Intelligent maintenance decision-making method and system for primary helium fan based on electromagnetic bearing monitoring

Abstract

The invention discloses a main helium fan intelligent maintenance decision-making method based on electromagnetic bearing monitoring, and the method comprises the steps: carrying out the self-adaptive feature extraction and fusion of multi-source monitoring data of an electromagnetic bearing through employing an edge-cloud cooperation architecture; artificial intelligence models such as a graph neural network and active federated learning are introduced, and bearing health state representation is constructed in a self-adaptive mode; early fault symptoms are captured through a space-time anomaly detection model, fault traceability is revealed through causal attention reasoning, and an interpretable maintenance strategy is formed. Predictive maintenance of key equipment of the nuclear turbine unit is enabled by using an artificial intelligence technology, and a new thought and a new method are provided for improving nuclear power operation safety and economical efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of nuclear power equipment status monitoring and fault diagnosis, and in particular to a main helium blower intelligent maintenance decision method and system based on electromagnetic bearing monitoring. Background Art

[0002] Electromagnetic bearings are a non-contact suspension support technology that uses electromagnetic force to balance the weight of the rotor and achieve frictionless high-speed rotation of the rotor. Compared with traditional mechanical bearings, electromagnetic bearings have the advantages of no lubrication, maintenance-free, high speed, and precision. They are widely used in high-end equipment such as nuclear power turbines, high-speed centrifugal compressors, and flywheel energy storage.

[0003] With the continuous development and maturity of electromagnetic bearing technology, its application in nuclear power units is becoming more and more in-depth. Taking the main helium fan of a nuclear power plant as an example, the use of electromagnetic bearings instead of traditional oil film bearings can significantly improve the operating stability and reliability of the main helium fan, extend the overhaul cycle, and reduce the losses caused by maintenance, which is of great significance for improving the safety and economy of nuclear power plants. However, due to the complexity of the structure and working principle of electromagnetic bearings, it still faces challenges such as difficult state monitoring, difficult fault diagnosis, and difficult operation and maintenance decision-making during actual operation. Traditional electromagnetic bearing state monitoring mainly relies on vibration, temperature, displacement and other sensors installed on bearings and rotor components to obtain key parameters during bearing operation. However, due to the compact internal structure and numerous parts of electromagnetic bearings, there is a strong coupling correlation between the various monitoring parameters. The abnormality of a single parameter is often difficult to directly determine the cause of the fault. It requires long-term experience accumulation of professionals to make an accurate diagnosis, and the diagnosis and decision-making process is easily affected by human factors. In addition, the offline analysis mode of monitoring data means that faults can only be discovered when the equipment has obvious abnormalities, which loses the opportunity for early warning and active intervention, resulting in high maintenance costs.

[0004] In view of the above problems, it is urgent to develop an intelligent electromagnetic bearing status monitoring and operation and maintenance decision-making system. The system should fully integrate the massive monitoring data of electromagnetic bearings from multiple sources and heterogeneity, use machine learning algorithms to adaptively extract health degradation characteristics, and build an end-to-end intelligent mapping model from data to health status to maintenance strategy. It can evaluate the health level of bearings online in real time, detect signs of degradation failures early and trace the key influencing factors, and assist professionals in formulating the best maintenance plan. The present invention aims to provide an intelligent maintenance decision-making system for main helium fans based on electromagnetic bearing monitoring, empowered by cutting-edge artificial intelligence technology, to establish an integrated intelligent operation and maintenance solution for electromagnetic bearings of nuclear power steam turbine units, and promote the intelligent upgrade of nuclear power equipment operation and maintenance mode. Summary of the invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides an intelligent maintenance decision method for the main helium blower based on electromagnetic bearing monitoring, which can accurately evaluate the health status of the main helium blower, predict failure risks in advance, and actively optimize maintenance strategies, playing a key role in ensuring the safe and economical operation of nuclear power units.

[0007] To solve the above technical problems, the present invention provides the following technical solutions, which are intelligent maintenance decision methods for main helium blowers based on electromagnetic bearing monitoring, including: adopting an edge-cloud collaborative architecture, edge computing nodes collect and preprocess multi-source monitoring data of electromagnetic bearings in real time, and upload the preprocessed data to the cloud through a secure communication network; the cloud uses a graph neural network to construct a spatiotemporal correlation graph for the time series of each monitoring parameter of the electromagnetic bearing, and at the same time adopts a federated active transfer learning framework to optimize the health assessment model; the cloud uses a spatiotemporal sequence anomaly detection model to construct a spatiotemporal anomaly score function, and adopts a spatiotemporal causal attention mechanism to achieve explainable maintenance decision reasoning, and comprehensively monitors the time dependence and causal relationship of parameters; the electromagnetic bearing monitoring data, health assessment results, fault warning information and maintenance decision suggestions are intuitively presented through a Web visualization platform to assist experts in formulating maintenance plans.

[0008] As a preferred solution of the main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, the preprocessing includes: the edge computing node preprocesses the electromagnetic bearing vibration data using an adaptive multi-scale fusion method, performs L-layer wavelet packet decomposition on the vibration signal, and obtains 2 L subband coefficient sequences, for each subband l (l = 1, 2, ..., 2 L ) Calculate the variance Var l and Skew l :

[0009]

[0010] Where N is the subband length, mean and std are the mean and standard deviation respectively, x l (i) represents the i-th coefficient in the l-th sub-band, and the significance measure of sub-band l is defined as:

[0011] Sig l =w1Var l +w2|Skew l |

[0012] Among them, w1 and w2 are weight coefficients, select Sig l The largest first K subbands extract time domain, frequency domain and time-frequency domain features, denoted as The softmax attention mechanism is used to adaptively fuse sub-band features:

[0013]

[0014]

[0015] in, represents the kth feature of the lth subband The attention weight, W and u are the parameters of the attention network, b is the bias, and v is the fused feature vector. Represents matrix transpose.

[0016] As a preferred solution of the main helium fan intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, wherein: the construction of the spatiotemporal correlation graph includes: using a graph neural network in the cloud to construct a spatiotemporal correlation graph for the time series of each monitoring parameter of the electromagnetic bearing, and extracting low-dimensional latent variables that comprehensively reflect the health status of the bearing through end-to-end representation learning; constructing the monitoring parameters into an undirected graph G = (V, E), and the vertex set V represents the time series of each parameter {v i , i=1,2,...,N}, the edge set E represents the strength of association between parameters, and the hidden layer of the vertex obtained after T rounds of iterative updates is represented as:

[0017]

[0018] Among them, W1 and W2 are graph convolution parameters, σ is the activation function, is the neighbor set of vertex i, α ii is the attention weight between vertices i and j, h i , They respectively represent concatenating the feature vectors of vertices i and j as the input of the attention network, calculating the correlation between the two vertices, and finally concatenating the representation of each vertex and mapping it to the health state space through the fully connected layer.

[0019] As a preferred solution of the main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, wherein: the federated active transfer learning framework includes: the cloud uses the federated active transfer learning framework to adaptively optimize and correct the health assessment model across devices and working conditions without directly sharing the original monitoring data, and continuously improve the model performance through active learning strategies;

[0020] Set the local data set of the i-th device to The model parameter is w n , the sample selection strategy for active learning is a n , the federation aggregation parameter is w, in each communication round t:

[0021] Each device is based on a i Select the sample set to be labeled Experts label each device using the labeled Update w through the gradient descent optimization algorithm i ; Each device uploads w i To the cloud, the cloud aggregates to obtain the global model:

[0022]

[0023] The cloud will w (t) Distribute to each device and update the next round of selection strategy Make the samples with the wrong classification and target domain style preferred, repeat the above steps until the global model converges, and select strategy a i Taking into account the uncertainty, representativeness and diversity of the sample, it is defined as the weighted sampling probability:

[0024] p(x)=λ1H(y|x;w i )+λ2sim(x,D i )+λ3div(x,S i )

[0025] Among them, H(y|x; w i ) is the sample x in the current model w i The predicted entropy under sim(x,D i ) is x and the local dataset D i The maximum similarity, div(x,S i ) is the relationship between x and the labeled set S i The minimum distance, λ1, λ2, λ3 are weight coefficients.

[0026] As a preferred solution of the main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, the spatiotemporal sequence anomaly detection model includes: the cloud uses a spatiotemporal sequence anomaly detection model to capture the weak signs of early electromagnetic bearing failure from the two dimensions of time and space; combines the long short-term memory network LSTM to model time dependence, and the convolutional neural network CNN to extract spatial features, and constructs a spatiotemporal anomaly score function:

[0027] A(x (t,n) )=p(x (t,n) |{x (t-k:t-1,n) ,x (t,n-m:n+m)})

[0028] Among them, x (t,n) It represents the value of the nth monitoring parameter at time t, k is the time window length, m is the spatial neighborhood radius, and p is the abnormal probability estimation function; the abnormal state is determined by threshold discrimination.

[0029] As a preferred solution of the main helium fan intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, the spatiotemporal causal attention mechanism includes: the cloud uses the spatiotemporal causal attention mechanism to realize explainable maintenance decision reasoning, integrates the time dependence and causal relationship of the monitoring parameters, and generates an attribution decision sequence matching the fault state; sets the monitoring parameter time series data as X={x1,x2,…,x T}, the causal graph is G c , modeling the spatial dependencies between parameters through a graph attention network:

[0030]

[0031] Where Pa(i) is the value of vertex i in G c The parent node set, h j' Represents the new feature representation after the features of parent node j are transformed by the graph attention layer;

[0032] Representing space-time causality as H tc Input decision network to integrate failure mode knowledge and maintenance program knowledge The weight of each knowledge unit is dynamically adjusted through the attention mechanism to generate a decision sequence that matches the current fault state;

[0033] The time dependency is modeled by a gated recurrent unit, and the state transition function is:

[0034] f t =σ g (W f [h t-1 ,x t ]+b f )

[0035] i t =σ g (W i [h t-1 ,x t ]+b i )

[0036] o t =σ g (W g [h t-1 ,x t ]+b o )

[0037]

[0038] h t =o t ⊙tanh(c t )

[0039] Among them, σ g is the sigmoid gate function, W f , W i , W g , W c and b f , b i , b o , b c is the parameter to be learned, ⊙ is the element-wise multiplication, represents the state of the candidate memory unit at time t, C t represents the final state of the memory unit at time t, f t represents the forget gate at time t, i t represents the input gate at time t, o t represents the output gate at time t, h t represents the hidden state at time t;

[0040] The causal relationship is learned from monitoring data through PC or GES causal discovery algorithm, and the causal dependency between variables is represented by a directed acyclic graph.

[0041] As a preferred solution of the main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring described in the present invention, wherein: the Web visualization platform includes, in the form of a dynamic dashbord, integrated display of trend graphs of various monitoring parameters, a health index dashboard, a fault association network, and a maintenance decision tree. The user selects the equipment, measuring points, and time periods of interest through an interactive menu. The system generates a visualization view in real time according to the query conditions. The platform embeds VR / AR components to realize three-dimensional panoramic modeling of the main helium blower, and overlays and renders the faulty parts in the form of a heat map to intuitively present the internal status of the equipment;

[0042] Encrypted data is transmitted between edge computing nodes and the cloud through the 5G high-speed and secure industrial Internet network to ensure data integrity and confidentiality; the fault diagnosis knowledge base adopts a knowledge representation method that combines ontology and rules to support semantic retrieval and logical reasoning.

[0043] Another object of the present invention is to provide an intelligent maintenance decision system for the main helium blower based on electromagnetic bearing monitoring, which can monitor the operating status of the electromagnetic bearing of the main helium blower online in real time, overcome the lag of traditional offline and post-diagnosis methods, and significantly improve the intelligence level of status monitoring and fault diagnosis.

[0044] As a preferred solution of the main helium blower intelligent maintenance decision system based on electromagnetic bearing monitoring described in the present invention, it includes: a field data acquisition unit, an edge computing unit, a cloud computing unit, and a Web visualization platform;

[0045] The field data acquisition unit is deployed at the main helium blower site to collect multi-source monitoring data of the electromagnetic bearing in real time through vibration, temperature and current sensors;

[0046] The edge computing unit is directly connected to the field data acquisition unit to pre-process the collected raw monitoring data;

[0047] The cloud computing unit is connected to the edge computing unit via the 5G industrial Internet, receives the pre-processed data and performs intelligent modeling analysis;

[0048] The Web visualization platform serves as a human-computer interaction interface and intuitively presents the health assessment results and maintenance decision suggestions output by the system.

[0049] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and is characterized in that when the processor executes the computer program, steps of an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring are implemented.

[0050] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, the steps of an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring are implemented.

[0051] Beneficial effects of the present invention: The present invention achieves high-quality collection and efficient transmission of electromagnetic bearing status data by deploying multi-source heterogeneous sensors at the main helium blower site and adopting an adaptive multi-scale fusion preprocessing method to reduce noise, extract features and compress monitoring data, laying the foundation for subsequent intelligent modeling and analysis. The introduction of graph neural networks to adaptively learn the spatiotemporal correlation between electromagnetic bearing monitoring parameters overcomes the limitations of traditional artificial feature engineering and extracts deep feature representations that more comprehensively and accurately reflect the health status of the bearing. An innovative active federated transfer learning paradigm is proposed to achieve joint modeling of monitoring data between different main helium blowers without directly sharing the original data, and an active learning mechanism is introduced to continuously optimize the model performance, meeting the data privacy protection requirements while improving the generalization ability of fault diagnosis. The evolution laws in the temporal and spatial dimensions can identify signs of bearing degradation as early as possible; and the propagation path and influencing factors of the fault are revealed from a causal perspective, making the intelligent diagnosis and decision-making process transparent and explainable, giving the system output results physical meaning. The machine learning model is deeply integrated with domain knowledge, and the implicit expert experience is introduced into the feature representation of the constrained data by ontology. The model output is constrained and standardized by inference rules, so as to achieve the improvement of diagnosis, prediction and decision-making effects in the complementarity of data and knowledge. The system adopts the edge-cloud collaborative architecture, and closely connects the on-site intelligence through high-speed industrial Internet such as 5G, so as to realize the full process of monitoring, diagnosis, prediction and decision-making and closed-loop optimization, and promote the leapfrog development of electromagnetic bearings from condition monitoring to intelligent operation and maintenance. A Web platform integrating monitoring, diagnosis, prediction, decision-making and visualization has been developed, and a VR / AR interactive interface has been introduced to provide intuitive and operational auxiliary decision-making basis for operation and maintenance personnel from a global perspective, thereby improving the efficiency and accuracy of operation and maintenance.

[0052] The present invention opens up a new technical route for the online monitoring and intelligent maintenance of electromagnetic bearings. The constructed end-to-end intelligent closed loop can capture the signs of bearing degradation in the shortest time, reveal the cause of the fault, and give the optimal maintenance suggestion, upgrading the maintenance mode from passive repair to active prevention. It has significant benefits in reducing operation and maintenance costs, extending the service life of bearings, and improving equipment reliability. It has important application value and broad application prospects for promoting the intelligent development of the nuclear power industry and ensuring the safe and economical operation of the units. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0054] Figure 1This is a general architecture diagram of a main helium blower intelligent maintenance decision system based on electromagnetic bearing monitoring provided by an embodiment of the present invention.

[0055] Figure 2 An edge computing unit data processing flow chart of an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring provided in accordance with an embodiment of the present invention.

[0056] Figure 3 A flowchart of intelligent modeling of a cloud computing unit for an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring provided in one embodiment of the present invention.

[0057] Figure 4 A functional architecture diagram of a Web visualization platform for an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0059] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0060] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0061] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0062] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0063] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0064] Example 1, reference Figure 2-Figure 4 , which is the first embodiment of the present invention, provides a main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring, including:

[0065] S1: Using an edge-cloud collaborative architecture, edge computing nodes collect and preprocess multi-source monitoring data of electromagnetic bearings in real time, and upload the preprocessed data to the cloud through a secure communication network.

[0066] Furthermore, the preprocessing includes: the edge computing node uses an adaptive multi-scale fusion method to preprocess the electromagnetic bearing vibration data, and performs L-layer wavelet packet decomposition on the vibration signal to obtain 2 L subband coefficient sequences, for each subband l (l = 1, 2, ..., 2 L ) Calculate the variance Var l and Skew l :

[0067]

[0068] Where N is the subband length, mean and std are the mean and standard deviation respectively, x l (i) represents the i-th coefficient in the l-th sub-band, and the significance measure of sub-band l is defined as:

[0069] Sig l =w1Var l +w2|Skew l |

[0070] Among them, w1 and w2 are weight coefficients, select Sig lThe largest first K subbands extract time domain, frequency domain and time-frequency domain features, denoted as The softmax attention mechanism is used to adaptively fuse sub-band features:

[0071]

[0072] in, represents the kth feature of the lth subband The attention weight, W and u are the parameters of the attention network, b is the bias, and v is the fused feature vector. Represents matrix transpose.

[0073] It should be noted that edge computing includes data preprocessing and data compression. Data preprocessing uses an adaptive wavelet packet decomposition algorithm to reduce noise and extract features from electromagnetic bearing vibration data. Data compression uses a convolutional autoencoder network. The encoder uses a convolutional layer to extract local features and downsample, and the decoder uses a deconvolution layer to gradually restore the original vibration signal. The autoencoder is trained by minimizing the reconstruction error so that it can adaptively retain key information in the vibration signal during the encoding stage.

[0074] S2: The cloud uses a graph neural network to construct a spatiotemporal correlation graph for the time series of each monitoring parameter of the electromagnetic bearing, and uses a federated active transfer learning framework to optimize the health assessment model.

[0075] Furthermore, the construction of the spatiotemporal correlation graph includes: using a graph neural network in the cloud to construct a spatiotemporal correlation graph for the time series of each monitoring parameter of the electromagnetic bearing, extracting low-dimensional latent variables that comprehensively reflect the health status of the bearing through end-to-end representation learning; constructing the monitoring parameters into an undirected graph G = (V, E), where the vertex set V represents the time series of each parameter {v i , i=1,2,...,N}, the edge set E represents the strength of association between parameters, and the hidden layer representation of the vertex obtained after T rounds of iterative updates is:

[0076]

[0077] Among them, W1 and W2 are graph convolution parameters, σ is the activation function, is the neighbor set of vertex i, α ii is the attention weight between vertices i and j, h i , They respectively represent concatenating the feature vectors of vertices i and j as the input of the attention network, calculating the correlation between the two vertices, and finally concatenating the representation of each vertex and mapping it to the health state space through the fully connected layer.

[0078] It should be noted that the federated active transfer learning framework includes: the cloud uses the federated active transfer learning framework to adaptively optimize and correct the health assessment model across devices and working conditions without directly sharing the original monitoring data, and continuously improve the model performance through active learning strategies;

[0079] Set the local data set of the i-th device to The model parameter is w n , the sample selection strategy for active learning is a n , the federation aggregation parameter is w, in each communication round t:

[0080] Each device is based on a i Select the sample set to be labeled Experts label each device using the labeled Update w through the gradient descent optimization algorithm i ; Each device uploads w i To the cloud, the cloud aggregates to obtain the global model:

[0081]

[0082] The cloud will w (t) Distribute to each device and update the next round of selection strategy Make the samples with the wrong classification and target domain style preferred, repeat the above steps until the global model converges, and select strategy a i Taking into account the uncertainty, representativeness and diversity of the sample, it is defined as the weighted sampling probability:

[0083] p(x)=λ1H(y|x;w i )+λ2sim(x,D i )+λ3div(x,S i )

[0084] Among them, H(y|x; w i ) is the sample x in the current model w i The predicted entropy under sim(x,D i ) is x and the local dataset D i The maximum similarity, div(x,S i ) is the relationship between x and the labeled set S i The minimum distance, λ1, λ2, λ3 are weight coefficients.

[0085] S3: The cloud uses a spatiotemporal sequence anomaly detection model to build a spatiotemporal anomaly score function, and uses a spatiotemporal causal attention mechanism to achieve explainable maintenance decision reasoning, and comprehensively monitor the time dependency and causal relationship of parameters.

[0086] Furthermore, the spatiotemporal sequence anomaly detection model includes: the cloud uses a spatiotemporal sequence anomaly detection model to capture the weak signs of early electromagnetic bearing failure from the two dimensions of time and space; combines the long short-term memory network LSTM to model time dependency, and the convolutional neural network CNN to extract spatial features, and constructs a spatiotemporal anomaly score function:

[0087] A(x (t,n) )=p(x (t,n) |{x (t-k:t-1,n) ,x (t,n-m:n+m)})

[0088] Among them, x (t,n) It represents the value of the nth monitoring parameter at time t, k is the time window length, m is the spatial neighborhood radius, and p is the abnormal probability estimation function; the abnormal state is determined by threshold discrimination.

[0089] Use LSTM to analyze the historical time series {x (t-k：t-1，n)}Learn long-term and short-term representations to obtain high-order features in the time dimension; use CNN to extract spatial correlation features between parameters in a sliding window manner. Concatenate the features of the two branches and input them into the fully connected layer to estimate the abnormal probability of the current spatiotemporal state. Determine abnormal alarms through threshold discrimination.

[0090] It should be noted that the spatiotemporal causal attention mechanism includes: the cloud uses the spatiotemporal causal attention mechanism to realize explainable maintenance decision reasoning, integrates the time dependency and causal relationship of monitoring parameters, and generates an attribution decision sequence matching the fault state; sets the monitoring parameter time series data as X = {x1, x2, ..., x T}, the causal graph is G c , modeling the spatial dependencies between parameters through a graph attention network:

[0091]

[0092] Where Pa(i) is the value of vertex i in G c The parent node set, h j' Represents the new feature representation after the features of parent node j are transformed by the graph attention layer;

[0093] Representing space-time causality as H tc Input decision network to integrate failure mode knowledge and maintenance program knowledge The weight of each knowledge unit is dynamically adjusted through the attention mechanism to generate a decision sequence that matches the current fault state;

[0094] The time dependency is modeled by a gated recurrent unit, and the state transition function is:

[0095] ft =σ g (W f [h t-1 ,x t ]+b f )

[0096] i t =σ g (W i [h t-1 ,x t ]+b i )

[0097] o t =σ g (W g [h t-1 ,x t ]+b o )

[0098]

[0099] h t =o t ⊙tanh(c t )

[0100] Among them, σ g is the sigmoid gate function, W f , W i , W g , W c and b f , b i , b o , b c is the parameter to be learned, ⊙ is the element-wise multiplication, Represents the state of the candidate memory unit at time t, which is determined by the current input x t and the previous hidden state h t-1 Calculated, it reflects the impact of the current input on the memory content, C t Represents the final state of the memory unit at time t, which is the memory state c at the previous moment t-1 and the current candidate memory state Through the forget gate t and input gate i t The control acquisition reflects the integration of past memory and current input, f t Represents the forget gate at time t, controlling the memory state c at the previous moment t-1 How much information in the system is retained to the current moment, which plays a role in forgetting old information. t Represents the input gate at time t, controlling the current candidate memory state How much new information is added to the current memory state C tIt plays a role in selecting and memorizing new information. t Represents the output gate at time t, controlling the current memory cell state C t How much information is output to the hidden state h t , plays the role of selective output, h t Represents the hidden state at time t, which combines the hidden state h at the previous moment t-1 , also includes the current memory cell state C t The information is used as the feature representation of the current moment and participates in the calculation of the next moment;

[0101] Furthermore, the hidden state of each time step is finally aggregated into a causal representation through attention pooling:

[0102]

[0103] Where L is the number of graph attention network layers, β t is the time step weight, w t is a learnable parameter. c As the causal representation of the current observation, the decision network is input to predict the optimal maintenance action. The decision network adopts a Seq2Seq structure. The Encoder converts h c Mapped to the decision hidden state z, the Decoder iteratively generates the maintenance action sequence {a1, a2, ..., a M}:

[0104]

[0105] In generating each action a m When and maintenance solution knowledge base The weight of each knowledge unit in the model is integrated with prior knowledge to guide the decision-making process. This model can trace the changes in key monitoring parameters that cause abnormalities based on the spatiotemporal propagation and causal graph of the fault, identify the root cause of the fault, and provide targeted maintenance measures.

[0106] The causal relationship is learned from monitoring data through PC or GES causal discovery algorithm, and the causal dependency between variables is represented by a directed acyclic graph.

[0107] S4: Through the Web visualization platform, the electromagnetic bearing monitoring data, health assessment results, fault warning information and maintenance decision suggestions are intuitively presented to assist experts in formulating maintenance plans.

[0108] Furthermore, the Web visualization platform includes, in the form of a dynamic dashbord, integrated display of trend graphs of various monitoring parameters, health index dashboards, fault association networks, and maintenance decision trees. Users select the equipment, measurement points, and time periods of interest through an interactive menu. The system generates a visualization view in real time according to the query conditions. The platform embeds VR / AR components to achieve three-dimensional panoramic modeling of the main helium blower, and overlays and renders the faulty parts in the form of a heat map to intuitively present the internal status of the equipment.

[0109] Encrypted data is transmitted between edge computing nodes and the cloud through the 5G high-speed and secure industrial Internet network to ensure data integrity and confidentiality; the fault diagnosis knowledge base adopts a knowledge representation method that combines ontology and rules to support semantic retrieval and logical reasoning.

[0110] Furthermore, the knowledge base uses a combination of ontology and rules to structure the domain knowledge. The ontology part defines the core concepts such as equipment, components, monitoring parameters, failure modes, maintenance plans, and the semantic associations between concepts. The rule part uses first-order predicate logic to formalize expert experience, for example:

[0111]

[0112] The knowledge base and data-driven model work together to improve semantic expression and reasoning efficiency. On the one hand, knowledge constraints help reduce the risk of overfitting of data-driven models and improve generalization; on the other hand, data-driven models can be used to supplement and optimize the knowledge base and achieve incremental learning. The combination of the two can significantly improve the accuracy and explainability of fault diagnosis and maintenance decisions.

[0113] Example 2, an embodiment of the present invention, provides an intelligent maintenance decision method for a main helium blower based on electromagnetic bearing monitoring. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.

[0114] (1) Obtain monitoring parameters such as electromagnetic bearing vibration and temperature through the on-site data acquisition unit and transmit them to the edge computing unit;

[0115] (2) The edge computing unit uses adaptive wavelet packet decomposition and attention fusion mechanism to reduce noise and extract features from vibration data, compresses the data through a convolutional autoencoder network, and finally uploads it to the cloud computing unit through 5G network security;

[0116] (3) The cloud computing unit first uses graph neural networks to extract the spatiotemporal correlation features of monitoring parameters and continuously optimizes the health assessment model through active federated learning;

[0117] (4) Combining LSTM and CNN to build a spatiotemporal anomaly detector to detect the signs of electromagnetic bearing degradation early;

[0118] (5) When an anomaly is detected, causal attention reasoning is initiated, and combined with the device ontology knowledge graph, the cause of the fault is traced from the three dimensions of time, space, and causality, and maintenance decision suggestions are given by analogy with the expert rule base;

[0119] (6) The health assessment, abnormal warning and maintenance decision results are transmitted to the Web visualization platform and displayed through a dynamic dashboard and VR / AR scenes to assist operation and maintenance personnel in intuitively analyzing equipment status and formulating maintenance plans.

[0120] In summary, this embodiment makes full use of multi-source monitoring data of electromagnetic bearings, integrates cutting-edge artificial intelligence technologies such as graph neural networks, active learning, spatiotemporal anomaly detection, causal reasoning, etc., and constructs a full-chain, multi-granular, and strongly generalized state monitoring and fault diagnosis model. It also introduces prior expert knowledge to constrain and optimize the model training and decision-making process, thereby achieving deep collaboration between data-driven and knowledge-driven.

[0121] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

[0122] Embodiment 3, the third embodiment of the present invention, is different from the first two embodiments in that:

[0123] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0124] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0125] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0126] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0127] Example 4, reference Figure 1 , which is an embodiment of the present invention, provides an intelligent maintenance decision system for a main helium blower based on electromagnetic bearing monitoring, which is characterized by comprising a field data acquisition unit, an edge computing unit, a cloud computing unit, and a Web visualization platform.

[0128] The field data acquisition unit is deployed at the main helium blower site to collect multi-source monitoring data of electromagnetic bearings in real time through vibration, temperature and current sensors.

[0129] The edge computing unit is directly connected to the on-site data acquisition unit to pre-process the collected raw monitoring data.

[0130] The cloud computing unit is connected to the edge computing unit through the 5G industrial Internet, receives pre-processed data and performs intelligent modeling and analysis.

[0131] The Web visualization platform serves as a human-computer interaction interface, intuitively presenting the health assessment results and maintenance decision recommendations output by the system.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent maintenance decision method for main helium blower based on electromagnetic bearing monitoring, characterized in that: include, Adopting the edge-cloud collaborative architecture, the edge computing nodes collect and preprocess the multi-source monitoring data of electromagnetic bearings in real time, and upload the preprocessed data to the cloud through a secure communication network; The cloud uses a graph neural network to construct a spatiotemporal correlation graph for the time series of each monitoring parameter of the electromagnetic bearing, and uses a federated active transfer learning framework to optimize the health assessment model; The cloud uses a spatiotemporal sequence anomaly detection model to construct a spatiotemporal anomaly score function, and uses a spatiotemporal causal attention mechanism to achieve explainable maintenance decision reasoning, integrating the time dependency and causal relationship of monitoring parameters; The Web visualization platform intuitively presents electromagnetic bearing monitoring data, health assessment results, fault warning information and maintenance decision recommendations, assisting experts in formulating maintenance plans.

2. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 1 is characterized in that: The preprocessing includes: the edge computing node uses an adaptive multi-scale fusion method to preprocess the electromagnetic bearing vibration data, and performs L-layer wavelet packet decomposition on the vibration signal to obtain 2 L subband coefficient sequences, for each subband l (l = 1, 2, ..., 2 L ) Calculate the variance Var l and Skew l : Where N is the subband length, mean and std are the mean and standard deviation respectively, x l (i) represents the i-th coefficient in the l-th sub-band, and the significance measure of sub-band l is defined as: Say l =w1Var l +w2|Skew l | Among them, w1 and w2 are weight coefficients, select Sig l The largest first K subbands extract time domain, frequency domain and time-frequency domain features, denoted as The softmax attention mechanism is used to adaptively fuse sub-band features: in, represents the kth feature of the lth subband The attention weight, W and u are the parameters of the attention network, b is the bias, and v is the fused feature vector. Represents matrix transpose.

3. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 2 is characterized in that: The construction of the spatiotemporal association graph includes: using a graph neural network in the cloud to construct a spatiotemporal association graph for the time series of each monitoring parameter of the electromagnetic bearing, extracting low-dimensional latent variables that comprehensively reflect the health status of the bearing through end-to-end representation learning; constructing the monitoring parameters into an undirected graph G = (V, E), where the vertex set V represents the time series of each parameter {v i , i=1,2,...,N}, the edge set E represents the strength of association between parameters, and the hidden layer of the vertex obtained after T rounds of iterative updates is represented as: Among them, W1 and W2 are graph convolution parameters, σ is the activation function, is the neighbor set of vertex i, α ii is the attention weight between vertices i and j, h i , They respectively represent concatenating the feature vectors of vertices i and j as the input of the attention network, calculating the correlation between the two vertices, and finally concatenating the representation of each vertex and mapping it to the health state space through the fully connected layer.

4. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 3 is characterized in that: The federated active transfer learning framework includes: the cloud uses the federated active transfer learning framework to adaptively optimize and correct the health assessment model across devices and working conditions without directly sharing the original monitoring data, and continuously improve the model performance through active learning strategies; Set the local data set of the i-th device to The model parameter is w n , the sample selection strategy for active learning is a n , the federation aggregation parameter is w, in each communication round t: Each device is based on a i Select the sample set to be labeled Experts label each device using the labeled Update w through the gradient descent optimization algorithm i ; Each device uploads w i To the cloud, the cloud aggregates to obtain the global model: The cloud will w (t) Distribute to each device and update the next round of selection strategy Make the samples with the wrong classification and target domain style preferred, repeat the above steps until the global model converges, and select strategy a i Taking into account the uncertainty, representativeness and diversity of the sample, it is defined as the weighted sampling probability: p(x)=λ1H(y|x;w i )+λ2sim(x,D i )+λ3div(x,S i ) Among them, H(y|x; w i ) is the sample x in the current model w i The predicted entropy under sim(x, D i ) is x and the local dataset D i The maximum similarity, div(x, S i ) is the relationship between x and the labeled set S i The minimum distance, λ1, λ2, λ3 are weight coefficients.

5. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 4 is characterized in that: The spatiotemporal sequence anomaly detection model includes: the cloud uses a spatiotemporal sequence anomaly detection model to capture the weak signs of early electromagnetic bearing failure from the two dimensions of time and space; combines the long short-term memory network LSTM to model time dependency, and the convolutional neural network CNN to extract spatial features, and constructs a spatiotemporal anomaly score function: A(x (t，n) )=p(x (t，n) |{x (t-k：t-1,n) ,x (t，n-m：n+m) }) Among them, x (t，n) It represents the value of the nth monitoring parameter at time t, k is the time window length, m is the spatial neighborhood radius, and p is the abnormal probability estimation function; the abnormal state is determined by threshold discrimination.

6. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 5 is characterized in that: The spatiotemporal causal attention mechanism includes: the cloud uses the spatiotemporal causal attention mechanism to realize explainable maintenance decision reasoning, integrates the time dependency and causal relationship of monitoring parameters, and generates an attribution decision sequence matching the fault state; sets the monitoring parameter time series data as X={x1, x2, ..., x T }, the causal graph is G c , modeling the spatial dependencies between parameters through a graph attention network: Where Pa(i) is the value of vertex i in G c The parent node set, h j′ Represents the new feature representation after the features of parent node j are transformed by the graph attention layer; Representing space-time causality as H tc Input decision network to integrate failure mode knowledge and maintenance program knowledge The weight of each knowledge unit is dynamically adjusted through the attention mechanism to generate a decision sequence that matches the current fault state; The time dependency is modeled by a gated recurrent unit, and the state transition function is: f t =σ g (W f [h t-1 ,x t ]+b f ) i t =σ g (W i [h t-1 ,x t ]+b i ) the t =s g (W g [h t-1 ,x t ]+b o ) h t =o t ☉tanh(c t ) Among them, σ g is the sigmoid gate function, W f , W i , W g , W c and b f , b i , b o , b c is the parameter to be learned, ⊙ is the element-wise multiplication, represents the state of the candidate memory unit at time t, C t represents the final state of the memory unit at time t, f t represents the forget gate at time t, i t represents the input gate at time t, O t represents the output gate at time t, h t represents the hidden state at time t; The causal relationship is learned from monitoring data through PC or GES causal discovery algorithm, and the causal dependency between variables is represented by a directed acyclic graph.

7. The main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring according to claim 6 is characterized in that: The Web visualization platform includes, in the form of a dynamic dashbord, integrated display of trend graphs of various monitoring parameters, health index dashboards, fault association networks, and maintenance decision trees. Users select the equipment, measurement points, and time periods of interest through an interactive menu. The system generates a visualization view in real time according to the query conditions. The platform embeds VR / AR components to achieve three-dimensional panoramic modeling of the main helium blower, and overlays and renders the faulty parts in the form of a heat map to intuitively present the internal status of the equipment. Encrypted data is transmitted between edge computing nodes and the cloud through the 5G high-speed and secure industrial Internet network to ensure data integrity and confidentiality; the fault diagnosis knowledge base adopts a knowledge representation method that combines ontology and rules to support semantic retrieval and logical reasoning.

8. A system using the main helium blower intelligent maintenance decision method based on electromagnetic bearing monitoring as described in any one of claims 1 to 7, characterized in that: It includes field data collection unit, edge computing unit, cloud computing unit and Web visualization platform; The field data acquisition unit is deployed at the main helium blower site to collect multi-source monitoring data of the electromagnetic bearing in real time through vibration, temperature and current sensors; The edge computing unit is directly connected to the field data acquisition unit to pre-process the collected raw monitoring data; The cloud computing unit is connected to the edge computing unit via the 5G industrial Internet, receives the pre-processed data and performs intelligent modeling analysis; The Web visualization platform serves as a human-computer interaction interface and intuitively presents the health assessment results and maintenance decision suggestions output by the system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images