologies often employ separate knowledge graphs and prediction models, lacking online learning mechanisms. This solution stores the knowledge graph as graph-structured data with confidence attributes, achieving continuous optimization of relational edge weights through closed-loop feedback between prediction results and actual observations, thus enhancing the model's generalization ability under new equipment and rapidly changing environments.

[0113] Through the above technical solutions, this application solves the problem of limited prediction accuracy caused by the separation of knowledge graph reasoning and performance prediction. It achieves fusion reasoning of data and knowledge by storing a relational network with confidence levels in a graph database. It overcomes the shortcomings of traditional methods that lack online learning capabilities by dynamically adjusting state transition probabilities and operating condition influence coefficients using a real-time prediction error backpropagation mechanism, thereby improving the accuracy of performance degradation prediction under complex operating conditions. It eliminates the poor adaptability of fixed threshold early warning mechanisms to nonlinear degradation processes by achieving reliable identification of early performance degradation through probabilistic path prediction and confidence assessment.

[0114] When implementing the embodiments of this application, please refer to... Figure 5 The five-platform system architecture refers to a hierarchical data processing architecture consisting of a user platform, a service platform, a management platform, a sensor network platform, and a production object platform. Specifically, a distributed computing framework can be used to realize data interaction and task scheduling between the platforms, and closed-loop control of data acquisition, transmission, processing, and feedback can be achieved through hierarchical division.

[0115] The user platform is configured to support users in obtaining perceived service information, processing it into user-understandable perceived information, and assisting users in decision-making. Users generate control commands based on the perceived information and their own needs, and submit service requests to the service platform through the information system. This includes user terminal devices such as mobile phones, computers, and dedicated terminals, which achieve service interaction through the integration of user information system software.

[0116] The service platform is configured as an API server or communication hub to establish service links between the management platform and the user platform. This includes various server clusters.

[0117] The management platform is configured as the central hub for IoT operations, performing core functions such as device status monitoring, data management, parameter configuration, and lifecycle management; it can be divided into multiple specialized management sub-platforms. This includes a high-performance server system, on which this system resides, and the aforementioned methods are performed.

[0118] The sensor network platform is configured to provide data communication, protocol parsing, identifier classification, transmission monitoring, and security management, undertaking tasks such as network management, command scheduling, and device status monitoring. Through a layered data processing mechanism, it avoids data redundancy and efficiency bottlenecks caused by directly connecting raw data to the management platform. This includes network infrastructure such as gateway devices and edge computing nodes.

[0119] The production platform is configured to perform actual production control, detection, metering, and other on-site operations. This includes terminal execution units such as production equipment and sensors.

[0120] Example 4 The fourth embodiment of the present invention differs from the previous embodiments in that: Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, or they may be electrical, mechanical, or other forms of connection.

[0122] The units described as separate components may or may not be physically separate. As will be apparent to those skilled in the art, the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0123] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0124] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or grid device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the performance of core components based on the Industrial Internet of Things, characterized in that, include: Obtain the original multidimensional feature vectors of core components of industrial equipment; Based on the original feature vector, feature alignment is performed to obtain a real-time multidimensional feature vector; Based on the aligned real-time multidimensional feature vector, a clustering algorithm is used to obtain prototype vectors of decay patterns representing different decay stages. Based on the prototype vector, combined with the equipment condition tags obtained synchronously with the component during operation, a knowledge graph of the performance degradation evolution path is obtained by modeling through a relationship network. Based on the real-time multidimensional feature vector, the performance prediction of the component is obtained through evidence reasoning and combined with the knowledge graph, and the confidence of the knowledge graph is updated according to the latest prediction. The process of modeling through relational networks to obtain a knowledge graph of performance degradation evolution paths specifically includes: Each of the prototype vectors is defined as a state entity in the knowledge graph, and the equipment condition label is defined as a condition entity in the knowledge graph. Establish directed edges from one state entity to another, with weights containing state transition probabilities and the time required for the transition to occur. And establish a causal relationship edge from a conditional entity to a directed relation edge, which is used to quantify the impact of the equipment operating condition on the state transition probability; Based on the conditional entities, the directed relation edges, and the causal relation edges, a weighted multidimensional relation network is obtained, and the multidimensional relation network is used as a knowledge graph. The process of obtaining component performance predictions based on the real-time multidimensional feature vectors through evidence reasoning and in conjunction with a knowledge graph, and updating the confidence level of the knowledge graph based on the latest predictions, specifically includes: Based on the latest acquired real-time multidimensional feature vectors, through evidence reasoning and in conjunction with the knowledge graph, the posterior probability of the component evolving into each prototype vector is calculated within a specified time period. Based on the posterior probability, the prototype vector with the highest posterior probability is obtained as the final performance prediction result; Based on the deviation between the final performance prediction result and the actual observation results of the core components of industrial equipment, the confidence of the state transition probability weights of the directed relation edges and the influence weights of the causal relation edges in the knowledge graph are updated.

2. The method for predicting the performance of core components based on the Industrial Internet of Things according to claim 1, characterized in that, In the process of modeling the relationship network, among which, The equipment operating condition label is a one-dimensional or multi-dimensional discrete or continuous variable, including at least one of equipment load rate, equipment temperature, ambient temperature, ambient humidity, and lubricating oil content; The influence weight of the causal relationship edge is modeled using a nonlinear mapping function, which maps the equipment condition label to the adjustment coefficient of the weight of the state transition probability.

3. The method for predicting the performance of core components based on the Industrial Internet of Things according to claim 1, characterized in that, The evidence reasoning specifically includes: using Bayes' theorem, combined with the state entities, condition entities and relation edges defined in the knowledge graph, to obtain the posterior probability of the core component evolving into each target prototype vector within a specified future time period; The Bayes theorem first transforms the knowledge graph into a Bayesian network, where the state entities are nodes in the Bayesian network and the directed edges are conditional probabilistic dependencies between nodes. The real-time multidimensional feature vector is input into the Bayesian network as observation evidence; In the knowledge graph, based on the observational evidence, the posterior probability of evolving into each prototype vector is calculated using a variable elimination algorithm; Select the prototype vector with the highest posterior probability as the most likely decay mode at present; Starting from the most likely decay mode in the knowledge graph, find the corresponding evolution path, combine the probability weights on the path, predict the decay state entity in the next stage and its occurrence time, and generate the performance prediction result.

4. The method for predicting the performance of core components based on the Industrial Internet of Things according to claim 3, characterized in that, The confidence update specifically includes: When the actual observation results or maintenance records of the core components are received, the feedback or maintenance records will be used as the true labels; Compare the consistency between the performance prediction results and the actual labels; If consistent, increase the confidence level of the causal relationship edge upon which the prediction depends; If there is a discrepancy, the confidence level of the causal relationship edge upon which the prediction depends is reduced, and the associated probability weights are adjusted according to the degree of difference.

5. The method for predicting the performance of core components based on the Industrial Internet of Things according to claim 4, characterized in that, The probability weights are adjusted according to the degree of difference. The gradient descent method is used to backpropagate, with the goal of minimizing the mean square error between the performance prediction result and the actual observation result, to obtain the state transition probability weights and the influence weights of the causal relationship edges.

6. A core component performance prediction system based on the Industrial Internet of Things, characterized in that, The system is applied to the core component performance prediction method based on the Industrial Internet of Things as described in any one of claims 1 to 5, comprising a data acquisition module, a data preprocessing module, a decay pattern recognition module, a knowledge graph construction module, and a performance prediction and update module connected in sequence; wherein: The data acquisition module is configured as a data acquisition terminal deployed on the core components of industrial equipment to acquire the original multidimensional feature vectors of the core components of industrial equipment. The data preprocessing module is configured to perform feature alignment operation on the original multidimensional feature vector based on a unified timestamp reference to obtain a real-time multidimensional feature vector. The decay pattern recognition module is configured to obtain prototype vectors representing decay patterns at different decay stages by using a clustering algorithm based on the aligned real-time multidimensional feature vectors. The knowledge graph construction module is configured to model the performance degradation evolution path by combining the prototype vector with the equipment condition tags obtained synchronously with the core component in the running state, and by modeling through a relationship network. The performance prediction and update module is configured to obtain a performance prediction of a component based on the real-time multidimensional feature vector, through evidence reasoning and in conjunction with a knowledge graph, and update the confidence level of the knowledge graph based on the latest prediction; based on the latest acquired real-time multidimensional feature vector, through evidence reasoning and in conjunction with the knowledge graph, calculate the posterior probability of the core component evolving into each prototype vector within a specified time period; and obtain the prototype vector with the highest posterior probability as the final performance prediction result based on the posterior probability. The process of modeling through relational networks to obtain a knowledge graph of performance degradation evolution paths specifically includes: Each of the prototype vectors is defined as a state entity in the knowledge graph, and the equipment condition label is defined as a condition entity in the knowledge graph. Establish directed edges from one state entity to another, with weights containing state transition probabilities and the time required for the transition to occur. And establish a causal relationship edge from a conditional entity to a directed relation edge, which is used to quantify the impact of the equipment operating condition on the state transition probability; Based on the conditional entities, the directed relation edges, and the causal relation edges, a weighted multidimensional relation network is obtained, and the multidimensional relation network is used as a knowledge graph. The process of obtaining component performance predictions based on the real-time multidimensional feature vectors through evidence reasoning and in conjunction with a knowledge graph, and updating the confidence level of the knowledge graph based on the latest predictions, specifically includes: Based on the latest acquired real-time multidimensional feature vectors, through evidence reasoning and in conjunction with the knowledge graph, the posterior probability of the component evolving into each prototype vector is calculated within a specified time period. Based on the posterior probability, the prototype vector with the highest posterior probability is obtained as the final performance prediction result; Based on the deviation between the final performance prediction result and the actual observation results of the core components of industrial equipment, the confidence of the state transition probability weights of the directed relation edges and the influence weights of the causal relation edges in the knowledge graph are updated.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 5.

Images